PARP1 Suppresses the Transcription of PD-L1 by Poly(ADP-Ribosyl)ating STAT3

Poly (ADP-ribose) polymerases (PARP) are enzymes that transfer ADP-ribose groups to target proteins, thereby affecting various nuclear and cytoplasmic processes (1). PARP1, the founding member of the PARP family, was characterized for a long time as a DNA damage sensor and a key factor in DNA-repair systems. However, it has become increasingly clear that PARP1 also plays an important role in gene expression through the poly(ADP-ribosyl)ation (PARylation) of sequence-specific DNA-binding transcription factors (2). For example, PARP1 can poly(ADP-ribosyl)ate NELF (negative elongation factor) and promote its release from paused RNA polymerase II, allowing productive elongation and RNA production (3). By poly(ADP-ribosyl)ating Smad3 and Smad4, PARP1 dissociates Smad complexes from DNA, which can attenuate Smad-specific gene expression (4). PARP1 can also ADP-ribosylate the regulatory T-cell (Treg)-specific transcription factor FOXP3, which subsequently promotes its degradation and, therefore, negatively regulates Treg function (5). C/EBPβ, an adipogenic transcription factor, is reported to be ADP-ribosylated by PARP1, resulting in the inhibition of its DNA-binding activity and the subsequent arrest of the adipogenic transcriptional program (6). These findings directly link PARylation to biological outcomes, including cell differentiation, the posttranscriptional regulation of RNA, and metabolic functions. Although PARP1 has been shown to play a critical role in gene regulation, very limited substrates have been identified because of the labile nature of the ADP-ribose linkage to the protein, the dynamic nature of the modification, the heterogeneity in its signal, and its rarity (7).

Over the last few years, there have been major breakthroughs in our understanding of tumor-associated immune suppression. Among the multiple mechanisms that can contribute to immune suppression in the tumor microenvironment, programmed death ligand 1 (PD-L1), an inhibitory member of the B7 family, plays a central role in the evasion of the immune system by many cancer types (8, 9). PD-L1 expression is upregulated in various epithelial tumors, including ovarian cancer (10), breast cancer (11), non–small cell lung carcinoma (12), and the colon cancer (13). PD-L1 binds to either PD-1 or CD80 receptors on T cells, B cells, dendritic cells, and natural killer cells to inhibit their proliferation, cytokine release, and cytolytic activity, whereas the blockade of coinhibitory ligation with monoclonal antibodies, such as PD-L1 or PD-1 antibodies, restores T-cell function and increases therapeutic efficacy (14). The impressive and durable clinical response of PD-L1 or PD-1 blockade immunotherapy led to the FDA approval of nivolumab, pembrolizumab, and atezolizumab for the treatment of multiple types of cancer (15–17).

The expression of PD-L1 is commonly elevated in cancer cells and is an important mechanism that negatively regulates T-cell activation to prevent autoimmune responses under physiologic conditions (14). Studies have also focused on how and when PD-L1 is upregulated during the pathogenesis of cancer, and several transcriptional factors have been shown to be involved in the expression of PD-L1 (18). HIF1α regulates PD-L1 by binding to a hypoxia response element (HRE) in the PD-L1 promoter to activate PD-L1 transcription (19). STAT3 has also been demonstrated to bind to the PD-L1 promoter to transcriptionally regulate its expression (20). PD-L1 is commonly expressed and transcriptionally can be regulated by STAT3 and MYC in ALK-negative anaplastic large cell lymphoma (21). NF-κB is involved in LMP1-induced PD-L1 expression and is also a major mediator of IFN-γ-induced PD-L1 expression (22). MYC regulates the expression of PD-L1 by binding to its promoter in murine and human leukemia and lymphomas (23). Although the transcriptional regulation of PD-L1 via NF-κB, STAT, or MYC has been studied, the molecular mechanism controlling the transcriptional activity is still poorly understood.

In this study, we introduce PARP1 as a negative regulator of the STAT3-mediated transcription of PD-L1 in various cancers. We showed that PARP1 poly(ADP-ribosyl)ated STAT3, inhibited its phosphorylation and transcriptional activities, and attenuated the expression of PD-L1. Our results provide a clear example of PARP1-regulated gene expression, uncover a posttranslational modification of STAT3, and improve the understanding of the mechanism that regulates the expression of PD-L1.

The antibodies against PD-L1 (#sc-50298), PARP1 (#sc-7150), STAT3 (#sc-482X), STAT3 (#sc-482), anti-rabbit IgG (#sc-2007), HA (#sc-7392), Actin (#sc-1615), STAT1 (#sc-346), MYC (#sc-56634), and protein A/G (#sc-2003) were purchased from Santa Cruz Biotechnology. Anti-PAR (#4335-AMC-050) was purchased from Trevigen. PE-conjugated phospho(p)-STAT3 (Tyr705) (DA37) rabbit monoclonal antibody (mAb; #8119), STAT3 (#4904S), p-STAT3 (#9145L), GSK3β (#12456), PD-L1 (#13684S), and p-GSK3β (#5558S) were purchased from Cell Signaling Technology. Anti-DYKDDDDK-tag (#A00187) and anti-DYKDDDDK IP resin (#L00425) were purchased from GenScript. The anti-HA affinity gel (#B23302) was purchased from BioTools. The siRNA-negative control and sequence-specific short RNA duplex pools for interference were purchased from GenePharma. The PARP inhibitors olaparib (#S1060), veliparib (#S1004), talazoparib (#S7048), A-966492 (#S2197), vanadate (#S2000), and AZD1480 (#S2162) were obtained from Selleck. The propidium iodide (#ZF-50-0001) was purchased from MultiSciences. FITC-conjugated CD47 (#556045), FITC-conjugated CD274 (#558065), and FITC mouse IgG were purchased from BD Biosciences. Sulforhodamine B (#230162) was purchased from Sigma-Aldrich.

All the cell lines were obtained from Cell Bank of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences (March 2013 and September 2016). Each new aliquot was passaged in our laboratory for fewer than three months after resuscitation, after which new seed stocks were thawed. All the cell lines were authenticated using DNA fingerprinting (variable number of tandem repeats), confirming that no cross contamination occurred during this study. Cell lines were tested for mycoplasma contamination at least every month when they were in use (SKOV3 and OVCAR8 were latest tested on July 3, 2018, and other cell lines were latest tested on September 9, 2017). SKOV3, OVCAR8, PC9, and SW620 were maintained in RPMI-1640 (Gibco, #31800) with 10% fetal bovine serum (FBS; ExCellBio, #FSP500). HEK293T was cultured in DMEM (Gibco, #12800) with 10% FBS (Gemini, #900-108). A549 was maintained in F12 (Sigma, #N670) and 10% FBS (Gemini). Colo205 was maintained in DMEM (Gibco) and 10% FBS (ExCellBio). All the media contained 1% penicillin/streptomycin (Gibco, #15140122), and all cells were maintained at 37°C in a 5% CO₂ incubator.

The mammalian expression vector PCMV6 (#PS100001) and the full-length of PARP1 (#RC207085) and STAT3 (#NM 139276) were all purchased from Origene. The mutation of PARP1 was performed by site-directed mutagenesis and verified by sequencing. A mutation-specific PCR amplification method was developed as follows: 100 ng/μL full-length PARP1 or STAT3: 1 μL; 10XPCR Buffer: 5 μL; 25 mmol/L MgSO₄: 3 μL; 2 mmol/L dNTPs: 5 μL; primers: 2 μL; KOD-Plus-Neo: 1 μL; double distilled water: 33 μL. All reagents used above contained in the KOD-Plus-Neo kit (#KOD-401) from Toyobo. The specific primers are detailed in Supplementary Table S1. HEK 293 cells were transfected with the indicated plasmids using JetPRIME (Polyplus, #114-15) for indicated time, according to the manufacturer's instructions.

SKOV3 or OVCAR8 cells (1.5 × 10⁵) were seeded in 6-well plate for each well and allowed to grow for 12 hours. Then, cells were transfected with transfection reagent JetPRIME (Polyplus, #114-15) and STAT3 (PARP1, MYC, and STAT1) siRNA or scrambled siRNA (siRNA-negative control, NC), JetPRIME Buffer (Polyplus, #712-60): 200 μL, jetprimer: 2 μL, 20 μmol/L siRNA: 2.5 μL for each well) for 24 or 48 hours. The siRNA sequences are shown in Supplementary Table S2.

SKOV3 or OVCAR8 cells (1.5 × 10⁵) were seeded in 6-well plate for each well and allowed to grow for 12 hours. Then, cells were treated for 0, 2, 4, 8, 12, or 24 hours with PARP inhibitors (10 μmol/L olaparib, 10 μmol/L veliparib, or 10 μmol/L A966492) or with DMSO alone (Con). For another study, cells were transfected with siRNA for 24 hours then treated with PARP inhibitors for another 12 or 24 hours. Total mRNA was collected by using TRIzol Reagent (Invitrogen, #15596026), and RNA purity (A260/280 ratio of 2.0–2.1) and quantity was evaluated using the NanoDrop ND-1000 Spectrophotometer (Thermo Fisher Scientific). First-strand cDNA was synthesized from 2 μg of total RNA using TransScript One-Step gDNA Removal and cDNA Synthesis SuperMix (TRAN, #AT311-03). Quantitative RT-PCR was performed on 7500 Fast Dx Real-Time PCR Instrument (Applied Biosystems). The reaction system (20 μL) was as follows: 2.5 μL forward and reverse primers (10 μmol/L), 0.1 μg cDNA obtained from reverse transcription, respectively, 10 μL iTaqTM Universal SYBR Green Supermix (Bio-Rad, #172-5124), and DEPC water (Sangon Biotech, #B501005-0500) to 20 μL. The PCR protocol consisted of thermal cycling was as follows: Initial denaturation at 95°C for 3 minutes, followed by 40 cycles at 95°C (3 seconds) and 60°C (31 seconds). To ensure that the targeted amplicon was amplified only, after the last extension step, melting curves were then generated, and the temperature was increased from 60°C to 95°C at 0.1°C/second. Fold changes in the expression of each gene were calculated using the formula 2^−(ΔΔCt) by a comparative threshold cycle (Ct) method. All reactions were performed in triplicate, and beta-actin was used as the normalizing gene. The primers used are listed in Supplementary Table S3.

OVCAR8 cells were transfected with siRNA targeting PARP1 for 48 hours, and a series of immunosuppressive factors were analyzed by qRT-PCR, including genes for PD-L1, IL10, IL13, CXCL8, MIF, NOS2, COX2, B7H3, IDO, CD47, TIGIT, and CSF2. Heat map representation of the fold change of 12 immunosuppressive factor mRNAs that were affected by PARP1 depletion was generated. Red represents increased expression, while blue represents reduced expression. The z-score is specified on the left. The heat map was created using Microsoft Excel 2013 (Microsoft Corporation) conditional formatting feature.

A ChIP assay was performed using the EZ ChIP chromatin immunoprecipitation kit (Millipore, # 17-371) according to the manufacturer's protocol. Briefly, SKOV3 cells were cross-linked by 1% (v/v) methanol-free formaldehyde (#252549, Sigma) for 10 minutes and then quenched using 125 mmol/L glycine (#G6201, Sigma). Cells were collected in lysis buffer (contained in the CHIP kit) and were sonicated for 10 minutes in high mood into 200 to 500 bp fragments using Bioruptor plus (Diagenode). Soluble chromatin-containing lysates obtained from aldehyde-fixed and sonicated SKOV3 cells were incubated with the STAT3 (#sc-482x) antibody, and rotated at 4°C overnight. Nonimmunoprecipitated lysates (input) and immunoprecipitates obtained with the anti-rabbit IgG (Santa Cruz Biotechnology, #sc-2007) served as a positive and negative control, respectively. Next, the DNA–protein immune complexes were precipitated with protein A–agarose beads (contained in the CHIP kit), and the DNA was then extracted with phenol/chloroform, precipitated with ethanol, and PCR-amplified using primers specific for the PD-L1 gene promoter: 5′-CAAGGTGCGTTCAGATGTTG-3′ and 5′-GGCGTTGGACTTTCCTGA-3′. PCR was performed on Mastercycler nexus gradient PCR instrument (Eppendorf). The reaction system (10 μL) was as follows: 2.0 μL DNA, 13.2 double distilled water, 2.0 μL 10× PCR buffer,1.6 μL 2.5 mmol/L dNTP, 0.8 μL primers, 0.4 μL Taq (5 U/μL), and the following PCR reaction program was used: Initial denaturation at 94°C (3 minutes), followed by 32 cycles at 94°C (20 seconds), 59°C (30 seconds), 72°C (2 minutes), and final extension at 72°C (2 minutes). Ten microliters of each PCR reaction was removed for analysis by 4% agarose (BioFroxx, #110GR100) gel electrophoresis.

The HEK293T cells were lysed in lysis buffer containing 25 mmol/L Tris-base (pH 7.4), 1% NP-40, 150 mmol/L NaCl, 1 mmol/L PMSF, 1 mmol/L Na₃VO₄, and leupeptin (5 μg/mL; Biosharp, BS140A). We used the BCA protein assay to quantitate protein using BCA kit (Yeasen, #20201), and then 500 μg of cell protein from supernatants were immunoprecipitated with anti-HA affinity gel or anti-DYKDDDDK IP resin for 12 hours at 4°C, followed by incubation for another 1 hour after mixing with 30 μL of protein A/G beads (Santa Cruz Biotechnology, #sc-2003). The whole immunocomplexes were then analyzed by Western blot. Briefly, the proteins were separated by ExpressPlus PAGE Gel (GenScript, #M42015C), transferred to PVDF membranes (Immobilon, #IPVH00010), blocked in TBST (150 mmol/L NaCl, 10 mmol/L Tris-HCl, pH 7.5, and 0.1% Tween 20) containing 5% bovine serum albumin (Amersco, #0332) for 1 hour, and incubated with the STAT3 (Santa Cruz Biotechnology, #sc-482) or PARP1 (Santa Cruz Biotechnology, #sc-7150) antibodies at a dilution of 1:1,000, then treated with anti-rabbit (Fdbio Science, #FDR007) secondary antibodies at a dilution of 1:5,000. The specific bands were analyzed by a Western blot infrared imaging system (Amersham Imager 600, #66210336).

HEK 293 cells transfected with HA-STAT3 and Flag-PARP1 were lysed in 4% SDS buffer (pH 8.0, 4% SDS, 150 mmol/L NaCl, 50 mmol/L triethylamine), followed by sonication at an output power of 500 W at 4°C for three 5-second pulses. Because the 4% SDS can prevent PARylated proteins from binding to antibodies against PAR during the IP procedure, the lysates were diluted 1:9 with 1% NP-40 lysis buffer (pH 7.4, 25 mmol/L Tris-base, 150 mmol/L NaCl, 10% glycerol) after sonication. The diluted lysates (input) were immunoprecipitated by incubating with anti-HA affinity gel or anti-PAR. After an overnight incubation, the immunocomplexes incubated with anti-PAR were collected following incubation with 30 μL protein A/G before extensive washing, or the anti-HA affinity gel was washed with wash buffer (pH 7.4, 25 mmol/L Tris-base, 150 mmol/L NaCl), and the immunoprecipitated proteins were analyzed by Western blot with the anti-PAR or anti-HA at a dilution of 1:1,000. To normalize for input protein, 3% to 4% of each cell lysate was subjected to Western blot and probed with anti-HA (Diagbio, #db2603) and Anti-DYKDDDDK-tag (GenScript, #A00187), following the protocol described above.

For immunostaining, 10,000 SKOV3 cells were fixed with 4% paraformaldehyde at room temperature for 20 minutes. The fixed cells were then rinsed with ice-cold 1× PBS and permeabilized with 200 μL PBS-T solution (0.1% Triton X-100 (BioFroxx, #2424) in PBS solution) at 4°C for 10 minutes. The cells were rinsed with PBS for 5 minutes and blocked for nonspecific interaction with 4% BSA in PBS solution at 37°C for 30 minutes. After blocking, the cells were incubated with the p-STAT3 antibody (1:100) overnight at 4°C. Secondary antibodies (1:200) used were green anti-rabbit (Invitrogen, #A-11008). The nuclei were counterstained with DAPI (Invitrogen, #D1306). The slides were imaged using an Olympus Fluoview-microscope (FV10i-O).

A section corresponding to the PD-L1 promoter [−281 to +43 base pair (bp) relative to the transcription start site] was obtained by PCR from the SKOV3 genomic DNA by Genomic DNA Kit (Invitrogen, #K182001). The construct was confirmed by DNA sequencing. The pGL4.14-PD-L1 plasmid was generated by inserting a KpnI/XhoI (KpnI, Biolab, #R3142S; XhoI, #1094A) fragment containing PD-L1 constructs into the KpnI–XhoI restriction site of the pGL4.14 vector (Promega, #E669A) as reported (24). Reporter activities were assayed using the dual-luciferase/renilla assay system. The Renilla-luciferase-expressing plasmid (Promege, #E2231) was used as the internal control to normalize the transfection efficiency. The cells were collected 36 hours after cotransfected with PGL4.14-PD-L1, Renilla, Flag-PARP1, and HA-STAT3 using JetPRIME (PolyPlus, #114-15). The luciferase activity was detected using the Dual-Luciferase Reporter Assay System (Promega, #E1910) following the manufacturer's protocol. The results are expressed as relative luciferase activity (firefly luciferase/renilla luciferase).

At the experimental endpoint, 4 × 10⁵ SKOV3 and OVCAR8 cells were fixed with 4% formaldehyde for 15 minutes at 37°C and permeabilized on ice for 30 minutes in 90% methanol. After washing with PBS, the cells were then stained with PE-conjugated p-STAT3 and FITC-conjugated CD274, and incubated for 2 hours at room temperature. After washing with PBS, cells were resuspended in 0.5 mL of PBS to measure the expression of PD-L1 and p-STAT3 by BD FACSuite flow cytometry (BD Bioscience).

SKOV3 and OVCAR8 were harvested, washed twice with PBS, and blocked with 3% BSA at 37°C for 45 minutes. The cells were then stained with FITC-conjugated CD274 at 37°C for 2 hours (5 μL/2 × 10⁵ cells in 100 μL 0.2% BSA). The instrument threshold was kept constant throughout the detection by BD FACSuite (BD Bioscience). Cells (10⁵) were analyzed for each sample.

Cell viability was analyzed by sulforhodamine B staining as previously performed (25). Apoptosis was detected using propidium iodide (PI) and flow cytometry. The cells were incubated at −20°C for 24 hours in 75% ethanol, and then resuspended in 500 μL PBS, 2.5 μL RNaseA (10 mg/mL), and 5 μL PI (1 mg/mL) and allowed to incubate in the dark at 37°C for 20 minutes. The cells were then measured by a BD FACSCalibur flow cytometer (BD Bioscience).

After treatment with Olaparib or DMSO alone for 24 hours, 10⁷ OVCAR8 cells were harvested, washed twice with PBS, and blocked with 3% BSA at 37°C for 45 minutes. The cells were then stained with FITC-conjugated CD274 at 37°C for 2 hours (5 μL/2 × 10⁵ cells in 100 μL 0.2% BSA). Then, cells were washed with 1× PBS and immediately sorted using BD FACSAria II (BD Biosciences) flow cytometer. The positivity of CD274 antibody stain was determined by quadrant analysis as compared with the control (cells treated with DMSO alone were stained with the FITC-conjugated CD274 as the control).

Using cBioPortal (http://cbioportal.org) to analyze the data, an ovarian serous cystadenocarcinoma (TCGA) research project based on 563 patients was analyzed as previously described (26). The correlation between CD274 and STAT3 target gene mRNA was analyzed using cBioPortal as described (27).

In this study, 53 primary tumor samples were collected from patients diagnosed with ovarian cancer who underwent primary ovarian cancer surgery at Zhejiang University Hospital (Hangzhou, China). Only patients without preoperative radiotherapy or chemotherapy were enrolled in the study. All samples were obtained from tumor section without any adjacent normal tissue. Then formalinfixed, paraffin-embedded tissue cases were cut at 4-μm thick sections and subjected to the following examinations. Tissue collection and analysis in this study were approved by the Ethics Committee of Women's Hospital of Zhejiang University, and written informed consent to perform the biological studies was obtained from all participants. All ovarian specimens were histologically verified by two pathologists independently.

For histologic analysis, human ovarian cancer tissues were paraffin-embedded. All slides were incubated with 3% hydrogen peroxide after deparaffinization and then were blocked by incubating in blocking buffer (5% goat serum in PBS) for 30 minutes. Slides were then probed with the anti-PARP1 diluted at 1:200 (Origene, #TA321380S), anti–PD-L1 diluted at 1:50 (Origene, #UM800121), anti-PAR, or anti–p-STAT3 followed by treatment with biotinylated secondary antibodies and horseradish peroxidase–conjugated avidin and were visualized with 3,3-diaminobenzidine following the manufacturer's protocol. All the reagents above were contained in anti-rabbit HRP-DAB Cell and Tissue Staining Kit (R&D Systems, #CTS005) and anti-mouse HRP-DAB Cell and Tissue Staining Kit (R&D Systems, #CTS002). The digital images were quantified with the assistance of Image-Pro Plus 6.0 software (IPP, version 6.0, Media Cybernetics).

All of the statistical data are presented as the mean ± SD. Statistical significance of the differences was determined using Student t test. Significance was defined as P < 0.05. For IHC analysis, the results were analyzed using Image-Pro Plus 6.0 software. For the evaluation of protein expression, the selected measurement parameter was integrated optical density (IOD). The optical density was calibrated systematically and the area of interest was extracted through the HSI mode, and the expression of each protein was calculated by IOD/tissue area, as described (28). Also, protein expression was evaluated by intensity scoring on scale of “–” to “+++” (–: no staining, +: weak staining, ++: moderate staining,+++: high staining) and proportion (0%, 1%–24%, 25%–50%, >50%; ref. 29). The P value was calculated using the Pearson correlation test. The Student t test and Pearson correlations were analyzed by Prism 6 software (GraphPad Software Inc.).

In addition to the biochemistry and molecular biology of PARP1 in DNA damage detection and repair, the mechanistic and functional understanding of the role of PARP1 in different biological processes has grown considerably (7). To investigate whether PARP1 is involved in cancer immune evasion, the ovarian cancer cell line OVCAR8 was transiently transfected with siRNA against PARP1, and we characterized a series of immunosuppressive factors, including genes encoding PD-L1, IL10, IL13, CXCL8, MIF, NOS2, COX2, B7H3, IDO, CD47, TIGIT, and CSF2, using qRT-PCR. PARP1 silencing led to an increase in PD-L1 mRNA (7.97-fold change, P = 0.004), as well as increases in expression of IL10 (2.47-fold change, P = 0.18), IL13 (2.25-fold change, P = 0.26), CXCL8 (1.88-fold change, P = 0.38), MIF (2.04-fold change, P = 0.44), COX2 (3.08-fold change, P = 0.25), and TIGIT (1.36-fold change, P = 0.51; Fig. 1A and B). Among these affected genes, we were particularly interested in the alteration of PD-L1 gene expression, which may play a central role in the evasion of immune system of many cancer types (30).

To further confirm the involvement of PARP1 in PD-L1 regulation, we generated another siRNA PARP1-silenced ovarian cell line, SKOV3. PARP1-depleted SKOV3 cells also exhibited an enhanced transcription of PD-L1 mRNA (Fig. 1B). In parallel, the PD-L1 protein expression and cell-surface PD-L1 expression were upregulated with PARP1 silencing (Fig. 1C and D). The impact of individual PARP1 siRNA on PD-L1 expression associated with the gene-silencing efficiency of the siRNA (Fig. 1B, C, and D), suggesting a PD-L1–associated defect. The increased PD-L1 expression in PARP1-depleted cells was also recapitulated in the breast cancer cell line MCF-7, the non–small cell lung carcinoma cell lines A549 and PC9, and the colon cancer cell lines Colo205 and SW620 (Supplementary Fig. S1). Taken together, these data suggest that PARP1 suppresses PD-L1 transcription in various cancer cells.

Considering that PARP1 controls the transcription of target genes, either with both catalytic-dependent and catalytic-independent mechanisms (31, 32), PARP inhibitors that could efficiently suppress the catalytic activity of PARP1 were used to confirm whether PARP1-mediated PD-L1 expression was correlated to its catalytic activity. A significant elevation of PD-L1 mRNA expression was observed when OVCAR8 and SKOV3 cells were treated with 10 μmol/L olaparib for various times (Fig. 2A). To further test whether PARP inhibitors enhanced the transcription of PD-L1, we cloned the promoter region of PD-L1 downstream of a luciferase reporter gene (pGL4.14-PD-L1). The luciferase activity of the cells transfected with pGL4.14-PD-L1 was significantly increased when PARP1 was inhibited (Fig. 2B), and the protein expression of PD-L1 was upregulated by multiple PARP inhibitors in a dose-dependent manner (Supplementary Fig. S2A). To exclude the impact of PARP1 on the degradation of PD-L1, cells were treated with olaparib, cycloheximide, or both for the indicated time. No significant difference in the kinetics of PD-L1 protein expression was observed between these two groups (Supplementary Fig. S2B). A similar upregulation of PD-L1 was observed with two additional PARP inhibitors, veliparib and A966492 (Fig. 2C and D). PARP inhibitors were unable to affect the fluorescence intensity of PD-L1 (Supplementary Fig. S2C and S2D). We also evaluated the expression of CD47, another cell-surface immunosuppressive factor, which was not altered at the mRNA level upon PARP1 silencing (Fig. 1A), and the cell-surface expression of CD47 was not significantly affected by the PARP inhibitor olaparib (Supplementary Fig. S2E). Neither cell viability nor apoptosis was affected by PARP inhibitors at indicated concentrations (Supplementary Fig. S3). The reconstitution of siRNA-resistant, wild-type PARP1 attenuated the expression of PD-L1 in siPARP1 (PARP1-silenced) cells to a level similar seen in scramble cells (Fig. 2E), whereas the reconstitution of the mutant PARP1 failed to attenuate PD-L1 expression. These data suggest that the inhibition of PD-L1 transcription by PARP1 depends on its catalytic activity.

We next asked how PARP1 regulated PD-L1 transcription. Studies indicate that the PARylation of transcriptional factors by PARP1 can alter or modulate their activities, playing a critical role in gene regulation (1). We hypothesized that this function might also apply to PARP1-modulated PD-L1 transcription. Thus, we silenced the transcription factors STAT1, STAT3, and MYC (33–35), which have been reported to regulate PD-L1 transcription by directly binding to its promoter region. We found that PD-L1 expression induced by PARP1 inhibition was not affected by STAT1 and MYC deficiency but was abolished with the silencing of STAT3 (Fig. 3A). In parallel, PARP1 inhibition promoted PD-L1 transcription, which was attenuated by STAT3 silencing (Fig. 3B). Similar results were obtained with the OVCAR8 and MCF-7 cells (Supplementary Fig. S4A and S4B). These data suggest a major contribution of STAT3 to PARP1-mediated PD-L1 transcription.

To further confirm the involvement of STAT3 in PARP1-mediated PD-L1 transcription, SKOV3 cells were transfected with pGL4.14-PD-L1, Flag-PARP1, and HA-STAT3, as indicated for 36 hours. The PD-L1 luciferase activity was upregulated more than 5-fold compared with the vector when transfected with HA-STAT3, whereas it was significantly downregulated when cotransfected with Flag-PARP1 (Fig. 3C). We also performed a ChIP assay using a PCR primer set capable of amplifying the promoter region of PD-L1 containing the STAT3 binding sites. The direct binding of STAT3 to the PD-L1 promoter was observed in SKOV3 cells, and this binding was significantly enhanced by PARP1 inhibition (Fig. 3D). We analyzed an ovarian serous cystadenocarcinoma data set containing 563 samples from TCGA, based on the large-scale cancer genomics data sets provided by cBioPortal. The mRNA expression of MCL1, CCL5, IFNγ, IL4R, and IL2RA, five well-known STAT3 target genes, was correlated with mRNA expression of CD274 (Supplementary Fig. S4C), further suggesting that the expression of PD-L1 is positively associated with STAT3 activity. These data collectively revealed that PARP1 regulates PD-L1 expression through STAT3.

We then asked whether PARP1 directly poly(ADP-ribosyl)ated STAT3 to inhibit its transcriptional activity. To test this hypothesis, we first determined if PARP1 could form complexes with STAT3 in cancer cells. Pulldown assays validated the direct endogenous interaction of PARP1 with endogenous STAT3 (Fig. 4A). This interaction was further confirmed with the ectopic expression of PARP1 or STAT3 (Fig. 4B). PARP1 inhibition also disrupted the interaction between PARP1 and STAT3 (Fig. 4B). Because endogenous PARylation was scarcely detected, we used an ectopic expression system to study the ADP-ribosylation of STAT3. Automodification of PARP1 is a well-documented phenomenon (36). To distinguish whether the PAR signal was coming from STAT3 and not from automodified PARP1, we extracted proteins using a 4% SDS lysis buffer to abolish the interaction between PARP1 and STAT3 but not the covalent PARylation of STAT3 (37). As expected, PAR was pulled down with STAT3 protein, and PARylation of STAT3 was observed, as the molecular weight smear signals began with 92 KD, indicating that PAR signal came directly from STAT3 (Fig. 4C). The PARylation of STAT3 was also observed in STAT3-containing immunoprecipitates from HEK 293T cells ectopically expressing HA-STAT3 and Flag-PARP1 (Fig. 4D), which was abolished by PARP1 inhibition. Taken together, these results demonstrated that STAT3 is PARylated by PARP1.

It remains to be determined if the role of PARP1-mediated PD-L1 transcription is due to the PARylation of STAT3. Previous studies have identified that the glutamate (E988) and histidine (H862) are the residues responsible for the polymerase activity of PARP1 (38). We generated point mutations in the PARP1 gene at positions histidine (H862) and glutamate (E988), respectively. Compared with the group transfected with wild-type PARP1, the PARylation of STAT3 was not observed when the cells were transfected with the H862 or E988 mutants (Fig. 5A). We then analyzed the effect of mutated PARP1 on the transcription of PD-L1. Mutated PARP1 could not suppress STAT3-mediated PD-L1 promoter activity due to loss of its polymerase activity (Fig. 5B), and in both the SKOV3 and OVCAR8 cell lines, PARP1 inactivation did not downregulate PD-L1 protein expression (Fig. 5C and D). These data indicated that PARP1 inhibits PD-L1 transcription through the PARylation of STAT3.

Tyrosine phosphorylation is an important posttranslational modification of STAT3, which functions as a switching signal to activate STAT3. The tyrosine phosphorylation of STAT3 is required for its dimerization, nuclear translocation, and DNA binding (39). Thus, we next assessed whether PARP1 affected the tyrosine phosphorylation of STAT3. The silencing or pharmacologic inhibition of PARP1 enhanced the phosphorylation of STAT3 in SKOV3 and OVCAR8 cell lines (Fig. 6A and B). In parallel, a significant increase in the nucleus localization of STAT3 was observed, further suggesting the activation of STAT3 (Fig. 6C), and the constitution of wild-type PARP1 significantly attenuated the phosphorylation of STAT3 caused by PARP1 silencing (Fig. 6D). Using two-color flow cytometry, we found that PARP1-silenced cells exhibited a significant increase in the proportion cells positive for both phosphorylated (p) STAT3 and PD-L1 (13.34% and 15.74% of the total cell count) compared with control groups (3.10% and 2.62%) after 24 hours (Supplementary Fig. S5A and S5B). The increased STAT3 phosphorylation of PARP1-deficient cells was also recapitulated in the breast cancer cell line MCF-7, the non–small cell lung carcinoma cell lines A549 and PC9, and the colon cancer cell lines Colo205 and SW620 (Supplementary Fig. S6). Shiping Jiao and colleagues also report that a PARP inhibitor upregulated the expression of PD-L1 in breast cancer (40), and they attribute the induction of PD-L1 to GSK3β inactivation, an off-target activity of PARP inhibitors. To determine whether this mechanism is also involved, we evaluated GSK3β activity in our system. Our data showed that the PARP inhibitor does not affect the phosphorylation of GSK3β in ovarian cancer (Supplementary Fig. S6E). These data suggest that PARP1 acts as a suppressor of STAT3 phosphorylation in various cancer cells.

STATs are phosphorylated by JAK2, and they then dimerize and translocate to the nucleus, where they activate gene transcription. Our data showed that JAK2 phosphorylation was not affected by PARP1 silencing (Fig. 6E). Thus, we inferred that PARP1-mediated PARylation might affect the dephosphorylation process of STAT3. Cells were first transfected with Flag-PARP1 for 24 hours, and then 500 μmol/L vanadate, a general inhibitor for tyrosine phosphatases, was added for 25 minutes. Overexpression of PARP1 could not reduce the phosphorylation of STAT3 in the presence of vanadate (Fig. 6F). These results indicated that PARylation increases dephosphorylation of STAT3.

We then generated STAT3 with a Y705F mutation (mutation of tyrosine 705 into phenylalanine, leading to activation deficiency) and a constitutively active STAT3 (STAT3C). This mutant has STAT3 constitutively active as a result of substituting the cysteine residues (C661A and C663N), allowing STAT3 dimerization and activation. The PARylation signal could not be detected when transfected with STAT3Y705F. However, a PARylation signal was detected in STAT3C cells. The smear signal of STAT3 was consistent with its phosphorylation (Fig. 6G). Alternatively, we used another assay, in which the cells were transfected with Flag-PARP1 and HA-STAT3 followed by treatment with 5 μmol/L AZD1480 (JAK2 inhibitor) for 12 hours. The phosphorylation of STAT3 was efficiently inhibited by AZD1480, while the PARylation of STAT3 was significantly reduced (Fig. 6H). Taken together, these results indicated that PARP1 was prone to bind the active form of STAT3.

To further validate our findings in human cancer patient samples, we analyzed the correlation between the expression of PARP1 and PD-L1 in human ovarian cancer specimens using IHC. High PARP1 protein was detected in 24 (58.5%) of the 41 specimens, of which 23 (56.1%) cases showed low PD-L1 expression (Fig. 7A). The Pearson χ² test further showed an inverse correlation between PARP1 and PD-L1 expression in human cancer patient specimens (P = 0.0002, r = −0.54783). These results suggest a connection between high PARP1 expression and low PD-L1 expression in primary ovarian cancer tissues.

To explore the correlation between PARylation and STAT3, cell sorting was used to separate low and high PD-L1 expression cells after PARP1 inhibitor administration. High PD-L1 expression associated with higher p-STAT3 (Supplementary Fig. S7A and S7B). After culturing for another 72 hours (to return to baseline), a stronger smear band was observed in high PD-L1–expressing cells (Supplementary Fig. S7C), suggesting that the response by cells to PARP1 inhibition might be due to their high basal PARP1 catalytic activity. IHC of ovarian cancer patient tissues (n = 53) was used to define PAR and p-STAT3 levels and found PAR and p-STAT3 both were found in the nucleus (Fig. 7B). An inverse correlation was found between PARylation and p-STAT3 expression (P < 0.0001, r = –0.8953). Our results suggest that PARP1-mediated PARylation might affect PD-L1 by suppressing STAT3 activity.

In the current study, we determined that STAT3, a key oncogenic transcriptional factor, could be poly(ADP-ribosyl)ated by PARP1. We discovered a cross-talk between PARylation and phosphorylation. Currently, very few examples of definitive biological roles for the PARylation of proteins exist in the literature. Our results provide a clear example of how the ADP-ribosylation of a key transcription factor can affect the molecular and biochemical function of a protein, as well as the biological outcomes that it controls. More broadly, our results connected PARP1-dependent PARylation to the regulation of cancer immune suppression.

Increasing evidence exists showing that the expression of PD-L1 is mediated by the oncogenic activation of signaling pathways and is also regulated by factors in the tumor microenvironment (12, 41, 42). Signaling pathways drive PD-L1 expression primarily by transcriptional upregulation. Among the transcriptional factors involved, STAT3 has been identified as a critical regulator of PD-L1 expression in various cell systems. PD-L1 is transcriptionally regulated by STAT3 and MYC in ALK-negative anaplastic large cell lymphoma (21). The AKT–STAT3 pathway regulates the expression of PD-L1 on non–small lung cancers with aberrant EGFR activity (43), and STAT3 regulates PD-L1 expression in chemotherapy-resistant lung squamous cell carcinoma (44). In nasopharyngeal carcinoma, latent membrane protein-1 enhances PD-L1 expression with a concomitant increase in phosphorylated STAT3 (45). Here, we reported that PARP1 suppressed PD-L1 expression through the PARylation of STAT3, which could be found in various cancer cells. This finding differs from previous studies and provides a negative regulation pattern of PD-L1 in cancers. In this regard, the regulation of PARP1 activity would be important for the expression of PD-L1. Although DNA strand breaks are potent stimuli of PARP1 enzymatic activity, the context-dependent mechanism regulating PARP1 activation and signaling is poorly understood. Studies suggest that, despite similar expression, PARP1 is differentially activated in cell lines under genotoxic conditions, which generates signaling outputs with substantial heterogeneity (46). This could indicate that the upregulation of PD-L1 differs in various cancer cells. Taken together, our findings suggest that PARP1 potentially safeguards PD-L1 by suppressing STAT3 activity and that the loss of PARP1 suppression might be another mechanism for elevated PD-L1 expression in cancer cells.

STAT3 is constitutively activated by the phosphorylation of tyrosine 705 and serine 727 in a variety of cancers because of the expression of oncoproteins and tumor-produced factors. Phosphorylated STAT3 monomers form homo- or heterodimers and translocate to the nucleus where they bind to specific sequences in the promoters of target genes to regulate gene transcription (47). The activity of STAT3 could also be negatively regulated by tyrosine phosphatases, which inhibit STAT3 phosphorylation (48). In this study, we observed that silencing or inhibition of PARP1 enhanced the phosphorylation of STAT3, which was reversed by reconstitution of wild-type PARP1. Phosphorylated JAK2 was not affected by PARP1 silencing or inhibition, but treating with vanadate, a general inhibitor for tyrosine phosphatases, blocked the effects of PARP1 on dephosphorylation of STAT3. On the basis of this, we concluded that PARP1-mediated PARylation might affect the dephosphorylation process of STAT3. However, how ADP-ribosylation affects the behavior of tyrosine phosphatases needs further investigation. Here, we showed that PARylation is another posttranslational modification of STAT3 that reduces the activity of STAT3. Whether this is an important negative regulation mechanism for STAT3 is worth further study.

Among the posttranslational modifications, only the cross-talk between ADP-ribosylation and ubiquitylation has been reported (49). An example of this is PAR-dependent ubiquitylation, a process in which the PARylation of a protein serves as a signal for its subsequent ubiquitylation, which may then lead to the ubiquitin-dependent degradation of the protein by the proteasome (50). PAR-dependent ubiquitylation is likely to be a general mechanism controlling the stability and degradation of many PARP substrate proteins. Our findings reveal a cross-talk between PARylation and phosphorylation. It will be of interest to determine whether this is also a general mechanism regulating the activity of PARP substrate proteins.

Over the decades, PARP1 has been studied as the key DNA damage repair enzyme and has now become an attractive target for cancer therapy (36). A large number of PARP inhibitors are in clinical studies, and several of them have been approved for the clinical therapy of ovarian cancer. However, our study suggests that the application of PARP inhibitors in various cancers promotes the transcription of PD-L1. Whether this effect will lead to cancer evasion during treatment with PARP inhibitors should be evaluated. Our data also showed that silencing of PARP1 enhanced the transcription of PD-L1 in two breast cancer cell lines. It should be noted that in Shiping Jiao's study, they showed an inverse correlation between the PARylation and PD-L1 expression in clinical breast cancer samples, which partially supports our findings concerning the regulation of PD-L1 by PARP-mediated ADP-ribosylation.

Taken together, our study revealed that PARP1 attenuates the transcription of PD-L1 through the PARylation of STAT3 and subsequently dephosphorylation of STAT3 (Fig. 7C). Our study not only provided insights into the biological function of PARP1, thereby uncovering a new regulatory mechanism of PD-L1, but also provided new ideas and strategies to improve the clinical therapies used for ovarian cancer.

No potential conflicts of interest were disclosed.

Conception and design: L. Ding, X. Chen, B. Yang

Development of methodology: X. Chen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Chen, G. Liang, F. Yao, H. Wu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Ding, X. Chen, F. Yao, Z. Yao, J. Zhang

Writing, review, and/or revision of the manuscript: L. Ding, X. Chen, G. Liang

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

Study supervision: L. Ding, Q. He, B. Yang

Other (primary tumor samples collection): Y. Qian

This work was supported by the National Natural Science Foundation of China for Distinguished Young Scholar (No. 81625024 to B. Yang) and the National Natural Science Foundation of China (No. 81773754 and No. 81573428 to L. Ding).

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.