Tumor Intrinsic PD-L1 Promotes DNA Repair in Distinct Cancers and Suppresses PARP Inhibitor–Induced Synthetic Lethality

The immune checkpoint molecule programmed death-ligand 1 (PD-L1, CD274, B7-H1) suppresses immunity when the cell surface–expressed protein engages programmed death-1 (PD-1) on antitumor immune cells in the canonical PD-L1/PD-1 signaling pathway (1, 2). Recent work from our lab and others now clearly demonstrates novel tumor-intrinsic PD-L1 signaling consequences from non–surface-expressed PD-L1 driving tumor progression and treatment resistance (reviewed in ref. 2). As there is no uniformly established definition, we define cell-intrinsic tumor PD-L1 signaling as those that mediate cellular functions from surface, cytosolic, or nuclear PD-L1 in that cell. Tumor-intrinsic PD-L1 signals have diverse cell-intrinsic biological effects including regulation of metabolism via mTORC1 and/or autophagy (3, 4), tumor stemness (4), epithelial-to-mesenchymal transition (3, 5), transcription of immune-related genes (6, 7), sister chromatid cohesion (8), and immunosuppressive cell recruitment through NLRP3 inflammasome signaling (9). As a result of these effects and other effects, tumor-intrinsic PD-L1 influences treatment responses to mTORC1 or autophagy inhibitors (3, 9), IFNs (9), anti-PD-1 immunotherapy (6, 7), chemotherapy (10–12), and therapeutic irradiation (13).

PARP inhibitors selectively kill tumors defective in homologous recombination (HR) DNA repair (14). Several PARP inhibitor drugs have received FDA approvals for breast, ovarian, prostate, and pancreatic cancers harboring mutations in genes encoding BRCA1 or BRCA2, two essential mediators of HR (15). Interestingly, PARP inhibitors are also effective in some patients regardless of tumor BRCA mutation status (16), suggesting additional PARP inhibitor response mechanisms and biomarkers requiring further investigation. Furthermore, it is now well established that DNA repair defective cancers are amenable to immune checkpoint blockade immunotherapy through multiple mechanisms including increased tumor mutational burden (17) and activation of cytosolic DNA-sensing/type-I IFNs from the accumulation of unrepaired, damaged DNA in cytosol (18). The induction of exogenous DNA damage by PARP inhibitors increases antitumor immune responses and can synergize with anti-PD-L1 immune checkpoint blockade in some immune checkpoint blockade–resistant murine cancer models using tumors harboring BRCA1 mutations or other models with or without known defects in DNA repair (19–21). Nonetheless, human clinical trials testing anti-PD-L1/PD-1 immune checkpoint blockade plus PARP inhibitors have demonstrated only limited treatment efficacy (22). Furthermore, significant treatment resistance to PARP inhibitors can develop in initially treatment responsive tumors usually due to restoration of DNA end resection, an initiating step of HR (23, 24). Whether cell-intrinsic PD-L1 signals can control specific DNA repair pathways or drive treatment resistance to PARP inhibitors and potential mechanisms are not known.

Immune checkpoint blockade using anti-PD-L1 or anti-PD-1 antibodies is a highly successful form of immunotherapy and has revolutionized the management of many cancers including those of bladder and breast, and melanoma. Mechanistically, immune checkpoint blockade is thought primarily to work by disrupting the engagement of surface-expressed PD-L1 with its cognate receptor PD-1 expressed on exhausted antitumor CD8⁺ T cells, thus resulting in reinvigoration of antitumor immunity (1, 25). Although some individuals experience durable or complete treatment responses, most patients with cancer fail to respond to PD-L1 immune checkpoint blockade (26). For example, only approximately 20% of patients with PD-L1⁺ bladder cancer responded to anti-PD-L1 in a pivotal trial (27). These clinical observations and anti-PD-L1 immune checkpoint blockade modeling in murine transplantable tumors suggest an incomplete mechanistic understanding of tumor PD-L1 biology and that PD-L1 regulates tumor pathogenesis beyond the canonical surface PD-L1/PD-1 pathway.

We now demonstrate that cell-intrinsic tumor PD-L1 signals increase HR DNA repair capacity in tumor cells. Specifically, tumor PD-L1 promoted DNA end resection, an early step in HR. As a result, PD-L1 deficiency, but not surface PD-L1 blockade with anti-PD-L1 immune checkpoint blockade, resulted in increased DNA damage accumulation and improved tumor control in response to PARP inhibition in BRCA1 wild-type tumors even in the absence of host immunity. Remarkably, tumor PD-L1 suppressed PARP inhibitor induced activation of cell-intrinsic TANK-binding kinase 1 (TBK1) and expression of the T-cell chemoattractant CCL5, suggesting additional tumor-intrinsic PD-L1 effects on immune evasion. Mechanistically, PD-L1 associated with BARD1, a HR mediator involved in promoting BRCA1 functionality and nuclear accumulation. Tumor PD-L1 depletion resulted in reduced BRCA1 nuclear accumulation without altering BRCA1 or BARD1 content, suggesting that tumors deficient in PD-L1 may exhibit BRCAness, a tumor that shares molecular features with BRCA1-mutant tumors and responds to similar therapeutic strategies (28). Our findings demonstrate that PD-L1 is a novel HR factor and shifts the paradigm of PD-L1 signals in tumor pathogenesis from surface expressed to more global consequences. Furthermore, our studies could explain treatment resistance to PARP inhibitors that cannot be overcome with anti-PD-L1/PD-1 immune checkpoint blockade and implicate tumor PD-L1 as a biomarker for HR proficiency and PARP inhibitor resistance.

Murine melanoma B16-F10 was described previously (3). Murine 4T1 triple-negative breast cancer cells were purchased from ATCC. These were cultured in RPMI1640 (Corning) containing 5% FBS, 1% penicillin/streptomycin, 1% l-glutamate, and 1% HEPES. Human T24 and UM-UC-3, and murine MB49 bladder cancer cell lines were gifts from R. S. Svatek [University of Texas Health at San Antonio (UTHSA), San Antonio, TX] and were cultured in McCoy's 5A (T24) or DMEM (MB49, UM-UC-3; both from GIBCO), respectively, plus 1% penicillin/streptomycin. MDA-MB-231 cells were a kind gift from Dr. Susan Mooberry (UTHSA) and cultured in DMEM. All lines were confirmed Mycoplasma free by Mycoalert PLUS (Lonza Bioscience). Cell lines were not reauthenticated specifically for these experiments, except T24.

PD-L1 knockout (PD-L1^KO) cell lines were generated using commercially available CRISPR/Cas9 plasmids. Single-guide RNA (sgRNA) sequences used to generate PD-L1^KO murine cells include CTCCAAAGGACTTGTACG, GCAAGTGATTCAGTTTG, and TGCTGCATAATCAGCTA. PD-L1^KO T24 cells were generated using human PD-L1 targeting sgRNA sequences TCCCAAGGACCTATATG, ATAGTAGCTACAGACAG, and CGCTGCATGATCAGCTA. Clones were validated using flow cytometry, sequencing, and immunoblots, which we reported previously (12). PD-L1 knockdown (PD-L1^lo) in 4T1 cells was by lentiviral vector transduction with an short hairpin RNA (shRNA; Sigma) against mouse PD-L1 or a scrambled shRNA control (CTRL) as we described previously (3). Pooled cell populations were used for all genetically altered cells except for PD-L1^KO B16, which was clonal and had signal and response data in line with polyclonal cultures for data shown here.

Gemcitabine was a gift from Dr. Robert S. Svatek. Olaparib or talazoparib (PARP inhibitors) and emricasan (caspase inhibitor) were purchased from Selleckchem and diluted in DMSO for in vitro studies. Irradiation was with CellRad hardware (Precision). Therapeutic blocking antibodies (anti-PD-L1) against murine PD-L1 (clone 10F.9G2) and respective isotype control were obtained from Bio X Cell. Anti-PD-L1 targeting human PD-L1 (atezolizumab) was obtained from Invivogen. Chemicals were authenticated by the manufacturer.

Cells were lysed using 1× RIPA buffer (20 mmol/L Tris-HCl, pH 8.0, 150 mmol/L NaCl, 1 mmol/L disodium EDTA, 1 mmol/L EGTA, 2.5 mmol/L sodium pyrophosphate, 1 mmol/L β-glycerophosphate, 1% triton-X100) plus Halt protease/phosphatase inhibitor cocktail (Thermo Fisher Scientific). Protein concentrations were measured by BCA (Thermo Fisher Scientific). Immunoblotting was performed as described previously (3, 4). Briefly, 30 µg of protein were run per lane on 4% to 15% SDS-PAGE Precast TGX gels (Bio-Rad) and transferred to nitrocellulose membranes with pore thickness of 0.2 µg (Bio-Rad). For chemiluminescent detection, membranes were incubated with species-appropriate horseradish peroxide–conjugated secondary antibodies for 2 hours at room temperature. The membranes washed with tris-buffered saline with 0.1% tween 20 (TBST) were incubated with Western Lightening Plus reagent (PerkinElmer) for chemiluminescence detection. Bands were quantified using image J and normalized to loading controls. Antibodies obtained from Cell Signaling Technology (1:1,000 dilution) were phospho-histone H2AX (#9718), cleaved PARP (#5625), phospho-Chk1^Ser345 (#2348), PD-L1 (#13684), phospho-ATR^Ser428 (#2853), total-Chk1 (#2360) total-ATM (#2873), total-ATR (#13934), NBS1 (#14956), Rad50 (#3427), vinculin (#13901), β-actin (#12620), GAPDH (#2118), p-TBK1^Ser172 (#5483), and TBK1 (#3013). Antibodies obtained from Abcam were PD-L1 (#233482) and RAD51 (#133534) and diluted 1:1,000. Antibodies obtained from Santa Cruz Biotechnology were BRCA1 (#6954) and BARD1 (#74559) and diluted 1:200. PE-labeled phospho histone H2AX^ser139 (Thermo Fisher Scientific, CR55T33) was used to detect intracellular DNA damage accumulation by flow cytometry. For measuring γH2AX by flow cytometry, cells were plated in 6-well plates (100,000 cells/well), treated with indicated drugs for 48 hours, and prepared for flow cytometry of an intracellular marker as we reported previously (29). Nuclear and cytoplasmic fractionations were prepared NE-PER kit (Thermo Fisher Scientific) and immunoblotting performed as described above. Antibodies were authenticated by the manufacturer.

Cells were lysed by immunoprecipitation (IP) lysis buffer (Thermo Fisher Scientific, #87787) plus Halt protease/phosphatase inhibitor cocktail (Thermo Fisher Scientific #78442). Cell lysates were incubated with anti-PD-L1 antibody (Cell Signaling Technology, #13684,) with gentle agitating at 4°C overnight. Samples were then further incubated with 20 µL Protein A/G PLUS-Agarose (Santa Cruz Biotechnology, sc-2003) for 2 hours at 4°C, and centrifuged at 7,500 × g for 30 seconds at 4°C. Pellets were washed four times with IP lysis buffer, resuspended with 40 µL 2× sample loading buffer with β-mercaptoethanol, and heated for 5 minutes at 95°C. The denatured samples were analyzed by immunoblotting. Inputs were 10% of initial coimmunoprecipitation (co-IP) lysate and blots were exposed following enhanced chemiluminescence incubation with X-ray film for equivalent times for both inputs and IP samples.

T24 control and PD-L1^KO cells were synchronized in early S-phase with a double thymidine block. First thymidine (2 mmol/L) block for 18 hours, followed by release for 9 hours, then second block with thymidine (2 mmol/L) for 17 hours. After second block, cells were released and collected at 0, 2, 4, 6, 8, 10, and 12 hours. The nonsynchronized population was also collected. For cell-cycle progression analysis by flow cytometry, asynchronized and synchronized cells were resuspended in 38 mmol/L trisodium citrate supplemented with 5 mmol/L EDTA (pH 8.0), 100 μg/mL RNase A, and 150 μg/mL propidium iodide (for DNA staining) and incubated for 1 hour at room temperature (protected from light). Data were collected using a BD FACSCalibur and results were analyzed with FlowJo software. For each sample, at least 50,000 events were collected, and aggregated cells were gated out.

Wild-type C57BL/6J (BL6), BALB/c, RAG2^KO (BL6 background), and NOD.Cg-Prkdc^scidIl2rg^tm1Wj1/Szj (nonobese diabetic/severe combined immunodeficiency, NSG) mice were either bred in our in-house facility or purchased from Jackson Labs and maintained under pathogen-free conditions. All animal studies were approved by the UT Health San Antonio Institutional Animal Care and Use Committee.

NSG mice were challenged with MB49 (0.3 × 10⁶ cells) subcutaneously on opposite flanks. Olaparib (daily 25 mg/kg, 4% DMSO) was injected intraperitoneally on days 3–11. 4T1 cells (0.5 × 10⁶ cells) were injected into mammary fat pads of BALBc mice with Matrigel (1:1, Corning). Two hundred μg of anti-PD-L1 antibody (10F.9G2) was administered intraperitoneally per mouse once every 3 days. B16 cells were injected subcutaneously on opposite flanks into BL6 (0.5 × 10⁶ cells) or RAG2^KO mice (0.4 × 10⁶ cells). Both males and females were studied. Olaparib or vehicle controls were administered daily intraperitoneally (50 mg/kg, 4% DMSO) once tumors reached 50 mm² on days 3–13 for wild-type mouse experiments. Tumors were measured with Vernier calipers and volume calculated as (length × width²)/2. Mice were sacrificed when tumors reached approximately 1,500 mm³ or if distressed.

Cells (1 × 10³) were plated in 96-well plastic culture plates in respective medium and treated for 5 days with olaparib or talazoparib at concentrations shown, or anti-PD-L1 (50 µg/mL) for 5 days. Drug concentrations were carefully optimized in preliminary work. Cell viability was by MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide) at 5 mg/mL. Absorbance was measured at 540 nm. Results are presented as a mean ± SEM. Total cells and viability were confirmed using a Vi-Cell-XR cell counter (Beckman Coulter).

The U2OS-puro-DR-GFP cell line has been described previously (30, 31). Cells were reversed cotransfected with 3 µg I-SceI expression plasmid (pCβASce) and 500 ng MISSION TRC2 pLKO.5-Puro empty vector or shRNA constructs, as indicated. Transfected cells analyzed by flow cytometry after 72 hours on a Celesta flow cytometer to measure the percentage of cells expressing GFP as described previously (32).

Total RNA was extracted from cell pellets with RNeasy (Qiagen) followed by cDNA synthesis using Superscript Vilo (Thermo Fisher Scientific). CCL5 cDNAs were assessed by SYBR green qPCR normalized to 18s-rRNA. Relative expression levels were calculated by the ΔΔC_t method. Deionized water as negative control was used.

Cells were seeded into 8-well IBIDI silicone-chambered slides and treated as indicated. Cells were fixed with 100% methanol (5 minutes) followed by 1× PBS wash. Staining was performed using MAXpack Immunostaining Medium Kit (Active Motif). Cells were incubated with MAXblock Blocking Medium for 1 hour at room temperature, followed by incubation with primary antibodies diluted in MAXbind Staining Medium at 4°C overnight. Primary antibodies were p-BRCA1^Ser1524 (1:200, Cell Signaling Technology, 9009), BRCA1 (1:100, SC-6954), p-RPA32-S4/8 (1:200, Abcam, ab87277), RAD51 (1:200, Abcam, ab3801), PD-L1 (1:1,000, Cell Signaling Technology, #29122), and 53BP1 (Abcam, ab175933). Cells were then washed with MAXwash and incubated with secondary antibodies (1:400 conjugated to Alexa Fluor 488 or Alexa Fluor 647). After washing three times for 10 minutes each with MAXwash, DAPI stain (0.02 µg/mL, Sigma, D9542) was applied for 10 minutes at ambient temperature, followed by MAXwash. Slides were mounted with ProLong Diamond (Thermo Fisher Scientific) and Slip-Rite cover glass (Thermo Fisher Scientific). Fluorescence was detected by a Zeiss LSM 710 confocal microscope with 63× magnification objective using identical settings and overlaid using ImageJ. Only discrete DNA repair foci (53BP1, Rad51) were quantified manually within the nuclear area using high-resolution zoomed in digital images, which was defined by DAPI-positive immunostaining, using Counter plugin in the ImageJ software. To calculate total BRCA1 nuclear to cytoplasmic ratio, protein intensities were measured over entire nuclei or cytoplasmic area, which was selected for each cell using the freeform drawing tool in ImageJ software within or outside of the DAPI-positive immunostaining, respectively. Mean background readings were subtracted from nuclear or cytoplasmic value for each cell using the following formula: Corrected nuclei/cytoplasm total BRCA 1 fluorescence = integrated density − (area of selected nuclei/cytoplasm × mean fluorescence of background readings) as described previously (33, 34). Control and PD-L1^KO cell lines were immunostained and imaged at the same time. But as shown on this zoomed image, discrete foci are nuclear as expected. For quantification, image J was used to quantify nuclear 53BP1 that colocalizing with DAPI nuclear stain (counting “red objects in blue”). Drug and X-ray concentrations were carefully optimized in preliminary work.

Statistical analyses were performed using Prism software (GraphPad). Data are represented as mean ± SEM unless specified otherwise. In vivo tumor growth was analyzed by two-way ANOVA plus Bonferroni post-tests. All other single measurement assays were assessed with unpaired Student t test.

Occasionally, datasets with suspected outliers were identified by Grubbs’ test (used only once for a given dataset) and removed from analysis. For all analyses, significance was based on a multiplicity-corrected α of 0.05.

Relevant data will be made available to qualified investigators upon reasonable request.

To study the role of PD-L1 in HR DNA repair, we generated PD-L1^KO human T24 and murine MB49 bladder cancer cells by CRISPR/Cas9 and corresponding control cells (CTRL), neither of which harbor germline BRCA mutations. We recently reported the validation of these cells and demonstrated that PD-L1^KO bladder cancer tumors were intrinsically more sensitive to the DNA-damaging cytotoxic chemotherapy agents gemcitabine or cisplatin (12), two standard-of-care drugs capable of inducing double-stranded DNA breaks, a lethal form of DNA damage. Because PD-L1 deficiency resulted in reduced cell viability in response to chemotherapy-induced DNA damage, we hypothesized that tumor PD-L1 could promote HR, a high-fidelity double-stranded DNA break repair mechanism utilized by cells when a homologous template is present (i.e., after DNA replication). Given the crucial role of BRCA1 in initiating HR, we assessed the ability of PD-L1^KO cells to form BRCA1 nuclear foci following ionizing radiation from X-rays. PD-L1^KO cells formed reduced DNA damage–induced BRCA1 foci (p-BRCA1¹⁵²⁴) compared with CTRL cells 1 hour after exposure to 2 Gy ionizing irradiation or 1 ng/mL gemcitabine in vitro by immunofluorescence microscopy (Fig. 1A).

BRCA1 initiates HR by facilitating DNA end resection, a nucleolytic process resulting in the production of 3′ single-stranded DNA at the double-stranded DNA break site (35). Consistent with reduced DNA damage–associated BRCA1 nuclear foci, PD-L1^KO cells had diminished p-RPA32^S4/S8, a surrogate marker for DNA end resection, in response to ionizing radiation (Fig. 1A). Specifically, RPA32 coats the single-stranded DNA freed by resection at the break and activates the ATR/Chk1 pathway (36). In further support of impaired DNA end resection, immunoblotting demonstrated reduced ionizing radiation–induced p-RPA32 and p-ATR/p-Chk1, in cells depleted for PD-L1 following irradiation (Fig. 1B). ATR is activated downstream of RPA32-coated single-stranded DNA (reviewed in ref. 37) and thus ineffective ATR activation would further lead to ineffective repair of resected double-strand DNA break lesions carried over by end resection effectors in the HR pathway like BRCA1. Reduced RPA32 foci and consequently reduced phosphorylation of ATR/Chk1 pathway in PD-L1^KO cells is thus consistent with impaired HR DNA damage repair at an upstream step (DNA end resection) in these cells following X-irradiation.

Consistent with a prior report in breast and colon cancer cells (13), we observed increased accumulation of DNA damage measured by γH2AX in PD-L1^KO T24 versus CTRL cells (Fig. 1B), an indicator of double-stranded DNA breaks. In further support of impaired double-strand break repair due to PD-L1 deficiency, PD-L1^KO failed to repair X-irradiation induced DNA damage even by a late timepoint of 24 hours (Fig. 1C). To demonstrate that increased γH2AX accumulation in PD-L1^KO cells was independent of apoptotic cell death, we repeated experiments in the presence of the pan-caspase inhibitor emricasan and again found increased γH2AX accumulation in PD-L1^KO versus CTRL cells after X-irradiation (Fig. 1D). Thus, failure to prevent DNA damage accumulation in these PD-L1^KO cells is likely caused by faulty DNA repair rather than increased intrinsic sensitivity to cell death. Altogether, these data suggest that tumor PD-L1 can promote the initiation of HR by regulating the formation of BRCA1 foci and consequently DNA end resection in response to exogenous DNA damage.

For downstream homology-directed repair of double-stranded DNA breaks to occur, the RPA-coated end-resected DNA must be replaced with the recombinase RAD51 that catalyzes pairing with a homologous DNA template (35, 38). Consistent with reduced DNA end resection, PD-L1^KO cells essentially failed to form ionizing irradiation–induced RAD51 foci (Fig. 2A and B), a marker of HR efficiency (39). This finding suggests that tumor PD-L1 can specifically promote the homology-directed repair of double-stranded DNA breaks and that PD-L1–deficient tumors could phenotypically mimic BRCA1-mutant tumors. In the context of reduced HR and in the absence of end resection, ligation-dependent double-stranded DNA break repair pathways predominate such as nonhomologous end joining (NHEJ; ref. 35). P53 binding protein 1 (53BP1) facilitates NHEJ by physically protecting double-stranded DNA breaks from being end resected (40). In BRCA1-proficient cells, BRCA1 facilitates the removal of 53BP1 to initiate HR. Consistent with reduced p-BRCA1 and p-RPA32 and with reduced RAD51 recruitment to nuclear foci, PD-L1^KO cells significantly accumulated 53BP1 foci compared with CTRL cells (Fig. 2C and D), suggesting an enhanced reliance on NHEJ due to PD-L1 deficiency.

We next utilized the direct repeat GFP reporter (DR-GFP) system to measure functional HR efficiency of DNA repair by a site-specific double-stranded DNA break (30). PD-L1 knockdown by shRNA resulted in a significant approximately 2-fold reduction in HR efficiency in U2OS cells (Fig. 2E). Utilizing the EJ5 reporter system for DNA end joining (41), we observed a moderate but significant increase in NHEJ efficiency in response to double-stranded DNA breaks following PD-L1 knockdown in U2OS cells (Fig. 2F). Together, these data suggest that PD-L1 promotes the efficiency of double-stranded DNA break repair by HR and that PD-L1 deficiency can elicit a compensatory increase in NHEJ. We additionally assessed cell-cycle progression following thymidine synchronization and found no significant impairment in progression of PD-L1^KO through G₁–S–G₂-phases compared with CTRL cells (Fig. 2G–I). Thus, as BRCA1-mediated HR predominantly occurs in S-phase (35), the PD-L1–dependent regulation of BRCA1 is unlikely due to a block in the G₁–S transition.

Given our findings that genetic PD-L1 depletion can induce HR deficiency by reducing BRCA1 nuclear foci formation and consequently DNA end resection, we assessed whether tumor cell–intrinsic PD-L1 signals could regulate the content of HR factors involved in initiating DNA end resection and execute homology-directed repair. We did not observe a significant difference in cell content of BRCA1 or its obligate interacting partner BARD1 in CTRL versus PD-L1^KO cells (Fig. 3A), suggesting that PD-L1 could regulate BRCA1 localization rather than BRCA1 gene expression. Similarly, expression of factors affecting DNA end resection and homology-directed repair including ATM, NBS1, RAD50, or RAD51 did not change in our CRISPR/Cas9 PD-L1^KO cells in culture (Fig. 3A).

We next considered whether cell-intrinsic PD-L1 could facilitate HR by forming protein complexes with BRCA1 or its obligate interacting partner BARD1. Interestingly, endogenous PD-L1 immunoprecipitated BARD1 but not BRCA1 in human tumor cells and the PD-L1-BARD1 association could be further increased following exposure to DNA damage (Fig. 3B and C). These data implicate intracellular PD-L1 as a novel protein involved in HR and suggest the possibility that PD-L1 could improve BRCA1 function through BARD1 interaction for DNA end resection promotion, but does not exclude more indirect mechanisms or lack of mechanistic, direct interactions.

BARD1 depletion can phenocopy BRCA1 loss in eliciting PARP inhibitor synthetic lethality (32), HR defects (32, 42), and cancer susceptibility risk (43). BARD1 facilitates BRCA1-mediated HR DNA repair by enhancing BRCA1 nuclear localization thereby allowing BRCA1 to accumulate in DNA damage foci (44). Given that PD-L1 did not alter the BRCA1 expression level in our cells but instead interacted with BARD1, we considered whether tumor PD-L1 deficiency could phenocopy BARD1 defects and alter BRCA1 accumulation in the nucleus. PD-L1 depletion diminished BRCA1 nuclear foci (Fig. 3D and E), consistent with significant reduction in the ratio of nuclear to cytoplasmic BRCA1 in PD-L1^KO versus CTRL cells before and after exposure to a high dose of 8 Gy X-irradiation assessed by confocal imaging (Fig. 3D and F) and immunoblots of subcellular fractionations (Fig. 3G).

Thus, PD-L1 deficiency can phenocopy aspects of BARD1 loss in reducing BRCA1 nuclear localization and the consequent reduction in formation of BRCA1 DNA damage–induced nuclear foci. Altogether, our data support a novel concept that tumor PD-L1 when intracellular (as opposed to surface-expressed) could regulate the activity of crucial molecules required for HR. Such effects could increase treatment resistance to small-molecule PARP inhibitors approved in cancer treatments.

Turner and colleagues originally defined “BRCAness” as any tumor that shares phenotypic features with germline BRCA1-mutated cancers (28). One such defining phenotype of BRCA1-mutated tumor cells is their striking sensitivity to PARP inhibitors (14). Because PD-L1–deficient cells had reduced BRCA1 functionality and HR efficiency compared with cells expressing PD-L1, we considered whether PD-L1 deficiency could induce BRCAness and result in enhanced sensitivity to small-molecule PARP inhibitors. Consistent with impaired functional HR by DR-GFP (Fig. 2E), PD-L1^lo U2OS cells were significantly more sensitive to PARP inhibitor treatment compared with CTRL cells in vitro, in a system devoid of any immune cells (Fig. 4A). Interestingly, neither human T24 nor the murine transplantable melanoma cell line B16-F10 exhibited appreciable PARP inhibitor synthetic lethality following PD-L1^KOin vitro (Fig. 4B and C). To assess the generalizability of immune-independent, cancer cell–intrinsic PD-L1–driven resistance to PARP inhibitors, we chose two additional bladder cancer cell lines, which we recently reported and characterized functional cell-intrinsic PD-L1 effects (12): human UM-UC-3 and murine MB49. In contrast to T24, PD-L1^lo UM-UC-3 and PD-L1^KO MB49 were both significantly more sensitive to PARP inhibitor treatment to respective CTRLs in vitro (Fig. 4D and E). Because, UM-UC-3 but not T24 cells exhibited increased synthetic lethality to the potent PARP inhibitor, talazoparib (compare Fig. 4B with D) following PD-L1 depletion, the translational consequences of cancer cell–intrinsic PD-L1 promotion of HR on PARP inhibitor treatment efficacy could depend on tumor type, histology, genetic context (e.g., TP53 status) or specific PARP inhibitors, which have different PARP trapping abilities (olaparib does not, whereas talazoparib does trap PARP protein at damaged DNA, enhancing cytotoxicity; ref. 45).

To dissociate tumor-intrinsic PD-L1 control of BRCAness formally from canonical, immune cell suppressive effects of PD-L1, we challenged severely immunodeficient NSG mice with CTRL or PD-L1^KO MB49 (which had robust in vitro PARP inhibitor sensitivity) and assessed PARP inhibitor treatment outcomes in vivo. Consistent with in vitro viability data using (Fig. 4E), PD-L1^KO but not CTRL MB49 tumor growth was robustly growth inhibited (Fig. 4F and G), corroborated by significantly lower tumor weights after PARP inhibitor (olaparib) treatment (Fig. 4H). These data altogether support that cell-intrinsic PD-L1 signals regulate BRCAness and that PD-L1 deficiency can confer PARP inhibitor synthetic lethality without the requirement of host immunity in some tumors. Furthermore, cell-intrinsic PD-L1 regulation of HR could be refractory to inhibition by currently FDA-approved anti-PD-L1 immune checkpoint blockade antibodies and our findings could explain treatment failure of immune checkpoint blockade and PARP inhibitor combinations seen in early-phase clinical trials (22).

Some cell-intrinsic tumor PD-L1 effects, including its promotion of mTORC1 signals (3) and MAPK signaling, are transmitted by surface-expressed PD-L1 and can be PD-1 dependent (46). Clinically available anti-PD-L1 antibodies block the PD-1 binding site of PD-L1 extracellularly and disrupt the PD-1/PD-L1 signaling axis to provoke changes in cell biology in an immune-independent manner (3, 47). In contrast to genetic PD-L1^KO, anti-PD-L1 antibodies did not significantly enhance T24, MB49, B16-F10, or 4T1 sensitivity to the PARP inhibitor olaparib in vitro (Fig. 5A,–D) at antibody concentrations we and others showed to be biologically active for other tumor functional outcomes (3, 47). These results suggest an additional role of tumor PD-L1 in regulating BRCAness separate from its known suppressive effect on antitumor CD8⁺ T cells and support our data presented here that intracellular but not surface PD-L1 could regulate HR.

To test genetic tumor PD-L1 deficiency effects on PARP inhibitor responses in vivo in immune competent hosts, we transplanted B16-F10 melanoma cells subcutaneously into wild-type, syngeneic C57BL6 mice. CTRL tumors were resistant to the PARP inhibitor olaparib, whereas PD-L1^KO B16-F10 tumors significantly responded (Fig. 5E and F). We were unable to compare MB49 responsiveness in vivo in wild-type mice due to lack of reliable PD-L1^KO MB49 tumor growth in wild-type mice (12). Host PD-L1 was not required for olaparib suppression of PD-L1^KO tumors seen in their olaparib sensitivity when transplanted into genetically PD-L1 depleted, syngeneic hosts (Fig. 5G) further consistent with PARP inhibitor effects from tumor PD-L1.

Next, we transplanted highly anti-PD-L1 immune checkpoint blockade–resistant 4T1 triple-negative breast cancer cells into mammary fat pads of wild-type, syngeneic BALBc mice and asked whether anti-PD-L1 could synergize with PARP inhibitor. Blocking surface PD-L1 in vivo with anti-PD-L1 antibodies did not increase olaparib efficacy in immunologically competent wild-type mice bearing 4T1 tumors (Fig. 5H), consistent with in vitro data (Fig. 5D). In striking contrast, genetic PD-L1 depletion significantly increased 4T1 olaparib sensitivity in vivo (Fig. 5I). Thus, genetic PD-L1 targeting, but not surface PD-L1 blockade with anti-PD-L1 antibodies, can enhance PARP inhibitor sensitivity in immune checkpoint blockade–resistant BRCA1 wild-type tumors, further suggesting novel cell-intrinsic tumor PD-L1 participation in HR.

Despite observing no significant synthetic lethality in vitro following olaparib exposure in PD-L1^KO versus CTRL T24 or B16 cells (Fig. 4B and C), we still measured a significant approximately 2-fold increase in accumulation of double-stranded DNA breaks measured by γH2AX in PD-L1^KO T24 cells in vitro after 48 hours compared with CTRL cells (Fig. 6A). Similarly, we observed PD-L1–dependent increases in γH2AX in response to olaparib treatment in the other BRCA sufficient murine syngeneic transplantable lines MB49 bladder cancer, B16 melanoma, and 4T1 triple-negative breast cancer (Fig. 6B,–D) also representing distinct genetic major histocompatibility complex backgrounds and consistent with cell-intrinsic PD-L1 promotion of double-strand DNA break repair by HR. Prior studies have demonstrated that PARP inhibitor efficacy in HR defective tumors in immunologically intact hosts in vivo is at least partially dependent on the activation of innate cytosolic DNA sensing like cGAS/STING (48, 49). Given that cell-intrinsic tumor PD-L1 signals promoted HR, we considered whether tumor PD-L1 could also suppress PARP inhibitor–induced downstream activation of such cell-intrinsic innate immune DNA-sensing activation.

PD-L1^KO in B16-F10 tumor cells exhibited significantly increased activation of p-TBK1, a downstream effector molecule of STING-driven tumor immunogenicity relative to untreated cells in PD-L1^KO and CTRL cells, respectively, following olaparib treatment (Fig. 6E). In line with p-TBK1 activation, we also observed approximately 3-fold increased CCL5 expression, a TBK1-induced T-cell chemoattractant, in PD-L1–deficient cells basally, which was further increased by olaparib exposure compared with identically treated CTRL cells (Fig. 6F). Although these signals prove innate immune activation in tumor cells (48), they might not mechanistically explain therapeutic responses to PARP inhibitor in vivo. Of note, DNA damage can generally provoke increased PD-L1 expression (50). Specifically, PARP inhibitors increase PD-L1 expression through multiple mechanisms, resulting in canonical inhibition of CD8⁺ T cells (19, 51, 52). Similarly, we found significantly increased tumor PD-L1 expression following olaparib treatment (Fig. 6E). Thus, these prior studies and our novel findings implicate dual consequences of DNA damage–induced PD-L1 in immune suppression by additionally antagonizing excess activation of innate DNA-sensing effector molecules in tumor cells. In further support of an immune component to PARP inhibitor treatment efficacy elicited by PD-L1 deficiency, we observed complete loss of olaparib efficacy against PD-L1^KO B16-F10 tumors in RAG2 knockout mice (RAG^KO; Fig. 6G) versus its high efficacy against this tumor in immune competent wild-type mice (Fig. 5F) including those that are PD-L1 deficient (Fig. 5G). These RAG2^KO mice lack T cells and B cells and thus lack all tumor-specific immunity (3). These results could explain lack of immune-independent treatment response to PARP inhibitors in PD-L1^KO B16 in vitro yet effective treatment response in vivo in wild-type mice and support potential immune-mediated cell-intrinsic PD-L1–driven PARP inhibitor resistance through DNA repair by HR. Our findings altogether suggest that tumor PD-L1 depletion can enhance tumor sensitivity to prototypical PARP inhibitor through cell-intrinsic mechanisms that can be immune independent or immune dependent in selective cancers and implicate PD-L1 deficiency as a treatment response biomarker.

Although canonical surface-expressed PD-L1 signaling extrinsically to PD-1⁺ antitumor immune cells is well appreciated (26), we demonstrate here that novel tumor-intrinsic PD-L1 signals promote HR in tumor cells. Specifically, PD-L1 promoted DNA end resection in response to DNA damage, an early step of HR. PD-L1 deficiency resulted in reduced BRCA1 nuclear accumulation and DNA damage–associated nuclear foci. Therefore, genetic PD-L1 reduction led to PARP inhibitor synthetic lethality, in vitro and improved PARP inhibitor treatment efficacy in vivo. We further demonstrate that genetic tumor PD-L1 depletion, but not anti-PD-L1 immune checkpoint blockade antibodies can sensitize highly immunotherapy resistant, BRCA-proficient tumors to PARP inhibitors showing distinct differences in outcomes from cell-intrinsic PD-L1 versus canonical surface-expressed PD-L1 targeted by current FDA-approved immune checkpoint blockade antibodies. Thus, our findings not only inform novel DDR biology and also show high translational potential, they show a means to improve PARP inhibitor efficacy where anti-PD-L1 does not, also helping explain a potential treatment resistance mechanism to anti-PD-L1/PARP inhibitor combinations. These data also suggest tumor PD-L1 as a novel BRCAness factor and/or treatment response biomarker.

Two prior studies reported PD-L1 that facilitated gene expression either by promoting mRNA stability or transcription (6, 13), suggesting the presence of immune checkpoint independent, functional tumor cell–intrinsic PD-L1 in the cytoplasm and nucleus. However, using stable CRISPR/Cas9 PD-L1^KO tumor cells, we observed no significant difference in protein content of several factors involved in the early initiating steps of HR including BRCA1 or BARD1 that are inconsistent with mRNA stability or transcriptional effects and support a nontranscriptional PD-L1–governed mechanism controlling HR, which appears to be through intracellular protein–protein interactions here. In that regard, PD-L1 immunoprecipitated with BARD1, the obligate interacting partner of BRCA1 for its nuclear transition. Our PD-L1^KO cells phenocopied known effects of BARD1 loss (44) including reduced BRCA1 nuclear accumulation and DNA damage–induced BRCA1 intranuclear foci. Although we observed a potential BARD1/PD-L1 interaction, additional biochemical work using PD-L1 mutants are required to confirm and understand the functional importance of such complexes in human cells fully and to distinguish nuclear versus cytoplasmic PD-L1 consequences. Nuclear expressed PD-L1 could facilitate the retention of BRCA1-BARD1 at sites of DNA breaks by tethering protein complexes to chromatin in a similar manner to that previously reported for the BARD1-interacting protein HP1 (42). Alternatively, nuclear PD-L1 could inhibit nuclear export, or cytoplasmic PD-L1 could enhance nuclear import of BRCA1-BARD1. PD-L1 could regulate BRCA1-BARD1 ubiquitin ligase activity among other considerations. Ubiquitination of local histones by BRCA1-BARD1 and in conjunction with the chromatin remodeler SMARCAD1, facilitates the eviction of 53BP1 from DNA ends (53). This latter possibility is consistent with our finding that PD-L1 depletion results in increased 53BP1 nuclear foci, reduced DNA end resection, and enhanced NHEJ in response to double-stranded DNA breaks. Altogether our findings support the novel concept that tumor cell–intrinsic PD-L1 participates in double-strand DNA break repair, which could also alter HR/NHEJ pathway choice, all areas requiring further investigations.

Upstream DNA damage–sensing kinases are required to signal the damaged DNA (e.g., ATM, ATR, DNA-PK activate γH2AX) but are triggered by specific types of DNA lesions to elicit specific responses. ATR responds to double-strand DNA breaks and replication lesions, and thus is classically activated by the product of end resection (single-stranded DNA coated by RPA32). ATR/Chk1 activation is considered here as an additional readout of end resection along with reduced p-RPA32 foci shown by imaging. The reduced activation (phosphorylation) of the ATR/Chk1 DNA damage–sensing pathway in PD-L1^KO cells is thus consistent with the impaired HR DNA damage repair that we demonstrated in these cells and could inform X-irradiation or PARP inhibitor combinations with other selected DNA damage response inhibitors (e.g., Chk1i or ATRi).

Aside from improving understanding of DNA repair biology, our data provide insights into clinical translational opportunities. For example, several PARP inhibitors are FDA approved to treat various cancers (14, 15). They were originally developed for use in BRCA-mutant cancers, but it is now appreciated that BRCA-sufficient cancers can also respond to PARP inhibitors, termed “BRCAness” (28). Our data suggest that BRCAness mechanisms could include loss of PD-L1 promotion of BRCA1 nuclear functions. Predictive biomarkers for PARP inhibitor response are lacking. Our data suggest that tumor PD-L1 status could be used in selected cancers as a treatment response biomarker. Resistance to PARP inhibitors remains a clinical challenge. Our data support the concept that tumor PD-L1 expression could be a PARP inhibitor cell-intrinsic resistance mechanism, but it will require additional work to understand the epistasis between PD-L1 and other BRCAness genes. Finally, clinical utility of combining PARP inhibitor with classic surface immune checkpoint blockade has thus far been only modest (22). On the basis of insights here, it may be possible to identify patients more likely to respond to such combinations.

In contrast, combining immune checkpoint blockade with cytotoxic chemotherapy has proven surprisingly effective, leading to several FDA approvals of cytotoxic chemotherapy plus either anti-PD-1 or anti-PD-L1 blockade in triple-negative breast cancer, for example. We propose that combining selected chemotherapy agents, such as gemcitabine we tested here, plus PARP inhibitor could be effective. These clinical data also suggest that PD-L1 could control other DNA damage sensing or repair pathways, implicating additional new DNA repair biology, and other potentially exploitable targets based on understanding the underlying biology. For example, the differential effect of tumor-intrinsic PD-L1 on Chk1 activation we observed suggests that PD-L1 depletion could be combined with small-molecule ATM/Chk2 or ATR/Chk1 inhibiting molecules now in clinical trials. The innate immune activation (e.g., p-TBK1/CCL5) elicited by PD-L1 depletion could be exploited in any of these scenarios. Our preliminary studies suggest that such is the case, but additional mechanistic in vivo studies are required to understand relative contribution of such signals to PARP inhibitor treatment efficacy in distinct tumor models. Finally, as genetic PD-L1 depletion on tumors is not yet feasible in human trials, pharmacologic PD-L1 depletion as we (10) and others (54, 55) have shown, are clinical strategies worth investigating.

PD-L1 is largely considered a surface bound molecule but can be observed in other subcellular compartments including the nucleus and cytoplasm (6, 56). How PD-L1 accumulates in subcellular compartments remains incompletely understood, but some reports implicate specific cellular stress factors like hypoxia (7), protein regulatory machinery such as CMTM4/CMTM6 (57, 58), or the cell cycle (59) as potent inducers or regulators of intracellular PD-L1 accumulation. Interestingly, the DNA-damaging chemotherapy agent doxorubicin can significantly induce PD-L1 nuclear localization (60) supporting a role for PD-L1 in genomic maintenance in response to DNA damage. In congruence, we observed increased PD-L1-BARD1 coimmunoprecipitation in some human cells following treatment with the cytotoxic chemotherapy agent gemcitabine. PD-L1 could physically colocalize in discrete nuclear foci with HR proteins like BARD1, an area requiring additional studies. There is a strong possibility that the fraction of PD-L1 that could translocate (induced by exogenous DNA damage) to the nucleus and mediate DNA repair is modified by phosphorylation or other posttranslational modifications, and would require a specific antibody directed against such modified PD-L1 to perform nuclear PD-L1 immunofluorescence efficiently. Examples in the field include γH2AX, p-RPA, and p-ATM. Such an assay would be of high interest clinically if PD-L1 nuclear translocation is proven functionally significant to DDR.

Interestingly, we did not see appreciable PARP inhibitor sensitivity in PD-L1–depleted T24 human bladder cancer cells or B16 mouse melanoma cells compared with respective control cells in vitro despite defective HR, increased accumulation of double strand breaks measured by γH2AX, and enhanced PARP inhibitor responses in wild-type mice to the latter. However, human U2OS sarcoma and UM-UC3 bladder cancer cells exhibited significant PARP inhibitor synthetic lethality, consistent with observed functional HR defects following PD-L1 depletion, although effects were influenced by the specific PARP inhibitor tested. One possibility is that cells like T24 could be intrinsically resistant to certain PARP inhibitors like olaparib even when HR defective. Another possibility is that PARP inhibitor treatment response in selected cancer cells depleted of PD-L1 could depend on specific tumor histologies or genetic backgrounds. Immune contributions could also assist the PARP inhibitor efficacy in vivo we observed in B16 tumors. PARP inhibitor efficacy in HR defective tumors in vivo is reported to depend on activation of the innate cGAS/STING/type-1 IFN cytosolic DNA-sensing pathway (48, 49). We detected augmented activation of TBK1 and expression of the T-cell chemoattractant CCL5, known effector molecules downstream of STING, in PD-L1^KO versus CTRL B16 cells following olaparib treatment, implicating an immune checkpoint–independent function of tumor PD-L1 in tumor immune evasion. However, whether the cGAS/STING pathway is involved in cell-intrinsic PD-L1 suppression of PARP inhibitor induced TBK1 and CCL5 and consequent PARP inhibitor treatment resistance in immunologically intact hosts requires additional mechanistic studies in vivo. Finally, specific PARP inhibitor mechanisms or potency could dictate treatment response in some human cancers following cell-intrinsic PD-L1 depletion. Tumor PD-L1 expression could be an important resistance mechanism to PARP inhibitors but may not supersede other known resistance mechanisms. Separate genetic studies are required to understand the epistatic relationship of PD-L1 with other double-strand break repair regulatory proteins or other oncogene/tumor suppressors.

Our study has limitations. Although we found reduced ATR signals in PD-L1^KO cells after X-irradiation, a mechanistic link to reduced HR DNA repair as we expect has not yet been demonstrated. PD-L1^KO MB49 tumors responded to olaparib in an immune-independent manner in vivo, whereas PD-L1^KO B16 tumors exhibited an immune component to their olaparib sensitivity in vivo. Our preliminary studies show that PD-L1 suppresses tumor STING activation, which could account for these immune-related efficacy differences, but more work is needed to define mechanistic details. Specific PD-L1 interactions with BARD1 were supported by our data, but not explicitly defined. Despite many intensive efforts, we encountered technical difficulties precluding reliable studies of PD-L1/BARD1 interactions in cells, largely related to poor performance of anti-PD-L1 antibodies in molecular/cell biology studies. Nonetheless, we show for the first time novel cancer cell PD-L1 control of a specific DNA repair pathway and demonstrate new DNA repair biology, clinical translation approaches and potential for novel treatment response biomarkers. Tumor cell–intrinsic PD-L1 signals are an important area for biology, drug, and biomarker discovery meriting much additional attention as we reviewed recently (2). It is also an immunoglobulin superfamily member, whose other immunoglobulin superfamily members share some signaling similarities (2) meriting investigations.

A.V.R. Kornepati reports grants from NCI during the conduct of the study; in addition, A.V.R. Kornepati has a patent for “Methods for diagnosing and treating cancers” pending. C.M. Rogers reports grants from NIH during the conduct of the study. R.S. Svatek reports other support from JBL, Merck, and Emtora and personal fees from Ferring Pharmaceuticals outside the submitted work. R. Li reports personal fees from Parthenon Therapeutics outside the submitted work. Y. Hu reports grants from NIH during the conduct of the study. J.R. Conejo-Garcia reports personal fees and other support from Alloy Therapeutics and grants and personal fees from Anixa Biosciences outside the submitted work; in addition, J.R. Conejo-Garcia has a patent for Anixa Biosciences issued, licensed, and with royalties paid from The Wistar Institute and a patent for Compass Therapeutics pending. E. Dray reports grants from NIH during the conduct of the study. T.J. Curiel reports grants from Clayton Foundation, NCI, and Owens Foundation during the conduct of the study; personal fees from Agenus, Xencor, Faron, ImVax, and Gilead outside the submitted work; in addition, T.J. Curiel has a patent for CLFR.P0482WO pending to C/A. No disclosures were reported by the other authors.

A.V.R. Kornepati: Conceptualization, data curation, formal analysis, supervision, validation, investigation, methodology, writing–original draft, writing–review and editing. J.T. Boyd: Investigation. C.E. Murray: Investigation. J. Saifetyarova: Investigation. B. de la Peña Avalos: Investigation. C.M. Rogers: Validation, investigation, writing–review and editing. H. Bai: Investigation. A.S. Padron: Investigation. Y. Liao: Validation, investigation. C. Ontiveros: Investigation. R.S. Svatek: Conceptualization, resources, supervision. R. Hromas: Conceptualization. R. Li: Conceptualization, formal analysis, methodology. Y. Hu: Conceptualization, investigation. J.R. Conejo-Garcia: Conceptualization, writing–review and editing. R.K. Vadlamudi: Conceptualization, formal analysis, supervision, funding acquisition, validation, methodology. W. Zhao: Conceptualization, supervision, investigation. E. Dray: Conceptualization, investigation, writing–review and editing. P. Sung: Conceptualization, formal analysis, supervision, writing–review and editing. T.J. Curiel: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

DR-GFP and EJ5 cells were a kind gift from Jeremy Stark and Maria Jasin (Memorial Sloan Kettering). MDA-MB-231 cells were a kind gift from Sue Mooberry, UTHSA. This study was supported by NIH grant awards F30-CA239390 (A.V.R. Kornepati), TL1 TR002647 (C.E. Murray), South Texas MSTP NIH-T32GM113896 (A.V.R Kornepati, C.E. Murray, J.T. Boyd, C. Ontiveros), R01-GM141091 (W. Zhao), R35 CA241801 (P. Sung), and R01CA205965, P30 CA054174, STARS, Owens Foundation, the UTHSA Daisy M. Skinner endowment, The Ovarian Cancer Research Alliance of Greater Cincinnati and the Clayton Foundation (T.J. Curiel). Flow cytometry data were generated in the UT Health San Antonio Flow Cytometry Shared Resource Facility, which is supported by the National Center for Advancing Translational Sciences, National Institutes of Health, through grant UL1 TR002645.

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.