Tumor-Intrinsic PD-L1 Signals Regulate Cell Growth, Pathogenesis, and Autophagy in Ovarian Cancer and Melanoma

PD-L1 (B7-H1, CD274), an immune co-signaling molecule in the B7-homology (B7-H) family (1), negatively regulates T-cell functions through PD-1 and CD80 interactions (2) and is immunopathogenic in diverse cancers (3). Anti–PD-L1 monoclonal antibodies (αPD-L1) are clinically efficacious against distinct cancers where the principal mechanism of action is thought to be protecting PD-1–expressing antitumor T cells from inhibition by tumor PD-L1 (3–7). αPD-L1 is FDA-approved for bladder and lung cancers. αPD-1 was recently FDA-approved to treat melanoma, renal cell carcinoma, non–small cell lung cancer, and head and neck cancers. It is also proposed to work by protecting PD-1–expressing antitumor T cells from inhibition by tumor surface–expressed PD-L1, among other potential mechanisms (7, 8).

Because of the promise of αPD-L1 and αPD-1 immunotherapy, much attention has focused on effects of the tumor PD-L1/T cell PD-1 axis in cancer immunotherapy and immunopathogenesis. Following the initial report that tumor PD-L1 kills antitumor T cells (4), most work on PD-L1 signals in cancer has focused on tumor-extrinsic PD-L1 effects, particularly on T cells. However, recent work shows that tumor PD-L1 prevents tumor apoptosis (9), reduces chemotherapy-mediated killing by altering mitogen-activated protein kinase signals (10), and regulates tumor glucose metabolism in sarcomas (11). Thus, tumor PD-L1 has important tumor-intrinsic signaling and survival effects, but these are little studied.

In our studies of αPD-L1 immunotherapy, we used RNAi technology to engineer low PD-L1–expressing tumors cells (termed PD-L1^lo) from parental PD-L1–expressing cell lines and made the unexpected observation that PD-L1^lo tumor cells proliferated more slowly than control cells in vitro, prompting us to study tumor-intrinsic PD-L1 signaling effects in more detail. We found that tumor PD-L1 had profound effects on expression of many genes that play important roles in tumor signaling and metabolism, including mTOR and autophagy pathways. Autophagy is a catabolic mechanism utilized by tumor cells to modulate cellular stress and metabolism and is targeted to treat some cancers (12). We demonstrated that αPD-L1 antibody used as immunotherapy had unexpected, important tumor-specific, immune-independent effects on tumor growth and metastatic spread and that tumor PD-L1 also affected sensitivity to pharmacologic autophagy and mTOR inhibitors and cytokines. These data shift the current paradigm regarding the role of PD-L1 effects on tumor immunopathogenesis and response to immune checkpoint blockade inhibitory antibodies to include tumor-intrinsic signaling consequences, indicate approaches to treatment on the basis of tumor PD-L1 status in novel ways, and suggest that PD-L1 could be a biomarker for treatments other than αPD-L1 or αPD-1.

PD-1, PD-L1, and B7-H3 are immunoglobulin superfamily members (13), and all regulate tumor mTOR (and our new data here; refs. 14, 15). Thus, this superfamily appears to have common tumor cell–intrinsic effects on important cell signaling and functional outcomes that require much additional study. Our work demonstrates additional tumor-intrinsic effects in this superfamily, with a focus on PD-L1.

Wild-type (WT) C57BL/6J (BL6), βδ TCR knockout (KO), and NOD.Cg-Prkdc^scidIl2rg^tm1Wj1/SzJ [nonobese diabetic/severe combined immunodeficiency (NOD/SCID)/IL2Rγ KO, NSG] mice were purchased from Jackson Laboratory. PD-L1 KO mice were a kind gift from Lieping Chen (16). All mice were syngeneic BL6 and maintained under specific pathogen-free conditions and given food and water ad libitum. Age- and sex-matched mice that were at least 8 weeks of age were used for all experiments. Only females were used for ID8agg ovarian cancer studies. All animal studies were approved by The University of Texas Health Science Center at San Antonio Institutional Animal Care and Use Committee.

Tumor challenge into all mice except NSG was injection of 5 × 10⁵ B16 cells subcutaneously as described (16) or 4 × 10⁶ ID8agg cells intraperitoneally. For NSG mice, 1.5 × 10⁵ B16 or 1 × 10⁶ ID8agg cell challenges were used. B16 growth was measured every other day with Vernier calipers and volume calculated as (length × width²)/2. Survival was determined as tumor >1,500 mm³ or distress (16). ID8agg tumor burden was determined by in vivo luciferase imaging, and survival was assessed by ascites formation, weight gain, or distress (17).

αPD-L1 (10F.9G2) and αPD-1 (RMP1-14) antibodies or respective isotype controls were injected intraperitoneally. In NSG mouse challenges, we gave 200 μg/mouse every other day starting 1 day before tumor challenge. In WT challenge, we gave 100 μg/mouse αPD-L1 or isotype control every 5 days for B16 or ID8agg cell starting on day 7 after tumor challenge for times indicated.

For pharmacologic autophagy inhibition in vivo, mice were injected intraperitoneally with 60 mg/kg chloroquine every other day (18, 19) or 25 mg/kg 3-methyladenine every 5 days (20) versus PBS control starting on day 7 following tumor challenge, regimens shown to inhibit tumor cell autophagy effectively in vivo.

In vivo luciferase detection was performed on an IVIS Lumina (Perkin-Elmer) with isoflurane-anesthetized mice 15 minutes after intraperitoneal injection with 200 μL luciferin potassium (15 mg/mL, Gold Biotechnology) with a 30-second exposure. An identical region of interest was drawn over each abdomen, and tumor burden was quantified as average radiance (photons/second/cm²/sr).

Mouse B16-F10 melanoma (herein “B16” for simplicity) and human ES2 ovarian cancer cells were purchased from the ATCC; ID8 were a kind gift from George Coukos (University of Pennsylvania, Philadelphia, PA). Cells were not revalidated for this work. We generated an aggressive ID8 line, ID8agg, by serial passage through WT hosts (Supplementary Fig. S1). Mouse cells were used in passages less than 5 and ES2 was passaged for less than 6 months. All cells were maintained in 5% FBS-containing RPMI-1640 medium supplemented with 1% penicillin/streptomycin, 1% l-glutamate, and 1% HEPES buffer. For serum starvation experiments, 0.1% FBS replaced 5.0%.

Stable PD-L1 knockdown in B16 and ID8agg lines was generated using lentivirus transduction particles containing validated PD-L1 shRNA (Sigma, TRCN0000068001) against murine Pdcd1Ig1 in pLKO.1-puro vector using puromycin selection per manufacturer's protocol (Supplementary Fig. S2C). Scrambled PD-L1 shRNA control sequence (Sigma, SHC216V) was used to generate control lines. Individual PD-L1^lo clones were selected in 2 μg/mL puromycin. For ES2, PD-L1 shRNA (Sigma, TRCN0000056914) against human Pdcd1Ig1 and controls were used as described above. Individual PD-L1^lo clones were selected in 4 μg/mL puromycin. The PD-L1–overexpressing (PD-L1^hi) B16 polyclonal line was generated by transfection of pCMV6-PD-L1-GFP plasmid (OriGene, MG203953) using Turbofect (ThermoFisher) per the manufacturer's protocol (Supplementary Fig. S2D). Pooled PD-L1^hi clones stably expressing PD-L1-GFP were obtained using G418 selection. Luciferase expression was engineered into ID8agg cells by transfection with pGL4.51 [luc2/CMV/neo] plasmid DNA (Promega) and Attractene transfection reagent (Promega) according to manufacturer's instructions, and individual clones were selected with G418.

Cells were stained and sorted as previously described (21), using LSR II and FACSAria hardware and analyzed by FACSDiva (BD Bioscience) and FlowJo software (FlowJo, LLC). Anti-mouse PD-L1 (10F.9G2), anti-mouse PD-1 (29F.1A12), and matched isotype control antibodies were purchased from BioLegend. Cells were treated with recombinant mouse IFNγ (0.1 ng/mL, R&D Systems) for 48 hours as noted to induce PD-L1.

Cells (6 × 10³) were plated in 96-well plastic culture plates in medium and treated 12 hours later with chloroquine (50 μmol/L), 3-methyladenine (5 mmol/L), rapamycin (5 nmol/L; Sigma for all), TNFα (10 ng/mL; R&D Systems), cisplatin (Sigma; 1–10 μmol/L), or paclitaxel (Sigma; 1–30 nmol/L). αPD-L1 (10F.9G2) or αPD-1 (RMP1-14) antibodies or respective isotype controls were used at 50 μg/mL for the entire culture period. Data shown are from optimized drug concentrations in preliminary work not shown. Cell proliferation rates were determined using MTT (5 mg/mL), which measures metabolic activity, 72 hours after treatment. Absorbance was measured at 540 nmol/L using a BioTek Synergy 2 Multi-Mode Plate Reader. Data are presented as mean absorbance versus medium control ± SEM. Proliferation was assessed in triplicate and compiled from at least 3 separate experiments. Cell viability was determined using a Vi-Cell XR (Beckman Coulter).

Total RNA was isolated from control ID8agg and ID8agg PD-L1^lo (clone 3) cells using RNeasy (Qiagen), and RNA quality was ensured on an Agilent Bioanalyzer. Fifty basepair single read sequencing was performed using an Illumina HiSeq 2000 system at The UTHSCSA Genome Sequencing Facility. Data are presented as gene sequence fold change in PD-L1^lo versus control. RNA-seq data were also analyzed by DAVID Bioinformatics and corrected for multiple comparisons. Genomic data will be deposited in the Gene Expression Omnibus database.

Cell lysates were prepared in 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 1 mmol/L phenylmethylsulphonyl fluoride and Halt protease/phosphatase inhibitor cocktail (Thermo Scientific). Protein concentration was measured by Bradford method (Thermo Scientific). Fifty micrograms of protein was separated by 4% to 15% SDS-PAGE (BioRad Hercules), transferred to polyvinylidene fluoride (PVDF) membranes (GE Water and Process Technologies), blocked in TBS (pH 7.4) plus 0.1% Tween-20 and 5% skim milk, and incubated overnight at 4°C with 1:1,000 diluted phospho- and/or total antibodies against indicated proteins (Cell Signaling) plus anti-mouse β-actin (Santa Cruz Biotechnology). Membranes were incubated with horseradish peroxide–conjugated antibodies for 1 hour. Proteins were detected by enhanced chemiluminescence (Pierce). Band quantification and normalization to total protein were by ImageJ software (22). Data show means of 3 individual blots with comparisons only made between like blots from the same gels.

Cells were plated at low confluence in 6-well plates (50,000 cells/well). On day 2, cells were exposed to serum starvation (0% FBS), normal medium (10% FBS), or chloroquine (50 μmol/L) for 24 hours. Medium was removed, cells were washed with PBS and treated with 4% paraformaldehyde/PBS for 20 minutes at room temperature, washed, and then permeabilized with 0.1% Triton X-100 for 10 minutes. Cells were then blocked with 5% normal goat serum (Cell Signaling Technology) containing 0.3% Triton X-100 in PBS for 60 minutes. Diluted primary antibody, anti-mouse LC3 A/B (Cell Signaling Technology), was applied in blocking buffer overnight at 4°C. Alexa Fluor-555 secondary antibody diluted in 1% normal goat serum in PBS were added for 1 hour at ambient temperature. Cells were fixed using Vectashield hard set mounting medium containing DAPI dye (Vector Laboratories). Images were acquired using confocal microscopy (Olympus FV-1000) and overlaid using ImageJ (22).

Total RNA was isolated from homogenized whole lung tissue using RNeasy (Qiagen). cDNA was synthesized with 1 μg of total RNA using the ImPromII Reverse Transcription System (Promega) and random primers. Quantitative PCR (qPCR) was conducted using the 7900HT Real-Time PCR System (Applied Biosystems), amplified with transcript-specific primers with SYBR Green (Thermo Scientific), according to manufacturer's instructions. Mouse primers were: Tyr, 5′-CTCTGGGCTTAGCAGTAGGC-3′ and 5′-GCAAGCTGTGGTAGTCGTCT-3′; gp100, 5′-ACATTTCATCACCAGCAGGGTGCC-3′ and 5′-AACAAGTGGGTGCTGGCC-3′; Trp-2, 5′-GTCCTCCACTCTTTTACAGACG-3′ and 5′-ATTCGGTTGTGACCAATGGGT; Trp-1, 5′ CCCCTAGCCTATATCTCCCTTTT-3′ and 5′-TACCATCGTGGGGATAATGGC-3′; and Gapdh, 5′-AACGACCCCTTCATTGAC-3′ and 5′-TCCACGACATACTCAGCAC-3′ as the internal control (23).

Statistical analyses were conducted with Prism software (GraphPad). Data in bar graphs are means ± SEM. For tumor growth, we used 2-way ANOVA plus Bonferroni post-tests to compare replicate means. Kaplan–Meier estimates and the log-rank test were used to analyze statistical differences in survival. For all other single measurement assays, we used an unpaired t test. P < 0.05 was considered significant.

The ID8 ovarian cancer cell line produces tumors that replicate important aspects of human disease, including local spread and ascites after intraperitoneal injection into syngeneic BL6 mice (24) but disease is not manifested for 10 to 15 weeks (16). We generated an aggressive ID8 line, ID8agg, by serial passage through WT hosts that produces rapid disease (Supplementary Fig. S1). Flow cytometry showed that ID8agg cells express PD-L1 that is upregulated by IFNγ (Fig. 1A) and immunoblot and confocal microscopy confirmed PD-L1 expression (Supplementary Fig. S2A). B16 melanoma cells also express basal PD-L1 that is further upregulated by IFNγ (Fig. 1B, Supplementary Fig. S2B).

We used plasmids stably expressing shRNA to reduce basal and IFNγ-inducible PD-L1 expression (PD-L1^lo) in ID8agg and B16 cells and a constitutively active vector to increase PD-L1 expression in B16 (PD-L1^hi, Fig. 1A and B, Supplementary Fig. S2). PD-L1^lo B16 cells proliferated significantly slower than control B16 cells and PD-L1^hi cells exhibited the highest rate of proliferation in vitro (Fig. 1C). Subcutaneous WT mouse challenge elicited a similar trend, with PD-L1^hi B16 cells producing the fastest tumor growth and PD-L1^lo cells the slowest (not shown). Tumor PD-L1 can inhibit antitumor immunity by impeding antitumor CD8⁺ T-cell function (4). Thus, to dissect tumor cell–intrinsic effects versus immune effects in vivo, we challenged severely immunodeficient NSG mice with distinct B16 cell constructs subcutaneously. Consistent with in vitro and WT challenge data, in the absence of tumor-specific immunity, PD-L1^hi B16 tumors grew fastest and PD-L1^lo tumors grew the slowest (Fig. 1D, left), which correlated with reduced overall survival (Fig. 1D, right). Furthermore, PD-L1^hi cells generated the greatest lung metastases following subcutaneous challenge, whereas PD-L1^lo cells produced the fewest as detected by PCR for tumor-specific gene products (Fig. 1E). In support, PD-L1^lo ID8agg cells grew more slowly in vitro (Fig. 1F) and in vivo compared with control ID8agg after intraperitoneal challenge in WT (not shown) and NSG mice (Fig. 1G). Together, these data clearly demonstrate that tumor-intrinsic PD-L1 controls tumor growth in vitro and in vivo in distinct tumors and anatomic compartments. PD-L1–mediated growth and promotion of bona fide metastatic tumor spread were all independent of antitumor immunity.

To determine whether tumor cell proliferation rate is influenced by PD-L1 or PD-1 surface expression, we first showed that B16 also expresses PD-1 as reported (15) and found that ID8agg also expresses PD-1 (Fig. 2A). Proliferation of control but not PD-L1^lo B16 cells in vitro was significantly reduced by both αPD-L1 and αPD-1 (Fig. 2B). CD80 was not detected (not shown). To assess whether αPD-L1 reduces tumor growth in mice lacking antitumor immunity, we challenged NSG mice with control B16 cells and treated with αPD-L1, which reduced control B16 (Fig. 2C) and PD-L1^hi B16 growth (Supplementary Fig. S3A), consistent with in vitro data. We confirmed that αPD-1 slows control B16 growth in NSG mice (not shown) as reported (15). To determine the role of tumor cell–intrinsic PD-L1 in metastatic propensity, primary lung metastases in NSG mice challenged subcutaneously with B16 were assessed as mRNA of melanoma-specific genes in lung homogenates as described (23). αPD-L1 or αPD-1 each reduced B16 lung metastases significantly and similarly (Fig. 2D, Supplementary Fig. S3B). Similar to B16, αPD-L1 and αPD-1 each reduced rates of control ID8agg cell proliferation in vitro but without significant effect on PD-L1^lo ID8agg (Fig. 2E). In contrast, αPD-L1 had negligible treatment effects on peritoneal ID8agg cell challenge into WT mice (Supplementary Fig. S4) and did not slow tumor growth or improve survival in NSG challenge (Fig. 2F and G). Thus, whereas certain PD-L1 signaling outcomes are similar in distinct tumors (and see data to follow), αPD-L1 effects in vivo can be dissociated from those in vitro, as well as from PD-L1 effects in distinct tumors as also seen in lack of αPD-L1 influence on PD-L1–expressing sarcoma growth in NSG mice (11).

It is possible that αPD-L1 acts on host PD-L1 to mediate tumor growth inhibition. We first showed that αPD-L1 treated control B16, but not PD-L1^lo B16 in WT mice (Supplementary Fig. S5A and S5B). To assess host PD-L1 effects specifically, we showed that αPD-L1 treated parental B16 challenge into PD-L1 KO mice as effectively as in WT mice, whereas αPD-L1 had no effect on PD-L1^lo B16 challenge in PD-L1 KO mice (Supplementary Fig. S5C and S5D). Therefore, tumor PD-L1 expression is necessary and sufficient for αPD-L1 effects on B16 growth in vivo, and host PD-L1 is not required, consistent with antiproliferative effects in vitro (Fig. 2B).

Data thus far clearly indicate a major cell-intrinsic, immune-independent role for tumor PD-L1 in both ovarian cancer and melanoma cells. We next investigated PD-L1–dependent signaling pathways. RNA-seq of control versus PD-L1^lo ID8agg cells identified 1,269 differentially expressed basal genes using a 2-fold change cutoff (Fig. 3A and data not shown), including significant gene expression differences in canonical and noncanonical autophagy pathways (Table 1A).

We assessed PD-L1 effects on autophagic flux as conversion of LC3-I to LC3-II in parental ID8agg or PD-L1^lo cells. Serum starvation for 24 hours augmented autophagy in control cells as expected. Autophagic flux was significantly higher in basal PD-L1^lo versus control ID8agg cells as assessed by LC3-II/LC3-I ratio (25), but serum starvation did not increase LC3-II further in PD-L1^lo cells (Fig. 3B), suggesting that tumor PD-L1 regulates basal and starvation-induced autophagy in ID8agg. In support of PD-L1–mediated autophagy suppression, LC3 foci formation was lower in control versus PD-L1^lo ID8agg (Fig. 3C). PD-L1 also blunted basal autophagic flux in B16, but in contrast to ID8agg, autophagy was significantly induced during serum starvation in PD-L1^lo B16 (Fig. 3D), the appropriate physiologic response. Consistent with Western blot data, there were more LC3 foci in PD-L1^lo B16 cells indicative of higher autophagosome formation versus control B16 under basal conditions and serum starvation (Fig. 3E), consistent with tumor cell PD-L1–mediated suppression of autophagy in B16.

Because PD-L1 depletion deregulates autophagy, including cell-specific effects (Fig. 3B–E), we tested effects of the pharmacologic autophagy inhibitor chloroquine (18, 19). In contrast to basal PD-L1-dependent proliferation effects, PD-L1^lo B16 cells were the most resistant to in vitro chloroquine-mediated proliferation inhibition, whereas PD-L1^hi cells were the most sensitive (Fig. 4A). Consistent with B16 cell data, PD-L1^lo ID8agg cells were significantly more resistant to chloroquine-mediated proliferation inhibition versus control ID8agg cells (Fig. 4A). Using the autophagy inhibitor 3-methyladenine (20), similar, but less pronounced, proliferation effects were seen in both B16 and ID8agg cells in vitro (data not shown).

Consistent with in vitro data, chloroquine and 3-methyladenine were significantly more effective in reducing control versus PD-L1^lo B16 challenge into WT mice (Fig. 4B–D). Because autophagy inhibitors could also affect antitumor immunity, we assessed effects in T-cell–deficient βδ TCR KO mice, where chloroquine retained its clinical activity against control B16 (Fig. 4E) and remained ineffective against PD-L1^lo (Fig. 4F). Similarly, in NSG mice, chloroquine was significantly effective against challenge with control B16 cells (Fig. 4G) but was ineffective in reducing tumor growth in mice challenged with PD-L1^lo B16 cells (Fig. 4H). Together, these data support the concept that tumor PD-L1 sensitizes B16 melanoma to autophagy inhibitors independent of T-cell immunity and that tumor PD-L1 expression predicts autophagy-dependent growth. Consistent with PD-L1-mediated dependence on autophagy that modulates cellular stress, serum starvation, which inhibits mTOR signaling and constrains cell dependence on autophagy, significantly reduced PD-L1^hi and control B16 cell viability in vitro under serum starvation conditions (with greatest effect in PD-L1^hi cells). We found reduced but significant effects in parental B16 and negligible effects in PD-L1^lo cells (Fig. 4I), suggesting tumor PD-L1 hinders autophagy that normally compensates effects of serum starvation or other cytotoxic conditions. In contrast, neither chloroquine nor 3-methyladenine improved survival against control or PD-L1^lo ID8agg challenge in vivo (Fig. 4J). These data further establish common and cell-specific PD-L1 signaling effects.

mTOR is a serine/threonine kinase regulating cellular growth and metabolism that is elevated in many cancers (26, 27) and recently shown to be affected by tumor-intrinsic PD-1 and PD-L1 (11, 15). As mTOR regulates important cellular processes including autophagy through distinct mTORC1 and mTORC2 complexes with distinct functions (28), we assessed additional mTOR signaling details. ID8agg RNA-seq identified potential mTOR signaling effects of PD-L1 (Table 1B). For example, the mTORC1 activator Rheb1 was significantly lower in PD-L1^lo cells (P = 5.35 × 10⁻⁷), consistent with reduced mTORC1 signaling. Prps6, a translation regulator controlled by mTORC1, was also significantly reduced (P = 7.97 × 10⁻⁷). mTORC2 activity is defined by target phosphorylation and cannot be assessed this way.

We used immunoblots to validate that tumor PD-L1 promotes basal mTORC1 signaling as assessed by P70S6K^T389 phosphorylation (Fig. 5A and B). PD-L1^lo ID8agg cells paradoxically increased mTORC1 during serum starvation whereas PD-L1^lo B16 cells did not (Fig. 5A and B). Rapamycin effectively suppressed mTORC1 in control B16 and ID8agg cells as expected and equally suppressed mTORC1 in PD-L1^lo B16 and ID8agg cells (Fig. 5D, Supplementary Fig. S6). PD-L1^lo B16 and ID8agg cells were more sensitive to rapamycin-mediated suppression of proliferation in vitro versus control cells, by MTT assay (Fig. 5C), suggesting that PD-L1 regulation of metabolic activity is mTORC1-dependent, although mTORC2 effects cannot be excluded (29, 30). Under basal conditions, PD-L1 inhibited phosphorylation of the mTORC2 substrate Akt^S473 in B16 and ID8agg (Fig. 5A and B), consistent with reduced mTORC2 activity (31–33). Rapamycin differentially affected Akt^S473 phosphorylation in B16 versus ID8agg in a PD-L1–dependent manner (Fig. 5D). mTORC1 and phosphorylated P70S6K^T389 can mediate a negative feedback loop on PI3K/Akt (34, 35), which is derepressed by rapamycin. Hence, rapamycin effects on Akt^S473 phosphorylation could be mTORC2-independent.

We next treated cells with chloroquine to inhibit autophagy and found that it inhibited mTORC1 as expected (36, 37) in control ID8agg but paradoxically increased mTORC1 in control B16 cells. Chloroquine had no effect on mTORC1 in PD-L1^lo ID8agg cells but significantly increased mTORC1 in PD-L1^lo B16 cells (Supplementary Fig. S6). Nonetheless, no observed mTOR effects readily explain how PD-L1 sensitizes cells to chloroquine-mediated cell proliferation inhibition. Thus, PD-L1–mediated mTOR and autophagy effects could be independent. Other effects of chloroquine or rapamycin or their combination varied between the 2 tumors on the basis of basal versus serum starvation conditions and PD-L1 status (Fig. 5D, Supplementary Fig. S6 and much additional data not shown), likely reflecting distinct tumor mutational landscapes in ovarian cancer (38) versus melanoma (39) and further supporting common versus cell-specific PD-L1 signaling effects. As mTOR and autophagy regulate responses to cytotoxic insults including chemotherapy and cytokines (40, 41), we explored additional effects. Tumor PD-L1 rendered B16 cells resistant to TNFα in vitro, whereas ID8agg cells were intrinsically more resistant and PD-L1 had no effect on TNF-α–mediated cytotoxicity (Supplementary Fig. S7A). Surprisingly, tumor PD-L1 did not significantly or consistently alter sensitivity of B16 or ID8agg to the cytotoxic agents cisplatin or paclitaxel (Supplementary Fig. S7B and S7C). Thus, tumor PD-L1 is a prosurvival agent but effects depend on the specific toxic insult.

To test human relevance, we first showed that the human ES2 ovarian cancer cell line expressed basal PD-L1 and then engineered PD-L1^lo ES2 cells (Fig. 6A and B). Some ES2 cells also express basal PD-1 (Fig. 6B). Consistent with B16 and ID8agg data, PD-L1^lo ES2 cells proliferated significantly slower than control ES2 (Fig. 6C). LC3-II versus LC3-I was increased in PD-L1^lo but not in control cells (Fig. 6D), consistent with increased autophagic flux when PD-L1 was reduced. In support, LC3 foci were reduced in control versus PD-L1^lo cells indicative of PD-L1–mediated decrease in autophagosome formation (Fig. 6E). Altogether these data are consistent with PD-L1–dependent suppression of autophagic flux in human cancer cells, similar to mouse cell data. Further consistent with B16 and ID8agg data, PD-L1 in ES2 cells augmented mTORC1 signals (P70S6K^T389 phosphorylation). In contrast to B16 and ID8agg data, PD-L1 knockdown eliminated ES2 mTORC2 activation (Akt^S473 phosphorylation; Fig. 6F).

Validating our observations in B16 and ID8agg, control ES2 cells were significantly more sensitive to proliferation inhibition by chloroquine compared with PD-L1^lo ES2 cells (Fig. 6G), further suggesting that PD-L1 regulation of cancer cell autophagy and autophagy dependence could be a common mechanism in PD-L1–expressing mouse and human cancer cells.

Tumor-expressed PD-L1 alters tumor immunopathogenesis by delivering negative signals to PD-1–expressing antitumor T cells (3–7), but emerging evidence shows that PD-L1 and PD-1 also have tumor-intrinsic functions (11, 15). Our work now clearly establishes that tumor PD-L1 has additional, important tumor intrinsic effects. We found that tumor PD-L1 promoted cell-intrinsic growth in 2 distinct tumor types (melanoma and ovarian cancer), in distinct anatomic compartments (skin and peritoneum) and in an immune-independent fashion for both. Thus, these properties are not specific to one single cell type or anatomic location.

αPD-L1 can alter PD-L1–mediated cell-intrinsic growth signals, as α-PD-L1 slowed B16 tumor growth in NSG mice incapable of mediating tumor-specific immunity or antibody-dependent cellular cytotoxicity. Melanoma PD-L1 promoted immune-independent bona fide metastases from subcutaneous melanoma, a more stringent model than intravenous cell challenge. Increased tumor growth and proliferation likely contribute to increased metastases, but additional factors could also contribute. Further work is needed to understand applicability to additional cancers and define mechanisms.

αPD-1 and αPD-L1 comparably inhibited B16 growth in vitro and comparably reduced B16 metastases in vivo. These data are consistent with tumor PD-L1/tumor PD-1 cooperation, as only about 5% to 10% of cells expressed PD-1. In vivo, αPD-L1 effects appears to be directed to tumor and not other PD-L1–expressing cells, as αPD-L1 (i) had no significant effect on PD-L1^lo B16 melanoma growth in WT mice and (ii) was fully protective in PD-L1 KO mice challenged with parental B16 cells. However, this treatment effect could be tumor- or compartment-specific, as αPD-L1 did not slow tumor growth or improve survival in intraperitoneal ID8agg ovarian cancer challenged NSG mice, despite reducing in vitro proliferation. In support of tumor-specific effects, a recent report showed that αPD-L1 did not slow PD-L1⁺ sarcoma growth in NSG mice (11). Specific tumors might be refractory to PD-L1–driven growth or survival signals owing to tumor-specific mutations (38, 39), degree of PD-L1 expression, antibody affinity and delivery differences, or other factors.

Tumor PD-L1 suppressed tumor autophagy in melanoma and ovarian cancer, suggesting that autophagy disruption could be a relatively general tumor PD-L1 effect. Tumor PD-L1 regulates glucose metabolism in sarcomas (11), suggesting that PD-L1 could have a variety of important cell-intrinsic metabolic effects. Furthermore, B7-H3 regulates glucose metabolism in breast cancer cells (42), suggesting that the B7-H superfamily, to which PD-L1 and B7-H3 belong (43) or the immunoglobulin superfamily to which these and PD-1 all belong (13), could have important tumor metabolic effects, which is an area meriting additional investigations. Our RNA-seq data demonstrated numerous potentially PD-L1–regulated genes. Ingenuity Pathway Analysis determined that many critical pathways were affected, including immune cell trafficking (chemokines and cytokines), inflammation, TGFβ signaling, metabolism, and cancer stem cell genes among others (our unpublished data), in addition to effects shown here. Thus, much additional, important PD-L1–mediated tumor cell signaling likely awaits definition.

We found that PD-L1–replete B16 melanoma and ID8agg ovarian cancer cells exhibit low basal autophagy, with high mTORC1 activity, and were disproportionately sensitive to autophagy inhibitor–mediated growth reduction compared with cancer cells with low PD-L1 expression. We postulate that further autophagy reduction in cancer cells with elevated mTORC1 activity and low autophagic activity (directly related to PD-L1 expression) is catastrophic. In support, autophagy levels can be fine-tuned to allow cells to meet metabolic demands (44) without compromising other functions. In vivo, this PD-L1–mediated balance augments clinical response to autophagy inhibitors. These data suggest that autophagy inhibitors could boost αPD-L1 treatment of PD-L1–positive cancers, yet other tumor- and/or immune-mediated effects may contribute. Despite high levels of autophagic flux, PD-L1^lo tumor cells might have a reduced autophagy requirement due to lower metabolic demand and/or cellular stress, as supported by reduced sensitivity to autophagy inhibitors. Alternatively, pharmacologic inhibitors could incompletely inhibit this high autophagy for clinical effect. Genetic approaches to autophagy manipulation are required to explore mechanisms further, but our data support the thesis that tumor PD-L1 expression and basal autophagy levels could be a biomarker for effective clinical autophagy inhibitor use. Our data also suggest that autophagy inhibition plus αPD-L1 merits clinical investigation.

As autophagy is a tumor survival mechanism, we expected that tumor PD-L1 could affect cell viability after specific insults and found differences in survival in serum starvation, mTOR, and autophagy inhibitors and TNF-α. Surprisingly, we did not detect a robust difference in sensitivity to cis-platinum or paclitaxel in vitro, despite potential indications for a role of tumor-intrinsic PD-L1 in chemoresistance. An abstract suggested that tumor PD-L1 protects from cytotoxic chemotherapy (10). That work used PD-L1 overexpression and did not specify cell lines or cytotoxic agents, making us unable to comment on potential differences in findings.

PD-L1 regulates tumor mTOR signals in melanoma and sarcoma (11, 15), but mechanistic details are lacking and unreported in ovarian cancer. We show that low PD-L1 in melanoma and ovarian cancer cells results in reduced phosphorylation of the mTORC1 substrate P70S6K^T389. PD-L1^lo cells proliferated more slowly than control lines, suggesting that lower mTORC1 activity in PD-L1^lo cells compromises their growth. Proliferation inhibition by the (predominantly) mTORC1 inhibitor rapamycin abolished P70S6K^T389 phosphorylation in control and PD-L1^lo B16 and ID8agg cells in vitro as expected, yet PD-L1^lo cells were more rapamycin-sensitive than control cells, suggesting that proliferation of PD-L1^lo tumors is more dependent on basal mTORC1 activity compared with PD-L1–replete tumor cells. PD-L1–mediated mTORC1 effects could differ in tumor cell metabolism, survival, and therapeutic outcomes. Nonetheless, these data support the concept that tumor PD-L1–dependent mTORC1 activity drives proliferation. Thus, PD-L1 expression could predict mTORC1 activity and rapamycin-sensitive tumor growth, which is clinically exploitable. For instance, these data suggest that αPD-L1 and/or αPD-1 immunotherapies combined with mTOR inhibitors in treating PD-1/PD-L1–replete tumors or mTOR inhibitors alone in treating PD-1^lo/PD-L1^lo tumors are rational approaches.

Other mTORC1 substrates, mTORC2, or other rapamycin effects could also mediate specific PD-L1–dependent effects. For example, mTORC2 and its substrate Akt drive tumor cell growth and survival (45). We found that phosphorylation of the mTORC2 substrate Akt^S473 was induced in PD-L1^lo B16 and ID8agg cells. In contrast to parental ID8agg cells, rapamycin did not further induce pAkt^S473 in PD-L1^lo cells, demonstrating a PD-L1–dependent effect, perhaps from low basal mTORC1 and P70S6K activities, in which case negative feedback on Akt could be inactive and thus unaffected by rapamycin. We speculate that PD-L1–driven mTORC1 activation accounts for reduced autophagy, but genetic and additional approaches are required for definitive mechanistic insights. Additional study of PD-L1–mediated mTOR signaling and outcomes of mTOR inhibition (e.g., metabolic activity), and cell-specific effects could help optimize mTOR inhibition strategies to treat selected cancers. As PD-L1, PD-1, and B7-H3 all regulate tumor mTOR signaling (11, 15, 42), this could be another general feature of selected immunoglobulin superfamily members and requires additional study.

Finally, we demonstrated human relevance by showing that cell-intrinsic PD-L1 in the human ES2 ovarian cancer cell line controls proliferation, mTORC1, autophagic flux, and sensitivity to autophagy inhibitors. PD-L1 in human melanoma cells controls PD-1 effects on melanoma growth and mTOR signals (15). Thus, human studies merit much additional attention.

In summary, we show that tumor cell–intrinsic PD-L1 signals mediate effects common to melanoma and ovarian cancer cells in mouse lines and effects extend to human ovarian cancer cells. PD-L1 promotes tumor cell proliferation and immune-independent growth in melanoma and ovarian cancer cells and metastatic melanoma spread in vivo. Tumor cell–intrinsic PD-L1 altered autophagy inhibitor and mTOR inhibitor efficacy. PD-L1 promoted basal mTORC1 activation in all cells tested and inhibited phosphorylation of the mTORC2 substrate Akt in mouse lines but not the human line tested. αPD-L1 inhibited in vivo melanoma growth but not ovarian cancer growth in an immune-independent fashion and melanoma but not ovarian cancer cells were slowed by autophagy inhibitors in vivo. Thus, tumor PD-L1 could be a biomarker for response to mTOR or autophagy inhibitors in selected cancers. Tumor PD-L1 appeared to cooperate with tumor PD-1 for selected effects. As our understanding of tumor cell–intrinsic PD-L1 effects increases, our ability to predict treatment responses to various agents and combine them in rationale ways for more effective clinical use will improve. Given the extraordinary effect that PD-L1 appears to have on diverse, critical cellular processes, much additional work in this area is merited.

No potential conflicts of interest were disclosed.

Conception and design: C.A. Clark, T.J. Curiel

Development of methodology: C.A. Clark, S. Pandeswara, B. Yuan, J.M. Drerup, R.K. Vadlamudi, T.J. Curiel

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.A. Clark, H. Gupta, G.R. Sareddy, S. Lao, J.M. Drerup, A. Padron, Y. Liu, V. Hurez

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.A. Clark, H. Gupta, G.R. Sareddy, B. Yuan, J.M. Drerup, K. Murthy, K. Thedieck, R. Li, R.K. Vadlamudi, T.J. Curiel

Writing, review, and/or revision of the manuscript: C.A. Clark, H. Gupta, J.M. Drerup, J.R. Conejo-Garcia, M.J. Turk, K. Thedieck, V. Hurez, R. Li, T.J. Curiel

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.A. Clark, S. Pandeswara, A. Padron, K. Murthy, T.J. Curiel

Study supervision: K. Murthy, T.J. Curiel

Work was supported by grants from the NIH, European governments and private foundations. T.J. Curiel received grants CA170491, CA54174, CA054174, CDMRP, The Holly Beach Public Library, and The Owens Foundation and the Skinner endowment. K. Thedieck received grants from Rosalind Franklin Fellowship, BMBF e:Med Young investigator network GlioPATH (01ZX1402B), BMBF e:Med Demonstrator project MAPTor-NET (31P9013A), Ubbo Emmius Funds, Stichting Michelle). M.J. Turk received grant CA120777. C.A. Clark received grants CA206398 and UTHSCSA MD/PhD Program support.

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.