Mechanisms regulating PD-L1 expression on tumor and immune cells
The PD-1/PD-L1 checkpoint is a central mediator of immunosuppression in the tumor immune microenvironment (TME) and is primarily associated with IFN-g signaling. To characterize other factors regulating PD-L1 expression on tumor and/or immune cells, we investigated TME-resident cytokines and the role of transcription factors in constitutive and cytokine-induced PD-L1 expression.
Thirty-four cultured human tumor lines [18 melanomas (MEL), 12 renal cell carcinomas (RCC), 3 squamous cell carcinomas of the head and neck (SCCHN), and 1 non-small-cell lung carcinoma (NSCLC)] and peripheral blood monocytes (Monos) were treated with cytokines that we detected in the PD-L1+ TME by gene expression profiling, including IFN-g, IL-1a, IL-10, IL-27 and IL-32g. PD-L1 cell surface protein expression was detected by flow cytometry, and mRNA by quantitative real-time PCR. Total and phosphorylated STAT1, STAT3, and p65 proteins were detected by Western blotting, and the genes encoding these proteins were knocked down with siRNAs. Additionally, the proximal promoter region of PDL1 (CD274) was sequenced in 33 cultured tumors.
PD-L1 was constitutively expressed on 1/17 cultured MELs, 8/11 RCCs, 3/3 SCCHNs, and on Monos. Brief IFN-g exposure rapidly induced PD-L1 on all tumor cell lines and Monos regardless of constitutive PD-L1 expression. PD-L1 mRNA levels were associated with protein expression, which was diminished by exposure to transcriptional inhibitors. siRNA knockdown of STAT1 but not STAT3 reduced IFN-g- and IL-27-induced PD-L1 protein expression on tumor cells. In contrast, STAT3 knockdown in Monos reduced IL-10-induced PD-L1 protein expression, and p65 knockdown in tumor cells reduced IL-1a-induced PD-L1 expression. Notably, constitutive PD-L1 expression was not affected by knocking down STAT1, STAT3, or p65. Differential effects of IFN-g, IL-1a, and IL-27 on individual tumor cell lines were not due to PDL1 promoter polymorphisms.
Multiple cytokines found in an immune-reactive TME may induce PD-L1 expression on tumor and/or immune cells through distinct signaling mechanisms. Factors driving constitutive PD-L1 expression were not identified in this study. Understanding complex mechanisms underlying PD-L1 display in the TME may allow treatment approaches mitigating expression of this immunosuppressive ligand, to enhance the impact of PD-1 blockade.
KeywordsPD-L1 Cytokines Interferon gamma Interleukins Tumor microenvironment Transcription factors Cancer immunotherapy
classical Hodgkin lymphoma
non-small-cell lung carcinoma
programmed death-ligand 1
quantitative reverse transcriptase polymerase chain reaction
renal cell carcinoma
squamous cell carcinoma of head and neck
Signal transducer and activator of transcription
Programmed death ligand 1 (PD-L1, CD274) expressed on tumor and/or immune cells in the tumor microenvironment (TME) interacts with PD-1 on tumor infiltrating lymphocytes, attenuating effector T cell responses and allowing tumors to escape immune attack [1, 2]. Understanding how TME-resident cytokines and signaling pathways regulate PD-L1 expression may provide therapeutic opportunities to mitigate PD-L1-induced intratumoral immunosuppression .
There are two general mechanisms by which tumor cells can express PD-L1, protecting them from immune elimination: “innate immune resistance” and “adaptive immune resistance” . Innate resistance refers to constitutive PD-L1 expression on tumor cells, resulting from PDL1 gene amplification or aberrant activation of oncogenic signaling pathways. Activation of ALK/STAT3 in T cell lymphoma , AP-1/JAK/STAT in classical Hodgkin lymphoma (cHL) , the microRNA-200/ZEB1 axis in non-small-cell lung cancer (NSCLC) , c-jun/STAT3 in BRAF inhibitor-resistant melanoma , and PI3K in glioma  have each been reported to upregulate PD-L1 expression on tumor cells. Additionally, Myc has been shown to regulate constitutive PD-L1 expression at the mRNA level in multiple tumors, such as T cell acute lymphoblastic leukemia, melanoma and NSCLC . Recently, post-transcriptional regulation of PD-L1 has also attracted attention, with reports that cyclin-dependent kinase-4 (CDK4) and glycogen synthase kinase 3 beta (GSK3B) can promote PD-L1 protein degradation in cultured tumors [11, 12].
In contrast to innate resistance, adaptive immune resistance refers to PD-L1 expression on tumor or immune cells in response to inflammatory factors secreted in the TME during antitumor immune responses. While IFN-g is generally thought to be the primary T cell derived cytokine responsible for adaptive PD-L1 expression, we have described several additional TME-resident cytokines that can upregulate PD-L1 expression on cultured human monocytes (Monos) and/or tumor cells, including IL-1a, IL-10, IL-27 and IL-32 g [13, 14, 15]. Transcripts for IFN-g, IL-10 and IL-32 g were over-expressed in PD-L1+ compared to PD-L1(−) melanoma biopsies; in vitro, IL-10 and IL-32 g induced PD-L1 expression on Monos but not on melanoma cells . IL-1a was upregulated in Epstein-Barr virus (EBV) negative PD-L1+ cHL, and IL-27 was upregulated in EBV+ PD-L1+ cHL. When combined with IFN-g, IL-1a and IL-10 further increased PD-L1 protein expression on human Monos in vitro, compared to the effects of IFN-g alone. IL-27 increased PD-L1 expression on Monos as well as dendritic cells, T cells, and some tumor cell lines [14, 16] . Others have reported that the transcription factors JAK/STAT1 , IRF-1  and NF-kB , involved in inflammatory cytokine production, can contribute to IFN-g-induced PD-L1 expression on hematopoietic tumors, lung cancer, and melanoma, respectively. In a murine medulloblastoma model, the cyclin-dependent kinase CDK5 appeared to regulate IFN-g-induced PD-L1 expression . Overall, existing evidence suggests that PD-L1 may be differentially regulated with respect to specific signaling pathways and transcription factors in different cell types, although IFN-g appears to be a dominant cytokine driving expression of this immunosuppressive ligand.
We undertook the current study to broadly examine mechanisms underlying constitutive and cytokine-induced PD-L1 expression in four human tumor types – melanoma (MEL), renal cell carcinoma (RCC), squamous cell carcinoma of the head and neck (SCCHN), and NSCLC – and to investigate the potential roles of STAT1, STAT3, and p65 activation in driving constitutive and inducible PD-L1 expression on tumor cells and Monos.
Cell culture and flow cytometry
Established cultures of human MELs, RCCs, SCCHNs, and NSCLC (Additional file 5: Table S1) were maintained in RPMI 1640 medium or DMEM with 10% heat-inactivated fetal calf serum. Human Monos were enriched by negative selection from cryopreserved peripheral blood mononuclear cells with the Pan Monocyte Isolation Kit (Miltenyi Biotec, San Diego, CA). Cells were cultured in the presence of recombinant IFN-g (100 or 250 IU/ml; Biogen, Cambridge, MA), IL-1a (10 ng/ml), IL-6 (20 ng/ml), IL-10 (100 ng/ml), IL-27 (50 ng/ml) or IL-32 g (100 ng/ml; all R&D Systems, Minneapolis, MN) for the indicated time periods (Additional file 6: Table S2). In some experiments, actinomycin D (ActD, 10 μg/ml) or cycloheximide (CHX, 2 μg/ml; both Thermo Fisher Scientific, Waltham, MA) was added to cultures 1 h before IFN-g treatment. Adherent cells were harvested with trypsin. To assess cytokine effects on PD-L1 expression, cells were stained with anti-human PD-L1 (clone MIH4, ThermoFisher Scientific, Carlsbad, CA) or an isotype control. HLA-DR (clone L243, Becton Dickenson, San Jose, CA) staining was performed simultaneously to provide a control for the effects of IFN-g. PD-L2 was stained with clone MIH18 (Thermo Fisher Scientific). Data were acquired on the BD FACSCalibur and analyzed with FlowJo Software (TreeStar, Ashland, OR). Expression level of a molecule was calculated as delta mean fluorescence intensity (∆MFI), which is MFI of specific staining – MFI of isotype control staining. Cytokine-induced expression of a molecule was calculated as ∆∆MFI, which is ∆MFI with cytokine exposure – ∆MFI without cytokine exposure.
Real time quantitative reverse transcriptase PCR (qRT-PCR)
mRNA was extracted from cells 6–16 h after cytokine treatment with the RNeasy Mini Kit (QIAGEN, Germantown, MD). Total mRNA from each sample was reverse-transcribed with the qScript™ cDNA SuperMix (Quanta Bioscience, Beverly, MA). Real-time PCR was performed in triplicate for each sample using commercial primers and probes for CD274, HLA-DRA, and housekeeping genes (Thermo Fisher Scientific). Forty cycles of PCR were conducted using a QuantStudio 12 K Flex Real-Time PCR System. Results were analyzed using the manufacturer’s software (Applied Biosystems). Fold change of mRNA expression before and after cytokine treatment was calculated as 2^(ΔCt before – ΔCt after), in which ΔCt = Ct specific probe – Ct internal control.
Lysates of whole cells or nuclear proteins were prepared with M-Per and NE-Per (Thermo Fisher Scientific) respectively, as described . Briefly, 20 μg protein per lane was separated by 4–12% Bis-Tris SDS-PAGE under reducing conditions and transferred to a polyvinylidene difluoride membrane, which was blocked with 5% dry non-fat milk. Membranes were stained with antibodies specific for signal transducer and activator of transcription (STAT)1 (polyclonal, catalog # 9172), phospho-STAT1 (clone 58D6), STAT3 (clone 124H6), phospho-STAT3 (pSTAT3; clone M9C6), p65 (clone D14E12), phospho-p65 (pp65; clone 93H1), c-jun (clone 60A8) and phospho-c-jun (pc-jun; clone D47G9) (all Cell Signaling Technology, Beverly, MA) at 4 °C overnight. Membranes were counterstained with anti-rabbit IgG-HRP (1:1000–1:12,000 dilution) or anti-mouse IgG-HRP (1:1000–1:5000) for 1 h at room temperature (GE Healthcare, UK or Kindle Bioscience, Greenwich, CT). Blots were also stained with anti-beta-actin-peroxidase (1:200,000 dilution; Sigma, St. Louis, MO, clone AC-15). Proteins were detected by ECL Western blotting detection reagents (GE Healthcare) or Hi/Lo Digital–ECL Western Blot Detection Kit (Kindle Bioscience) and the density of the target molecule was quantified with the ImageJ program (https://imagej.nih.gov/ij/) . Normalized density was calculated as the ratio of target molecule density to beta-actin density.
Short inhibitory RNA (siRNA) transfection
ON-TARGET plus SMART pool siRNAs for STAT1, STAT3, and p65 were purchased from Dharmacon (Lafayette, CO). siRNA transfection was done with the Nucleofector II or 4D-nucleofector device (Lonza, Basel, Switzerland) following the Amaxa Cell Line Nucleofector Kit, Human Monocyte Nucleofector Kit, or SF/SE Cell Line 4D Nucleofector X kit protocols. Briefly, 1 × 106–4 × 106 tumor cells or 1 × 107 Monos were suspended in 100 μl transfection solution supplemented with 100–300 pmol specific or scrambled siRNA. Electroporation was done with transfection programs recommended in the Lonza Knowledge Center (https://knowledge.lonza.com/) . Two days after transfection, cells were incubated with cytokines. Knockdown effects and transcription factor phosphorylation were detected 15 min later by Western blotting. Percentage of knockdown was calculated based on the actin-normalized density of the target molecule in Western blotting, by the formula (scrambled siRNA - specific siRNA)/scrambled siRNA × 100. The average targeted knockdown achieved in this study was 70%. PD-L1 and HLA-DR expression at the cell surface was detected and quantified 24 h later by flow cytometry, and the effects of knockdown with target-specific siRNAs were calculated with reference to scrambled siRNA.
PDL1 promoter region sequencing
Genomic DNA from cultured tumor cell lines or cryopreserved peripheral blood lymphocytes was extracted from 1 × 106 cells using the PureLink Genomic DNA kit (Thermo Fisher Scientific, K1820–00). Based on the public PDL1 (CD274) gene sequence (GenBank NC_000009.12), three primers (PDLP-F1, 5’GTTTCCAGGCATCACCAGATGCT; PDLP-F2, 5’TCCTCATGGGTTATGTGTAGTTTG; PDLP-R, 5’CCTCATCTTTCTGGAATGCCCTA) were designed to amplify 2.1 kb and 1.1 kb regions that are immediately upstream of the ATG translation start site. These two regions were amplified using an Expand TM High Fidelity PCR system (Sigma, catalog # 11732650001). Amplified PCR products were purified by a QIAquick PCR Purification kit (Qiagen, catalog # 28104) and sent to the Johns Hopkins University Core Facility for Sanger sequencing. Amplicons were sequenced using the following primers: PDLP-seq, 5’TGCTGAATTCAGTCCTTAATGG and PDLP-seqR, 5’CCATTAAGGACTGAATTCAGCA; PDLP-seq2, 5’CAGATACTCTGGAAGAGTGGCT and PDLP-seq2R, 5’AGCCACTCTTCCAGAGTATCTG.
IFN-g-induced PD-L1 protein expression on tumor cells is associated with de novo PD-L1 (CD274) mRNA transcription
To further explore this phenomenon, we incubated cultured MELs with ActD, a mRNA transcription inhibitor, or CHX, a protein synthesis inhibitor, prior to IFN-g exposure. Six h after IFN-g exposure, we found that each chemical completely blocked the emergence of PD-L1 protein on the cell surface. As expected, in the same cells, ActD suppressed IFNg-induced PDL1 mRNA transcription while CHX did not (Additional file 1: Figure S1). These data suggest that IFN-g drives new PD-L1 transcription and translation, and that translocation of preexisting intracellular PD-L1 protein stores is not a major mechanism underlying IFN-g-induced PD-L1 expression on the cell surface.
STAT1 but not STAT3 mediates IFN-g-induced PD-L1 protein expression on tumor cells
IL-1a and IL-27 induce PD-L1 expression on tumor cells, associated with new PD-L1 mRNA transcription
Similar to our findings with IFN-g, changes in PD-L1 protein expression induced by IL-1a or IL-27 corresponded with changes in PDL1 gene expression, in 2 of 2 RCC lines tested (Fig. 3e). This suggests that new mRNA transcription mediated by IL-1a or IL-27 exposure contributes to PD-L1 regulation. In contrast to the findings described above, the Th17 cytokines IL-17A and IL-23, which we previously detected in the microenvironment of some human cancers but which did not enhance PD-L1 protein expression on Monos , also failed to induce PD-L1 on tumor cells (not shown).
p65 and STAT1 respectively mediate IL-1a- and IL-27-induced PD-L1 expression on tumor cells
PDL1 gene promoter sequence variations do not correlate with quantities of PD-L1 protein induced on tumor cells by IFN-g, IL-1a or IL-27
To determine whether sequence variations in the promoter region of the PDL1 gene, where transcription factors would be expected to bind, are associated with different levels of tumor cell PD-L1 protein expression induced by cytokines, we sequenced a 650 bp or 2 Kb region upstream of the PDL1 transcription initiation codon in 33 tumor cell lines and 12 autologous normal tissues. Nine of 33 tumor cell lines harbored -482C and 3 of 33 harbored -382G, which have been reported as SNPs (https://www.ncbi.nlm.nih.gov/snp) . Neither gene alteration correlated with the level of PD-L1 protein expression induced by IFN-g, IL-1a or IL-27 exposure (Additional file 3: Figure S3).
STAT1 and STAT3 play distinct roles in cytokine-induced PD-L1 expression on monocytes
There is currently heightened interest in understanding mechanisms that drive expression of the immunosuppressive ligand PD-L1 in the TME, since the PD-1:PD-L1 pathway is now recognized as a dominant immune checkpoint in cancer. While this pathway has been targeted with some success in cancer therapy, current drug development strategies aim to overcome the failure of many tumors to respond to PD-1 pathway blocking drugs, and to address relapses that can occur following initial tumor regression. PD-L1 can be expressed by diverse cell types in the TME, including tumor, immune and endothelial cells. It is assumed that PD-L1 expression by any cell type in the TME can function locally to dampen antitumor immunity. This assumption has been borne out by the development of several predictive biomarkers for the therapeutic effects of anti-PD-1 drugs, that score PD-L1 protein expression on tumor cells, tumor-infiltrating immune cells, or both .
IFN-g secreted by tumor-reactive T cells, signaling through the transcription factor STAT1, is the single major cytokine that induces PD-L1 protein expression. This is associated with the phenomenon of adaptive tumor immune resistance . Here we show that the effect of IFN-g in enhancing PD-L1 expression by tumor cells and Monos occurs as a result of new mRNA transcription, rather than translocation of preexisting intracellular protein stores to the cell surface. We also show that this adaptive phenomenon can increase PD-L1 expression in cells already having constitutive expression. This raises the possibility that drugs targeting STAT1 might be deployed against IFN-g-induced PD-L1 expression, to enhance anti-PD-1 therapies. Furthermore, our data indicate that targeting STAT1 might also mitigate PD-L1 expression induced by IL-27. The broad spectrum of biological roles for STAT1 suggests that it could be difficult to target this factor specifically or selectively in tumor cells. However, a recent report from Cerezo et al. suggests that drugs inhibiting eukaryotic initiation factor (eIF)4A can down-modulate STAT1 transcription in a tumor-selective manner, indirectly reducing PD-L1 expression and mediating tumor regression in murine models . Further, these authors demonstrated in vitro that eIF4A chemical inhibition can decrease IFN-g-inducible PD-L1 expression in cell lines from a variety of human tumor types, including melanoma, breast and colon cancer, suggesting the potential for broad applicability of this approach.
In our previous studies of the TMEs of several different cancer types, we found that elevated levels of transcripts for the cytokines IL-1a, IL-10, IL-27 and IL-32 g, in addition to IFN-g, were associated with PD-L1 protein expression. As shown in the current report, each of these cytokines can induce PD-L1 expression on tumor cells and/or Monos in vitro, although to a lesser extent than IFN-g. Furthermore, some cytokines such as IL-1a and IL-27 can have an additive or synergistic effect on PD-L1 expression when combined with IFN-g (Fig. 3, Additional file 8: Table S4). Here we show that IL-27, similar to IFN-g, induces PD-L1 by activating STAT1. However, IL-10 induces PD-L1 by activating STAT3, and IL-1a by activating the p65 transcription factor. This demonstration of the involvement of distinct signaling pathways in driving PD-L1 expression suggests new strategies for targeting diverse transcription factors, or their upstream cytokines or receptors, to mitigate PD-L1 expression in the TME. For instance, STAT3 inhibitors, which are already in clinical testing, have been proposed to synergize with anti-PD-1/PD-L1 through their immunomodulatory effects, based on data from murine models . Furthermore, because the signaling pathway by which IL-1a drives PD-L1 expression is non-overlapping with IFN-g and IL-27, our findings suggest that genetic defects in tumor cell STAT1 signaling, which can be acquired under the selection pressure of anti-PD-1 therapy , would not interfere with the ability of IL-1a to sustain tumor cell expression of PD-L1. Such tumors would maintain the ability to evade immune attack from PD-1+ T cells. Ongoing efforts to compare the immune microenvironments of tumors that are responsive or resistant to anti-PD-1 therapies will explore these hypotheses.
Finally, there appears to be a unique set of cytokines, including IL-10 and IL-32 g, which are capable of promoting PD-L1 expression on Monos but not on tumor cells, as studied in our previous report  and in unpublished data. The failure of tumor cells to express the IL-10 receptor may explain the failure of IL-10 to promote PD-L1 expression on them (data not shown). Regarding IL-32 g, because its receptor has not yet been identified, potential mechanisms underlying its Mono-selective PD-L1-inducing activity are unknown at this time. PD-L1 expression by Monos may be an important source of immunosuppression in the TME, and antibodies blocking cytokines or cytokine receptors mediating this expression should be considered as potential adjuncts to PD-1 pathway blockade .
Factors driving the expression of the immunosuppressive ligand PD-L1 in the TME are diverse and can vary according to cell type. Both tumor and immune cells are important sources of PD-L1 expression. Cytokines regulating PD-L1 expression, including IFN-g, IL-1a, IL-10, IL-27 and IL-32 g, signal through diverse transcription factors and have variable effects on tumor cells and Monos. Understanding the complex mechanisms underlying intratumoral PD-L1 expression will open new opportunities for developing rationally targeted combination therapies to enhance the effects of anti-PD-1 drugs.
The authors thank Hao Wang for advice on statistical analyses, Paige Damascus for technical support, Chirag Patel for advice on siRNA transfection procedures and blocking mRNA and protein synthesis in vitro, and Daria Gaykalova for providing SCCHN cell lines (all from Johns Hopkins University School of Medicine, Baltimore, MD); and James C. Yang (National Institutes of Health, Bethesda, MD) for providing RCC cell lines.
SC designed, conducted and analyzed experiments, and wrote the manuscript. GAC, TSP, TLM and PW acquired and analyzed data. DMP and FP supervised and analyzed portions of this study. SLT designed and supervised the study and co-wrote the manuscript. All authors reviewed and approved the manuscript.
This study was supported by NCI R01 CA142779 (DMP and SLT), the Barney Family Foundation (SLT), Moving for Melanoma of Delaware (SLT), the Laverna Hahn Charitable Trust (SLT), R01 AI089830 (FP), Melanoma Research Alliance (FP) and the Johns Hopkins Bloomberg~Kimmel Institute for Cancer Immunotherapy.
Ethics approval and consent to participate
The use of human tissues in this study was approved by the Johns Hopkins Institutional Review Board.
Consent for publication
DMP and SLT report stock and other ownership interests from Aduro Biotech, Compugen, DNAtrix, Dragonfly Therapeutics, ERVAXX, Five Prime Therapeutics, FLX Bio, Jounce Therapeutics, Potenza Therapeutics, Tizona Therapeutics, and WindMIL; consulting or advisory role with AbbVie, Amgen, Bayer, Compugen, DNAtrix, Dragonfly Therapeutics, Dynavax, ERVAXX, Five Prime Therapeutics, FLX Bio, lmmunomic Therapeutics, Janssen Oncology, Medlmmune, Merck, Tizona Therapeutics, and Wind MIL; research funding from Bristol-Myers Squibb, Compugen, and Potenza Therapeutics; patents, royalties, and other intellectual property from Aduro Biotech, Bristol-Myers Squibb, and lmmunonomic Therapeutics; and travel, accommodations, expenses from Bristol-Myers Squibb, and Five Prime Therapeutics. SC, GAC, TSP, PW, TLM and FP have no conflicts of interest to disclose.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.