Melanoma Extracellular Vesicles Generate Immunosuppressive Myeloid Cells by Upregulating PD-L1 via TLR4 Signaling

Malignant melanoma is characterized by a rapid progression, metastasis to distant organs, and poor survival of patients (1). Despite the therapeutic success achieved by the immune checkpoint inhibitors such as antibodies targeting cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) and programmed cell death protein 1 (PD-1), most patients fail to respond to treatment (2). One of the major reasons for this poor response rate is the development of immunosuppressive tumor microenvironment (TME), in which myeloid-derived suppressor cells (MDSC) play a crucial role (3–6). They accumulate in preclinical melanoma mouse models and melanoma patients and strongly inhibit antitumor functions of T and natural killer (NK) cells, promoting tumor progression (4–8). MDSC represent a heterogeneous population of monocytic and polymorphonuclear cells that are derived from immature myeloid cells (IMC) and activated by soluble inflammatory factors constantly produced by tumor and host cells (3–8). One of the major mechanisms of MDSC-mediated immunosuppression is linked to the upregulation of programmed cell death ligand 1 (PD-L1) interacting with its receptor PD-1 expressed on tumor-infiltrating T cells (3–5, 9).

It has been recently demonstrated that MDSC could be also derived from IMC or differentiated myeloid cells by the exposure to extracellular vesicles (EV) secreted by tumor cells (10–14). The term EV is currently applied to define all kinds of vesicles released by various cell types including erythrocytes, platelets, leukocytes, and cancer cells (15). The process of EV secretion is particularly active in proliferating cells such as cancer cells (16). Small vesicles (50–150 nm) are released from the cell surface (microvesicles) or from the endosomal system (exosomes). By the fusion of late endosome with the plasma membrane, exosomes are secreted into the extracellular space (17, 18). In contrast, apoptotic vesicles (bodies) are larger (1,000–5,000 nm) and can be separated by size and density from smaller vesicles (17, 18). Depending on the cell origin, EV contain different biologically active molecules such as proteins, mRNA, miRNA, and lipids and are considered as mediators of intercellular communication (14, 19, 20).

Although the conversion of IMC into immunosuppressive MDSC by tumor-derived EV was already described, the molecular mechanism underlying MDSC induction and EV involvement in the generation of the TME is poorly understood. We suggested previously that both the EV-mediated signaling via myeloid cell receptors and the delivery of biologically active molecules by EV (cargo function) might be important in this process (8, 20). Here, we investigated the molecular mechanisms of interaction between normal myeloid cells (IMC and monocytes) and melanoma-derived EV in mouse and human setting, leading to the generation of immunosuppressive MDSC. Murine EV were isolated from the Ret melanoma cell line (Ret-EV) that was established from skin melanomas isolated from RET transgenic mice (21). This model closely resembles human melanoma in terms of clinical development and tumor–stroma interactions (22). We found that Ret-EV induced the upregulation of PD-L1 expression on bone marrow (BM)–derived murine IMC and immortalized myeloid suppressor cell line MSC-2 (23). EV-treated IMC strongly suppressed the activation of CD8⁺ T cells, which was restored by antibodies against PD-L1 blocking the PD-1/PD-L1 axis. Importantly, the observed effects were mediated by Toll-like receptor (TLR) signaling (because Ret-EV failed to induce immunosuppressive capacity of IMC from Tlr4^−/−, Tlr2^−/−, and Tlr7^−/− mice) as well as by the heat-shock protein 86 (HSP86) expressed on Ret-EV. Moreover, normal human CD14⁺ monocytes exposed to EV produced by human melanoma cells also displayed elevated PD-L1 expression in a HSP86/TLR4-dependent manner and acquired a strong immunosuppressive capacity, blocking T-cell activation. Our results suggest that EV-induced PD-L1 expression mediates the conversion of normal myeloid cells into immunosuppressive MDSC in tumor-bearing hosts via TLR signaling, indicating a promising target for cancer immunotherapy.

Healthy C57BL/6 mice (6–8 weeks) were purchased from Charles River. Tlr4^−/−, Myd88^−/−, and Myd88^−/−/Trif^−/− C57BL/6 mice were delivered from University of Essen (Germany). Tlr2^−/− and Tlr7^−/− C57BL/6 mice were kindly provided by Beatrix Schumak (University of Bonn, Germany) and Stefan Bauer (University of Marburg, Germany). Mice were kept under pathogen-free conditions in the animal facility of the University Medical Center (Mannheim, Germany). Animal studies have been conducted in accordance with an Institutional Animal Care and Use Committee.

Murine Ret melanoma cell line was established from skin melanomas isolated from RET transgenic mice (22) and cultured in RPMI-1640 (Gibco) supplemented with 10% heat-inactivated FBS (Gibco) and 1% penicillin/streptomycin (Sigma-Aldrich). Human melanoma cell lines HT-144 and SK-MEL-28 obtained from the ATCC were cultured in DMEM supplemented with 0.1 mmol/L β-mercaptoethanol (both from Gibco), 10% heat-inactivated FBS, and 1% penicillin/streptomycin. Immortalized myeloid suppressor cell lines MSC-1 and -2 (23) were kindly provided by Stefano Ugel (University of Verona, Italy) and grown in RPMI-1640 supplemented with 10 mmol/L sodium pyruvate (Gibco), 10% heat-inactivated FBS, and 1% penicillin/streptomycin. All cell lines were maintained under 5% CO₂ at 37°C and routinely tested for Mycoplasma contamination using the Mycoplasma Detection Kit for Conventional PCR (Minerva Biolabs). Cells were at passages not greater than seven between thawing and use in the described experiments.

Lentiviral shRNA plasmid DNA for stable knockdown of HSP86 and Mission TRC pLKO.5-puro nonmammalian shRNA control plasmid, as well as KNK-437 (an inhibitor of inducible HSP), actinomycin D, Bay11-7082 (an inhibitor of NF-κB), and 5-(N,N-Dimethyl) amiloride hydrochloride (DMA) were purchased from Sigma-Aldrich. FcR Blocking Reagent was provided by BD Biosciences. Dynabeads mouse and human T-Activator CD3/CD28 were from Thermo Fisher Scientific. Anti-human CD14 MicroBeads and CD8⁺ T cell Isolation Kit were purchased from Miltenyi Biotec. Anti-mouse mAbs CD9, PD-L1, and isotype control mAbs (IgG2a) were obtained from Thermo Fisher Scientific; CD81, p65, phospho-p65 (S536), STAT3, p-STAT3 (Y705), and calreticulin were provided by Cell Signaling Technology; Histon H3 and GAPDH were from Biolegend; ALIX were from Santa Cruz Biotechnology. Fluorochrome-conjugated anti-mouse mAbs CD11b-APC-Cy7, Gr1-PE-Cy7, and PD-L1-BV421 were purchased by BD Biosciences. Anti-human mAbs PD-L1 (405.9A11; Cell Signaling Technology) and ALIX were provided by Cell Signaling Technology; GAPDH and HSP90 were obtained from Biolegend; CD9 were from BD Biosciences; CD81 from Invitrogen. Anti-human TLR2- and TLR4-neutralizing mAbs and TLR agonists (LPS, PAM3CSK4, and R848) were obtained from InvivoGen. Fluorochrome-conjugated anti-human mAbs CD14-FITC and PD-L1-PE-Cy7 were from BD Biosciences.

Fibroblasts were obtained as previously described (24). Briefly, mouse hearts depleted from blood and mechanically disrupted in cold Hank's Balanced Salt Solution containing 100 U/mL collagenase and 0.1% trypsin were added followed by a constant shaking at 37°C for 10 minutes. After centrifugation at 300 g, 4°C for 5 minutes, cells were resuspended in DMEM/F12 medium (Gibco), containing 10% FBS and 1% penicillin/streptomycin, and plated into cell culture dishes. Two hours later, only fibroblasts were adhered, and the medium was changed.

EV secreted by these cell types were isolated as described before (25, 26). Briefly, serum-free conditioned medium from cultured tumor cells or fibroblasts was sterile filtered through 0.22 μmol/L Steritops (Merck Millipore) followed by size exclusion filtration through Amicon-Ultra-15, 100 kDa (Merck Millipore) at 3,500 g for 30 minutes. EV were pelleted from supernatants by ultracentrifugation at 100,000 g for 90 minutes (Sorvall Discovery) and resuspended in sterile PBS. The number/size distribution of particle was analyzed by NTA (Nanosight). For this, EV samples were prepared at the concentration 5 ng/mL, and particles were measured 5 times for 1 minute. Each EV preparation was tested for endotoxins by limulus amebocyte lysate assay (Thermo Fisher Scientific).

EV were isolated from patients' plasma after informed written consent (ethics committee approval 2010-318M-MA) as described (27). Briefly, heparinized blood samples were subjected to the density gradient centrifugation using Biocoll (Biochrom). After removal of peripheral blood mononuclear cells (PBMC), plasma was collected, aliquoted, and stored at −80°C. For EV isolation, plasma was thawed, diluted with cold PBS, and filtered through a 0.22 μm Syringe Filter (Merck). The filtered plasma was centrifuged at 10,000 g for 30 minutes to remove debris and then pelleted by ultracentrifugation at 100,000 g for 70 minutes. The pellet was resuspended in sterile-filtered PBS and frozen at −20°C until use. In some experiments, plasma was depleted from EV by ultracentrifugation at 100,000 g.

Ret melanoma cells with stably knocked down HSP86 (shHSP86) or treated with scrambled sequence shRNA construct (shSCR) were seeded in 96-well plates (Gibco) at a density of 2,500 cells/well. Alamar blue (10% of the culture medium volume) was added after cell attachment for 4 hours followed by the measurement of fluorescence at an emission wavelength of 535 nmol/L and an excitation wavelength of 590 nmol/L using a SpectraMax M5 microplate reader (Tecan Infinite F200 PRO). Cells were incubated further for 24, 48, and 72 hours; Alamar blue was added for 4 hours before the end of each time point, and the fluorescence was measured.

The analysis was performed as described before (27). Briefly, Ret-EV were stained with anti-CD81 mAbs (dilution 1:50) followed by the treatment with Protein A-Gold (10 nm) kindly provided by George Posthuma (University Medical Center Utrecht, Utrecht, the Netherlands). Images were taken with the electron microscope EM910 (Carl Zeiss) at 80 kV and were registered with a CCD-Camera (TRS-system, Tröndle) using the manufacturer's software ImageSP.

CD11b⁺Gr1⁺ cells were isolated from the BM of healthy 6- to 8-week-old C57BL/6 mice using MDSC Isolation kit (Miltenyi Biotec) with the purity above 90%. Note that 10⁶ cells were cultured in RPMI-1640 supplemented with 10% FBS depleted from EV, 10 mmol/L sodium pyruvate (Gibco), 1% penicillin/streptomycin, and 0.05 μmol/L β-mercaptoethanol with or without EV (50 μg/mL) in a total volume of 0.1 mL for 16 hours followed by various analyses.

Buffy coats isolated from healthy donors were purchased from the German Red Cross Blood Service Baden Württemberg-Hessen. PBMC were isolated by density gradient centrifugation using Biocoll (Biochrom). Then, CD14⁺ monocytes were sorted by anti-human CD14 MicroBeads according to the manufacturer's protocol. Monocytes (2 × 10⁵) were cultured in RPMI-1640 supplemented with 10% FBS depleted from EV, 10 mmol/L sodium pyruvate (Gibco), 1% penicillin/streptomycin, and 0.05 μmol/L β-mercaptoethanol with or without EV (15 μg/mL) in a total volume of 0.1 mL for 16 hours followed by FACS analysis.

Effects of EV derived from Ret melanoma cells or cardiac fibroblasts were evaluated using a transwell system. Note that 2 × 10⁵ cells were incubated for 24 hours at 37°C with or without 15 μmol/L DMA. After 24 hours, IMC were added to the upper chamber of a polycarbonate Transwell culture insert (Costar) with a pore size of 0.4 μm (Sarstedt) and cocultured for another 24 hours. Then, IMC were analyzed by flow cytometry.

Cells were treated with FcR Blocking Reagent and stained with mAbs for 30 minutes at 4°C. Acquisition was performed by multicolor flow cytometry using FACSCanto II with FACSDiva 6.0 software (BD Biosciences). Dead cell exclusion was based on the scatter profile. The compensation was performed with BD CompBeads set (BD Biosciences) using the manufacturer's instructions. FlowJo software (Tree Star) was used to analyze at least 100,000 events.

Murine IMC or immortalized myeloid suppressor cells MSC-1 and -2 were incubated with Ret-EV (50 μg/mL) for 16 hours in a total volume of 0.1 mL. In some experiments, MSC-2 cells were treated with Ret-EV in the presence of 1 μg/mL actinomycin D or preincubated for 3 hours with the NF-κB inhibitor Bay (5 or 10 μmol/L). Then, cell pellets were solubilized for 30 minutes on ice in lysis buffer (250 mmol/L NaCl, 50 mmol/L HEPES, 0.5% NP-40, 10% glycerol, 2 mmol/L EDTA, 10 mmol/L NaF, 1 mmol/L Na-orthovanadate, 1 mmol/L PMSF, and 10 mg/mL of each leupeptin and aprotinin) or radioimmunoprecipitation assay buffer (Merck) containing 1% protease inhibitor (Calbiochem). Lysates were cleared by centrifugation and boiled with reducing or nonreducing SDS loading buffer. Protein concentrations were determined using Pierce BCA Protein Assay Kit (Thermo Fisher Scientific). Samples were separated on SDS-PAGE gels and transferred to polyvinylidene fluoride membranes using iBlot2 (Thermo Fisher Scientific). After blocking with 3% BSA in Tween-20/TBS, membranes were treated with primary antibodies followed by horseradish peroxidase–conjugated secondary antibody and ECL detection (Thermo Fisher Scientific).

Note that 10⁵ CD11b⁺Gr1⁺ IMC from the BM of wild-type or TLR-deficient mice were treated with 50 μg/mL EV in a total volume of 0.1 mL for 16 hours at 37°C. Then cells were washed twice at 300 g for 5 minutes to remove the rest of the EV and treated with PD-L1–neutralizing or isotype control mAbs for 15 minutes followed by centrifugation at 300 g for 5 minutes. CD8⁺ T cells were isolated from spleens of healthy C57BL/6 mice using the naïve CD8⁺ T cell Isolation Kit (Miltenyi Biotec) according to the manufacturer's instructions and labeled with 1 μmol/L carboxyfluorescein succinimidyl ester (CFSE). After washing CFSE out by the centrifugation at 300 g for 5 minutes, cells were stimulated with Dynabeads mouse T-Activator CD3/CD28 (Thermo Fisher Scientific) and incubated with CD11b⁺Gr1⁺ cells at indicated ratios for 72 hours. T-cell proliferation was evaluated by CFSE dilution by flow cytometry. In addition, IFNγ secretion was evaluated in the coculture supernatants using the Mouse IFN-γ ELISA MAX Standard kit (Biolegend) according to the manufacturer's instructions.

Note that 2 × 10⁵ CD14⁺ monocytes from healthy donors were treated with 0.5 to 50 μg/mL EV in a total volume of 0.2 mL for 16 hours at 37°C. Then cells were washed twice at 300 g for 5 minutes to remove the rest of EV. CD8⁺ T cells were obtained from healthy donors using the CD8⁺ T cell Isolation Kit, cultured in X-VIVO 20 serum-free medium (Biozym) for 16 hours at 37°C and labeled with 1 μmol/L CFSE. After washing CFSE out by the centrifugation at 300 g for 5 minutes, CD8⁺ T cells were stimulated with Dynabeads human T-Activator CD3/CD28 (Thermo Fisher Scientific) and cocultured with CD14⁺ monocytes at indicated ratios for 72 hours. T-cell proliferation was evaluated by flow cytometry.

HEK293T cells were used for lentiviral particle production. For transfection, plasmid containing respective shRNA (11 μg) was incubated with the packaging plasmids VSV-G (5.5 μg) and pCMV-dR 8.91 (8.25 μg) in DMEM and X-treme GENE (Roche) solution for 30 minutes and added to HEK293T producer cells. After incubation for 12 hours, the supernatant was discarded. Upon further culture for 12, 24, and 36 hours, the supernatant was collected and virus particles were concentrated by ultracentrifugation. Then mouse Ret or human SK-MEL-28 melanoma cells were incubated with concentrated virus for 24 hours. Upon the first infection, Ret or SK-MEL-28 cells were reinfected with the same virus in fresh medium, and, after 48 hours of transduction, the cells were washed twice with PBS and cultured. To select transduced cells, 2 μg/mL puromycin was added for 3 days. The following plasmids were used: plasmid shRNA HSP86: KD 1: TRCN0000321086; KD 2: TRCN0000321084; KD 3: TRCN0000321007; KD 4: TRCN0000321085; KD 5: TRCN0000321083 and control plasmid: Mission TRC pLKO.5-puro nonmammalian shRNA control plasmid.

C57BL/6 mice were injected s.c. with 5 × 10⁵ Ret cells transduced with shRNA HSP86 (shHSP86 Ret) or with scrambled control (shSCR Ret). Tumor development was monitored daily by measuring tumor diameter.

The method was performed as described before (26). Briefly, total RNA was isolated by Trizol and transcribed with SensiFast cDNA Synthesis Kit (Bioline) followed by the quantification by a NanoDrop Spectrophotometer (ND-2000; Thermo Fisher Scientific). Primers were designed using Primer3 web tool and were produced by Metabion. S18RNA was applied as an internal standard. The sequences of primers used are shown in Supplementary Table S1.

Statistical analyses were performed using GraphPad Prism (GraphPad Software) on at least three independent experiments if not indicated differently. Data were analyzed with a one-way ANOVA test for multiple groups or an unpaired two-tailed Student t test for two groups. A value of P < 0.05 was considered statistically significant.

We isolated EV from the supernatant of murine Ret melanoma cells (Ret-EV) using ultrafiltration followed by ultracentrifugation (25, 26). Nanoparticle tracking analysis (NTA) showed that the size of Ret-EV was 99.1 nm ± 7.2 nm (Fig. 1A). Moreover, EV expressed typical markers such as CD9, CD81, and ALIX but were negative for calreticulin, a marker of endoplasmic reticulum (ER; Fig. 1B). Interestingly, the immunosuppressive molecule PD-L1 was found in Ret cell lysate. However, it was only weakly detectable in Ret-EV (Fig. 1B). In addition, isolated EV were visualized by electron microscopy using immunogold staining for CD81 as an EV marker (Fig. 1C). To investigate the uptake by myeloid cells, EV were labeled with CFSE and incubated with immortalized myeloid suppressor cell line MSC-1 (23). Fluorescent microscopy analysis revealed that labeled Ret-EV were internalized and localized inside the cells (Fig. 1D). Furthermore, using flow cytometry, we observed the uptake of CSFE-labeled Ret-EV by CD11b⁺Gr1⁺ IMC isolated from the BM of C57BL/6 mice (Fig. 1E).

To test the effects of Ret-EV on myeloid cells in vitro, we isolated murine BM-derived CD11b⁺Gr1⁺ IMC and exposed them to Ret-EV. In accordance with previous reports (10, 12, 26), we observed a significant upregulation of the expression of inflammatory and immunosuppressive mediators such as IL1β, IL6, IL10, TNFα, and cyclooxygenase-2 (COX-2) in IMC measured by RT-PCR (Fig. 2A). Importantly, we detected also a strong upregulation of PD-L1 both at the mRNA and protein levels (Fig. 2A and B). To exclude possible contamination by endotoxins, we applied the limulus amebocyte lysate assay and found Ret-EV samples free of endotoxin.

Using flow cytometry, we confirmed Ret-EV–mediated elevation of the frequency of PD-L1 expressing IMC and the level of its expression measured by the median fluorescence intensity (MFI; Fig. 2C and D). In parallel, CD11b⁺Gr1⁺ IMC were treated with EV isolated from normal cardiac fibroblasts (Fibro-EV). No significant increase in PD-L1 expression was observed, suggesting that this PD-L1 modulation was due to Ret-EV (Fig. 2C and D).

To exclude possible changes in EV samples due to the isolation procedure, we established a coculture system. Ret melanoma cells or fibroblasts were seeded into 24-well plates followed by treatment with DMA, an H⁺ proton-pump inhibitor known to block EV secretion (12). After 24 hours of incubation, IMC were put into the upper compartment of transwell chambers to exclude any direct cell–cell contact. The PD-L1 expression on IMC was measured after 48 hours. As shown in Fig. 2E and F, DMA-treated Ret cells induced significantly lower elevation of PD-L1 expression than nontreated Ret cells. As expected, the incubation with cardiac fibroblasts did not change the frequency of PD-L1–expressing IMC.

Next, we investigated if tumor-derived EV could also stimulate PD-L1 expression on myeloid cells in vivo. To this end, we transduced Ret cells with the vector-expressing EV marker CD81 linked to GFP and injected them into syngeneic C57BL/6 mice to follow the EV production and uptake in vivo. Fourteen days later, tumor single-cell suspensions were prepared and flow cytometry analysis revealed the presence of CD11b⁺Gr1⁺ MDSC expressing GFP, suggesting that CD11b⁺Gr1⁺ cells acquired CD81-GFP via the uptake of Ret-EV (Fig. 2G). Importantly, GFP⁺CD11b⁺Gr1⁺ MDSC showed a tendency for higher intensity of PD-L1 expression as compared with their GFP⁻ counterparts (Fig. 2H).

Next, we studied the possibility that EV-educated CD11b⁺Gr1⁺ IMC not only increased PD-L1 expression but also affected the activation of T lymphocytes. Upon incubation with Ret-EV for 16 hours, IMC were added to purified autologous mouse spleen CD8⁺ T cells that were labeled with CFSE and activated with anti-CD3/CD28 antibodies. We observed a dose-dependent inhibition of T-cell proliferation in the presence of EV-educated IMC measured by flow cytometry (Fig. 3A). Importantly, blocking antibodies against PD-L1 could significantly prevent such suppression of T-cell proliferation (Fig. 3B). In addition, the production of IFNγ by stimulated T cells was found to be restored in the presence of PD-L1–blocking antibodies (Fig. 3C). These results suggest a pivotal role of PD-L1 expression induced on EV-treated myeloid cells in their inhibition of T lymphocyte functions. Because the immunosuppressive ability represents a key feature of MDSC, in which PD-L1 expression plays an important role (3–5, 9), such myeloid cells could be considered to be converted to MDSC (3, 4) under the influence of melanoma-derived EV.

We observed that Ret-EV increased in vitro the expression of PD-L1 on the murine MSC-2 but not MSC-1 cells (Supplementary Fig. S1A). Studying the mechanism of this upregulation, we found that such PD-L1 elevation on immortalized myeloid suppressor cell line MSC-2 was completely abrogated in the presence of actinomycin D, an inhibitor of mRNA synthesis (Supplementary Fig. S1B). This observation suggests that Ret-EV rather trigger the synthesis of PD-L1 de novo than deliver this protein to myeloid cells. Deciphering Ret-EV–induced signaling, we demonstrated a strong activation of transcription factor NF-κB as detected by phosphorylation of the p65 subunit in MSC-2 cells shown via Western blot (Supplementary Fig. S1C). Next, MSC-2 cells were incubated with the NF-κB inhibitor Bay11-7082. We noticed a dose-dependent suppression of the induction of PD-L1 expression, suggesting a critical role of NF-κB signaling in this induction mediated by Ret-EV (Supplementary Fig. S1D).

Next, we isolated IMC from C57BL/6 mice deficient for the expression of adapter protein myeloid differentiation primary response 88 (MyD88) involved in the TLR/NF-κB signaling. In addition, IMC were obtained from mice deficient for both MyD88 and TIR-domain-containing adapter-inducing interferon-β (TRIF), another adapter protein that participates in TLR/NF-κB signaling. As expected, the treatment with Ret-EV failed to increase the frequency of PD-L1⁺CD11b⁺Gr1⁺ IMC from both types of knockout mice (Supplementary Fig. S1E) as well as the PD-L1 expression on these cells expressed as MFI (Supplementary Fig. S1F).

To identify which TLR could be involved in PD-L1 upregulation, CD11b⁺Gr1⁺ IMC were treated with various TLR ligands such as Pam3/CSK4 (TLR2), LPS (TLR4), and R848 (TLR7/8). We found that all three ligands were able to induce PD-L1 expression on these cells (Fig. 4A). Moreover, IMC from Tlr2^−/−, Tlr4^−/−, or Tlr7^−/− mice displayed significantly lower upregulation of PD-L1 expression upon the incubation with Ret-EV as compared with their counterparts from wild-type mice (Fig. 4B). However, the strongest reduction after the treatment was observed in IMC from Tlr4^−/− mice, showing almost no elevation of the frequency of PD-L1⁺ cells and the intensity of PD-L1 expression (Fig. 4B and C). In agreement with the protein expression data, the induction of PD-L1 mRNA by Ret-EV was also completely abolished in IMCs from Tlr4^−/− mice (Fig. 4D) and to a less extent in cells from Tlr2^−/− and Tlr7^−/− mice (Supplementary Fig. S2A). Importantly, IMC from Tlr4^−/− mice (in contrast to IMC from wild-type mice) were unable to inhibit T-cell proliferation upon the treatment with Ret-EV (Fig. 4E). Similar results were obtained using IMC from Tlr2^−/− or Tlr7^−/− mice (Supplementary Fig. S2B), suggesting a critical role of TLR for the induction of immunosuppressive activity of myeloid cells by tumor-derived EV.

Some HSP on EV were previously described as potent stimulators of TLR signaling (12, 28–30). We examined various preparations of Ret-EV for the expression of different TLR ligands such as HSP86, HSP72, HSP60, and HMGB1 by Western blot analysis. Only HSP86 was consistently observed in all Ret-EV samples (Fig. 5A). To determine if HSP86 was expressed on the EV surface, we adsorbed them on latex beads and stained for HSP86 followed by flow cytometry. HSP86 was found to be expressed on the surface of Ret-EV (Fig. 5B). Next, we treated Ret melanoma cells with KNK-437, a potent inhibitor of the synthesis of inducible HSP, and proved a dose-dependent reduction in HSP86 expression in these cells (Fig. 5C). EV isolated from KNK-437–treated (KNK-EV) or control (DMSO-treated) Ret cells were incubated with CD11b⁺Gr1⁺ IMC derived from wild-type mice. Interestingly, KNK-EV induced significantly lower upregulation of the frequency of PD-L1⁺ cells (Fig. 5D) and the intensity of PD-L1 expression (Fig. 5E) than EV isolated from control Ret cells. Importantly, the NTA revealed no differences between KNK-EV and control Ret-EV regarding EV concentration and size (Fig. 5F).

In another set of experiments, the HSP86 expression in Ret cells was stably knocked down using lentivirus-mediated shRNA delivery (shHSP86 Ret). Then these cells were cultured in 24-well plates followed by the addition of BM-derived CD11b⁺Gr1⁺ IMC as described above (see Fig. 2E). Coculture with shHSP86 cells mediated only very slight upregulation of PD-L1 expression on IMC that was significantly lower than that induced by Ret cells treated with scrambled sequence shRNA construct (shSCR Ret; Fig. 5G and H). Furthermore, wild-type IMC were treated with EV isolated from shHSP86 (shHSP86 Ret-EV) or shSCR (shSCR Ret-EV) Ret cells followed by IMC incubation with activated autologous CD8⁺ T cells. As shown in Fig. 5I, IMC treated with shHSP86 Ret-EV (in contrast to IMC-incubated shSCR Ret-EV) were not able to inhibit T-cell proliferation. Collectively, these findings indicate that HSP86 represents an essential component for TLR-mediated signaling leading to the upregulation of PD-L1 expression on IMC treated with Ret-EV.

To study the impact of HSP86 expression in tumor cells on their growth in vivo, we injected shHSP86-Ret cells into C57BL/6 mice. Tumor growth was found to be delayed as compared with mice inoculated with control shSCR-Ret cells (Fig. 6A). We tested also shHSP86-Ret and shSCR-Ret cells for their proliferation capacity in vitro and found no difference between the two cell types (Fig. 6B), indicating that the growth inhibition of shHSP86-Ret cells in vivo was not due to the impaired cell division. However, we observed a decrease in the frequency of CD11b⁺Gr1⁺ MDSC infiltrating HSP86-deficient tumors (Fig. 6C). In addition, the frequency of PD-L1⁺ MDSC within the total MDSC population was significantly reduced as compared with MDSC-infiltrating control Ret tumors (Fig. 6D). A similar strong reduction in the frequency of PD-L1⁺ MDSC was also observed among total MDSC in the BM (Fig. 6E).

These data suggest that HSP86 from Ret melanoma cells could be expressed in Ret-EV and play an important role in the generation of immunosuppressive MDSC in tumor-bearing mice.

Next, we addressed the question of whether also human melanoma–derived EV could induce PD-L1 expression on myeloid cells. Therefore, we isolated EV from the human melanoma cell line HT-144 (HT-144-EV) using the protocol described above for mouse Ret melanoma cells. HT144-EV were characterized by a size of approximately 120 nm measured by NTA (Supplementary Fig. S3A), the expression of typical EV markers, and the absence of the ER marker calreticulin (Supplementary Fig. S3B). Similarly to mouse melanoma–derived EV, human HT-144-EV also induced an upregulation of several inflammatory and immunosuppressive factors as well as PD-L1 expression in CD14⁺ monocytes isolated from healthy donors measured by RT-PCR (Fig. 7A). Importantly, a strong increase in the frequency of PD-L1⁺ cells and in the intensity of its expression was also confirmed at the protein level (Fig. 7B; Supplementary Fig. S4A). Moreover, we found that EV isolated from another human melanoma cell line SK-MEL-28 (SK-MEL-28-EV) induced a similar upregulation of PD-L1 expression on healthy monocytes (Fig. 7B; Supplementary Fig. S4A).

Next, we investigated if EV-educated CD14⁺ monocytes could acquire an immunosuppressive potential. To this end, similar to our mouse experiments, normal human monocytes were exposed to HT-144-EV followed by coculture with activated human CD8⁺ T cells labeled with CFSE. These experiments confirmed that CD14⁺ monocytes educated by melanoma-derived EV could strongly inhibit CD8⁺ T-cell proliferation, which was dependent on the EV concentration and the ratio between EV-educated monocytes and T cells (Fig. 7C).

In the murine system, TLR4 and (to a lesser extent) TLR2 were found to be the main triggers of EV-dependent PD-L1 upregulation on myeloid cells. To test whether these TLR were also involved in the EV-dependent upregulation of PD-L1 on human CD14⁺ monocytes, we incubated them with blocking anti-TLR4 and anti-TLR2 antibodies prior to the HT-144-EV treatment. Like in our mouse experiments, inhibiting TLR4 signaling resulted in the strongest abrogation of EV-mediated stimulation of the PD-L1 expression on monocytes (Fig. 7D). Interestingly, both lysates of HT-144 melanoma cells and EV isolated from these cells displayed very weak PD-L1 expression (Fig. 7E), suggesting that the observed PD-L1 induction was rather a result of new protein synthesis mediated by TLR signaling.

Next, we studied whether HSP could play a role in the EV-induced PD-L1 upregulation in human monocytes. HT-144 and SK-MEL-28 melanoma cells as well as HT-144-EV and SK-MEL-28-EV were characterized by the expression of HSP86 (Fig. 7E). Similar to mouse experiments, the HSP86 expression in HT-144 cells was stably knocked down using lentivirus-mediated shRNA delivery followed by the isolation of EV from these cells (shHSP86 HT-144 EV). Importantly, the cell lysate and shHSP86 HT-144 EV showed a drastic reduction in the expression of HSP86 (Supplementary Fig. S5A and S5B). The coculture of CD14⁺ monocytes with these EV led to a significantly lower elevation of the frequency (Fig. 7F) and intensity (Supplementary Fig. S4B) of PD-L1 expression as compared with monocytes incubated with EV isolated from HT-144 cells treated with scrambled sequence shRNA construct (shSCR HT-144 EV). Furthermore, monocytes incubated with shHSP86 HT-144 EV were significantly less efficient in the inhibition of T-cell proliferation than monocytes treated with shSCR HT-144 EV (Fig. 7G).

To address the question of whether EV circulating in melanoma patients could induce the same effect on CD14⁺ monocytes from healthy donors, EV were purified from the plasma of stage IV melanoma patients followed by their coculture with monocytes. We found a strong increase in the frequency of PD-L1⁺ cells as compared with these values in nontreated monocytes (Fig. 7H). The level of PD-L1 expression under these conditions was also significantly elevated (Supplementary Fig. S4C). Importantly, monocytes incubated with plasma devoid of EV (containing only soluble factors) significantly lower upregulation of PD-L1 expression (Fig. 7H; Supplementary Fig. S4C).

Taken together, our results suggest that the interaction of melanoma-derived EV with myeloid cells in both human and murine systems leads to the upregulation of PD-L1 expression on these cells and to the acquisition of their capacity to inhibit T-cell activity.

Although initial reports demonstrated that EV released by tumor cells could be a unique source of tumor-associated antigens and stimulate T-cell–mediated antitumor immune responses (28, 29), numerous studies over the last decade have demonstrated that tumor-derived EV play a significant role in tumor progression (16, 17, 30, 31). EV have been reported to be important for preparing metastatic niches and for the development of therapy resistance (13, 31, 32). Furthermore, the potential of tumor-derived EV to dampen immune functions of the host has been brought into focus. In particular, tumor EV have been found to impair the function of T cells and NK cells (19, 32, 33), induce regulatory T cells (34), and convert normal myeloid cells into potent immunosuppressive MDSC (10, 12–14, 35).

However, the molecular mechanism of such EV-mediated MDSC generation from normal myeloid cells is still unclear. Here, we found that normal murine IMC strongly upregulated PD-L1 and acquire the capacity to inhibit T-cell functions upon the treatment with EV derived from Ret melanoma cells, suggesting that these EV-educated cells could be defined as MDSC (3). Furthermore, blocking PD-L1 led to a substantial attenuation of their immunosuppressive potential, indicating an importance of PD-L1 induction in the conversion of IMC into MDSC.

PD-L1 expression was demonstrated to be involved in MDSC-mediated T-cell inhibition through the binding to PD-1 expressed on effector T cells (3–5, 9). Several recent publications reported that tumor-derived EV could express PD-L1 at the surface and directly signal to T cells, thereby inhibiting their functions and mediating immune evasion (33, 34, 36–38). Moreover, PD-L1 expression on EV from head and neck cancer patients was reported to correlate positively with disease progression (32, 39). However, we found a weak expression of PD-L1 in EV isolated from mouse and human melanoma cells. Moreover, Ret-EV could induce a strong PD-L1 upregulation at the protein and mRNA levels in IMC and MSC-2 but not in MSC-1 cells. In addition, blocking mRNA synthesis by actinomycin D in MSC-2 cells led to a complete abrogation of PD-L1 expression. All these data indicate that, in our system, PD-L1 was rather induced de novo via myeloid cell receptors but not transferred by EV to these cells.

HSP are known as chaperons for controlling the correct folding of newly synthesized proteins. However, under certain conditions such as stress or cancer, they can be released and act as damage-associated molecular patterns, activating the immune system via binding to TLR (40). HSP have been previously described to be expressed on EV (15, 16). Thus, EV-mediated HSP72 was found to stimulate MDSC functions through STAT3 activation and cytokine secretion mediated by TLR2 signaling (12). Several other HSP such as HSP70, HSP90, or HSP105 were also reported to be associated with EV and to trigger TLR2 or TLR4 (41–43). It was demonstrated that TLR2 could be activated via exosomal prostaglandin E2 (11). In addition, EV-induced stimulation of PD-L1 expression on monocytes could be mediated by TLR7 signaling (35). Here, we identified the HSP86/TLR4 axis as a major driver of PD-L1 upregulation, although the contribution of TLR2 or TLR7 could not be excluded. In agreement with our data, Cheng and colleagues (44) reported that EV derived from hepatocellular carcinoma cells induced PD-L1 expression in human monocytic cell line THP-1 and mouse macrophage cell line RAW264.7; however, the authors did not address the mechanism for this induction. Because they observed a reduction of EV-mediated PD-L1 stimulation by treating tumor cells with melatonin, which was previously reported to decrease the induction of several HSP after oxidative stress (45), we believe that HSP86 might stimulate PD-L1 expression also in their system. Furthermore, it has been recently demonstrated that EV isolated from glioblastoma stem cells induced PD-L1 on normal human monocytes via STAT3 activation (46).

We have recently reported that a well-defined set of miRNA expressed in melanoma-derived EV could convert normal human monocytes into MDSC (47). Because certain miRNAs were previously demonstrated to activate endosomal TLR such as TLR7 and TLR8 (48), it is conceivable that the miRNA described by Huber and colleagues (47) might contribute to MDSC conversion by signaling via TLRs. Thus, it seems that different molecules (e.g., lipids, RNA, proteins) could stimulate diverse TLR, leading to the induction of PD-L1 expression.

Investigating PD-L1 upregulation on myeloid cells by melanoma-derived EV in vivo, we injected Ret cells transduced with the CD81-GFP vector into wild-type mice. We demonstrated the presence of CD11b⁺Gr1⁺ MDSC expressing GFP in the TME, suggesting that MDSC obtained CD81-GFP most likely via the uptake of Ret-EV. The ex vivo isolated, GFP⁺ MDSC showed a tendency for higher intensity of PD-L1 expression as compared with their GFP⁻ counterparts. Although we cannot exclude that MDSC acquire GFP signal via phagocytosis, these results are in line with our previous publication, using a Cre-Lox system to track the uptake of EV secreted by Lewis lung carcinoma cells (14), and suggest that the induction of PD-L1 on MDSC could also happen in vivo.

HSP in cancer have emerged as novel therapeutic targets. It has been recently demonstrated that HSP90 inhibition enhanced cancer immunotherapy by upregulating IFN response genes (49). Furthermore, inhibiting HSP90 in melanoma cells prevented the induction of functional MDSC, displaying an impaired capacity to inhibit T-cell proliferation in vitro (50). In the mouse MCA205 sarcoma model, the application of HSP90 inhibitor 17-DMAG in vivo resulted in slower tumor development and reduced levels of MDSC in the TME (51). We reported here that HSP86 inhibition or knockdown in Ret cells or human melanoma cell lines abolished the ability of released EV to induce PD-L1 upregulation. Although this observation is most likely due to the depletion of HSP86 from the EV surface, we cannot rule out the possibility that the HSP machinery is involved in the sorting of various compounds into EV. Future experiments are needed to clarify this point.

Taken together, our data highlight the molecular mechanism of the conversion of normal mouse and human myeloid cells into MDSC by melanoma-derived EV. It involves the triggering of mainly TLR4 on myeloid cells by inducible HSP86 in EV, leading to NF-κB activation and upregulation of PD-L1 expression. Blocking TLR4 signaling or directly PD-L1 expression resulted in the abrogation of the immunosuppressive capacity of EV-educated myeloid cells, suggesting a critical role for this PD-L1–inducing pathway in the acquisition of immunosuppressive properties. Based on characteristics of these EV such as size, and expression of an endocytic marker ALIX, it is probable that exosomes rather than microvesicles are mainly responsible for the described effects. Our data suggest the possibility of MDSC generation in cancer not only by inhibiting IMC differentiation via inflammatory mediators (3–8) but also by the reprogramming of normal myeloid cells through tumor-derived EV that could be supported by the high plasticity of these cells in tumor-bearing hosts (52).

J. Utikal reports receiving honoraria from the speakers' bureau of, and is a consultant/advisory board member for, Melanoma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: V. Fleming, X. Hu, P. Altevogt, V. Umansky

Development of methodology: V. Fleming, X. Hu, R. Weber, C. Groth, V. Nagibin, P. Altevogt, V. Umansky

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): V. Fleming, X. Hu, C. Weller, R. Weber, C. Groth, Z. Riester, L. Hüser, V. Nagibin, C. Kirschning, V. Umansky

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): V. Fleming, X. Hu, C. Weller, R. Weber, C. Groth, L. Hüser, V. Bronte, J. Utikal, P. Altevogt, V. Umansky

Writing, review, and/or revision of the manuscript: V. Fleming, X. Hu, R. Weber, C. Groth, Z. Riester, J. Utikal, P. Altevogt, V. Umansky

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Q. Sun, J. Utikal, V. Umansky

Study supervision: P. Altevogt, V. Umansky

Other (punctual text editing and TLR-specific inputting): C. Kirschning

The authors thank S. Uhlig for help with cell sorting and E. Cordell for help with article preparation. This work was supported, in part, by grants from the German Research Council (RTG2099 to J. Utikal and V. Umansky), the DKFZ-MOST Cooperation in Cancer Research (CA181 to V. Umansky), and the Italian Association for Cancer Research (IG grant 18603 and Special Program Molecular Clinical Oncology 5 per mille, 12182 to V. Bronte).

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.