The tumor microenvironment is characterized by regulatory T cells, type II macrophages, myeloid-derived suppressor cells, and other immunosuppressive cells that promote malignant progression. Here we report the identification of a novel BDCA1+CD14+ population of immunosuppressive myeloid cells that are expanded in melanoma patients and are present in dendritic cell–based vaccines, where they suppress CD4+ T cells in an antigen-specific manner. Mechanistic investigations showed that BDCA1+CD14+ cells expressed high levels of the immune checkpoint molecule PD-L1 to hinder T-cell proliferation. While this BDCA1+CD14+ cell population expressed markers of both BDCA1+ dendritic cells and monocytes, analyses of function, transcriptome, and proteome established their unique nature as exploited by tumors for immune escape. We propose that targeting these cells may improve the efficacy of cancer immunotherapy. Cancer Res; 76(15); 4332–46. ©2016 AACR.
The immunosuppressive tumor microenvironment is a major obstacle hampering the efficacy of cancer immunotherapy (1). Not only do tumor cells express immune-inhibitory molecules like programmed cell death-1 ligands 1 and 2 (PD-L1 and PD-L2; ref. 2), but they also actively recruit immunosuppressive leukocyte populations like regulatory T cells (Treg; ref. 3), myeloid-derived suppressor cells (MDSC; ref. 4), and M2 macrophages (5). This immunosuppression is not restricted to the tumor milieu, as tumor cells were shown to secrete several factors promoting the accumulation of MDSCs in peripheral tissues (6). Altogether, these tumor-triggered immunosuppressive components prohibit effective antitumor immune responses by inhibiting dendritic cells (DC) and T-cell effector functions (7).
Tumor-induced immunosuppression may be accountable for the low efficacy of antitumor immunotherapeutics, including DC vaccines (8, 9). Another important determinant of the clinical success of DC vaccines, which utilize the patient's own DCs (10), is the type, quality, and stimulation status of applied DCs (11). Thus far, most clinical studies have been performed with ex vivo differentiated monocyte-derived DCs (MoDC). Upon activation, however, MoDCs show limited capacity to sufficiently produce T cell-polarizing cytokines and display signs of exhaustion (12–14). Recent studies clearly demonstrated that vaccination with primary blood DCs, such as plasmacytoid DCs (pDC; ref. 15), or BDCA1+ DCs (16) is more efficient than MoDC vaccines in significantly enhancing overall survival of treated melanoma patients. Thus, an optimal antitumor DC vaccine can only be realized by combining the most effective DC population with strategies to neutralize tumor-promoted immunosuppression.
In the current study, we characterized a novel blood myeloid cell population that uniquely coexpresses the DC marker BDCA1 and the monocytic maker CD14. We have shown that BDCA1+CD14+ cells exert antigen-specific immune suppression and, when present in BDCA1+ DC-vaccines, may severely hamper vaccine efficacy. Moreover, we have demonstrated that this population is elevated in the blood of melanoma patients. Despite sharing some properties with BDCA1+ DCs, BDCA1+CD14+ cells are weak inducers of T cells, which could be attributed to their elevated expression of PD-L1. Thus, preparing BDCA1+CD14+ cell-free DC vaccines, and systemically neutralizing the immunosuppressive functions of this population may be crucial to warrant the success of antitumor DC-based vaccines.
Materials and Methods
Peripheral blood mononuclear cells (PBMC) were isolated from healthy donors or stage III or stage IV melanoma patients using Lymphoprep (Axis-Shield PoC AS). When indicated, the total BDCA1-expressing population (that includes BDCA1+ DCs and BDCA1+CD14+ population) was isolated from PBMCs using the BDCA1+ DC isolation kit (Miltenyi Biotec), followed by monocyte isolation by applying MACS using anti-CD14 microbeads (Miltenyi Biotec). The two BDCA1-expressing subsets were discriminated on the basis of CD14 expression, by using anti-CD14-PerCP (BD Biosciences).
To separate BDCA1+ DCs, BDCA1+CD14+ cells, monocytes, and MDSCs, we resorted to FACS, which was preceded by a depletion of T, B, and NK cells from PBMCs. First PBMCs were treated with an FcR blocker (Miltenyi Biotec) to avoid any aspecific binding of antibodies. Then, the PBMCs were treated with anti-CD3-FITC, anti-CD20-FITC, anti-CD56-FITC (all from BD Biosciences) and anti-CD19-FITC (Dako Cytomation) mAbs followed by treatment with anti-FITC microbeads (Miltenyi Biotec) and MACS. The depleted PBMCs were then prepared for FACS by labeling them with the following antibodies: lineage cocktail (lin1)-FITC (includes antibodies against CD3, CD19, CD20, CD56, CD14 & CD16), anti-CD14-PerCP, anti-HLA-DR-PE-Cy7 (all from BD Biosciences), and anti-BDCA1-PE (Miltenyi Biotec). The four subsets were isolated using a FACSAria II (BD Biosciences). The FACS gating strategy is displayed in Supplementary Fig. S1.
Naïve CD4+ T-cell population was isolated from PBMCs by first isolating the total CD4+ T-cell population using MACS CD4+ T cell isolation kit (Miltenyi Biotec). Subsequently, naïve CD45RA+CD45RO− CD4+ T cells were separated from memory T cells by applying anti-CD45RO-PE (Dako Cytomation) and anti-PE beads (Miltenyi Biotec) followed by MACS. Purity levels higher than 98% were achieved, determined by flow cytometry.
Ascites were collected from patients suffering from stage III or IV epithelial ovarian cancer of serous histology. Ascites was filtered over a 100-μm cell strainer (BD Biosciences) before separating mononuclear cells by centrifuging over a density gradient. To determine the percentages of BDCA1+ DCs and BDCA1+CD14+ cells in the ascites, the isolated mononuclear cells were stained with anti-CD45-FITC, anti-BDCA1-PE, anti-BDCA3-APC (all from Miltenyi Biotec), anti-CD19-PerCP, anti-CD16-PE-Cy7, and anti-CD14-APC (all from BD Biosciences) mAbs. Cells were acquired on a CyAnTM ADP flow cytometer (Beckman Coulter). All flow cytometry data in this study were analyzed using FlowJo software (Tree Star).
The phenotype of BDCA1+ DCs, BDCA1+CD14+ cells, and monocytes was compared by flow cytometry. The following antibodies were used to determine the phenotype: anti-CD1a-FITC (Biolegend), anti-HLA-DR-PE-Cy7, anti-CD11c-PE, anti-CD33-APC, anti-CD206-FITC (both from BD Biosciences), anti-CD16-APC (Miltenyi Biotec), and anti-CD209-PE (Beckman Coulter). Flow cytometry was performed using either FACSCalibur or FACS Verse (both from BD Biosciences).
The antigen uptake capacity of BDCA1+ DCs, BDCA1+CD14+ cells and monocytes was determined by measuring the uptake of Alexa Fluor 488–labeled BSA. A total of 1 × 105 of BDCA1+ cells (isolated by MACS as described above) or monocytes were cultured in X-VIVO 15 medium (Lonza) in the presence or absence of 100 μg/mL of BSA-Alexa Fluor 488 (Invitrogen) at 37°C for 5, 15, 30, or 60 minutes. Similar cultures were performed at 4°C to measure any background readings resulting from spontaneous binding of BSA. After the culture period, BDCA1+ cells were stained with anti-CD14-PerCP, to discriminate BDCA1+CD14+ cells from BDCA1+ DCs. Uptake of BSA-Alexa 488 by different cell subsets was measured by flow cytometry (FACSCalibur, BD Biosciences).
Cellular subset activation
After isolation, the cellular subsets were cultured in a round-bottom 96 well-plate (50 × 103 cells for cytokine detection and 10 × 103 for coculture with T cells) using X-VIVO 15 medium supplemented with 5% human serum (HS; Bloodbank Rivierenland). These cells were either left unstimulated or they were stimulated with 450 U/mL GM-CSF (CellGenix), Pam3CSK4 (10 μg/mL; EMC), 1 μg/mL lipopolysaccharide (LPS), 20 μg/mL pIC (both from Sigma-Aldrich), or a mix of 20 μg/mL pIC and 4 μg/mL R848 (Axxora), and incubated overnight at 37°C. The expression of costimulatory molecules after activation was determined using anti-CD80-APC and anti-CD86-APC (both from BD Biosciences) mAbs. Cytokine production by these subsets was determined by measuring IL6 (Sanquin Reagents), IL12 (Thermo Fisher Scientific), TNFα (BD Biosciences), and IL10 (eBioscience) using a standard sandwich ELISA in 24-hour supernatants.
Mixed lymphocyte reaction
The ability of the subsets to induce T-cell proliferation and cytokine production was tested in a mixed lymphocyte reaction (MLR). A total of 1 × 104 of unstimulated or stimulated cells (as above) were added to 1 × 105 freshly isolated allogeneic nonadherent peripheral blood lymphocytes (PBL) from a healthy donor. IFNγ and IL10 production by those T cells was determined in 48-hour supernatants by a standard sandwich ELISA. Proliferation was determined by [3H]-thymidine (MP Biomedicals) incorporation. The incorporated [3H]-thymidine was measured after 16 hours by liquid scintillation spectroscopy.
The capacity of the subsets to induce naïve CD4+ T-cell proliferation was determined by coculturing 1 × 104 of unstimulated or stimulated cells (as above) with 5 × 104 allogeneic naïve CD4+ T cells. When mentioned, anti-PD-L1 (R&D Systems) or isotype control antibodies were added. Proliferation of T cells was determined by [3H]-thymidine incorporation after 3 to 4 days of coculture.
To determine the CD4+ T-cell polarization capacity of the subsets, 1 × 104 of unstimulated or stimulated cells, were cultured with 4 × 104 allogenic naïve CD4+ T cells in the presence of 10 pg/mL Staphylococcus aureus enterotoxin B (SEB, Sigma-Aldrich). T-cell cultures were maintained in the presence of IL2 (20 IU/mL, Novartis) till T cells were resting around day 11. Resting T cells were restimulated by PMA (Calbiochem) and ionomycin (Sigma Aldrich) in the presence of brefeldin (Sigma Aldrich). The percentage of IFNγ+ and IL4+ cells was determined by intracellular staining using anti-IFNγ-PerCP.Cy5.5 and anti-IL4-PE (both from BD Biosciences).
To assess the Th17-polarizing capacity of the subsets, 2 × 104 cells of either subsets, isolated from melanoma patient PBMCs, were cultured with 5 × 104 autologous CD4+ T cells, in the presence of 0.5 μg/mL SEB for 18 hours with the additional presence of brefeldin for the last 3 hours. The percentage of IFNγ+ and IL17+ cells was determined by intracellular staining using anti-IFNγ-PerCP.Cy5.5 and anti-IL17-PE (eBioscience).
Monocyte and BDCA1+ DC differentiation assay
Peripheral blood monocytes were isolated by MACS-positive selection using CD14 microbeads (Miltenyi Biotec) after BDCA1+ cell depletion (Miltenyi Biotec). Peripheral blood BDCA1+ DCs were isolated by MACS-positive selection using BDCA1+ cell isolation kit after monocyte depletion with CD14 microbeads (Miltenyi Biotec). Purified CD14+ BDCA1− monocytes or BDCA1+ CD14− were seeded at 1 × 105 cells/well in a 48-well plate and cultured in X-VIVO medium supplemented with 40% HS from healthy donors or 40% serum from melanoma patients. At day 3, the cells were harvested and analyzed for the expression of BDCA1, CD14, CD11c, and HLA-DR by flow cytometry as described above.
Keyhole limpet hemocyanin–specific suppressor assay
BDCA1+ DCs, BDCA1+CD14+ cells, monocytes, and MDSCs were isolated by FACS from melanoma patients as mentioned above. A total of 2 × 104 cells of each subset, in triplicates, were cultured overnight at 37°C in X-VIVO 5% HS. Meanwhile, the extra monocytes were pulsed by 10 μg/mL keyhole limpet hemocyanin (KLH; Biosyn) overnight at 37°C. Autologous CD4+ population was also isolated by MACS CD4+ T cell isolation kit (Miltenyi Biotec) and kept in X-VIVO 5% HS at 4°C. The next day, the isolated CD4+ T cells, which include KLH-specific T cells, were stimulated by KLH-pulsed monocytes in the presence or absence of one of the four subsets. T-cell proliferation was determined by [3H]-thymidine incorporation after 24 to 48 hours of coculture. When indicated, anti-PD-L1 (R&D Systems) or isotype control antibodies were added.
BDCA1+ DCs, BDCA1+CD14+ cells, and monocytes were isolated by FACS and total RNA was extracted using TRIzol (Invitrogen), following manufacturer's protocol. The quality of the isolated RNA (concentration, RIN, 28S/18S and size) was determined by Agilent 2100 Bioanalyzer (Agilent Technologies). RNA sequencing, raw data analysis, and the generation of RPKM (Reads Per Kilobase of transcript per Million mapped reads) values of every gene were then performed by BGI TECH SOLUTIONS as described previously (17). Reads were aligned to human genome version 19. Further RNA sequencing data processing was done by first normalizing RPKM values across subsets based on total signals obtained. Thereafter, only genes with a RPKM value above 1 (see Supplementary Fig. S2 for determination of this cutoff) for at least one subset in all three donors were taken into account. Missing values were replaced by the lowest overall RPKM value obtained in the whole study. On the basis of RPKM values for each gene the expression in each cell type with respect to the donor mean of all three cell types was calculated to obtain relative values that were subsequently log2 transformed. RNA sequencing data are deposited at the Gene Expression Omnibus (GEO; accession number GSE75042).
Hierarchical clustering and principal component analysis
Relative log-transformed data were used as input for hierarchical clustering and principal component analysis (PCA). Similarity matrices and hierarchical clustering for specific gene sets were generated using Euclidian distance in GENE-E software using standard settings (Broad Institute, Cambridge, MA; http://www.broadinstitute.org/cancer/software/GENE-E/index.html). PCA analysis was performed in the MEV(4.9.0) software environment (http://www.tm4.org/mev.html), as described previously (18).
Gene-set enrichment analysis
De novo gene sets were generated for inflammatory DCs (infDCs) and CD14+ monocytes from processed and normalized microarray data, downloaded from the Gene Expression Omnibus website (GSE40484; http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE40484), and used as input for a 3-group ANOVA in the MEV(4.9.0) software environment. Differentially expressed genes (DEG) between groups (P < 0.05 after applying a Bonferroni correction for multiple testing) were assigned post hoc to a gene set for a particular cell type if the mean expression exceeded that of the other two cell types. In addition, we used the MoDC signature and BDCA1+ DC signature gene sets as previously generated by Segura and colleagues (19). The detailed composition of all these gene sets, as well as the possible overlap between them, is provided in Supplementary Data 1 and Supplementary Fig. S3, respectively. An irrelevant collection of gene sets (the C2 gene set from mSig database; http://software.broadinstitute.org/gsea/msigdb/collections.jsp#C2) was used in addition to the four test gene sets to evaluate the Normalized Enrichment Scores (NES) and the false discovery rate (FDR) more properly.
These gene sets were used as input for a Gene Set Enrichment analysis (GSEA), using the GSEA webtool version 2.2.1 (http://www.broadinstitute.org/gsea/index.jsp; refs. 20, 21). Genes were ranked by taking 2log expression ratios of classes, and enriched signatures were calculated using the weighted statistic. A minimum presence in the RNAseq data of 5 and a maximum of 500 genes was required for each gene set. A 1,000 permutations per gene set were performed. The GSEA analysis was visualized using Bubble GUM software (http://www.ciml.univ-mrs.fr/applications/BubbleGUM/index.html; ref. 22).
Multiplex (antibody microarray) analysis
To analyze the differential expression of CD markers and other relevant proteins among BDCA1+ DCs, BDCA1+CD14+ cells and monocytes these subsets were isolated as described above. The cells were washed, frozen, and sent to Sciomics GmbH for further analysis. In short, proteins were extracted by adding Scio-Extract buffer (Sciomics), quantified, and labeled with fluorescent dyes. All samples were analyzed in a reference-based dual color approach on Scio-CD Antibody microarrays (Sciomics) targeting 81 CD-marker as well as additional relevant cytokines, chemokines, and other proteins (Supplementary Table S1). Each antibody is represented in six replicates on the array. The arrays were blocked and each sample was incubated competitively with a reference sample on one array. Resulting data were analyzed using the linear models for microarray data (LIMMA) package of R-Bioconductor after uploading the mean signal intensities for differential protein expression.
Immunofluorescence staining was performed on 4-μm thick sections of paraffin-embedded resection specimens of healthy skin, skin and colon lesions of metastatic melanoma, and draining lymph nodes. Slides were dried overnight at 37°C. Next, they were deparaffinized in xylene, followed by graded alcohols and finally tap water. Heat-induced antigen retrieval was performed in 10 mmol/L citrate acid buffer (pH 6.0) by microwaving for 15 minutes after boiling. Blocking was performed with 1% BSA in TBST for 10 minutes at room temperature. Samples were stained by Opal TSA Plus multiplex tissue staining kits (Perkin Elmer) according to the standard protocol provided. Slides were subsequently stained with mouse anti-CD14 (IgG2a, Leica Biosystems) labeled with Cyanine 5 fluorophore, mouse anti-HLA-DR (IgG2b, Thermo Fisher Scientific) labeled with Cyanine 3 fluorophore, mouse anti-CD1c (BDCA1, IgG1, Abcam) labeled with fluorescein, and a cocktail to identify tumor cells (melanoma mix) consisting of mouse anti-HMB45 (IgG1, Dako), mouse anti-Mart-1 (IgG1, Thermo Immunologic), mouse anti-Tyrosinase (IgG2a, Monosan), and rabbit anti-SOX-10 (IgG, Cell Marque) labeled with Cyanine 3.5. BrightVision poly-HRP-anti Ms/Rb/Rt IgG (Immunologic BV) was used for the secondary antibody incubations. Finally, slides were counterstained and mounted using DAPI Fluoromount-G (Southern Biotechnology Associates, Inc). Whole slide multispectral scan was performed using the Vectra slide scanner (version 2.0.7; PerkinElmer). Multispectral images were unmixed using spectral libraries built from images of single stained tissues for each reagent and unstained tissue using inForm (version 2.1.1; PerkinElmer). A selection of 20 to 30 representative original multispectral images of the particular tissue type (skin, colon, or lymph node) were used to train inForm for quantitative image analysis; segmentation of tumor/stromal tissue based on cyanine 3.5 signal, cell segmentation based on DAPI and membrane stain signals (HLA-DR, CD14 and BDCA1), and finally phenotyping of different cell types. All settings applied to the training images were saved within an algorithm. Vectra Review (Version 2.0.8, PerkinElmer Inc.) was used to select the areas for analysis; this consisted of the tumor and stromal tissue around it or all stromal tissue of control tissues. Batch analysis of the tumor and control tissues was performed with the same algorithm but separate algorithms were generated between tissue types.
FACS-sorted cells were subjected to cytospin and stained with May-Grünwald/Giemsa. In short, slides were fixed in methanol for 10 minutes and stained with May-Grünwald (Merck) for 5 minutes, washed, and stained with Giemsa (Merck) for 15 minutes, respectively. Pictures were taken with a Leica DM6000 B microscope equipped with a Leica DFC 480 camera (63× objective), using Leica Application suite V4.3.0.
Student t tests were performed for paired measurements with GraphPad Prism software (GraphPad La Jolla, CA). Values of P < 0.05 were considered significant.
BDCA1+CD14+ cell population is associated with lower response toward BDCA1+ DC–based vaccines
In a recent clinical trial performed by our group, BDCA1+ DC vaccines were utilized to treat melanoma patients (16). The vaccine preparations included another BDCA1+ population coexpressing CD14, in addition to BDCA1+ DCs (Fig. 1A). The frequency of this BDCA1+CD14+ population varied between 6% and up to 45% of the whole vaccine preparation. The BDCA1+ cells used in the vaccine were loaded with melanoma-specific antigens, as well as KLH as an immunogenic control antigen (23). Interestingly, we observed that patients who received vaccines with BDCA1+CD14+ cellular content below 25% demonstrated higher T-cell responses to KLH when compared with patients who received vaccines with BDCA1+ CD14+ cellular content above 25% (Fig. 1B). This result indicates that the presence of this cell population may hamper the induction of an effective antitumor immune response, possibly by active immunosuppression. To test this hypothesis, we compared the suppressive qualities of BDCA1+CD14+ cells, monocytes, BDCA1+ DCs, and MDSCs (defined as CD14+HLA-DRlo/−; Supplementary Fig. S1A). Only melanoma patient–derived MDSCs were able to suppress αCD3/αCD28–induced proliferation of autologous CD4+ T cells (Supplementary Fig. S4A). However, when KLH-specific CD4+ T cells were stimulated with KLH-loaded monocytes, only autologous BDCA1+CD14+ cells were able to suppress the proliferation of these T cells (Fig. 1C; left panel depicts the proliferation fold change of 3 patients and the right panel depicts the individual data of these patients; Supplementary Fig. S4B). Thus, BDCA1+CD14+ cells are capable of suppressing T-cell responses in an antigen-specific manner.
Tumors are infiltrated by cells resembling the blood BDCA1+CD14+ myeloid population
To determine whether progressive tumors may have an influence on BDCA1+CD14+ cell counts in the circulation, the frequency of this population was assessed in the peripheral blood of stage III/IV melanoma patients and healthy donors by flow cytometry (Supplementary Fig. S1). Interestingly, we noticed that the frequency of BDCA1+CD14+ cells was significantly elevated in melanoma patients in comparison with healthy donors (Fig. 2A). This increase concurred with an already reported rise in circulating MDSCs (24, 25). Also, a trend of lower BDCA1+ DC frequency was observed in melanoma patients (Fig. 2A). Cells coexpressing CD14 and BDCA1 were also detected in metastatic melanoma lesions from skin, their draining lymph nodes, and from colon metastases (Fig. 2B). Interestingly, the melanoma skin metastatic lesions contained significantly higher numbers of CD14+ (defined as HLA-DR+BDCA1−CD14+) and BDCA1+CD14+ (defined as HLA-DR+BDCA1+CD14+) cells in comparison with healthy skin, whereas the numbers of BDCA1+ cells (defined as HLA-DR+BDCA1+CD14−) did not change (Fig. 2C). There was a tendency of increased numbers of those three populations in metastatic melanoma lesions (lymph nodes and colon), in comparison with healthy tissues (Supplementary Fig. S5). Moreover, BDCA1+CD14+ cells were also detected in inflammatory tumor ascites from ovarian cancer patients, where they were even significantly more abundant than BDCA1+ cells (Fig. 2D), as previously reported (19).
Although the existence of a CD14+ DC subset has been described in the skin dermis (26), in the synovial fluid of inflamed arthritic joints and in ascites from breast or ovarian cancer patients (19), such a population has never been investigated in blood or in relation to melanoma. Therefore, we first analyzed the morphology of BDCA1+CD14+ cells in relation to BDCA1+ DCs and monocytes. Microscopic analysis, following May-Grünwald/Giemsa staining, revealed resemblance between BDCA1+CD14+ cells and BDCA1+ DCs, as demonstrated by the dendrites and the shape of the nucleus (Fig. 2E and larger view with multiple cells in Supplementary Fig. S6). Furthermore, we determined the phenotype of this population in comparison with BDCA1+ DCs and monocytes. In addition to the high expression of CD11c and HLA-DR, typical for BDCA1+ DCs, the BDCA1+CD14+ population also expressed the monocytic marker CD11b at higher levels than BDCA1+ DCs, yet lower than monocytes (Fig. 2F). Moreover, the BDCA1+CD14+ population lacked CD16 expression that is characteristic for a subset of nonclassical monocytes (27) and a CD16+ subset of DCs (28). All three populations shared high expression of the myeloid marker CD33, and virtually lacked CD1a, CD209, and CD206 expression (Fig. 2F). Collectively, these data show that BDCA1+CD14+ population shares phenotypical markers of both BDCA1+ DCs and monocytes.
Transcriptome and multiplex analyses of BDCA1+CD14+ cells identifies them as a distinct population
To further investigate the relation of the BDCA1+CD14+ population to monocytes and BDCA1+ DCs, we compared the transcriptomes of these three cell types isolated from the blood of three healthy individuals using a sensitive RNA sequencing approach. A similarity matrix and PCA of the subsets clearly singled out BDCA1+CD14+ cells as a unique population that is most closely related to monocytes (Fig. 3A). To gain more insight into the origin and function of BDCA1+CD14+ cells, the subsets were clustered on the basis of the expression of specific gene groups. Hierarchical clustering of the three subsets, based on the expression of myeloid lineage–specific transcription factors and growth factor receptors (29), suggested a close relation between BDCA1+CD14+ cells and monocytes (Fig. 3B). RNA expression of IRF4, IRF8, BATF3, ZBTB46, and FLT3, all crucial for DC development in mice (29, 30), was highest in BDCA1+ DCs. BDCA1+CD14+ cells, however, did demonstrate higher expression of these genes when compared with monocytes. The long noncoding RNA LOC645638 (Inc-DC), specifically expressed by DCs (31), was almost exclusive to BDCA1+ DCs (Fig. 3B). In contrast, expression of transcription factors involved in macrophage differentiation in mice, MAFB, EGR1, EGR2 was similar between BDCA1+CD14+ cells and monocytes, yet much lower in BDCA1+ DCs (Fig. 3B). Interestingly, the expression of the growth factor receptor CSF1R, characteristic for macrophages, was highest in BDCA1+CD14+ cells.
To substantiate that BDCA1+CD14+ cells are a distinct cell type, we next addressed the functional (dis)similarities between the three cell types concerning pattern recognition, antigen uptake, and antigen presentation. BDCA1+CD14+ cells proved most related to monocytes when considering the expression of toll-like receptors (TLR), C-type lectins, and Fc receptors (Fig. 3C). However, the expression of antigen-presenting class II HLA molecules and CD1 molecules was similar to that of BDCA1+ DCs (Fig. 3C). Taken together, these data further support the notion that BDCA1+CD14+ cells form a distinct population with a unique gene expression profile.
Recently, Segura and colleagues (19) described the presence of an inflammatory DC (infDC) resembling in vitro–generated moDCs in inflammatory fluids. To find out whether the BDCA1+CD14+ cells are homologous to these inflammatory cells, we performed a gene-set Enrichment analysis (GSEA). We retrieved three signatures described in Segura and colleagues for moDCs, BDCA1+ DCs, and monocytes. In addition, because a ready-to-use signature for the infDCs was lacking, we generated a gene set for these cells using the published microarray data. In parallel, we also generated gene sets for monocytes and BDCA1+ DCs from these data. Together, this yielded 6 gene sets (3 preexisting and 3 newly generated signatures). Unfortunately, 2 gene sets (the existing monocyte signature and the newly generated BDCA1+ DC high set) were not covered in our RNA sequencing data with more than 5 genes and thus excluded. The remaining 4 sets were analyzed with GSEA (Table 1; raw GSEA data in Supplementary Fig. S7; Supplementary Table S2; refs. 20–22). As expected, the BDCA1+ DC signature was enriched in BDCA1+ DCs with respect to monocytes and vice versa. Genes reflecting moDCs or infDC gene sets were more enriched in BDCA1+ DCs with respect to monocytes. The BDCA1+CD14+ cells displayed clear enrichment of BDCA1+ DC genes with respect to monocytes and enrichment of monocyte genes when compared with BDCA1+ DCs, indicating again these cells share gene expression with both cells types. The infDC signature was only enriched in BDCA1+CD14+ cells when compared with monocytes but not when compared with BDCA1+ DCs, indicating the BDCA1+CD14+ cells do not resemble infDCs more than BDCA1+ DCs. The moDC gene set was not found significantly enriched in any of the comparisons made. This indicates that BDCA1+CD14+ cells share commonalities with different types of DCs (Table 1). As a second approach, we used the genes from the 4 gene sets, used for GSEA, for hierarchical clustering of our RNA sequencing data. As expected the monocyte and BDCA1+ DC gene sets were clearly expressed highest in monocytes and BDCA1+ DCs, respectively. On the other hand, BDCA1+CD14+ cells clustered with monocytes for these gene sets again (Fig. 3D). MoDC signature genes were not highly expressed in any of the three cell types, indicating that moDCs do not resemble one cell type in particular but rather share elements with all three of them. Finally, the set of genes highest expressed by infDCs with respect to monocytes and BDCA1+ DCs in the study by Segura and colleagues (19), separated into two distinct gene clusters, a predominant cluster with highest expression in BDCA1+ DCs and a minor cluster overexpressed in monocytes. In BDCA1+CD14+ cells, moderate to high expression levels were observed for the infDC gene set, but contrasting the infDC reported by Segura and colleagues, these genes were hardly highest expressed in BDCA1+CD14+ cells but rather predominant in either monocytes or BDCA1+ DCs. Overall, the enrichment profile of infDC genes in BDCA1+CD14+ cells thus further corroborates the “in-between” BDCA1+ DCs and monocytes profile of BDCA1+CD14+ population, we find by both hierarchal clustering and PCA (Fig. 3A). Together, these analyses demonstrate that the BDCA1+CD14+ cells most resemble monocytes but also share some degree of homology with BDCA1+ DCs as well as infDCs. Hence, the precise relationships between blood BDCA1+CD14+ cells and monocytes, moDCs, BDCA1+ DCs, or infDCs remain to be investigated further, keeping in mind that imprinting by the microenvironment of development and/or residency has a strong impact on the gene expression of any given cell type and complicates alignment of cell subsets between tissues or when comparing steady-state versus inflamed conditions.
To further complement the RNA expression analysis, we studied the differential protein expression of CD molecules, HLA molecules, chemokines, and cytokines (Supplementary Table S1) in the three cell populations using antibody microarrays (Multiplex analysis). Hierarchical clustering of differentially expressed protein revealed a similar pattern to RNA expression data with BDCA1+CD14+ cells clustering closer to monocytes (Fig. 3E). Thus, protein analysis further supports the uniqueness of the BDCA1+CD14+ population.
BDCA1+CD14+ cells possess DC functional features
Although RNA and protein data insinuate a relation between BDCA1+CD14+ cells and both BDCA1+ DCs and monocytes, similarities in these two subsets require further functional validation. DCs are by definition professional antigen-presenting cells that can mount adaptive immune responses by activating T cells. Therefore, we initially assessed the capacity of BDCA1+CD14+ cells, isolated from peripheral blood of healthy donors, to take up antigens from the surrounding environment. With respect to BDCA1+ DCs, BDCA1+CD14+ cells displayed a significantly higher uptake of fluorescently labeled BSA, which was already apparent after 15 minutes of incubation (Fig. 4A, left). Monocytes were the best in antigen uptake, although the differences between BDCA1+CD14+ cells and monocytes were insignificant (Fig. 4A, right).
A major hallmark of DCs is their ability to sense danger signals and to respond by expressing costimulatory molecules and secreting cytokines, a process also referred to as maturation (32). Therefore, we determined to what extent BDCA1+CD14+ cells can mature in response to LPS (a TLR4 ligand) and pIC (a TLR3 ligand). In addition, we stimulated cells with GM-CSF, commonly used in cancer immunotherapy. BDCA1+ DCs responded to both LPS and pIC by upregulating the costimulatory molecules CD86 and CD80. BDCA1+CD14+ cells also responded to both stimuli, yet surprisingly only upregulated CD80 expression. Monocytes on the other hand modestly upregulated CD80 expression only in response to LPS and less efficiently than the other two subsets (Fig. 4B). Furthermore, all three populations responded to TLR stimulation by cytokine secretion. The BDCA1+CD14+ population displayed a unique cytokine profile with significantly higher amounts of IL6 and TNFα in comparison with BDCA1+ DCs. In addition to those proinflammatory cytokines, TLR-activated BDCA1+CD14+ cells were superior in producing the anti-inflammatory cytokine IL10 (Fig. 4C). This cytokine production profile was similar upon triggering TLR2 and TLR8 by Pam3CSK4 and R848, respectively (Supplementary Fig. S8). As for IL12 secretion, characteristic for BDCA1+ DCs, it was lacking in the two other subsets (Supplementary Fig. S9). Collectively, BDCA1+CD14+ cells are endowed with DC functions, as demonstrated by a high antigen uptake capacity, as well as the ability to mature in response to TLR stimulation.
The BDCA1+CD14+ population is capable of inducing T-cell responses that are dampened by PD-L1 expression
To determine whether BDCA1+CD14+ cells are capable of inducing T-cell responses, the T-cell–stimulatory capacity of this population, isolated from the peripheral blood of healthy donors, was compared with that of BDCA1+ DCs and monocytes in two settings. The first setting is a mixed leukocyte reaction (MLR) with allogenic PBLs. In agreement with their phenotypical maturation, the ability of BDCA1+CD14+ cells to induce PBL proliferation was higher than that of monocytes and lower than that of BDCA1+ DCs, regardless of the type of treatment (Fig. 5A). The levels of IFNγ in the supernatant of MLR cultures containing LPS-stimulated BDCA1+CD14+ cells were low in comparison with BDCA1+ DCs, yet increased to similar levels after pIC stimulation. On the other hand, IL10 was produced at comparable levels by PBLs stimulated by any of the cellular subsets, although pIC-stimulated monocytes triggered lower IL10 production by PBLs in comparison with BDCA1+ DCs and BDCA1+CD14+ cells (Fig. 5A).
In a more refined setting, we determined their capacity to induce the proliferation of allogenic naïve CD4+ T cells. Similar to the MLR setting, BDCA1+CD14+ cells were better at inducing CD4+ T-cell proliferation than monocytes, but lesser than BDCA1+ DCs (Fig. 5B). To determine the T-cell polarizing capacity of the three subsets, naïve CD4+ T cells were used. To our surprise, BDCA1+CD14+ cells were as efficient as BDCA1+ DCs, or even better after pIC stimulation, in inducing IFNγ-producing T cells. This was accompanied by a higher induction of IL4-producing T cells (Supplementary Fig. S10A). To reveal whether BDCA1+CD14+ cells are capable of inducing IL17-producing CD4+ T cells, we applied the same experimental setup described by Segura and colleagues (19). In this setup, BDCA1+CD14+ cells, BDCA1+ DCs, and monocytes that were isolated from the peripheral blood of melanoma patients failed to induce noteworthy amounts of IL17-secreting CD4+ T cells (Supplementary Fig. S10B). Thus, BDCA1+CD14+ cells can trigger T-cell proliferation more efficiently than monocytes, albeit to a lesser extent, in comparison with conventional BDCA1+ DCs, which maintain the lead in inducing functional T cells.
The observed differences in T-cell stimulatory capacity between BDCA1+ DCs and BDCA1+CD14+ cells could be attributed to the lower expression of costimulatory molecules that are vital for T-cell activation (Fig. 4B), but could also be mediated through coinhibitory molecules (33), such as PD-L1 and PD-L2. We did not detect PD-L2 expression on any of the subsets, with or without stimulation (data not shown). PD-L1, however, was expressed at higher levels by BDCA1+CD14+ cells in comparison with BDCA1+ DCs and monocytes (Fig. 5C, left), suggesting it may indeed have a role in their T-cell stimulatory capacity. Indeed, neutralizing PD-L1 molecules, using a blocking antibody, significantly enhanced BDCA1+CD14+ cell–induced T-cell proliferation (Fig. 5C, right). In contrast, blocking the effects of IL10, produced by BDCA1+CD14+ cells did not yield any effects on T-cell proliferation (Supplementary Fig. S11). Interestingly, blocking PD-L1 in the KLH-specific suppressor assay did not reverse the observed suppressive effect exerted by BDCA1+CD14+ cells (Supplementary Fig. S12). Thus, these results demonstrate that a higher PD-L1 expression by BDCA1+CD14+ cells plays a role in hampering the functionality of this cell population.
Systemic cancer–induced factors drive the differentiation of monocytes and BDCA1+ DCs into BDCA1+CD14+ cells
The conspicuous increase in the percentage of circulating BDCA1+CD14+ cells in melanoma patients (Fig. 2A) suggests that tumor-associated factors could invoke the emergence of this population from another precursor population. Monocytes are well known for their plasticity and capacity to differentiate into several cellular types, and may therefore be the most plausible candidate as the ancestral population of BDCA1+CD14+ cells. We tested this hypothesis by culturing monocytes, isolated from the peripheral blood of healthy donors, in the presence of melanoma patient–derived serum. Monocytes cultured in serum obtained from healthy individuals did not express any BDCA1, whereas, and in line with our hypothesis, melanoma patient–derived serum readily stimulated the expression of BDCA1 on the cultured monocytes (Fig. 6A). Similar to BDCA1+CD14+ cells found in melanoma patient blood, the ex vivo–generated cells also displayed reduced autofluorescence (tinted histograms) and maintained the expression of CD14. The serum of the melanoma patients also increased CD11c and HLA-DR expression, thus leading to a cell surface profile similar to that of primary BDCA1+CD14+ cells isolated from circulation. Interestingly, using the same setup with BDCA1+ DCs instead of monocytes, similar results were obtained with a subpopulation clearly coexpressing BDCA1 and CD14 (Fig. 6B). Taken together, these data indicate that the BDCA1+CD14+ population may have a mixed origin, emerging from either monocytes or BDCA1+ DCs, due to tumor-related factors, representing yet another, but unique, mode of tumor-related immune suppression.
Tumor-induced immunosuppression is the Achilles heel of both naturally arising immune responses and responses induced by anticancer immunotherapeutics. Tumors can suppress at many levels. They have been suggested to disrupt chemokine circuits that are pivotal for T-cell attraction (34, 35), can manipulate vascular endothelium to form a physical barrier in the face of tumor-reactive T cells (36), and can actively suppress these cells (37). Once T cells manage to transmigrate through the endothelial barrier, they have to negotiate a hostile immunosuppressive tumor stroma before encountering tumor cells. Tumors recruit and promote the expansion of immunosuppressive leukocyte populations like Treg cells (3), MDSCs (4), and M2 macrophages (5). In this study, we characterized another immunosuppressive element, the BDCA1+CD14+ cells. Although we identified and characterized this population in peripheral blood, cells with a similar phenotype as assessed by several markers were readily found in tumor tissue and in ovarian cancer ascites fluid. Similar to MDSCs (38), we found that this population is significantly elevated in peripheral blood of stage III and stage IV melanoma patients, supporting the notion that tumors promote a systemic accumulation of suppressive elements (6). A common feature between MDSCs and BDCA1+CD14+ cells is the expression of CD14, which may imply a joint origin. However, these two suppressive populations vary in the mode of suppression. Whereas MDSCs universally suppress the proliferation of bystander T cells, BDCA1+CD14+ cells uniquely suppress T cells in an antigen-specific manner. This difference may be linked to the high expression levels of HLA-DR on BDCA1+CD14+ cells, which is by definition low on MDSCs.
Transcriptome analysis revealed a possible relation between BDCA1+CD14+ cells and macrophages as demonstrated by the expression of genes that are pivotal for macrophage differentiation in mice, such as CSF1R (29). Nevertheless, CSF1R was also reported to be highly expressed by infDCs in both mice (39) and humans (19). Macrophages are by definition tissue-resident cells, whereas BDCA1+CD14+ cells are found in both circulation and tissues, highlighting another discrepancy between the two cell subsets and arguing against classifying them in the same group. Yet the similarities revealed by transcriptome data urge for further delineation of the link between these two subsets.
CD14 expression is not restricted to monocytes. Also, DCs in certain compartments or under certain conditions express CD14. Indeed, steady-state human dermis harbors a CD14+ subset of DCs (DDC; ref. 26). Similar to BDCA1+CD14+ cells, CD14+ DDCs express BDCA1 allowing their distinction from dermal macrophages (40). In analogy to BDCA1+CD14+ cells, CD14+ DDCs express low levels of costimulatory molecules rendering them poor inducers of T-cell proliferation (41). Furthermore, CD14+ DDCs have been implicated in the induction of immune tolerance (42, 43). We demonstrate that BDCA1+CD14+ cells could be generated from monocytes under the effect of melanoma patient serum. Similarly, a recent study showed that human CD14+ DDCs could be generated from circulating monocytes following coculture with primary human endothelial cells (44), adding yet another layer of resemblance to circulating BDCA1+CD14+ cells and strongly suggesting a link between the two populations. In addition to CD14+ DDCs, upregulated CD14 expression by human MoDCs is also observed following treatment with Mycobacterium tuberculosis (45), vitamin D (46), dexamethasone (47), and IL10 (48). The common feature of these CD14+ MoDCs is the low expression of costimulatory molecules, poor induction of T-cell responses, and promotion of immune tolerance. Our in vitro experiments show that the BDCA1+CD14+ cells are in some aspects related to monocytes or may be even derived from them. Yet, their resemblance to primary blood BDCA1+ DCs by transcriptome analysis, together with their in vitro antigen presentation capacities, suggests that they are, in fact, DC-like cells. Together with their immunosuppressive capacity, our data argue that BDCA1+CD14+ cells represent a circulating suppressive DC-like cell population, which may arise from either monocytes or BDCA1+ DCs.
Tumor-associated immune suppression functions in part through immune-inhibitory molecules that sabotage tumor-specific immune responses. Among these inhibitory molecules is the PD-1/PD-L1 pathway. High PD-L1 expression levels by tumor cells, tumor-infiltrating lymphocytes, or both, are associated with aggressive tumor behavior, poor prognosis, and elevated risk of mortality (2). Interestingly, the PD-L1/PD-1 checkpoint can be targeted by several clinical-grade antibodies currently being tested. Thus far, clinical trials targeting PD-1 (49, 50) or PD-L1 (51) by blocking antibodies not only proved tolerable, but showed durable antitumor responses in up to 50% of all treated melanoma patients. In this respect, checkpoint inhibitors may act on PD-L1 expressed by both tumor cells and BDCA1+CD14+ cells.
Collectively, we characterized a novel cell subset, which may contribute to the systemic state of immunosuppression in cancer patients, and thus impairing the efficacy of cancer immunotherapy. Thus, a direct opportunity to enhance the efficacy of DC vaccination lies in the depletion of BDCA1+ DC-based vaccines of suppressive BDCA1+CD14+ cells and systemically annihilating the suppressive functions of this population.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: G. Bakdash, S.V. Hato, E.L. Smits, C.G. Figdor, I.J.M. de Vries
Development of methodology: G. Bakdash, S.V. Hato, C.G. Figdor, I.J.M. de Vries
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Bakdash, M.A.J. Gorris, A. Halilovic, A.E. Sköld, G. Schreibelt, S.P. Sittig, R. Torensma, T. Duiveman-de Boer, C. Schroder, E.L. Smits
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Bakdash, S.I. Buschow, M.A.J. Gorris, A. Halilovic, T. Duiveman-de Boerm, C. Schroder, C.G. Figdor, I.J.M. de Vries
Writing, review, and/or revision of the manuscript: G. Bakdash, S.I. Buschow, M.A.J. Gorris, A. Halilovic, G. Schreibelt, S.P. Sittig, R. Torensma, E.L. Smits, C.G. Figdor, I.J.M. de Vries
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Bakdash, A. Halilovic, G. Schreibelt
Study supervision: G. Bakdash, C.G. Figdor, I.J.M. de Vries
The authors wish to thank Dr. Angela Vasaturo, Drs. Inge Reinieren-Beeren, and Rob Woestenenk for valuable technical assistance.
This work was supported by grants from the Netherlands Organization for Scientific Research (NWO 951.03.002 and NWO-Vici 016.140.655), grants from the Dutch Cancer Society (KWF2009-4402, KWF2010-4722, KWF2013-5958), a Radboud UMC PhD grant, and the Swedish Research Council and lmmunoTools GmbH (Germany) for Multiplex award.
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.