Research ArticleTUMOR IMMUNOLOGY

IFN-III is selectively produced by cDC1 and predicts good clinical outcome in breast cancer

See allHide authors and affiliations

Science Immunology  17 Apr 2020:
Vol. 5, Issue 46, eaav3942
DOI: 10.1126/sciimmunol.aav3942

cDC1 and cancer

The conventional dendritic cell 1 (cDC1) subset has been shown to play a critical role in antitumor responses, and here, Hubert et al. show that human cDC1 contribute to antitumor responses through production of type III interferon (also known as IFN-λ). They analyzed primary breast tumors and publicly available transcriptomic data and pinpointed IFN-λ1 production by cDC1 as being associated with favorable patient outcomes, and this cytokine promoted a microenvironment with TH1-associated cytokines and chemokines. By treating breast cell tumor suspensions with a TLR3 agonist, they were able to induce TH1-polarized responses and IFN-λ1 production by cDC1. These findings provide insight into how cDC1 cells contribute to antitumor immunity and how they might be targeted for potential therapeutic strategies.

Abstract

Dendritic cells play a key role in the orchestration of antitumor immune responses. The cDC1 (conventional dendritic cell 1) subset has been shown to be essential for antitumor responses and response to immunotherapy, but its precise role in humans is largely unexplored. Using a multidisciplinary approach, we demonstrate that human cDC1 play an important role in the antitumor immune response through their capacity to produce type III interferon (IFN-λ). By analyzing a large cohort of breast primary tumors and public transcriptomic datasets, we observed specific production of IFN-λ1 by cDC1. In addition, both IFN-λ1 and its receptor were associated with favorable patient outcomes. We show that IFN-III promotes a TH1 microenvironment through increased production of IL-12p70, IFN-γ, and cytotoxic lymphocyte–recruiting chemokines. Last, we showed that engagement of TLR3 is a therapeutic strategy to induce IFN-III production by tumor-associated cDC1. These data provide insight into potential IFN- or cDC1-targeting antitumor therapies.

INTRODUCTION

The use of antitumor immunotherapies such as monoclonal antibodies targeting immune checkpoints (ICPs) has provided promising results for the treatment of several cancers. Despite favorable outcome of responding patients, the overall response rate remains relatively low, and an ongoing challenge is the identification of new immunotherapy targets. Hence, dendritic cells (DCs) represent promising targets owing to their central role in the initiation and the control of immune responses. Their functions encompass a wide range of mechanisms and responses mediated by different subsets, namely: the plasmacytoid DCs (pDCs), the two subsets of classical/conventional DCs (cDCs) called CD141/BDCA3high cDC1 and CD1c/BDCA1+ cDC2, as well as Langerhans cells (LCs), which display DC functions although they belong to the macrophage lineage.

The cDC1 population is of particular interest because of their role in the activation of cytotoxic antitumor responses. In mice, these cells were shown to act as professional antigen (Ag) cross-presenting cells to CD8+ T cells and are considered to be essential for the induction of antitumor immunity (14) and responses to immunotherapies (58). The superiority of cDC1s relative to other DCs to activate cytotoxic immune responses through Ag cross-presentation has also been shown in humans (914), but their role in antitumor immunity is largely understudied unlike that in mice (15). This DC subset was identified in several tumors (3, 1618), and transcriptomic analyses revealed that a high cDC1 infiltration score is associated with favorable patient prognoses (3, 4, 1820), as well as improved clinical responses to anti–PD-1 (programmed cell death 1/CD279) therapy in small cohorts of patients with metastatic melanoma (19). However, the mechanisms underlying the impact of human cDC1 in patient outcomes have not been elucidated.

We previously showed that human cDC1 represent a major source of type III interferon [IFN-III; also called IFN-λ1/2/3 or interleukin-29/28A/28B (IL-29/28A/28B)] produced in response to Toll-like receptor 3 (TLR3) engagement (21, 22). IFN-III shares the same signaling pathway as IFN-I, leading to the transcription of multiple interferon-stimulated genes (ISGs). Similar to IFN-I, IFN-III play a crucial role in autoimmune diseases (23) and viral infections (24). Their antitumor activity has also been reported in several mouse models (2528). In humans, antiproliferative (29, 30) and proapoptotic (31, 32) activities of IFN-III have only been demonstrated in vitro. Owing to their specificity of action on a narrow range of cell types, such as epithelial cells and some immune populations, IFN-III–based therapies may potentially lead to fewer toxic side effects compared with IFN-I–based treatments. Thus, investigating the potential production of IFN-III by tumor-associated cDC1 (TA-cDC1) is crucial to unveiling the mechanisms underlying their protective role in antitumor immunity.

Here, we demonstrate that IFN-III is selectively produced by cDC1 in human tumors and is associated with favorable outcomes in breast cancer. Furthermore, we describe the association between cDC1-derived IFN-III and the presence of crucial cytokines and chemokines, including IL-12p70, IFN-γ, CXCR3-L, and CX3CR1-L, which promote effector T cell recruitment and activation. Last, we propose that TLR3 activation of intratumoral cDC1 could be used as a potential therapeutic strategy leading to IFN-III, as well as type 1–related cytokine and chemokine production. These data support the development of therapies targeting cDC1 to trigger IFN-III release resulting in a cytokine microenvironment conducive to the induction of cytotoxic antitumor immune responses.

RESULTS

cDC1 are enriched in breast tumors compared with peripheral blood

To evaluate the infiltration of primary breast tumors by DC populations, we performed multiparametric flow cytometry analysis of freshly dissociated and digested tissues. Among human leukocyte antigen DR–positive (HLA-DR+) lineage cells, four discrete TA-DC populations were distinguishable in most tumors (Fig. 1A), namely: pDCs (CD11c CD123+), cDC1 (CD11c+ BDCA1 BDCA3hi), cDC2 (CD11c+ BDCA1+ CD207), and LCs (CD11c+ BDCA1+ CD207hi). These DC phenotypes were similar to DC subsets previously found in several noncancerous tissues (33) and lung tumors (17, 18). With the exception of LCs that are rarely present in blood, all other DC subsets were detected among peripheral blood mononuclear cells (PBMCs) of patients (fig. S1A). We assessed the phenotype of TA-DCs and blood DCs (Fig. 1B and fig. S1B) and observed that classical markers such as BDCA2/CLEC4C and CD11b were expressed by pDCs and cDC2, respectively. As in blood, signal regulatory protein α (SIRP-α), which is used by CD47-expressing tumor cells to inhibit phagocytosis by antigen-presenting cells (34), was expressed by all TA-DCs except cDC1. In contrast, expression of CLEC9A and XCR1 was restricted to TA-cDC1, reinforcing the selection of these two proteins as cDC1-specific markers. cDC1 also expressed the highest level of BTLA (B- and T-lymphocyte attenuator), but expressed lower levels of DC lysosome-associated membrane protein (DC- LAMP)/CD208 compared with cDC2 and LCs, suggesting a moderate maturation stage (Fig. 1B and fig. S2A). Last, an unsupervised viSNE analysis highlighted the homogeneity of the CLEC9A+ cluster, representing the cDC1 subset in tumor and blood (Fig. 1C and fig. S1C), whereas cDC2 and LC populations were more heterogeneous.

Fig. 1 cDC1 infiltrate human breast tumors.

(A) Fluorescence-activated cell sorter (FACS) gating strategy allowing the identification of four TA-DC subsets among viable CD45+ HLA-DR+ Lineage (CD3/14/15/19/56) populations, namely: pDCs (CD11c CD123+), cDC1 (CD11c+ BDCA1 BDCA3hi), cDC2 (CD11c+ BDCA1+ BDCA3−/low CD207), and LCs (CD11c+ BDCA1+ BDCA3−/low CD207hi). Representative results of n = 21 breast tumors. FSC, forward scatter. (B) Phenotypic characterization of TA-DC subsets for indicated markers by FACS. Color histogram, indicated marker; dashed line, isotype control. Representative results of n > 8 breast tumors. (C) viSNE analysis of viable CD45+ HLA-DR+ Lin for the expression of CD11c, CD123, BDCA1, BDCA3, CD207, CLEC9A, CD11b, SIRP-α, and BTLA markers, color-coded according to the relative expression of markers (top), with populations indicated. Representative result of n = 4 breast tumors.

Flow cytometry analyses were conducted on a larger cohort of 90 patients harboring all subtypes of breast tumors. As expected, the CD45+ infiltrate varied greatly between tumors (fig. S2B). Although cDC1 constitute a discrete population, they were always detectable within CD45+ cells (Fig. 2, A and B). Unexpectedly, the ratio of cDC1 to all TA-DCs was markedly higher in tumors compared with blood (3.5-fold increase), unlike that of other DC subsets (Fig. 2C), highlighting their likely selective recruitment within tumors and suggesting a potential role of cDC1 in antitumor immunity. By focusing on breast tumor subtypes, we demonstrated a preferential cDC1 infiltration in triple-negative breast cancers (TNBCs), which are the most aggressive breast cancers and display the highest level of immune cell infiltration (fig. S2C). TNBCs also displayed a larger proportion of cDC2 and pDCs compared with other breast cancer subtypes, although this difference was not marked, likely because of a small sample size (fig. S2, D and E). Conversely, the SBR (Scarff-Bloom and Richardson) grade did not appear to influence the proportion of TA-cDC1, although this may also reflect the small sample size of the SBR1 group (fig. S2, C to E).

Fig. 2 cDC1 are the only DC subset enriched in breast tumors compared with patient PBMCs.

(A and B) Proportion of each TA-DC subset (identified with the gating strategy defined in Fig. 1) among viable CD45+ cells in n = 23 patient PBMCs (A) and in n = 21 breast tumors (B). Statistical analysis by Friedman test. (C) Ratio of one DC subset to all DCs among viable CD45+ cells. Statistical analysis: Mann-Whitney test. Horizontal bars represent the median of each group of samples. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. ns, not significant.

cDC1 are positively correlated with patient survival in many human cancers

We investigated the prognostic value of cDC1 compared with other DC populations. On the basis of available transcriptomic datasets of human DCs (18, 3537), we defined human DC signatures composed of CLEC9A and XCR1 for cDC1, CLEC10A and CD1E for cDC2, CD1A and CD207 for LCs, and CLEC4C and LILRA4 for pDCs. The abundance score for each DC population was estimated using gene expression datasets from The Cancer Genome Atlas (TCGA) (38) with the microenvironment cell populations (MCP)–counter algorithm developed by Becht et al. (39). We assessed the overall survival of patients stratified according to the median of each expression score. This approach revealed the strong association between cDC1 infiltration and a good prognosis in breast cancer (Fig. 3A). cDC2 and LCs, but not pDCs, were also positively correlated with patient survival although to a lesser extent. We extended our analysis to 13 other human cancers and revealed that cDC1 represented the only DC subset associated with a prolonged overall survival in most of the solid tumors (8 of 14) (Fig. 3B and fig. S3).

Fig. 3 TA-cDC1 are strongly associated with an increased overall survival in multiple human cancers.

(A) Kaplan-Meier analysis of the overall survival of patients stratified according to the median for the expression score of each DC subset infiltrating breast tumors (calculated with the MCP-counter algorithm) using TCGA datasets. Statistical analysis: log-rank test. (B) Summary of P values associated with the log-rank test evaluating the prognostic impact of each DC subset in 14 human TCGA cancer datasets.

Breast tumors highly infiltrated by cDC1 are characterized by an enriched IFN signature

Because cDC1 is the key DC population responsible for effector CD8+ T cell activation, we investigated their precise localization within breast tumors using CLEC9A and CD8A probes by in situ hybridization, combined to an opal-based immunofluorescence detection of cytokeratin-positive tumor cells (Fig. 4A). Using the Halo software, 16 zones were randomly defined and quantified for eight breast tumors. This approach revealed a strong correlation between cDC1 and CD8+ T cells (Fig. 4B). We also observed that cDC1 (CLEC9A+ cells) were predominantly localized in the stroma than in the tumor bed (Fig. 4C). Image analysis also highlighted close contacts between cDC1 and CD8+ T cells in breast tumors (Fig. 4, A and C). cDC1 (70 ± 10%) in stroma are in contact with at least one CD8+ T cell (Fig. 4, A and C). Using TCGA transcriptomic database of breast cancer, we validated a positive correlation between cDC1 and CD8+ T cell infiltration scores (Fig. 4D). However, this was not a specific feature of cDC1, because this correlation was shared with cDC2 and pDCs. In contrast, LCs and CD8+ T cell infiltration scores were not correlated in tumors, although LCs, but not pDCs, were associated with a good prognosis (Figs. 3A and 4D).

Fig. 4 Enrichment of IFN signatures is a specific feature of cDC1-infiltrated tumors.

(A) Visualization of CD8+ T cells and cDC1 respectively by CD8A (green) and CLEC9A (red) probes in situ hybridization, combined to an opal-based immunofluorescent staining of cytokeratin-positive tumor cells (white). Nuclei were counterstained with DAPI. White scale bars represent 50 μm for 40× images and 25 μm for 100× images. (B) Quantification of CD8+ T cells and CLEC9A+ cDC1 in eight breast tumors using the Halo software. Pearson correlation test. (C) Proportion of both cDC1 in stroma versus tumor bed and of cDC1 in close contact with at least one CD8+ T cell, quantified with the Halo software. (D) Scatterplots showing the Spearman correlation between CD8+ T cell and each TA-DC subset scores (calculated with the MCP-counter algorithm) in n = 1100 breast tumors (TCGA datasets). (E) High-throughput GSEA by BubbleGUM in breast tumors enriched in only one DC subset (TCGA dataset, n = 48 for cDC1, n = 18 for cDC2, n = 95 for LCs, and n = 67 for pDCs). Bubble enrichment patterns (black boxes) highlighted by the selection of gene sets [from Hallmark collection (H) or homemade gene sets] and pairwise comparisons of interest. NFκB, nuclear factor κB; NES, normalized enrichment score; FDR, false discovery rate; NS, not significant. (F) Heat maps illustrating genes extracted from the overlap of the GSEA leading edges identified by each pairwise comparison between breast tumors enriched only in cDC1 and those enriched only in one of the three other DC subsets. Gene expression values were averaged across tumors enriched in only one DC subset and then log2-transformed.

To gain further insight into the functional specificity of TA-cDC1 associated with their positive prognostic impact, we used the DC infiltration scores defined by MCP-counter to design groups of tumors enriched only in one DC subset. We analyzed their association with gene signature pathways using BubbleGUM (40) for high-throughput gene set enrichment analysis (GSEA) of homemade gene sets and of the MSigDB collections. Type I/III and type II IFN signatures were strongly enriched in tumors highly infiltrated with cDC1 compared with other groups (Fig. 4E and fig. S4A). Hierarchical clustering of type I/III IFN signature genes specifically enriched in cDC1high tumors compared with all other tumor types revealed two groups of genes. The first one (group I) is shared with pDCshigh tumors and mostly composed of ISGs, and a second one (group II) containing genes only specific of cDC1-enriched tumors such as LPAR6, UBA7, or CSF1 (Fig. 4F). In the type II IFN signature, many genes involved in Ag processing and presentation were highly expressed in cDC1high tumors (B2M, PSMB8, HLA-A, HLA-B, TAPBP, PSMB9, and PSME1), in addition to genes important for the cross-talk between natural killer (NK) cells and DCs such as HLA-A and HLA-B, IL-15R, and IL-15. CD274/PD-L1 was particularly enriched in cDC1high tumors (Fig. 4F). Although no pathway was specifically enriched in cDC2high tumors, these tumors seem to have the lowest IFN signatures (Fig. 4E). The signatures related to G2M checkpoint and hypoxia were enriched in LChigh tumors compared with other groups (Fig. 4E). On the other hand, whereas the tumor necrosis factor–α (TNF-α) signaling pathway was predominantly associated with cDC1, cDC2, or LCs compared with pDCs, oxidative phosphorylation (OXPHOS) and immunosuppressive pathways were distinctive of pDChigh tumors (Fig. 4E). This last result is in line with the correlation between pDCs and regulatory T cells (Tregs) in breast tumors compared with other DC subsets (r = 0.6042) (fig. S4B) and corroborates our previous observations (41). Last, in comparison with cDC1high tumors, tumors highly infiltrated by pDCs, LCs, or cDC2 presented with a more mesenchymal phenotype, as evidenced by the enriched epithelial-to-mesenchymal transition (EMT) signature (Fig. 4E and fig. S4A). This in silico analysis revealed an association between cDC1 and the presence of a strong IFN signature in human breast tumors.

IFN-λ1 is selectively produced by cDC1 in breast tumors

IFN-III production is a well-known feature of cDC1 (21, 22) that has been characterized during viral infections, but neither the presence of this cytokine nor its role has been investigated in human tumors. The enrichment in the type I/III IFN signature in cDC1-exclusive tumors prompted us to analyze the link between IFN-III and TA-cDC1. We initially demonstrated the up-regulation of IFNL1 gene expression in tumoral compared with normal adjacent tissue (NAT), on the basis of TCGA transcriptomic datasets of multiple cancers (Fig. 5A), confirming the presence of IFN-III in tumors. This differential expression between tumor and NAT was highly marked for breast, head and neck, and lung cancers in which cDC1 are strongly associated with favorable patient outcome (Fig. 3B). We confirmed the presence of IFN-λ1 at the protein level in 87 of 106 soluble tumor milieu (STM) samples obtained by mechanical dissociation of breast tumors, with a concentration ranging from 10 to 800 pg/ml in half of the tumors (Fig. 5B). IFN-λ2 was also detected in these STMs and highly correlated with IFN-λ1 (fig. S5A). In addition, IFN-λ1 was the most abundant IFN subtype in breast tumors compared with type I IFNs (fig. S5B). IFN-α was completely absent and IFN-β was detected at a very low concentration (< 50 pg/ml) in only 30% of STMs (fig. S5B), consistent with our previous demonstration of the inability of TA-pDCs to produce IFN-α in breast tumors (41). These results are in complete agreement with mRNA levels of all IFN subtypes in the TCGA breast cancer dataset, showing high expression of genes coding for IFN-III, whereas all of the IFNA genes were below the detection threshold (fig. S5C). Intracytoplasmic flow cytometry analysis revealed spontaneous production of IFN-λ1 that is restricted to cDC1 in the absence of any ex vivo stimulation in a third of the tumors (4 of 12) (Fig. 5, C and D). We also detected TNF-α production, mostly by TA-cDC2, but no IFN-α (fig. S5D). To confirm these findings, we performed double fluorescent RNA in situ hybridization with IFNL1 and CLEC9A probes (Fig. 5E). Using Halo software, 16 zones of 0.64 mm2 were randomly defined and quantified for six breast tumors (representing a total area per tumor of ~12 mm2). This approach confirmed the presence of IFNL1 transcripts in CLEC9A+ TA-cDC1 in two of six tumors (Fig. 5, E to G). Tumors in which IFN-λ1 was detected appeared to be more infiltrated by cDC1 (fig. S5E). In addition, IFNL1 was absent from cytokeratin-positive tumor cells (Fig. 5E). These results indicate that IFN-λ1 production is a specific feature of cDC1 in a human tumor context and that this cytokine may play a central role in their antitumor activity.

Fig. 5 IFN-λ1 is specifically produced by cDC1 in breast tumors.

(A) Differential expression analysis of the IFNL1 gene in multiple transcriptomic TCGA datasets between the tumor (gray boxes) and the NAT (white boxes). Statistical analysis: Wilcoxon test. (B) IFN-λ1 quantification by ECLIA multiplex assay in n = 107 STMs of human breast tumors. (C) Intracellular IFN-λ1 FACS staining of one representative breast tumor suspension positive for IFN-λ1 (gray contour plot, isotype control). (D) Proportion of IFN-λ1 producing cells in n = 12 breast tumor suspensions. Horizontal bars represent the median of each group of samples. (E) In situ detection of IFN-λ1 producing cells: CLEC9A (red) and IFNL1 (green) mRNA were stained by duplex RNAscope, combined to an opal-based immunofluorescent staining of cytokeratin-positive tumor cells (white). Nuclei were counterstained with DAPI. White scale bars represent 50 μm for 40× images and 25 μm for 100× images. (F and G) Quantification of IFNL1 and CLEC9A simple and double-positive cells in 16 randomly selected zones of 0.64 mm2 per tumor using the Halo software in six breast tumors. The mean for each tumor is represented in (F) and the exact quantification of the two of six IFNL1+ tumors (#01 and #03) in (G). Bars and error bars respectively represent the mean and the SEM for each group of samples. Statistical analysis: Friedman test. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

IFN-λ1 is associated with a better survival and induces a T helper 1 soluble microenvironment in human breast tumors

We explored the prognostic impact of IFNL1 and of IFNLR1, the specific chain of its heterodimeric receptor, using public transcriptomic datasets. High expression level of these two genes correlated with greater relapse-free survival (Fig. 6A). To understand the mechanisms underlying the beneficial impact of type III IFN on cancer, we dissected the soluble tumor microenvironment by quantifying multiple cytokines and chemokines in the STM of more than 100 dissociated breast tumors. IFN-λ1 was strongly correlated with CXCR3-L (CXCL11 /CXCL10 /CXCL9) and CX3CL1 chemokines, as well as TNF-α and IL-12p40 (Fig. 6B and fig. S6). These results raised the hypothesis that the production of IFN-λ1 by cDC1 in tumors could be associated with cytokines and chemokines involved in the recruitment and activation of cytotoxic lymphocytes (NK cells and CD8+ T cells). To test this hypothesis, we assessed the impact of IFN-λ1 on the tumor microenvironment by treating human breast tumor suspensions with IFN-λ1 for 24 hours. We showed that IFN-λ1 is a potent inducer of IFN-β but does not induce IFN-α2 (Fig. 6C). In addition, the level of cytotoxic lymphocyte–attracting chemokines was largely increased, in particular, for two CXCR3-Ls (CXCL10 and CXCL11) and for CX3CL1 (Fig. 6C). IFN-λ1 activation alone induced IL-12p70 production, a key cytokine for the differentiation and activation of T helper 1 (TH1) lymphocytes and of effector CD8+ T cells, as well as high amounts of IFN-γ (Fig. 6C). Together, these results demonstrate a role for type III IFNs in the induction of a TH1 immune soluble microenvironment in human breast tumors.

Fig. 6 IFN-λ1 is associated with a favorable outcome and with a TH1 tumor microenvironment in breast cancer.

(A) Kaplan-Meier analysis of the relapse-free survival of patients stratified according to the median for the expression of IFNL1 and IFNLR1 genes (KMplot transcriptomic datasets). Statistical analysis by log-rank test. (B) Spearman correlation factors between IFN-λ1 and the other cytokines and chemokines quantified in n = 107 STMs of human breast tumors by ECLIA multiplex assay. (C) Cytokine and chemokine quantification by ECLIA multiplex in the supernatants of n = 6 tumor cell suspensions treated or not with IFN-λ1 for 24 hours. Bars and error bars respectively represent the mean and the SEM for each group of samples. Statistical analysis: Paired t tests, *P < 0.05, **P < 0.01, and ***P < 0.001. ns, not significant.

TLR3 stimulation is a potent strategy to induce the production of IFN-λ1 by cDC1 and TH1-related immune responses in breast tumors

Given the positive prognostic impact of IFN-λ1 in breast cancer and its putative role in the recruitment and activation of cytotoxic immune cells, we speculated that the induction of IFN-III could be a potential therapeutic strategy in combination with other immunotherapies. Several studies reported the involvement of TLR3 signaling in the activation of IFN-λ production by cDC1 (21, 22). Thus, we treated patient PBMCs or breast tumor cell suspensions with polyinosinic-polycytidylic acid (PolyI:C, double-stranded RNA), a TLR3 agonist to target cDC1, in addition to resiquimod (R848, single- stranded RNA) to stimulate other DC subsets through TLR7/8. As expected, this combined activation led to TNF-α production by all patient blood DC subsets (fig. S7), IFN-α by pDCs, as well as IFN-λ1 both by cDC1 and by pDCs (Fig. 7, A and B). In tumor cell suspensions, we confirmed the impairment of TA-pDC to produce IFN-α and demonstrated their inability to produce IFN-λ1 as well (Fig. 7, A and B), in contrast to patient blood pDCs. Unlike TA-pDCs, TA-cDC1 were responsive to TLR stimulation and could efficiently produce IFN-λ1 in 11 of 12 tumors (Fig. 7, A and B). This activation also led to TNF-α production by all TA-DCs (fig. S7). We also evaluated the global impact of such TLR3 stimulation on the tumor microenvironment. Multiple cytokines and chemokines were quantified in the supernatant of ex vivo PolyI:C-stimulated fresh breast tumor thick sections instead of tumor cell suspensions to conserve the tissue architecture. We observed a strong induction of IFN-λ1 upon TLR3 triggering (Fig. 7C). The production of IFN-γ, CXCL9, CXCL10, and CX3CL1 was markedly increased (Fig. 7C), thus highlighting the potential of cDC1 stimulation in tumor tissue via TLR3-L to induce a microenvironment favorable to recruitment and activation of cytotoxic effector cells.

Fig. 7 cDC1 activation by TLR3-L stimulated their IFN-λ1 production and the induction of TH1 immune responses.

(A) Representative FACS plots of ex vivo IFN-λ1 production by DC subsets from patient PBMCs and breast tumor suspensions treated or not with TLR-L (PolyI:C + R848) for 5 hours. DC subsets were identified using the gating strategy defined in Fig. 1 and fig. S1. (B) Proportion of IFN-λ1 and IFN-α producing DC subsets of n = 20 patient PBMCs and n = 12 fresh breast tumor suspensions treated or not with TLR-L (PolyI:C + R848) for 5 hours. The medium condition corresponds to Fig. 5C. Bars and error bars respectively represent the mean and the SEM for each group of samples. Statistical analysis: Paired t test. (C) Multiplex quantification by ECLIA assay of cytokines and chemokines in the supernatants of n = 6 fresh tumor thick sections treated or not with TLR3-L (PolyI:C) for 48 hours. Bars and error bars respectively represent the mean and the SEM for each group of samples. Statistical analysis: Wilcoxon test, *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. ns, not significant.

DISCUSSION

Our study highlights a crucial function of human cDC1 in antitumor immunity through their capacity to specifically produce type III IFN. Although little or no type I IFN is present in human breast tumors, IFN-III is produced in more than 50% of patients. Here, we uncover a key role for IFN-III in the induction of IL-12p70, IFN-γ, CXCR3-L and CX3CL1, cytokines, and chemokines involved in the recruitment and activation of NK and effector T cells and reveal that the expression of IFNL1 or IFNLR1 genes is associated with a favorable patient outcome.

We demonstrated that cDC1 are the source of IFN-III in the breast tumor environment. This was observed by flow cytometry showing that among immune and nonimmune cells only cDC1 spontaneously expressed IFN-λ1 in one of three of tumors and, to a less extent, in patient blood. This was validated by in situ hybridization, demonstrating that IFN-λ1 was only expressed by cDC1 in primary breast tumors. In accordance with the specific expression of IFN-λ in cDC1, our in silico analysis showed that the type I/III gene signature was enriched in tumors highly infiltrated with cDC1 compared with other DC types. Whereas IFN-III expression was reported in epithelial cells, hepatocytes, monocyte-derived DCs, pDCs, and cDCs in response to many viral infections such as hepatitis B and C viruses, herpes simplex virus (4244), and nonviral nucleic acids (45), we and others have shown that cDC1 produce large amounts of IFN-III upon TLR3 stimulation (21, 22). The spontaneous production of IFN-III specifically by cDC1 tumors suggest that they are activated in the breast tumor environment, possibly through TLR3 ligands such as double-stranded RNA, including endogenous retroviruses (46). Other intracellular sensors such as Ku70 are involved in IFN-III production in response to exogenous DNA through stimulator of interferon gene (STING) activation and may also be implicated in tumors (47).

By analyzing a number of cytokines and chemokines secreted in a large cohort of patients with breast cancer, we highlighted the strong correlation between IFN-λ1 and CXCL9/10/11, the three ligands of CXCR3, a chemokine receptor strongly expressed by TH1, CTL, NK, and NKT cells (48). In this context, molecules increasing paracrine expression of these CXCR3-Ls have been shown to initiate antitumor immunity in many models (49). We observed that IFN-λ1 is also correlated in the TME (tumor microenvironment) with CX3CL1, a chemokine involved in the recruitment of effector T cells endowed with particularly high cytotoxic activity (50). These last results raise the hypothesis of IFN-III involvement in cytotoxic immune cell recruitment in the tumor. By inducing the same core of genes as IFN-I (51), IFN-III may play a key role in the antitumor immune response. IFN-I is already known to be an inducer of cDC maturation (52), and it appears to strongly enhance cross-presentation, which is critical for the induction of CD8+ T cell responses against cancer (53). Here, we show that treatment of tumor cell suspensions with recombinant IFN-λ1 induces production of IFN-β and IFN-γ, revealing an amplification loop between these different IFN subtypes. The capacity of IFN-III to induce IFN-γ and IL-12p70 suggests a direct or indirect effect on T or NK effector cells, leading to their activation. In mice, it was shown that IFN-III acts directly on NK cells (26, 27, 54), but the effects of IFN-λR1 on human NK or T cells have not been reported, suggesting a more indirect effect in human tumors. In line with these observations, we demonstrated through in silico analyses that cDC1-enriched tumors are associated with a type II IFN signature. Thus, these data suggest that the endogenous activation of cDC1 not only will favor cytotoxic effector recruitment through CXCR3-L or CX3CL1 but also will induce their activation. We have previously demonstrated that IL-12p70 production induced by TLR activation of human DCs was dependent on autocrine type I IFN (55). Thus, a loop of IFN-III signaling may be necessary in tumors for bioactive IL-12p70 secretion by cDCs. Last, although IFN-I and IFN-III activate the same signaling pathway, they induce different responses in tumors, likely through different expression patterns of their receptors. In humans, few papers reported the expression of IFN-λR1 in epithelial cells of the airways and intestinal tract and in blood on B cells and pDCs (56). Human pDCs respond to IFN-III (5759) through the canonical Janus kinase–signal transducer and activator of transcription pathway (60, 61), leading to the up-regulation of ISG expression (6062) and type I IFN production (61). The cross-talk between pDCs and cDC1 through type I/III IFNs for antitumor immunity has previously been established in mouse tumor models (63). Human B cells also respond to IFN-III by up-regulating ISGs (44, 56, 60). In addition, it was reported that epithelial cell polarization and differentiation can influence the expression of IFN-λR1 in mice (64) and that histone deacetylase inhibitors confer IFN-III responsiveness to previously nonresponsive human cell lines (65). These findings suggest that the variable expression of the IFNLR1 gene in the tumor microenvironment could also modulate its responsiveness to IFN-III. Therefore, it will be important to precisely identify the intratumoral cell population directly responding to IFN-III.

We have also shown that cDC1 display the highest positive prognostic impact in breast cancer compared with other DC subsets. This observation extends results from two studies suggesting the association between breast tumor infiltration by cDC1 and a favorable prognosis (3, 4). We demonstrated that the cDC1 signature has a positive prognostic impact in several other tumor types (8 of 14), in particular in lung, head and neck, and metastatic melanoma, all responding to T cell ICP therapies. No prognostic impact was found in aggressive cancers, such as ovarian or pancreatic tumors, which have been found to be less sensitive to immunotherapy thus far. With respect to other DC subsets, a high infiltration of breast tumors by cDC2 and LCs is associated with an increased overall survival, although this association is of a lower significance compared with cDC1. In contrast with their association with a favorable outcome of patients with breast cancer, LCs are not correlated with CD8+ T cells but are associated with a hypoxic gene signature in breast tumors. This may be due to their presence within the tumor islets (66), in contrast to all other DC subsets preferentially localized in T cell aggregates. Unlike cDCs and LCs, the pDC signature has no impact on breast cancer patient outcome. Furthermore, in GSEAs, immunosuppression and OXPHOS pathways were higher in pDChigh tumors, which corroborates our previous demonstration of the role of pDCs in breast cancer progression by promoting Treg accumulation (67) and with another study supporting the implication of OXPHOS in the survival of Tregs (68). We also observed the EMT signature enrichment in cDC2/LC/pDChigh compared with cDC1high tumors, which is relevant with data highlighting the EMT as an immune evasion mechanism contributing to metastatic dissemination (69). Nevertheless, all those in silico analyses of patient survival have to be validated by in situ staining on large cohorts. In this context, the in situ identification of cDC1 was hampered by the lack of specific monoclonal antibody against CLEC9A or XCR1. Here, we visualized CLEC9A+ cDC1 in human tumors by in situ fluorescent hybridization. cDC1 have been characterized in non–small cell lung cancer, colorectal cancer, and melanoma using the staining of IFN regulatory factor 8 or CD141/BDCA3 staining (16, 17), two markers also respectively expressed by pDCs and cDC2. Thus, we believe that CLEC9A is a better cDC1 marker based on our extended flow cytometry analysis and by the recent characterization of Zilionis et al. (18) who performed single-cell RNA sequencing of lung tumors and demonstrated that TA-cDC1 have a high level of CLEC9A and XCR1 mRNA. We also observed that cDC1 are localized in lymphoid aggregates in in situ, where they establish direct interactions with CD8+ T cells. This is in accordance with the correlation that we observed by in silico analysis between cDC1 and CD8+ T cell scores, which was also previously reported in multiple solid tumors (4) including metastatic melanoma (6). However, we reported here that the strength of this association is comparable to the other DC subsets and thus could not explain by itself the better prognostic impact of cDC1, strengthening the importance and specificity of type III IFN production by cDC1.

Last, in contrast to TA-pDC impairment for their IFN-λ1 production in response to TLR7/8 ligand, as we previously showed for IFN-α (41, 70), we now highlight the potency of a TLR3 agonist to induce IFN-λ1 production by TA-cDC1. We demonstrated the induction of CX3CL1 and CXCL9/10 production through tumor cell suspension stimulation with a TLR3 agonist. These chemokines could be produced by cDC1 themselves, as previously demonstrated in blood (22). Their activation in tumors may directly lead to the secretion of both IFN and cytotoxic lymphocyte–recruiting chemokines, thus creating a favorable environment to boost antitumor immunity. TLR3 triggering presents promising results for the development of new combined therapies, including with ICP blockers. In melanoma-engrafted mice, the intratumoral administration of PolyI:C, combined with the cDC growth factor FLT3-L, markedly enhances the immunotherapeutic response of ICP-inhibiting treatments. Furthermore, the beneficial effect of this treatment was demonstrated to be IFN-I dependent (7). Because IFN-I and -III share the same signalization pathway, and as suggested by our results, the induction of IFN-III production by TLR3-L might be involved in this process as well. Last, the treatment of tumors by PolyI:C could also be beneficial to activate the MDA5/RIG-I pathway, which is known to directly induce the production of IFN-I by nonhematopoietic cells and to directly act on tumor cells (71). Because of the small size of fresh human breast tumors, it seems for now infeasible to perform functional analyses of IFN-III producing cDC1, for example to investigate their Ag cross-presentation ability. Thus, the particular role of those activated cDC1 would have to be further explored in animal studies. Furthermore, the kinetics of this IFN-III production during the early stage of immune surveillance would be very important to analyze, using, for example, IFN-III green fluorescent protein reporter mice. Last, further studies will be necessary to test whether our observation extends to other solid tumors.

Overall, these data suggest that cDC1 have a positive impact on patient survival, likely involving type III IFN production. This synthesis may be initiated by TLR3-mediated detection of endogenous double-stranded RNA (46) released during tumor remodeling. The endogenous signal leading to the IFN-III production by TA-cDC1 remains to be identified. However, our demonstration of potential antitumor functions of IFN-III provides valuable evidence to support the development of new therapeutic strategies targeting cDC1 to amplify the response to immunotherapies, especially in breast cancer.

MATERIALS AND METHODS

Study design

The objective of this study was to investigate the role of cDC1 and their IFN-III production in the antitumor immune responses. To this end, freshly resected breast tumors were dissociated into cell suspensions or cut into thick sections before performing flow cytometry assay and electrochemiluminescence assay (ECLIA). We also used formalin-fixed paraffin-embedded (FFPE) tumor samples for in situ hybridization and immunofluorescence assays. Last, we analyzed publicly available transcriptomic datasets of various solid tumors. The sample size per group, the experimental replicates, as well as the statistical methods, are described in each figure legend.

Study approval

All human samples (blood and tumors) were obtained after approval from the Institutional Review Board and Ethics Committee of the Centre Léon Bérard (CLB; L-06-36 and L-11-26) and patient-written informed consent, in accordance with the Declaration of Helsinki.

Patients with breast cancer

We enrolled patients diagnosed with primary breast carcinoma. Fresh tumors and blood samples (collected in EDTA anticoagulant-containing tubes) were obtained from the Biological Resources Center (BRC) of the CLB (BB-0033-00050, Lyon, France) and from the TUMOROTHEQUE (BRC of the Hospices Civils de Lyon, France). Tumors were used for single-cell suspension preparation, and ex vivo culture of thick section or FFPE for in situ stainings. Healthy human blood (collected in EDTA anticoagulant-containing tubes) was purchased anonymously from the Etablissement Français du Sang (Lyon, France).

PBMC isolation and tumor cell suspensions

Sections of the resected tumor were selected by the pathologists. Five hundred micrograms of fresh tissues was mechanically dissociated in 1 ml of RPMI 1640 medium (Gibco) with antibiotics [penicillin (100 IU/ml) and streptomycin (100 μg/ml), Invitrogen]. The supernatant of dissociated tumors, referred thereafter as STM, was immediately collected and stored at −80°C for subsequent cytokine and chemokine quantification. Tissues were then digested for 45 min at 37°C in RPMI 1640 with antibiotics, collagenase IA (1 mg/ml) and deoxyribonuclease I (20 μg/ml) (Sigma Aldrich). Digested samples were then filtered on a 70-μm cell strainer and resuspended in RPMI 1640 with antibiotics and supplemented with 10% fetal calf serum (complete RPMI) for further analysis. PBMCs were isolated from blood samples of patients or healthy donors through Ficoll density gradient centrifugation (Eurobio).

Ex vivo stimulation

Tumor cell suspensions and PBMC were cultured at 1 × 106 cells/ml for 5 hours in complete RPMI with different activators: R848 (5 μg/ml; Invivogen) and PolyI:C (30 μg/ml; InvivoGen). GolgiPlug (BD Biosciences) was added after 1 hour. At the end of the activation, harvested cells were stained for membrane markers, fixed, permeabilized, and stained for intracellular cytokines. Tumor cell suspensions were also activated for 48 hours with IFN-λ1 (100 ng/ml; R&D Systems) for cytokine quantification by ECLIA, which is the most sensitive assay for protein quantification.

To preserve subcellular architecture, we cut 200-μm sections of fresh tumors using a vibratome. These sections were incubated for 48 hours with PolyI:C (100 μg/ml; InvivoGen), and supernatants were collected for cytokine quantification by ECLIA.

Cell staining and flow cytometry

Single-cell suspensions were stained using antibodies listed in table S1. Dying cells were excluded by Zombie Violet/Yellow staining (BioLegend) depending on the experiment. Lymphocytes, NK cells, neutrophils, and other myeloid cells (monocytes, macrophages, and inflammatory monocytes) were also excluded using respectively anti-CD3/56/15/14 antibodies in the lineage. Intracellular cytokine or DC-LAMP staining was performed after fixation and permeabilization (Fix/Perm buffers, eBioscience). All flow cytometry acquisitions were done on an LSRFortessa Cell Analyzer (BD Biosciences), and data were processed in FlowJo 10.4 (Tristar). Some flow cytometry data were visualized using viSNE (Cytobank) (72), a dimensionality reduction method which uses the Barnes-Hut acceleration of the t-distributed stochastic neighbor embedding (t-SNE) algorithm. viSNE plots were generated separately for each patient.

Cytokine and chemokine quantification by Meso Scale Discovery (MSD) assay

The following cytokines and chemokines were quantified in STMs or in supernatants of activated thick tumor sections and tumor cell suspensions, using ECLIA and MSD technology according to the U-plex protocol (MSD): IL-1α, IL-1β, IL-6, IL-8, IL-12/IL-23p40, IL-12p70, IL-17A, IL-18, IL-23p19, IL-33, IFN-α2a, IFN-β, IFN-γ, IFN-λ, CX3CL1, TNF-α, CXCL9, CXCL10, CXCL11, transforming growth factor–β1 (TGF-β1), TGF-β2, and TGF-β3.

In situ hybridization combined with immunofluorescence on FFPE tumor sections

FFPE tumors were cut into 4-μm sections. In situ probe hybridization combined with immunofluorescence was performed on the Leica BOND RX System. This procedure is based on the standard RNAscope LS Multiplex Fluorescent Assay. Tumor slides were rehydrated and deparaffinized before fixation, protease pretreatment, probe hybridization (Hs-CLEC9A-C2 probe in combination with Hs-CD8A-C1 or Hs-IFNL1-C1 probes), amplification, and anti-cytokeratin staining. Opal dyes (PerkinElmer) were used for fluorescent detection of probes and anti-cytokeratin antibody. Nuclei were then counterstained with 4′,6-diamidino-2-phenylindole (DAPI), and slides were scanned using the Vectra Polaris automated quantitative pathology imaging system (PerkinElmer). Last, the Halo software (Indica Labs) was used to randomly define 16 zones of per tumor (0.64 mm2 per zone, 11.88 mm2 total for CLEC9A/IFNL1; 0.99 mm2 per zone, 15.84 mm2 total for CLEC9A/CD8A) and to quantify positive cells. Contacts between each cDC1 and at least one CD8+ T cells were manually quantified.

Survival analysis

The clinical outcome data and RSEM (RNA-seq by expectation-maximization) normalized expression datasets from TCGA were downloaded from the cBioPortal (February 2018 version) for 14 tumor types: bladder carcinoma (BLCA), breast invasive carcinoma (BRCA), colorectal adenocarcinoma (COAD), brain lower grade glioma (LGG), head and neck squamous cell carcinoma (HNSC), kidney renal papillary cell carcinoma (KIRP), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma (PAAD), prostate adenocarcinoma (PRAD), skin cutaneous metastatic melanoma (SKCM), stomach adenocarcinoma (STAD), and thyroid cancer (THCA). MCP-counter (39) was used to estimate the relative abundance of several populations of immune cells. This algorithm originally allows the quantification of the absolute abundance of eight immune populations and two stromal cell populations in heterogeneous tissues from transcriptomic data. Here, in addition to the original cell population signatures defined by Becht et al., the following gene signatures were used to run MCP-counter: cDC1 (CLEC9A, XCR1), cDC2 (CLEC10A, CD1E), pDCs (LILRA4, CLEC4C), and LC (CD207, CD1A). Overall survival analyses and plots were performed with R, using the packages survival and survminer. For each immune population, we compared patients displaying the top 50% highest level of infiltration by the given immune cell type and those with the 50% lowest level. The log-rank test was used to determine statistical significance for overall survival between these two groups of patients. Analysis of progression-free survival was performed for the top 50% and bottom 50% of IFNL1 and IFNLR1 gene expression–ranked values using the Kaplan-Meier plotter software (73).

Gene expression analysis

RNA-sequencing data for 14 types of solid human cancers and matched normal samples were downloaded using the TCGABiolinks R package (open-access data from the TCGA data portal, version March 2018) with the harmonized option (data aligned to hg38). For each cancer type, HTSeq raw read counts were normalized using the DESeq2 R package and log2-transformed. Wilcoxon tests were performed to assess whether the IFNL1 gene was differentially expressed between tumor and normal samples.

Heat maps and hierarchical clustering

Heat maps of log2-normalized expression values of selected genes were performed using the Morpheus website from the Broad Institute (https://software.broadinstitute.org/morpheus/). Hierarchical clustering was performed using the One-Pearson correlation as a metric and the complete linkage as a clustering method for genes.

Gene set enrichment analysis

To gain insight into the functional specificity of TA-DC subsets, we used the DC infiltration scores defined by MCP-counter to design groups (n = 48 for cDC1, n = 18 for cDC2, n = 95 for LCs, and n = 67 for pDCs) with a high infiltration score for one DC subset (score > median of all tumors) and low for the three other subsets (score < median of all tumors). Using these groups, high-throughput gene set enrichment analyses were performed using the BubbleMap module of BubbleGUM (40). BubbleMap analysis was performed with 1000 gene set–based permutations and with “Signal2noise” as a metric for ranking the genes. The results are displayed as a bubble map, where each bubble is a GSEA result and summarizes the information from the corresponding enrichment plot. The color of the bubble corresponds to the tumors from the pairwise comparison in which the gene set is enriched. The bubble area is proportional to the GSEA normalized enrichment score. The intensity of the color corresponds to the statistical significance of the enrichment, derived by computing the multiple testing–adjusted permutation-based P value using the Benjamini-Yekutieli correction. Enrichments with a statistical significance above 0.25 are represented by empty circles. Public gene sets of the Hallmark collection (v6.1) were downloaded from MSigDB (74), and homemade gene sets are detailed in table S2. Because IFN-III share the same signaling pathway and induce similar ISGs than IFN-I, we renamed the type I IFN Hallmark signature into type I/III IFN signature.

Statistics

Wilcoxon or paired t tests for paired samples, and Mann-Whitney tests for unpaired samples, were performed for the comparison of two groups. To compare more groups, Kruskal-Wallis tests were performed for unpaired samples and Friedman tests for paired samples. All graphs show each sample value. The horizontal bars represent the median for each group of samples. The bar plot represents the mean with the SEM for each group of samples. Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

SUPPLEMENTARY MATERIALS

immunology.sciencemag.org/cgi/content/full/5/46/eaav3942/DC1

Fig. S1. Identification of DC subsets in patient PBMCs.

Fig. S2. Proportion of DCs among the immune infiltrate of breast.

Fig. S3. Kaplan-Meier analysis of the overall survival of patients with cancer (parts 1 and 2).

Fig. S4. In silico analysis of pathways associated to each DC infiltration score.

Fig. S5. Quantification of IFNs in breast tumors.

Fig. S6. Correlations between the soluble factors present in human tumors.

Fig. S7. Analysis of the TNF-α production by DC subsets infiltrating breast tumors.

Table S1. FACS antibodies.

Table S2. Immunosuppression gene signature.

Data file S1. Raw data file (in Excel spreadsheet).

REFERENCES AND NOTES

Acknowledgments: We wish to thank the staff of the core facilities at the Cancer Research Center of Lyon (CRCL) for technical assistance and the BRC of the CLB for providing human samples, as well as N. Gadot and the research anatomopathology platform of the CLB. We are very grateful to S. Leon and the Ex Vivo platform (Department of Translationnal Research and Innovation of the CLB, Lyon, France) for help on thick tumor slice analysis. We thank I. Durand and P. Battiston-Montagne for assistance for flow cytometry and T. Andrieu for viSNE analyses. We would also like to thank. B. Manship for critical reading of the manuscript. Funding: This work was supported by funding from INSERM, INCA (PLBIO INCa_4508 and PLBIO INCa_11155), ANRS, ARC, Ligue contre le Cancer (Régionale Auvergne-Rhône-Alpes et Saône-et-Loire, Comité de la Savoie), SiRIC LYriCAN (INCa-DGOS-Inserm_1263), and the Auvergne-Rhône-Alpes region (IRICE). We would like to thank our financial supports: the Auvergne-Rhône-Alpes region and the ARC Foundation for M.H., the ESMO for E.G. (any views, opinions, findings, conclusions, or recommendations expressed in this material are those solely of the authors and do not necessarily reflect those of ESMO), the SIRIC project (LYRIC, grant no. INCa_4664) and the FP7 European TumAdoR project (grant no. 602200) for C.R. and C. Caux, as well as the LABEX DEVweCAN (ANR-10-LABX-0061) of the University of Lyon, within the program “Investissements d'Avenir” organized by the French National Research Agency (ANR), for V.O., A.-C.D., N.B.-V., C. Caux, and J.V.-G. Author contributions: M.H. designed and performed experiments and bioinformatics, analyzed results, did statistical analysis, and wrote the manuscript. E.G., C. Couillault, A.-C.D., C.R., V.O., and C.S. performed experiments and analyzed results. J.B. and B.D. set up a pipeline for the quantification of in situ stainings. T.-P.V.M. and J.K. performed bioinformatics and wrote the manuscript. I.T. and O.T. contributed to clinical project management and pathology review and provided clinical samples. M.D., N.B.-V., and C. Caux provided strategic advice and revised the manuscript. J.V.-G. designed experiments, supervised the research, and wrote the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.

Stay Connected to Science Immunology

Navigate This Article