Mice were bred at the CNIC and UCSF-specific pathogen-free conditions. Mouse strains include C57BL/6 mice, Clec9a^gfp/gfp mice backcrossed more than 10 times to C57BL/6 (DNGR-1-deficient, B6(Cg)-Clec9atm1.1Crs/J),20 Rag1^–/– mice (B6.129S7-Rag1tm1Mom/J, The Jackson Laboratory), Rag1^–/–Clec9a^gfp/gfp mice23 and Batf3^–/– mice (B6.129S(C)-Batf3tm1Kmm/J).6 To obtain OVA-specific CD8⁺ T cells, we used OTI transgenic mice (C57BL/6-Tg(TcraTcrb)1100Mjb/J) mated with B6-SJL (Ptprca Pepcb/ BoyJ) expressing CD45.1 allele.11 We used 7-week-old to 10-week-old mice (males or females) for all experiments in a sex-matched manner. Experiments were repeated two to four times and results were pooled.

All cell lines (table 1) were cultured in DMEM media supplemented with 10% fetal bovine serum, β-mercaptoethanol, L-glutamine, non-essential amino acids, sodium pyruvate, HEPES and penicillin and streptomycin. Passages were done with trypsin EDTA. One day before injection, cells were passed from confluent plates.

Cells were washed twice with phosphate-buffered saline (PBS), and 5×10⁵ cells were inoculated subcutaneously in 50 µL apyrogenic PBS (GIBCO) in the flank of isofluorane-anesthetized mice. Tumor measurements were obtained three times a week with an electronic caliper, and tumor size was calculated as the product of the longest dimension and its orthogonal. B16F10, B16ZsGreen, B16GFPOVA and B78mChOVA tumors were allowed to grow for 13 days, when mice were euthanized for analysis. Growth of B16Flt3L and MC38 tumors was allowed until compliance of experimental endpoint conditions, tumor ulceration or tumor size greater than 300 mm².

When indicated, tumor-free mice received a hydrodynamic injection of Flt3L-encoding plasmid (FL) or empty vector. Hydrodynamic injection was performed by intravenous administration into the tail vein of 2 mL room temperature PBS containing 10 µg plasmid in less than 8 s. Where indicated, mice were inoculated with tumors 24 hours after the hydrodynamic injection. Intraperitoneal administration of 100 µg anti-DNGR-1-blocking antibodies (7H11, BioXcell) or isotype control (Sigma) was performed at days 5, 7, 9 and 12 after tumor inoculation. When indicated, 30 mg/kg maraviroc (MVC; Sigma) or vehicle were intraperitoneally administered concurrent with antibody administration. Cyclophosphamide (140 mg/kg; Sigma) was administered following a metronomic regime at days 8 and 14 of tumor development.

Tumors, lymph nodes (LNs), ears and spleens were collected at indicated time points for flow cytometry analysis. Tumors and ears were finely minced with scissors and digested in 0.5 mg/mL Collagenase IV (Sigma-Aldrich) and 0.2 mg/mL DNase I (Sigma-Aldrich) for 60 min at 37°C with agitation. Spleens were minced and incubated for 15 min with 50 µg/mL Liberase TL (Sigma-Aldrich) and 0.2 mg/mL DNase I (Sigma-Aldrich). All these samples were smashed through a 100 µm mesh filter with a syringe plunger. LNs were disaggregated with forceps and digested in 0.5 mg/mL Collagenase IV (Sigma-Aldrich) and 0.2 mg/mL DNase I (Sigma-Aldrich) for 30 min at 37°C, pipetted at minute 15 of incubation. At the incubation endpoint, one volume of EDTA-containing Fluorescence-activated cell sorting (FACS) buffer was added to all samples. Digested LNs were filtered through a 100 µm mesh.

Single-cell suspensions were stained for 30 min at 4°C with LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Life Technologies). After washing with PBS, cells were stained in FACS buffer containing anti-CD16/32 (BioXcell), 3% FBS and 0.05% EDTA with the corresponding antibody cocktail for 30 min on ice. Cells were washed again and resuspended in FACS buffer for data acquisition using an LSRFortessa SORP (Becton Dickinson) or a FACSAria Fusion (Becton Dickinson) flow cytometry equipment. The quantification of absolute numbers in immune infiltrates was performed by acquiring a controlled volume of sample corresponding to a fixed weight of digested tissue at a constant speed for the same duration. This method was validated with the use of quantification beads in some of the experiments. Determination of absolute numbers allowed to pool data from the same experimental conditions obtained in independent experiments.

Tumors and tumor-draining lymph nodes (TdLNs) were collected and processed as previously described. Tumor single-cell suspensions were enriched in CD11c⁺ cells with anti-CD11c microbeads (Miltenyi Biotec), for further purification of subpopulations of antigen-presenting cells. Enriched CD11c⁺ cells from tumors were further sorted into MHCII^hiCD24⁺CD11b⁻CD103⁺ cDC1s, MHCII^hiCD24⁺CD11b⁺CD103⁻ cDC2s and MHCII^hiCD24⁻F4/80⁺ TAMs. Single-cell suspensions from TdLNs were enriched in DCs by negative selection using a cocktail of biotinylated antibodies (anti-B220, CD3, NK1.1) and streptavidin microbeads (Miltenyi Biotec). Then, CD11c^med MHCII^hi CD11b^lo CD103⁺ migratory cDC1s, CD11c^med MHCII^hi CD11b⁺ CD103⁻ migratory cDC2s, CD11c^hi MHCII^med CD11b^lo CD8α⁺ resident cDC1s and CD11c^hi MHCII^med CD11b⁺ CD8α^lo resident cDC2s were sorted. A FACSAria Sorter (Becton Dickinson) and a Synergy 4L Cell Sorter (Sony) were used for purification of antigen-presenting cell subpopulations.

Spleens and inguinal LNs from CD45.1⁺ OTI mice were collected for purification of CD8⁺ T cells. After incubation with 50 µg/mL Liberase TL (Sigma-Aldrich) and 0.2 mg/mL DNase I (Sigma-Aldrich), lymphoid organs were smashed through 100 µm mesh filters to obtain single-cell suspensions. CD8⁺ T cells were purified using the EasySep mouse CD8⁺ T cell isolation kit (StemCell). Then, CD8⁺ T cells were stained with CellTrace Violet (Invitrogen, Molecular Probes).

Antigen presenting cells (APCs) were incubated at different cell-cell ratios with 22,500 CellTrace Violet-stained OTI cells for 72 hours. T cell proliferation was quantified by CellTrace Violet dilution by flow cytometry.

Tumor-infiltrating cDC1s were collected from pools of two to three mice with bilateral tumors from four experimental conditions (WT mice pretreated with an empty plasmid or FL and Clec9a^gfp/gfp mice pretreated with an empty plasmid or FL), as previously described using the gating strategy shown in online supplemental figure S1A. cDC1s were directly collected in RLT buffer (Qiagen) containing 10 µmol/mL β-mercaptoethanol and was subsequently purified using the RNeasy Plus Micro Kit (Qiagen). Barcoded RNA-seq libraries were prepared using the Nextera XT DNA Library Preparation Kit (Illumina). RNA sequencing was performed on three pools per condition using the HiSeq 2500 System (Illumina). For quantitative PCR, the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) was used to generate cDNA. Quantitative PCR was performed in a 7900-FAST-384 instrument (Applied Biosystems) by using the GoTaq qPCR master mix from Promega (Madison, Wisconsin, USA).

RNA-seq data analysis was performed by the Bioinformatics Unit of CNIC. Sequencing reads were processed with a pipeline that used FastQC to evaluate their quality, and cutadapt to trim sequencing reads, eliminating Illumina and SMARTer adaptor remains, and to discard reads that were shorter than 30 bp. Resulting reads were mapped against mouse transcriptome GRCm38.91, and gene expression levels were estimated with RNA-Seq by Expectation Maximization (RSEM). Around 88% of the reads from any sample participated in at least one reported alignment. Expression count matrices were then processed with an analysis pipeline that used Bioconductor package limma for normalization (using TMM (trimmed mean of M values) method) and differential expression testing. Changes in gene expression were considered significant if associated to Benjamini and Hochberg adjusted p value <0.1.

The expression profiles of a collection of 2731 genes detected as differentially expressed in any of the six performed contrasts were clustered using k-means with n=10. Several clusters, representing genes with altered expression in control versus mFlex comparisons (independently of genotype), were selected and merged into two metaclusters (MCs).

Comparative pairwise functional analyses of genome-wide transcriptome profiles were performed with gene set enrichment analysis (GSEA) to identify gene sets that had a tendency to be more expressed in either of the conditions being compared. Gene sets, representing functional categories or pathways from the Hallmark were analyzed. Enriched gene sets with family-wise error rate (FWER) <0.1 were considered of interest.

TCGA gene expression data for breast carcinoma (BRCA), cervical squamous carcinoma (CESC), human skin cutaneous melanoma (SKCM), LUAD, LUSC, HNSC and BLCA datasets was downloaded from the Broad Institute Firehose portal (gdac.broadinstitute.org). Normalized gene expression matrices ending ‘.uncv2.mRNAseq_RSEM_normalized_log2.txt’ were used for downstream analyses. Summarized pan-cancer clinical data was obtained from the SAGE synapse database (syn12026747) and additional information from clinical data files downloaded from cBioPortal. Prior to analysis, all normal tissue samples were removed (those with patient ID ending in 11 or 12). Gene signatures of interest were computed as mean log2-normalized gene expression of signature genes. Patients without overall survival data (OS.time=0 or NA) were removed. Patient stage was annotated as a continuous variable as follows: Stage I=1, Stage II=2, Stage III=3, Stage IV=4. Male patients with BRCA were removed. Patients were stratified according to quantiles by signature scores. Survival analysis was carried out using the survival (3.1–12) and survminer (0.4.6) R packages. We compared the product of the gene expression of FLT3LG and either CCL5 or CCR5 with signature scores for cDC1s CLNK, BATF3, XCR1, CLEC9A 4 and KIT, CCR7, BATF3, FLT3, ZBTB46, IRF8, BTLA and MYCL1.9

Prism (GraphPad Software) was used for statistical analysis. Tukey’s range test was used to detect outliers. After validating normal distribution of samples by the Shapiro-Wilk test, two-tailed unpaired Student’s t-test was used to evaluate statistical significance between two conditions or one-way analysis of variance (ANOVA) followed by Fisher’s least significant difference test for three or more conditions. To evaluate the statistical significance of the cross-presentation capacity of antigen-presenting cells and tumor growth curves, two-way ANOVA was performed. Figure legends indicate the number of repetitions for each experiment and the total number of independent mice included (n).

To assess the role of DNGR-1 in antitumor immunity, we inoculated B16F10 melanoma cells subcutaneously in the flank of WT and DNGR-1-deficient (Clec9a^gfp/gfp) mice. B16F10 melanomas developed similarly in WT and Clec9a^gfp/gfp mice (figure 1A). To address whether this result was cell line-specific or tumor type-specific, we inoculated WT and Clec9a^gfp/gfp mice with MC38 colon carcinomas. Again, we found that MC38 tumors grew similarly in WT and Clec9a^gfp/gfp mice (figure 1B). Following the flow cytometry-gating strategy detailed in online supplemental figure S1A,1 we first defined that DNGR-1 expression was restricted to cDC1s among tumor myeloid infiltrates by monitoring the GFP fluorescence derived from the Clec9a^gfp/gfp knockin cassette (online supplemental figure S1B). The restricted expression of DNGR-1 on cDC1s was further confirmed with anti-DNGR-1 antibodies (online supplemental figure S1C). Then, the analysis of immune cell infiltrates in B16F10 tumors showed that tumors from WT and Clec9a^gfp/gfp mice contained similar numbers of cDC1s and cDC2s (online supplemental figure S2A). In this line, TdLNs from WT and Clec9a^gfp/gfp displayed a similar composition in terms of resident and migratory cDC1s and cDC2s (online supplemental figure S2B). These data suggest that DNGR-1 does not affect tumor growth in the steady state.

We next tested whether DNGR-1 contributes to handling of tumor antigens in vivo. With this aim, we injected WT mice with B16ZsGreen tumors. ZsGreen is a green fluorescent protein resistant to lysosomal acid pH, allowing for tracing tumor antigen loading. Next, we treated them with a DNGR-1-blocking antibody intraperitoneally11 20 23 or its isotype control, starting at day 5 after tumor inoculation. After 13 days of tumor development, we analyzed the delivery of tumor-derived ZsGreen into DC subsets from TdLNs. We found that DNGR-1 blockade with monoclonal antibodies did not alter the loading of ZsGreen into migratory CD103⁺ or CD103^– DCs nor their resident counterparts (online supplemental figure S2C). Also, the frequency of ZsGreen⁺ cells in different DC subsets is consistent with previous studies,31 32 with a lower frequency of CD8α⁺ resident DCs bearing tumor-associated antigens (TAAs) (online supplemental figure S2C). We also injected WT and Clec9a^gfp/gfp mice with B78mCherryOVA tumors to analyze the delivery of mCherry, which is sensitive to lysosomal degradation, in DCs retrieved from TdLNs. In accordance with other studies,8 32 we found that migratory cDC1s are the only cells that can retain native tumor-derived mCherry and that the lack of DNGR-1 did not affect it (online supplemental figure S2D).

To directly address a potential role for DNGR-1 in cross-presentation of TAAs, WT and Clec9a^gfp/gfp mice were injected with B16OVA tumors and, at day 13 after tumor inoculation, we purified antigen-presenting subpopulations from tumors and TdLNs and cocultured them ex vivo with OVA-specific CD8⁺ T (OTI) cells. We observed that tumor-infiltrating or migratory cDC1s were slightly superior than cDC2, tumor-associated macrophages or LN-resident DCs at cross-presenting TAAs, but in a DNGR-1-independent manner (online supplemental figure S2E,F). These data indicate that absence of DNGR-1 does not affect tumor growth, DC infiltrates or the capacity of DCs to deliver TAAs to TdLNs or cross-present them.

Since Flt3L-mediated expansion of Batf3-dependent DCs boosts tumor-specific CD8⁺ T cells,7 we tested whether Flt3L effect on cDC1s could be affected by DNGR-1 in the context of tumors. As an initial approach, we tracked the growth of Flt3L-overexpressing B16 tumors (B16Flt3L). Notably, we found that tumor development was delayed in Clec9a^gfp/gfp compared with WT mice (figure 1C). Because Flt3L expression by cancer cells themselves might differentially regulate immune cell infiltration as soon as tumors become different in size, we performed gene therapy based on the hydrodynamic administration of a secreted Flt3L-encoding plasmid (FL), which results in liver production of the growth factor,7 to separately test its effect in parental B16F10 growth. After 14 days, systemic overexpression of FL in tumor-free WT or Clec9a^gfp/gfp mice did not affect the more primitive monocyte-DC progenitors (online supplemental figure S3A), but expanded common DC progenitors and pre-dendritic cells (pre-DCs) in bone marrow (online supplemental figure S3B,C). Furthermore, FL also drove a prominent expansion of circulating cDC1s and cDC2s (online supplemental figure S3D,E). FL administration promoted the expansion of cDC1s in both spleen (online supplemental figure S3F) and ear skin (online supplemental figure S3G) compared with mice receiving an empty plasmid. Also, cDC2s expanded to a lower extent in both spleen (online supplemental figure S3H) and periphery (online supplemental figure S3I). In both organs, immune populations that do not rely on Flt3L for their expansion, such as neutrophils or macrophages, remained at similar levels in mice treated with FL (online supplemental figure S3J,K). Therefore, FL expanded both cDC1s and cDC2s and their related progenitors. Importantly, FL-mediated DC expansion was equivalent between WT or Clec9a^gfp/gfp mice (online supplemental figure S3). We then treated WT or Clec9a^gfp/gfp mice with FL and, 24 hours later, injected them subcutaneously with B16F10 or MC38 tumors. Of note, Clec9a^gfp/gfp mice displayed a reduced tumor growth (figure 1D,E). These results suggest that DNGR-1 absence limits tumor growth on Flt3L therapy.

As DNGR-1 is a cell surface receptor that possibly senses necrosis in the tumor context, this interaction could be blocked by the use of anti-DNGR-1-blocking antibodies.11 20 23 We pretreated WT mice with FL and, 1 day later, inoculated them subcutaneously with B16F10 or MC38 tumors. As tumors grew, we inoculated mice with anti-DNGR-1-blocking antibodies or isotype control. DNGR-1 blockade in the presence of FL delayed tumor growth (figure 1F,G). These results indicate that DNGR-1 blockade combined with Flt3L can be therapeutically targeted as a tumor immunotherapy.

To explore the cellular mechanisms underlying protection in DNGR-1-deficient settings following FL treatment, we studied the immune infiltrates of tumors from WT and Clec9a^gfp/gfp mice pretreated with FL. Consistent with improved antitumor immunity, Clec9a^gfp/gfp mice contained more CD8⁺ T cells within their tumors (figure 2A), but not more CD4⁺ T cells (figure 2B), when compared with WT controls. Similarly, therapeutic blockade of DNGR-1 resulted in increased CD8⁺ T cell infiltrates in FL-treated mice (figure 2C), but not in CD4⁺ T cells (figure 2D).

Since cDC1s are a key cell type for recruitment of CD8 T cells into tumors,33 we evaluated the role of cDC1s in the increased CD8⁺ T cell infiltration of tumors on DNGR-1 blockade, in the presence of FL. For this purpose, we used Batf3^–/– mice, which have a deficient cDC1 compartment.6 7 In contrast with WT mice, the administration of anti-DNGR-1-blocking antibodies did not increase the accumulation of CD8⁺ T cells in Batf3^–/– mice (figure 2E), suggesting that DNGR-1 blockade mediates its effect through cDC1s. These differences were only observed in CD8⁺ T cells, as CD4⁺ T cell infiltration remained similar in all experimental groups (figure 2F). Furthermore, the adaptive immune compartment was necessary for the protection against tumor growth, as B16F10 tumors grew at similar rates in FL-treated Rag1^–/– and Rag1^–/–Clec9a^gfp/gfp mice (figure 2G). Together, these data indicate that, on FL overexpression, DNGR-1 restricts antitumor CD8⁺ T cell responses that are dependent on cDC1s.

Given the restricted expression of DNGR-1 in cDC1s (online supplemental figure S1B,C), we studied its impact in cDC1 infiltration within tumors. Contrary to the similar DC infiltrates in steady state (online supplemental figure S2A,B), on FL treatment, the accumulation of CD103⁺ cDC1s within tumors in Clec9a^gfp/gfp mice was boosted compared with WT mice (Figure 3B and online supplemental figure S4A), without affecting tumor infiltration by cDC2s (Figure 3B and online supplemental figure S4B). The increased infiltration of cDC1s, but not cDC2s, within B16F10 tumors was recapitulated in FL-treated WT mice receiving anti-DNGR-1-blocking antibodies (figure 3C,D, online supplemental figure S4C,D). On the contrary, the numbers of the different DC populations at TdLNs remained comparable between WT and Clec9a^gfp/gfp mice treated with FL (online supplemental figure S4E–H).

To evaluate whether the increased cDC1 infiltration was dependent on the increased CD8⁺ T cell recruitment found on DNGR-1-deficient settings, we analyzed the DC compartment within B16F10 tumors from Rag1^–/– and Rag1^–/–Clec9a^gfp/gfp mice treated with FL. Despite having the same size (figure 2G), B16F10 tumors growing in Rag1^–/–Clec9a^gfp/gfp mice contained more tumor-infiltrating cDC1s when compared with Rag1^–/– controls (figure 3E), whereas numbers of cDC2s were similar (figure 3F). Thus, the accumulation of tumor-infiltrating cDC1s driven by DNGR-1 deficiency occurs independently of the tumor size and of the adaptive immune compartment. This result suggests that the increased infiltration of cDC1s on FL in DNGR-1-deficient settings precedes and drives adaptive antitumor immunity.

To test whether the effect of DNGR-1 on tumor growth and cDC1 infiltration was directly dependent on cDC1s, we treated WT or Batf3^–/– mice with FL and inoculated them subcutaneously with B16F10 tumor cells. While WT mice showed delayed tumor growth on DNGR-1 blockade, this protection was lost in Batf3^–/– mice (figure 3G). Accordingly, Batf3^–/– mice failed to increase the infiltration of cDC1s within their tumors (figure 3H), while cDC2s remained at similar levels across all experimental conditions (figure 3I). These results indicate that, in the presence of FL, DNGR-1 expression on cDC1s restricts the recruitment of cDC1s and CD8 T cells into the tumor.

To characterize the molecular mechanisms by which DNGR-1-deficient cDC1s enhance antitumor responses on FL treatment, we performed transcriptomics analysis on cDC1s purified from B16F10 tumors from WT and DNGR-1-deficient mice, either treated with FL or an empty plasmid (control). GSEA revealed seven gene sets modulated by FL in both WT and Clec9a^gfp/gfp cDC1s, TNFα signaling via NFκB, epithelial mesenchymal transition, inflammatory response, myogenesis, G2M checkpoint, E2F and Myc targets (figure 4A upper panel, online supplemental figure S5A, B). Among these, gene sets related to cDC1 activation, such as the TNFα signaling via NFκB and inflammatory response, were enriched on FL treatment in cDC1s from Clec9a^gfp/gfp mice compared with those from WT mice (figure 4A lower panel), as well as other immune-related gene sets such as the IFNα response (online supplemental figure S5C). These data suggest that the absence of DNGR-1 enhances some of the activation signatures imprinted by FL.

Out of the total genes differentially expressed in any condition, and according to their behavior across samples, k-means clusters were unified into two MCs of relevance for this study (online supplemental figure S6A). On the one hand, MC1 identified genes whose expression is decreased in WT cDC1s on FL injection, and further downregulated in cDC1s from FL-treated Clec9a^gfp/gfp mice (figure 4B,C). On the other hand, MC2 comprised clusters 2 and 9 and identified genes that are induced in WT mice on FL injection and are further upregulated in Clec9a^gfp/gfp cDC1s also on FL (figure 4B,C). Notably, MC2 includes cDC1 costimulatory molecules such as CD80, CD83 or CD70, cytokines such as IL12b or IL15 and chemokine receptors such as CCR7 and chemokines including CCL5. We validated by flow cytometry some relevant features from the RNA-Seq data. Using FL-treated mice that received anti-DNGR-1-blocking antibodies, we confirmed the increase in IL12p40 and CCR7 expression in cDC1s (online supplemental figure S6B,D), but not in cDC2s (online supplemental figure S6C,E). Altogether, these results suggest that DNGR-1 restrains the activation of tumor-infiltrating cDC1s induced by FL.

Based on our transcriptomic analysis of tumor-infiltrating cDC1s, we focused on chemokine genes as the potential mediators of cDC1s infiltration within the tumor bed. We identified Ccl5 and Ccl3 as chemokines induced in DNGR-1-deficient cDC1s versus WT cDC1s on FL administration (figure 5A). However, global Ccl3 expression was low (figure 5A), suggesting that the highly expressed Ccl5 could play a more relevant role in cDC1 recruitment. Indeed, analysis of dataset GSE1590734 indicates that peripheral migratory mDC1s express Ccl5 levels comparable to NK cells, while the expression level of Ccl3 by cDC1s is negligible (figure 5B). Indeed, we confirmed that Ccl5 was highly expressed in cDC1s from DNGR-1-deficient mice on FL administration compared with cDC1s from WT mice under the same conditions, but also compared with cDC1s from either WT or DNGR-1-deficient mice administered with an empty plasmid (figure 5C).

CCR5 is the chemokine receptor for CCL5, but also the main infection route for the HIV. Several drugs that block the entry of the virus have been developed such as maraviroc (MVC), which also blocks CCR5 signaling and migration in response to CCL5.35 To assess whether the CCL5/CCR5 axis is responsible for the enhanced accumulation of cDC1s in tumors from FL-treated mice in DNGR-1-deficient settings, we tested the effect of MVC in tumor growth and DC infiltration. WT mice were treated with FL and injected subcutaneously with B16F10 tumors. During tumor development, mice were treated with either anti-DNGR-1-blocking antibodies or their isotype control and, simultaneously, with MVC or vehicle. Notably, coadministration of MVC reverted the protection provided by anti-DNGR-1 antibodies and FL (figure 5D) and, consistently, impeded the enhanced accumulation of cDC1s on blockade of DNGR-1 (figure 5E). Besides, the infiltration of cDC2s was not affected in any treatment regime (figure 5F). Together, these data indicate that, in the presence of Flt3L, DNGR-1 deficiency in cDC1s boosts CCL5 expression, suggesting that cDC1s may contribute to the recruitment of more cDC1s through the production of CCL5.

Both FLT3LG and CCL5 have been separately associated with the amount of cDC1s within the TME,4 9 but based on our recent findings we hypothesized that the interplay of both factors might impact the infiltration of cDC1s into human cancers and, thus, affect their prognosis. For this reason, we generated an FLT3LG/CCL5 expression signature for patients from TCGA and stratified them into low (first quartile), intermediate (second and third quartiles) and high (fourth quartile). Notably, we found a strong correlation of FLT3LG and CCL5 expression in the case of metastatic human skin cutaneous melanomas (M-SKCM), BRCAs and CESCs (figure 6A). In addition, we calculated a cDC1 score based on the gene expression signature based on expression of KIT, CCR7, BATF3, FLT3, ZBTB46, IRF8, BTLA and MYCL1 as explained in Methods and described by Barry et al.9 The cDC1 signature score progressively increased with the FLT3LG/CCL5 expression signature (figure 6B). Importantly, patients that coexpressed high transcript levels of FLT3LG and CCL5 experienced longer overall survival compared with patients bearing an intermediate and low expression of the FLT3LG/CCL5 signature (figure 6C). These data support the involvement of CCL5 in cooperation with FLT3LG in the cDC1 infiltration and antitumor response.

Of note, when we replaced CCL5 by CCR5 in the analysis, the behavior was similar. FLT3LG expression also correlated with CCR5 expression within tumors from patients with M-SKCM, BRCA or CESC (online supplemental figure S7A). Segregating patients into low, intermediate and high FLT3LG/CCR5 expressors based on quartiles as indicated above also identified a correlation between the FLT3LG/CCR5 signature and the cDC1 signature score (online supplemental figure S7B) and an association of high expression of the FLT3LG/CCR5 signature with a better overall survival (online supplemental figure S7C). To compare the predictive power of our FLT3LG/CCL5 or FLT3LG/CCR5 signatures with previously described cDC1 gene signatures,4 9 we selected the highest and lowest quartiles of patients with TCGA with M-SKCM, BRCA and CESC according to Böttcher et al4 (online supplemental figure S7D) or Barry et al9 (online supplemental figure S7E) and analyzed their prognostic power. We observed that the HR and p value of FLT3LG/CCL5 or FLT3LG/CCR5 signatures were similar to those of cDC1 scores and even better in the case of M-SKCM (online supplemental figure S7F). These data indicate that the coexpression of FLT3LG and both CCL5 and CCR5 is a simplified biomarker for cDC1 tumor score and overall survival in several types of human cancers with similar or even improved predictive power than whole cDC1 signatures.

Given that some chemotherapies, such as cyclophosphamide, are linked to increased expression of Flt3L,36 we hypothesized that DNGR-1 absence could enhance tumor therapy in these settings. First, we confirmed that the administration of the chemotherapeutic agent cyclophosphamide increased serum levels of Flt3L (figure 6D). Of note, under these pharmacological settings of enhanced Flt3L concentration, the absence of DNGR-1 synergized with the intrinsic effect of cyclophosphamide, delaying B16 tumor growth (figure 6E) and extending survival (figure 6F). These data represent a proof-of-concept for a potential chemo-immunotherapy based on the administration of an Flt3L-inducing drug such as cyclophosphamide in combination with DNGR-1 deficiency, as a boost for the intrinsic antitumor effect of the chemotherapy.