Transcriptional and Functional Analysis of CD1c+ Human Dendritic Cells Identifies a CD163+ Subset Priming CD8+CD103+ T Cells

1Human dendritic cells (DCs) are sentinel cells of the immune system specialized in controlling T cell function. The mouse model has brought important concepts to our understanding of DCs and suggests that multiple DC subsets arising from specialized ontogenetic pathways are endowed with specific immune functions.

2Definition of human DC subsets is a prerequisite to understanding the division of labor underpinning induction of various types of immune responses. At homeostasis, conventional DCs (cDCs) include cDC1s (CD141+XCR1+CLEC9A+IRF8+) and cDC2s (CD1c+CD11c+CD172a+IRF4+). cDC1s and cDC2s arise through a specialized ontogenetic pathway from a common DC precursor (CDP) or from early IRF8+ multipotent lympho-myeloid progenitors (MLPs). Bone marrow progenitors for cDCs generate a common circulating precursor that progressively diverges into pre-cDC1 and pre-cDC2. This is further complicated by inclusion of AXL+SIGLEC6+CD11c+CD1c+ cells (AS-DCs also called type 5 DCs), which have been proposed to act as precursors for cDCs (pre-cDCs) or a lineage on its own.

3An additional layer of complexity in the DC network lies in its responsiveness to perturbations. For instance, inflammation affects hematopoiesis and phagocyte trafficking, resulting in leukocyte mobilization and tissue infiltration. Specifically, inflammation affects DC diversity and triggers mobilization of CD14+CD1c+ DCs, called inflammatory DCs (iDCs). The expression of CD1c lectin is shared between iDCs and cDCs. However, CD1c+ iDCs also express multiple monocytic markers, such as CD14, CCR2, and FcγRI/CD64. iDCs have been reported in inflamed skin, synovial fluid, ovarian cancer ascites, solid tumor infiltrates, and lymph nodes. The developmental pathway of human CD1c+CD14+ iDCs is poorly understood. In vitro studies suggest that iDCs obtained after differentiation of CD14+ monocytes in granulocyte macrophage colony-stimulating factor (GM-CSF) and interleukin-4 (IL-4) might correspond to in vivo iDCs. In this context, IL-4 acts through induction of the transcriptional regulator NCOR2. In addition, triggering the aryl hydrocarbon receptor in monocytes supports activation of IRF4-dependent differentiation of iDCs. Together, these studies support the prevailing notion that CD14+ monocytes act as immediate precursors for iDCs.

4Re-evaluation of circulating mononuclear phagocyte diversity has been enabled by single-cell RNA sequencing (scRNA-seq). Recent studies have revealed that a subset of DC-like cells, called DC3s, express mRNA for the CD14 and CD1c genes. However, this analysis was performed after excluding cells expressing the highest amount of CD14. As a consequence, this approach renders a problematic distinction between DC3s and bona fide CD14+ monocytes. This discrimination is further complicated by previous reports of CD14+CD1c+ “inflammatory” DCs recruited at inflammatory sites.

5Here we intended to re-evaluate the definition of DC3s using unbiased scRNA-seq and high-dimensional flow cytometry by exploring the full spectrum of CD14 and CD1c expression. In addition, we identify DC3 growth factor requirements and developmental pathways. Finally, we show that DC3s activate CD103+ T cells and that DC3 infiltration in human breast tumors correlates with the abundance of CD8+CD103+CD69+ tissue-resident memory (TRM) T cells.

6To probe the diversity of CD16−CD141−CD123− blood mononuclear phagocytes, we developed a sorting strategy including all phenotypic intermediates between CD14hiCD1clo and CD14loCD1chi cells. The proportions between cell populations were compensated to enrich in less abundant CD14loCD1chi cells (Figure S1A). Flow cytometry-sorted cells isolated from blood were analyzed using a droplet-based scRNA-seq approach (Figure 1A; Figure S1A). We found that cells expressing CD14 and/or CD1c could be separated into four CD33+ clusters (A, B, C, and D) (Figure 1A; Figure S1B). Contaminating clusters containing B and T lymphocytes and neutrophils were excluded from the analysis (Figure S1B). Hierarchical clustering performed on averaged single cell expression data within clusters showed that A and B were closer to each other than any of the other subsets (Figures 1B–1D). Cluster D fell between the group of clusters A and B and cluster C (Figure 1B). Classical cDC2 markers, such as CLEC10A, FCER1A, and major histocompatibility complex (MHC) class II genes (HLA-DQA, HLA-DPA, HLA-DRA, and the MHC class II-associated invariant chain CD74) were expressed prominently in clusters A, B, and D (Figures 1D and 1E; Figure S1C;;; ). In contrast, monocytic markers such as CD14, S100A8, S100A9, S100A12, and VCAN were more expressed in clusters C and D, with higher expression in C compared with D (Figures 1D and 1E). Finally, expression of the C5 receptor C5AR1 (CD88) was found to be restricted to cluster C together with SOD2 and RBP7 (Figures 1D and 1E).

7To determine whether our clustering reflects previously published data, we decided to evaluate the expression of gene signatures obtained from. We found that signature genes discriminating cDC2 within CD14lo cells were mostly represented in clusters A and B, confirming their identity (Figure 1F). Likewise, signature genes defining DC3s within CD14lo cells and CD14+ monocytes were significantly enriched in cluster D and cluster C, respectively (Figure 1F). Of note, genes enriched in DC3s compared with cDC2s (S100A8, S100A9, and CD14) were highly expressed in clusters C and D (Figure 1E; Figure S1E).This underlines the need to integrate a monocyte reference in any comparison aiming to define DC3s. We conclude that scRNA-seq analysis identifies DC3s (cluster D) as a specific subset sharing transcriptional features of cDC2s (clusters A and B) and monocytes (cluster C).

8Next we sought to define a flow cytometry-based strategy enabling analysis and prospective isolation of DC3s in blood. To this end, we performed an unsupervised flow cytometry data analysis based on genes identified by scRNA-seq (e.g., CD88, CD14, FcεRI, and CD1c) as well as markers previously associated with cDC2s (e.g., BTLA and CD5) and DC3s (e.g., CD163), even though they were not detected in the scRNA-seq analysis. We identified three main clusters (1, 2, and 3) together with a rarer cluster (4) of CD5hi cells (Figure 1G; Figure S1D). Cluster 4 appeared to be CD123hi contamination of AS-DCs (Figure S1D). Cluster 1 highly expressed CD88, aligning with monocyte cluster C identified by scRNA-seq. Cluster 2 did not express the monocyte-associated markers CD14 and CD88 but was characterized by expression of CD1c, FcεRI, and B and T lymphocyte attenuator (BTLA) (Figure 1G). In addition, cluster 2 showed heterogeneous expression of CD5 (Figure 1G). Similar to cluster 2, cluster 3 did not express the monocytic marker CD88 and displayed higher amounts of CD1c and FcεRI. However, cluster 3 could be distinguished from cluster 2 by higher expression of CD163 and heterogeneous expression of CD14 (Figure 1G). The unsupervised flow cytometry analysis allowed us to define a simple gating strategy enabling prospective isolation of monocytes (CD88+CD14+), cDC2s (CD5+ and CD5−CD88−CD1c+BTLA+CD163−), and DC3s (CD88−CD1c+BTLA−CD163+) (Figure 1H).

9To further validate the flow cytometry-based identification of cell subsets, we performed bulk RNA-seq analysis of sorted monocytes, cDC2s (CD5+ and CD5−), DC3s, plasmacytoid DCs (pDCs), and AS-DCs (Figure S1E). Principal-component analysis (PCA) indicated that, overall, CD5+ and CD5− cDC2s, DC3s, and monocytes separated from pDCs and AS-DCs along the PC1 axis, accounting for 61% variance (Figure 1I). CD5+ and CD5− cDC2 clustered closely together, and DC3s separated clearly from cDC2s and monocytes (Figure 1I; Figure S1F). Hierarchical clustering (HC) and differentially expressed gene analysis led to the same conclusion, with DC3s sitting between monocytes and cDC2s (CD5+ and CD5−) (Figure 1J; Figure S1G). Overall, DC3s are closer to cDC2s than monocytes (Figures 1I and 1J).

10Together, this validates that cellular clusters isolated by a flow cytometry-based approach align to clusters identified by unbiased scRNA-seq (Figure 1K).

11We conclude that DC3s are a separable entity within CD1c+ cells, defined by a distinct gene expression profile, and that they can be prospectively isolated using CD88, CD1c, CD163, and BTLA (Figure S1H). scRNA-seq and bulk gene expression profiling identify markers shared between DC3s and cDC2s (e.g., CLEC10A and FCER1A) and markers shared between DC3s and monocytes (such as S100A8, S100A9, CD14, and CD163).

12Tumor-infiltrating CD14+CD1c+ DCs have been reported in multiple instances, including ovarian cancer ascites, breast cancer, and melanoma. Therefore, we asked whether CD14+CD1c+ cells would align with DC3s. To this end, we analyzed mononuclear phagocytes infiltrating luminal breast cancer primary tumors. Using the gating strategy described in Figure 1, we found that, after exclusion of CD88+ monocytes and macrophages, the remaining CD45+HLA-DR+CD123−CD88− fraction contained cDC1s, CD14−CD1c+CD163−CD5+ cDC2s (CD5+ cDC2s), and CD14+CD1c+CD163+CD5− DC3s (CD14+ DC3s) (Figure 2A; Figure S2A). At this stage, we observed that CD163 and BTLA were particularly susceptible to enzymatic digestion of solid tissue, preventing consistent and reliable quantification of cells throughout the cohort of samples. Hence, we adopted a more restrictive definition of cDC2s and DC3s as CD1c+CD14−CD5+ and CD1c+CD14+CD5− cells, respectively (Figure 2A). This strategy enabled isolation of CD5+ cDC2s and CD14+ DC3s in peripheral tissues even though it might result in underestimation of their absolute numbers. Nevertheless, both subsets aligned phenotypically with their blood counterparts (Figures S2B and S2C) and expressed markers reported previously for CD1c+CD14+ iDCs, such as CD11c and FcεRI (Figure 2A; ).

13DC3s were consistently identified in 25 samples of primary tumors of clinical stages I, II, and III (Figure S2D). Macrophages represented by far the most abundant population. DC3s outnumbered cDC1s but were on par with bona fide CD5+ cDC2s (Figure 2B). The relative abundance of DC3s did not correlate with disease progression (Figure S2D) or with macrophages or cDCs (Figure 2C). In contrast, cDC1 infiltration correlated with cDC2s (Figure 2C).

14Bulk RNA-seq analysis of CD1c+CD14+ cells sorted from tumor-invaded lymph nodes indicated that they displayed a similar expression profile as blood DC3s (high expression of cDC2 markers such as CLEC10A and FCER1A combined with low expression of monocyte-associated markers such as C5AR1 and SOD2) (Figure 2D). Gene set enrichment analysis (GSEA) revealed that CD14+CD1c+ were more enriched for the DC3 > cDC2 signature compared with CD1c+ cells (normalized enrichment score [NES], 2.06; p = 0.0;; ). In addition, CD14+CD1c+ were more enriched for the DC3 > monocyte (Mono) signature compared with CD14+ cells (NES, 1.42; p = 0.02;; Figure 2E). Conversely, compared with CD1c+CD14+, the cDC2 > DC3 and Mono > DC3 gene signatures were enriched in CD1c+ (NES, −1.57; p = 0.004) and CD14+ (NES, −2.07; p = 0.0) cells, respectively (Figure 2E).

15We conclude that CD11c+FcεRI +CD14+CD1c+ iDCs infiltrating breast cancer align with DC3s.

16Inflammatory cues promoting mobilization of CD14+CD1c+ cells at the site of inflammation are not fully defined. Mouse studies define GM-CSF as a likely candidate. For this reason, we decided to test whether GM-CSF was sufficient to mobilize human CD14+CD1c+ DCs in a humanized mouse metastatic lung model. We generated B16 mouse melanoma engineered to overexpress human GM-CSF (B16_huGM) or FLT3L (B16_huFLT3L) (Figure S3A). Immunodeficient NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice were injected intravenously with B16 control (CTRL), B16_huGM, or B16_huFLT3L (Figure 3A). Lung metastasis-bearing mice were engrafted with human peripheral blood mononuclear cells (PBMCs). Two days later, human CD45+ leukocytes were found in lung tumor foci and juxta-tumor areas (Figure 3B; Figures S3B and S3C). Flow cytometry analysis of metastatic lungs showed that FLT3L, but not GM-CSF, promotes expansion of CD1c+CD5+ cells aligning with blood cDC2s (Figure 3C, blue). In contrast, GM-CSF, but not FLT3L, led to accumulation of CD1c+CD14+ cells aligning with blood DC3s (Figure 3C, orange; Figure S3D, orange). As shown for circulating peripheral blood subsets in Figure 1, FLT3L-dependent cDC2s and GM-CSF-dependent DC3s shared expression of Clec10A, CD11c, and FcεRI (Figure 3C; Figure S3E). All tumor-bearing lungs contained some monocytes and/or macrophages (CD14+CD88+) (Figure 3C; Figure S3B). To further establish alignment of GM-CSF-mobilized DC3s, we performed an unbiased scRNA-seq on human CD45+HLA-DR+ cells expressing CD14 and/or CD1c (Figures S3F and S3G). Two major clusters could be identified. Cluster 0 was characterized by expression of genes associated with DC3s and cDC2s, such as CLEC10A, FCER1A, and CD74 (Figures S3H and S3I), and was enriched in DC3 differentially expressed transcripts (; Figure S3J). Conversely, cluster 1 expressed markers defining monocytes (SOD2, C5AR1, and G0S2) and CXCL2, CXCL3, and CXCL8 chemokines (Figures S3H and S3I). We conclude that GM-CSF drives mobilization of CD14+CD1c+ cells aligning phenotypically with circulating DC3s.

17We next wondered whether GM-CSF could induce trans-differentiation of circulating cDC2s or monocytes into CD14+CD1c+ DC3s. To test this hypothesis, cDC2s, DC3s, and monocytes were sorted by flow cytometry from blood and cocultured in vitro in the presence of GM-CSF-expressing stromal cells (MS5_GM). After 2 days, cDC2s did not acquire CD163 or CD14 expression (Figure 3D). In contrast, DC3s did upregulate CD14 (Figures 3D and 3E). We found that monocytes differentiated into CD16+ and CD16− macrophages (Figure 3D). Importantly, CD88+CD1c−CD14+ monocytes did not give rise to CD1chiCD163hi DC3s (Figure 3D). Addition of IL-4 to GM-CSF culture did not affect the outcome (Figure S4A). The lack of trans-differentiation of cDC2s or monocytes into DC3s was also confirmed in vivo upon adoptive transfer into immunodeficient NSG mice carrying GM-CSF-expressing tumors (Figures S4B and S4C). Overall, we conclude that CD88−CD14+CD1c+ cells can differentiate from DC3s independent of cDC or monocytic lineages.

18We next wondered how DC3s would differentiate from bone marrow progenitors. To this end, we sought to define a culture system capable of generating DC3s together with cDC2s and macrophages. We cocultured human cord blood-derived CD34+ hematopoietic stem and progenitor cells (HSPCs) with stromal cells engineered to overexpress cDC-promoting factors (membrane-bound FLT3L together with stem cell factor (SCF) and CXCL12 [MS5_FS12]; ) in the presence or absence of GM-CSF (Figure 4A; Figure S5A). We found that GM-CSF increased differentiation of CD1c+CD14+CD163+ cells (orange) phenotypically aligning with blood DC3s (Figures 4A and 4B). CD1c+CD14− cells did not express CD163, suggesting that they mostly align with cDC2s. In addition, CD163 expression was restricted to CD1c+CD14+ cells (Figure 4B). GM-CSF alone (MS5_GM), but not FLT3L, was sufficient to induce differentiation of DC3s in vitro (Figure 4C). In contrast, FLT3L (MS5_FL) was sufficient to induce differentiation of cDC2s (Figure 4C). We conclude that cDC2s and DC3s have distinct growth factor requirements.

19We next intended to determine whether the transcriptional landscape of in-vitro-generated DC3s from stromal cell cocultures aligned with their in vivo counterparts. Bulk RNA-seq analysis of in-vitro-generated cells showed that CD14+CD1c+ DC3s (orange) sat between cDC2s (blue and turquoise) and macrophages (gray and brown) (Figure 4D). Of note, CD1c+CD14− cells generated in vitro were heterogenous for CD206 expression and were therefore analyzed as two independent subsets (CD1c+CD206− [blue] and CD1c+CD206+ [turquoise]; Figure S5B). However, both subsets displayed very similar transcriptomes regardless of CD206 expression, and the CD1c+CD14−CD206+ and CD1c+CD14−CD206− fractions strongly resembled circulating cDC2s.

20Further analysis of differentially expressed genes (DEGs) showed that in vitro DC3s differed from in vitro cDC2s by expression of monocyte-associated markers such as S100A8, S100A9, S100A12, and CD14 (Figure 4E; Figure S5C). Conversely, in vitro cDC2s displayed higher expression of CD1C, LAMP3, CD52, and LTB, as reported for lung cDC2s (; Figure 4E; Figure S5C). Using the GSEA methodology, we found that the set of genes upregulated in primary DC3s compared with cDC2s (DC3 > cDC2 from ) was enriched in GM-CSF-dependent in vitro DC3s compared with in vitro cDC2s (Figure 4F). Of note, GM-CSF-exposed cDC2s did not convert to DC3s (Figures S5D and S5E). We found that markers common for primary cDC2s and DC3s (CLEC10A, Figure 1) were higher in in vitro DC3s compared with in vitro macrophages (Figure 4E; ). Conversely, in vitro macrophages expressed more FCGR1A, C5AR1 (CD88), CXCL8, CXCL1, CXCL2, CXCL3, CCL2, CCL3, and CCL7 compared with in vitro DC3s (; Figure 4E; Figures S5C and S5F). Genes upregulated in primary DC3s compared with CD14+ monocytes (DC3 > CD14+ Mono from ) were significantly enriched in in vitro DC3s compared with macrophages (Figures 4F and 4G). We conclude that GM-CSF drives, in vitro, the differentiation of DC3s aligning to primary blood DC3s.

21The developmental relationship between DC3s and cDCs or monocyte lineages is not known. A classical view defines the development of phagocytes as a stepwise and ordered loss of developmental potential concomitant with lineage commitment. Historically, this process has been identified using prospective isolation of progenitor populations of decreasing potential. Specifically, early granulocyte-monocyte and DC progenitors (GMDPs) carries a tri-lineage potential. Loss of neutrophil potential defines monocyte and DC progenitors (MDPs). Loss of monocyte potential defines CDPs, which generate cDCs via a pre-cDC intermediate. Finally, loss of DC and neutrophil potential defines monocyte-committed progenitors (cMoPs). Having established the growth factor requirements for DC3 development, we decided to test the contribution of CDPs and cMoPs to generation of DC3s. Flow cytometry-sorted CDPs and cMoPs and the remaining GMDP-containing fraction (CD34+CD38+CD45RA+CD123−CD64−) isolated from cord blood-derived HSPCs (see the gating strategy in Figure S5G) were cocultured with stromal cells supporting cDC (MS5_FS12) or DC3 (MS5_GM) differentiation.

22Flow cytometry analysis after 7 days of culture showed that CDP gave rise exclusively to cDC1s and cDC2s in MS5_FS12 cocultures, as described previously (Figure 4H; Figure S5H; ). cMoP cultures gave rise solely to CD14+CD1c− cells, as described previously (Figure 4H; Figure S5H; ). As expected, the GMDP-containing fraction gave rise to granulocytes, macrophages, and cDC2s. Importantly, the GMDP-containing fraction also gave rise to DC3s in MS5_GM coculture (Figure 4H; Figure S5H). Therefore, we asked whether DC3s would arise directly from a multipotent progenitor or via formation of an intermediate DC3-committed progenitor devoid of any other lineage potential.

23To address this question in unbiased settings, we developed a single cell culture of CD34+CD38+CD123−CD64− progenitors distinct from CDPs or cMoPs (Figure 5A). We chose to combine MS5_FS12 with soluble GM-CSF for two reasons: (1) GM-CSF alone did not support growth of individual progenitors (Figure S5I), and (2) MS5_FS12 coculture was found to more efficiently support cDC and monocyte differentiation (Figure 4G).

24Flow cytometry analysis of 14-day progeny of individual progenitor cultures revealed multiple patterns of developmental potential (Figure 5A). Overall, the granulocyte potential was segregated from the potential for mononuclear phagocytes (Figure 5B). We found that only 0.3% of progenitors could differentiate into all four lineages corresponding to the GMDP functional definition (Figure 5D). Most individual progenitors gave rise to single lineage progeny (Figures 5C and 5D). Unipotent wells containing only neutrophils represented the most abundant outcome (39.4%), followed by macrophage-only (26.8%) and cDC-only (5.63%) wells (Figure 5C). Most individual progenitors endowed with DC3 potential had multi-lineage potential. DC3 potential was more associated with mononuclear phagocytes rather than granulocyte potential (61.4% and 19.3% of DC3-containing wells, respectively) (Figure 5D). Importantly, we also identified a minor fraction of individual progenitors giving rise exclusively to DC3s (Figures 5B–5D).

25We next aimed to further define the cell surface phenotype of progenitors endowed with DC3 potential. To this end, we attempted to establish correlations between the cell surface phenotype of single sorted cells (inferred from index flow cytometry sorting) and their developmental potential. A posteriori identification of the cell surface phenotype of DC3-committed progenitors revealed that they had a CD34+CD38+CD45RAintCD123−CD64−SIRPαloCD10−Clec12A+ phenotype (Figure 5E). As validation, we showed that the potential for DC3s lay in the Clec12A+ population of the GMDP-containing fraction (Figure 5F). We conclude that DC3s can develop via a DC3-restricted intermediate distinct from cDC-restricted CDPs or monocyte-restricted cMoPs. Even though the existence of a DC3-commited unipotent progenitor is not formally proven, our data are compatible with the notion that DC3 specification arises downstream of MDP.

26We next aimed to understand the immunological function of DC3s and to compare it with cDC2s and monocytes. First, we decided to test the responsiveness of DC3s, cDC2s, and monocytes to a cocktail of Toll-like receptor (TLR) agonists. cDC2s, DC3s, and monocytes were sorted by flow cytometry from blood and stimulated overnight ex vivo. PCA analysis of the total transcriptome of unstimulated and stimulated populations evidenced that all subsets underwent a certain degree of convergence in their transcriptome (Figure 6A). In support of this, we found an important overlap in the set of activation-induced genes defined for each subset (1,344 genes; Figure 6B). Despite the relative convergence of activated cells, we found that overnight activation did not compromise cell surface discrimination of DC3s from cDC2s and monocytes (Figure S6A). Indeed, TLR-activated DC3s could still be discriminated from TLR-activated cDC2s by 437 DEGs or from TLR-activated monocytes by 1,293 genes (Figure 6C). The same was true for the pairwise comparison of activated cDC2s and circulating DC3s (Figure S6B). In sum, we conclude that innate activation does not trigger conversion of cDC2s or monocytes into DC3s despite induction of a common transcriptional response to TLR stimulation.

27From the perspective of adaptive immunity, activated DC3s shared a lot of common features with activated cDC2s but less with activated monocytes: (1) stimulated DC3s and cDC2s upregulated CCR7 upon activation (Figure 6D), potentially enabling their ability to migrate toward T cell zones; (2) activated DC3s and cDC2s upregulated cell surface co-stimulatory molecules (CD80, CD86, CD70, and CD40; Figure 6E and 6F); (3) activated DC3s and cDC2s efficiently increased the expression of T cell-attracting chemokines such as CCL5 (Figures 6G and 6H), CCL19, CCL17, CCL22, CXCL9, CXCL10, CXCL11, and CXCL13 (Figure S6C); and (4) activated DC3s and cDC2s produced higher amounts of IL-12p70, IL-23, IL27, and IL-10 (Figures 6G and 6H; Figure S6E).

28In addition, activated DC3s shared some common features with activated monocytes that are less pronounced in activated cDC2s: (i) activated DC3s and monocytes secreted more inflammatory cytokines, such as tumor necrosis factor alpha (TNF-α) and IL-1β (Figures 6G and 6H); and (2) activated DC3s and activated monocytes upregulated inflammatory chemokines, such as CCL2 (Figures 6G and 6H), CCL1, and CCL3 (Figure S6C) or granulocytes attracting CXCL1, CXCL3, and CXCL5 (Figures S6C and S6D).

29Collectively, our data suggest that the transcriptome and secretome of activated DC3s, unlike the ones of activated monocytes, are consistent with a function in priming of naive T cells.

30To directly test the T cell priming capabilities of DC3s, we performed 5-day cocultures of flow cytometry-sorted, activated DC3s, cDC2s, and monocytes (see the cell sorting strategy in Figure S7A) with allogenic CD45RA+ naive T cells in the presence of a synthetic superantigen. We found that activated DC3s and cDC2s, unlike monocytes, triggered proliferative expansion and effector differentiation in CD4+ and CD8+ T cells probed by CD45RO acquisition (Figure S7B). The same results were also obtained using in-vitro-generated DC3s (Figure S7C). In contrast to monocytes, activated DC3s and cDC2s induced interferon γ (IFN-γ)- and TNF-α-producing CD4+ and CD8+ T cells but not IL- 17A (Figure 7A; Figure S7D). We found that DC3s had a specific ability to efficiently trigger CD103 expression in CD8+ T cells (Figure 7B), even without the presence of superantigen (Figure S7E). CD103 expression is a hallmark of tissue-resident memory T (TRM) cells because of its interaction with E-cadherin. Mechanistically, multiple factors, including transforming growth factor β (TGF-β), have been proposed to induce CD103 expression on T cells. Here we showed that anti-TGF-β neutralizing antibodies, but not others, blocked the expression of CD103 on CD8+ T cells (Figure 7C; Figure S7F).

31We next wondered whether CD103+ T cells aligned with bona fide TRM cells isolated from human tissue. To this end, we evaluated the gene expression profile of CD103− and CD103+ CD8+ T cells obtained after coculture of naive T cells with activated DC3s (Figure S7G). Using the GSEA methodology, we found that the signatures obtained for breast cancer or lung tissue CD103+ TRM cells were enriched in CD103+ T cells compared with CD103− T cells induced by DC3s (Figure 7D). DEG analysis revealed that human TRM cell markers such as NUSAP1, DUSP4, CXCR6, and FASLG (Figure 7E;;; ) were upregulated in CD103+ compared with CD103− CD8+ T cells. In addition, DC3-activated CD103+ CD8+ T cells expressed core components of the cytotoxic machinery (PRF1 and GZMA), as reported earlier for breast cancer-invading TRM cells (Figure 7E; ).

32CD8+CD103+ TRM cell infiltration has a protective prognosis value in breast cancer. To test the physiological relevance of DC3-dependent induction of CD8+CD103+ T cells, we analyzed CD103 and CD69 expression in CD3+CD8+ T cells from 18 samples of primary luminal breast cancer (Figure 7F). We found that CD8+CD103+CD69+ T cells were present across different cancer stages (Figure 7G). The frequency of CD8+CD103+CD69+ T cells was positively correlated with DC3 infiltration but not with other mononuclear phagocytes (Figure 7H). Despite the statistical significance of this correlation, it is important to underline that this was mainly driven by a subset of the samples. This suggests that further stratification of patients could help improve our understanding of the relationship between TRM cells and DC3s in breast cancer.

33Together, our results anchor DC3 function within the DC lineage; DC3s, just like cDC2s but unlike monocytes, are competent for priming and polarization of CD45RA+ naive T cells. In addition, we define induction of bona fide TRM cell-like CD103+CD8+ T cells as a specific privilege of DC3s but not cDC2s.

34Using scRNA-seq and high-dimensional flow cytometry, we provided evidence that DC3s represent a DC subset that can be separated and isolated from other DC subtypes. Regarding gene expression, our results are largely congruent with findings reported in previous studies. In addition, we provide a robust flow cytometry strategy to identify and purify DC3s by taking in account CD14+ monocytes and cDC2s. In agreement with previous studies, our results challenge the widely accepted notion that CD14 is a specific marker for monocytes. Indeed, we clearly demonstrate that a large fraction of DC3s expresses cell surface CD14. We identified CD88 (encoded by the C5AR1 gene) as a proper monocyte marker, enabling prospective purification of monocytes devoid of CD14+CD1c+ DC3s. We found that ex vivo GM-CSF cultures of pure CD88+CD14+ monocytes did not give rise to CD88-CD14+CD1c+ iDCs. Therefore, our findings provide an incentive to carefully revisit the prevailing notion that CD14+CD1c+ iDCs arise exclusively from monocytes in vivo.

35From the DC perspective, we refined the strategy to analyze the functional heterogeneity of CD1c+ DCs, including cDC2s and DC3s. Indeed, previous studies have reported heterogeneous expression of CD5 in CD1c+ DCs. However, our findings highlight that CD5−CD1c+ DCs contain CD163− cDC2s and CD163+ DC3s, which are transcriptionally distinct. Overall, our scRNA-seq and bulk RNA-seq analyses cluster together CD5+ and CD5− cDC2s. However, the developmental relationship between CD5− cDC2s, CD5+ cDC2s, and AS-DCs remains to be clarified.

36Because of CD1c expression and their relative similarity to cDC2s, DC3s have been embedded in the group of cDCs. Here we have provided substantial evidence arguing against this notion. Indeed, a conservative definition of cDCs includes (1) dependence on the FLT3L growth factor and (2) reliance on the CDP and pre-cDC developmental pathway. Here we have shown that DC3s do not meet any of these criteria. First, FLT3L alone was poorly active in stimulating the production of DC3s from CD34+ HSPCs in a controlled setting in vitro. Conversely, GM-CSF drives the commitment of DC3s under the same conditions. Moreover, we have shown that CDP did not give rise to DC3s although they are competent to generate cDC1s and cDC2s. Taking in account the heterogeneity of defined progenitor populations, we developed single-cell cultures enabling analysis of multiple lineage populations. Our single-cell cultures demonstrated that DC3s develop from CLEC12A+ DC3-restricted progenitors. In addition, the most frequent progenies differentiating along with DC3s from a bi-potent progenitor were cDCs and macrophages. This strongly suggest that DC3 progenitors diverge downstream of the MDP stage. Further studies will define more precisely the cell surface and molecular phenotype of the DC3-restricted progenitor. In support of a distinct regulation of cDC2s and DC3s, we have shown that cDC2 infiltration in breast cancer is correlated with cDC1s but not DC3s. Also, have shown recently that DC3s, but not cDC2s, expand in the blood of systemic lupus erythematosus patients. Further delineation of the inflammatory cues and transcription factors underpinning the development of the DC3 lineage is needed.

37Identification of DC3 as a cellular entity arising from a specific lineage brings forward the question of their specific immune function. We have shown that activated DC3s, just as cDC2s but unlike monocytes, secrete high amounts of T cell-polarizing cytokines (IL-12p70 and IL-23) and T cell-attracting chemokines (CXCL9, CXCL10, CXCL11, and CCL5). In addition, DC3s also secrete other cytokines (IL-10, IL-6, and TNF-α), some of which were poorly secreted by cDC2s (e.g., TNF-α). DC3s are polyvalent phagocytes with a cytokine pattern encompassing T cell and inflammation cues.

38A quintessential defining feature of DCs is their capacity to activate naive T cells. For instance, infiltration of CD1c+ DCs is associated with priming of T cell effectors when regulatory T (Treg) cell infiltration is low or Treg cell-mediated suppression is alleviated by checkpoint blockade. However, even when the heterogeneity of CD1c+ cells is appreciated, including recruitment of CD14+CD1c+ DCs in tumor-draining lymph nodes, little is known about the function of CD1c+ subtypes. Therefore, we asked whether the functions attributed to CD1c+ DCs would be carried out by cDC2s and/or DC3s. We found that DC3s, like cDC2s but unlike monocytes, primed and drove robust activation of naive T cells into IFN-γ- and TNF-α-secreting polyfunctional effectors.

39αE integrin (CD103) is a key marker of TRM cells through its interaction with E-cadherin, participating in retention of T cells at epithelial and mucosal sites. have identified a feature of total CD1c+ DCs in their ability to drive acquisition of CD103 in CD8+ T cells. Here we have shown that DC3s, but not cDC2s, induced expression of CD103 on CD8+ and CD4+ T cells. In addition, we have shown that CD103+ T cells primed by DC3s ex vivo align with bona fide TRM cells isolated from lung or breast cancer. This finding is consistent with the existence of early imprinting of the TRM cell program at the level of T cell priming by DC3s. This notion is supported by a recent in vivo study in a mouse model, evidencing early imprinting of the TRM cell program during T cell priming in the lymph nodes by αV integrin-expressing migratory DCs. Further supporting this view, we found that (1) DC3s upregulated CCR7 when activated by TLR agonists, potentially acting as migratory DCs, and (2) blocking studies indicated that TGF-β was required to prime CD103+ T cells. This role of priming in lymph nodes does not exclude that tumor-infiltrating DC3s might also provide TGF-β and other signals important for maintenance of TRM cells, as demonstrated in a mouse model. In support of this view, we found that infiltration of DC3s was selectively associated with the abundance of CD8+CD103+ T cells in luminal breast cancer primary tumors. Further studies are needed to delineate the full molecular mechanisms of CD103 induction in T cells by DC3s and their physiological in vivo relevance during immune responses. Indeed, CD103+CD8+ T cells are a protective biomarker in triple-negative breast cancer, and lung CD103+CD8+ T cells are a hallmark of protective immunity afforded by influenza vaccination. This underlines the potential of DC3s to regulate tissue immunity and defines them as targets for vaccines and immunotherapeutic interventions.

40We identify DC3s as CD88-CD1c+CD163+CD14+/-. Further studies will be needed to determine if this phenotype corresponds to all CD1c+CD14+ cells that had been reported in various inflammatory settings.

41We showed that DC3 differentiation is driven by GM-CSF in humanized mouse metastatic lung model. However, human circulating PBMCs were used in these experiments, preventing the assessment of GM-CSF impact on bone marrow DC3 progenitors in vivo. Hence, we cannot exclude a role of GM-CSF on DC3s survival instead of differentiation. In addition, we showed that GM-CSF drives DC3 differentiation from CD34+ umbilical cord HSPCs in vitro. This finding does not exclude that other growth factors control DC3 development in vivo.

42We aligned CD1c+CD14+ cells infiltrating breast tumor-draining lymph node with blood DC3s compared to cDC2s and monocytes. Due to limited availability of healthy human lymph nodes, we did not investigate if CD1c+CD14+ cells are infiltrating secondary lymphoid organs in homeostatic conditions. Blood DC3s upregulate CCR7 upon TLR stimulation suggesting they have a migratory potential. However, we did not investigate whether lymph nodes DC3s originate from non-lymphoid peripheral tissues via lymphatic vessels or directly from blood.

43A defining feature of cDCs lies in their ability to prime naïve T cells. We showed that DC3s can induce the proliferation of naïve T cell in a non-autologous priming context, i.e. the mixed leukocyte reaction. The capability of DC3s to uptake, process and present antigens by MHCI and MHCII molecules remains to be addressed.

44Our in vitro experiments showed that i) DC3s induce efficiently TRM differentiation from naïve CD8+T cells and, ii) DC3 infiltration correlates with TRM abundance in vivo. However, this does not provide a direct evidence that DC3s control TRM specification within tumor draining lymph nodes where they are likely to interact with naïve T cells. Indeed, we cannot rule out that, in vivo, DC3s act selectively in the tissue to maintain TRM populations primed in lymph nodes by a different DC subtype. Further studies will be needed to address how DC3s control CD8+ TRM populations in vivo.

45Additional information and request for resources and reagents should be directed to and will be made available by the Lead Contact, Pierre Guermonprez (pierre.guermonprez@kcl.ac.uk).

46The reagents generated in this study will be made available on request, but we may require a payment and/or a completed Materials Transfer Agreement if there is potential for commercial application.

47The scRNA-seq and the bulk RNA-seq datasets are deposited in the Genome Expression Omnibus under the SuperSeries accession numbers GSE151095.

48Human umbilical cord blood units were obtained from Anthony Nolan Cell Therapy Centre (ANCTC). Leukophoretic adult blood (buffy coats or leukocyte cones) were obtained from healthy volunteers through NHS.

49Tumor-invaded lymph nodes (tdLN) and primary tumors were collected from luminal breast cancer submitted to surgical resection at the Institut Curie Hospital (Paris, France), in accordance with institutional ethical guidelines. Patients’ clinical and pathologic characteristics are summarized in Table S2.

50NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ) mice were bred and maintained in specific pathogen-free animal facility in accordance with institutional KCL guidelines. All procedures involving animals were conducted according to requirements of UK Animals (Scientific Procedures) Act 1986.

51Peripheral blood mononuclear cells (PMBCs) were obtained by gradient centrifugation using Ficoll-Paque (GE Healthcare). Progenitor cells were enriched using CD34+ microbead isolation kit (Miltenyi).

52B16_CTRL, B16_huGM-CSF and B16_huFLT3L were generated by retroviral transduction of B16-F10 (C57BL/6 melanoma cell line) with an empty pMX-IRES-GFP vector or coding for human GM-CSF and human Flt3L respectively. Tumor cell lines were cultured in RPMI 1640 medium (Thermo Fisher) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher), penicillin-streptomycin (Thermo Fisher) and b-mercaptoethanol (Thermo Fisher) (complete RPMI) and maintained at 37°C and 5% CO2.

53Engineered B16-F10 cells were counted and resuspended in RPMI 1640 medium. NSG mice (8-12 weeks, males and females) were injected intravenously with 106 B16-F10 at day 0. 108 human blood PBMCs were injected intravenously at day 7 and 8. Mice were culled at day 9 and lungs were harvested.

54Mouse lungs were fixed with 1% PFA (Alfa Aesar) for 1hr at 4°C, washed and incubated in 34% sucrose solution (Sigma-Aldrich) overnight at 4°C. Lungs were embedded in Cryomatrix (Thermo Fischer) and frozen for cryostat sectioning (9 μm-thick). Sections were permeabilized using 0.5% saponin (Sigma-Aldrich), 2% BSA (Sigma-Aldrich), 1% FBS (Life Technologies) for 30 minutes at room temperature. Sections were labeled overnight at 4°C with mouse anti-human purified CD45 (HI30, Biolegend) followed by incubation for 1hr at room temperature with goat anti-mouse Cy3 (Jackson laboratory). All sections were labeled with Hoechst (Molecular Probes, Thermo Fisher) for nuclei staining 5 minutes at room temperature and mounted with Prolong diamond (Thermo scientific). Slides were imaged using a SP5 (Leica) and analyzed with Fiji software.

55Mouse lungs were harvested and transferred to 3ml digestion buffer (Hank’s Balanced Salt Solution (HBSS) with calcium and magnesium (Thermo Fisher) and with 0.1 mg/ml of Liberase TL (Roche) and 0.02μg/ml DNase I (Thermo Fisher). Lungs were dissociated using gentleMACS Octo Dissociator (Miltenyi) and incubated at 37°C for 45 minutes. The cell suspension was passed through a cell strainer (70μm, Corning) and red blood cells were lysed using ACK lysing buffer (Thermo Fischer). The absolute number of cells in the resulting cell suspension was assessed using AccuCheck Counting Beads (Thermo Fisher) on BD FACSCanto II (BD Biosciences).

56Mouse bone marrow-derived MS5 stromal cells engineered to express human membrane bound FLT3L alone (MS5_FL) or in combination with SCF and soluble CXCL12 (MS5_FS12) or human GM-CSF alone (MS5_GM) generated as previously described. Stomal cell lines were cultured in IMDM medium supplemented with 10% heat-inactivated FBS, penicillin/streptomycin, 50 μM β-mercaptoethanol (complete IMDM) and maintained at 37°C and 5% CO2.

57For in vitro co-culture experiments with stromal cells, MS5_FL, MS5_FS12 and MS5_GM feeders were seeded at 104 cells/well density in a 96-well plate (Thermo Fischer) and maintained overnight at 37°C and 5% CO2.

58Total enriched-CD34+ progenitors and flow cytrometry-sorted progenitor subsets (GMDP, CDP and cMoP) were plated on top of stromal cells, supplemented or not with 100ng/ml of recombinant human GM-CSF (Peprotech), for 14 days and 7 days, respectively. Blood phagocytes subsets were flow cytometry-sorted and plated on MS5_GM for 2 days.

59For bulk RNA sequencing analysis, total enriched-CD34+ progenitors were plated on top of MS5_FS12 feeders and with or without 100ng/ml of recombinant human GM-CSF (Peprotech). On day 5 and 10 of differentiation, half the medium was replaced with fresh complete IMDM or complete IMDM containing 100ng/ml of recombinant human GM-CSF. Cells were collected at day 15.

60All cells were collected with a solution of phosphate buffered saline (PBS) (GIBCO) 5mM EDTA (Thermo Fisher) at 4°C for 10min.

61PBMCs from healthy donors were stained and sorted as described previously. Blood cDC2 (105) and monocytes (3 × 105) were flow cytometry-sorted and injected subcutaneously along with B16_huGM (105) in 200 μL of ice-cold Matrigel® (BD Biosciences). Mice were sacrificed at day 4 by cervical dislocation and Matrigel® plugs were collected. Subcutaneous Matrigel® plugs were recovered, cut in pieces and incubated in HBSS (Life Technologies) 1% FBS, 0.37 U/ml Collagenase D (Roche), 10 μg/ml DNaseI (Roche) and 1 mg/ml Dispase (Sigma-Aldrich) for 30 minutes at 37°C. After digestion, plugs were smashed on a 70 μm strainer (Corning) and cells were collected and resuspended in flow cytometry buffer for flow cytometry analysis.

62Cells were stained in flow cytometry buffer (PBS 5mM EDTA 1% BSA) according to antibody panels (Key Resources Table) for 30 min. Dead cells were stained using DAPI or Live/Dead Blue staining (Thermo Fischer). Lineage (Lin) included CD3, CD19, CD20, NKp46, CD56, CD203c and CD66b, all conjugated with biotin. Multiparameter analysis was performed on LSRFortessa and Symphony (BD Biosciences) flow cytometers and analyzed using FlowJo software (Tree Star). The flow cytometry-sort was performed on BD FACS Aria II or BD FACS Aria Fusion at the Biomedical Research Centre (BRC) Flow Core Facility (Guy’s and St Thomas’ NHS Foundation Trust and King’s College London).

63Unsupervised clustering of flow cytometry data (Figure 1F) was performed using Self-Organizing Map clustering algorithm FlowSOM (Cytofkit R package). Data was then extracted as FCS files and further analyzed in FlowJo software.

64For bulk sequencing of in vitro differentiated subsets, up to 100 flow cytometry-sorted cells from three individual donors were collected directly in Lysis buffer (Takara Clontech, Cat# 635015) containing RNase inhibitors. RNaseq libraries were prepared on the contactless liquid handling system Labcyte Echo 525 (Labcyte Inc). In brief, ERCC was added to each sample and first strand full length cDNA was generated with a modified protocol of the SMARTseq v4 Ultra Low Input RNA Kit (Takara Clontech, Cat# 634891) using poly dT primers and a template switching oligo. Full length cDNA was amplified using SeqAmp DNA Polymerase (Takara Clontech, Cat# 638509). 12 ng of amplified cDNA from each sample was used to generate non-stranded RNA libraries using a modified protocol of the Ovation Ultralow System V2 1-96 kit (NuGEN, Cat# 0347-A01). In brief, amplified cDNA was fragmented through sonication on Covaris E220 (Covaris Inc), repaired and polished followed by ligation of indexed adapters. Adaptor ligated cDNA were pooled before final amplification to add flow cell primers. Libraries were sequenced on HiSeq2500 (Illumina Cambridge) for 100 cycles PE in Rapid mode.

65The raw sequencing data was initially processed using open source, web-based platform Galaxy (version 18.05.rc1) (https://usegalaxy.org). Reads were filtered for quality with more than 80% of the sequence having quality score > 33 using FastQC tool. Mapping against reference genome was performed with Hisat2 to the hg38 human genome. Adaptor sequences were detected automatically with TrimGalore!. Reads under 20bp were discarded. All processed sequencing files were imported in Partek® Genomics Suite software®, version 7.0©; 2017 (PGS), where they were processed further.

66mRNA was quantified using PGS built in RNA-seq workflow. Normalization method used was Reads Per Kilobase per Million mapped reads (RPKM) and mRNA was quantified against RefSeq Transcripts 2018-11-20 database. Hierarchal clustering on average expression within the group was performed on all identified protein coding genes (19791 genes). Based on the CD14-CD1c+CD206+/− two blue cluster similarity, they were both considered as cDC2-like cells and their datasets were merged for further analysis. The same was done for the two macrophage-like cells (gray and brown). Differentially expressed genes (Fold-Change ≤ −2 or ≥ 2 and p value < 0.05) were determined using one-way ANOVA in all pairwise comparisons with three donors grouped and visualized as Volcano plots in PGS. Individual samples were visualized via principal component analysis (PCA) using 500 most variable genes, which were determined based on median absolute deviation (MAD). The expression patterns of selected gene lists were displayed in the form of heatmaps, where rows and/or columns were ordered based on hierarchical clustering using Euclidean distance and average linkage in Morpheus (Broad Institute; https://software.broadinstitute.org/morpheus/). A few gene names are depicted next to the heatmap.

67PBMCs from three healthy donors or human CD45+ cells isolated from metastatic mouse lungs were isolated as previously described. Sorted cells from each donor were pooled together and CD1c+ and CD14+ cells were mixed at 80:20 ratio. 3 × 103 cells from the resulting cell suspension were partitioned into an emulsion of nanoliter-sized droplets using a 10X Genomics Chromium Single Cell Controller and RNA sequencing libraries were constructed using the Chromium Single Cell 3′ Library & Gel Bead Kit v2 (10X Genomics, Cat# PN-120237). Briefly, droplets containing individual cells, reverse transcription reagents and a gel bead were loaded with poly(dT) primers that include a 16 base cell barcode and a 10 base unique molecular index (UMI). Reverse transcription reactions were engaged to generate barcoded full-length cDNA followed by the disruption of emulsions using the recovery agent and cDNA clean up with DynaBeads MyOne Silane Beads (Thermo Fisher Scientific, Cat# 37002D). Bulk cDNA was amplified, and indexed sequencing libraries were constructed using the reagents from the Chromium Single Cell 3′ v2 Reagent Kit. Libraries were sequenced on NovaSeq 6000 Sequencing System (Illumina Cambridge).

68Cell Ranger (version 2.1.1) (from 10x genomics) was used to process Chromium single cell 3′ v2 RNA-seq output files. First, we generated fastq files for the Read1 for cell barcode and UMI and Read2 for transcript applying cellranger mkfastq (with default parameters). Second, we aligned the Read2 to the human reference genome GRCh38 using STAR (version 2.5.1) with cell ranger count (with default parameters) (https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/using/mkfastq). Further analysis was performed using Seurat package (version 2.3.4) in R (version 3.4.0). Before performed analysis, we applied the following filtering step: only genes expressed in 3 or more cells have been preserved and cells with less 200 unique genes and more than 4000 unique expressed genes were discarded (as they are potentially cells doublets). After filtering step, we used an expression matrix resulting in 14933 genes across 1622 cells (among 1625 cells) for the rest of the analysis. The matrix was normalized using genes expression values for each cell were divided by the total number of transcripts and multiplied by median of UMI counts. Then, these values were natural log-transformed before downstream analysis. For dimensionality reduction analysis, we first identified 3288 genes as highly variable genes across the single cells (cutoff value for dispersion = 0.5; cutoff value for average expression = 0). PCA performed using the variable genes as input and determined 10 PCs as significant PCs. These principal components were used as input for t-Distributed Stochastic Neighbor Embedding (tSNE). We used the shared nearest neighbor (SNN) modularity optimization-based clustering algorithm from the Seurat package (FindClusters function with default parameters) to identify the clusters of cells, following by Clustree analysis (clustree R package, version 0.2.2) by changing the resolution parameters from 0 to 2. Finally, we kept a resolution parameter at 0.8 and defined distinct 9 clusters. After controlling expression of some quality control genes, we excluded clusters E, F, G, H and I as contamination. Finally, we identified 4 relevant clusters. We identified cell specific marker by comparing cells in a specific cluster with cells in all other clusters using FindAllMarkers from Seurat package (MAST; logFC threshold = 0.5; only positive markers). Heatmap, feature plots and violin plots were performed using Seurat package.

69Single-cell RNA sequencing matrix for clusters A, B, C and D was created with gene signatures from. Signatures were defined as mean expression of discriminative markers for cDC2s and DC3s among lin-CD14- cells (cDC2 enriched and DC3 enriched, respectively) and of discriminative markers for CD14+ monocytes within lin-CD14+/− and/or CD16+/− monocytes (CD14+ mono enriched). The average expression of signature genes for each cell was calculated and plotted as a violin plot using R package ggplot2 (version 3.1.0).

70To statistically evaluate the enrichment of previously reported gene signatures (Gene Sets) in our dataset, we used pairwise comparisons using the gene set enrichment analysis (GSEA) method from the Massachussets Institute of Technology (https://www.broadinstitute.org/gsea). GSEA tests the relative position of a collection of genes (Gene Set) within an independent, ranked dataset (GeneList). Statistical analysis was performed by evaluation of nominal p value and false discovery rate (q value) based on 1,000 random permutations. Results were considered significant when the p value was below 0.05 and when the q value was below 0.25 (false discovery rate below 25%) accordingly to the recommendation from the software developers. For each pairwise comparison, the GSEA output can be represented as a bar code where each bar corresponds to the projection of one Probe Sets of the Gene Sets on the list of all the Probe Sets of the gene chips ranked from those having a high signal in one cell population (represented in red) to those having a high signal in the other cell population (represented in blue). Each bar code can be characterized by two parameters. The normalized enrichment score (NES) represents the number and differential expression intensity of the genes enriched in the corresponding cell subset. The NES is positive if the Gene Set is enriched in the first cell population and negative if it is enriched in the second cell population. The false discovery rate (FDR) statistical value (q) represents the likelihood that the enrichment of the Gene Set represents a false positive finding (for example if q = 0.25, 25% of the Gene Sets found enriched can be false positives). An absolute value of the NES below or around 1 means no enrichment as confirmed with associated q-values above 0.25.

71To simultaneously visualize pairwise comparisons of transcriptomes from cord blood-derived DC2s (light and dark blue) DC3s (orange) and macrophages (gray and brown), the BubbleMap module of BubbleGum was used. Results were considered significant when the p value was below 0.05 and the FDR (false discovery rate, q) value was below 0.25. The BubbleGum was performed using previously published gene signatures of pairwise comparison between DC2s and DC3s and DC3s and CD14+ monocytes.

72Tumor-draining lymph nodes (tdLN) and primary tumors were collected in CO2 independent medium (GIBCO; Cat# 18045-054) within few hours after the primary surgery. Tissue were cut into small fragments and submitted to enzymatic digestion using 0.1 mg/ml of Liberase TL (Roche) and 0.1 mg/ml of DNase (Roche) for 30 min. Cells were filtered on 40-μm cell strainer (BD), washed using CO2 independent medium (GIBCO; Cat# 18045-054) containing 0.4g/ml of human albumin and resuspended for cell counting.

73After tissue processing, cells obtained from tdLN were submitted to myeloid cells enrichment accordingly to prior flow cytometry-sorting. In brief, T and B lymphocytes, NK cells, erythrocytes and myelomonocytic cells were depleted using monoclonal antibodies against: CD3, CD19, CD56, CD235a and CD15, respectively. Subsequently, cell suspensions were stained for 30 min with antibody-conjugated as the following: HLA-DR, CD11c, CD14, CD1c, CD304, CD1a, CD206, CD141. Around 1,000 cells of each DC subset were sorted by flow cytometry using BD FACS ARIA II cell sorter, (purity > 98%). Cells were centrifuged and lysed with TCL buffer (QIAGEN) containing 1% of beta-mercaptoethanol before storage at −80°C. RNA were extracted and isolated using the Single Cell RNA purification kit (Norgen, Cat#51800) according to the manufacturer’s instructions and the RNA integrity number was evaluated with an Agilent RNA 6000 pico kit.

74PBMCs and in vitro generated cells were stained and sorted as described previously. In total 3 × 105 of blood cell subsets (cDC2, DC3 and Mono) or 105 in vitro generated cells (cDC2-, DC3 and Macro-like) were flow cytometry-sorted and cultured with TLR agonists cocktail containing 25 μg/ml Poly I:C (InvivoGen, Cat# 31852-29-6), 1 μg/ml R848 (InvivoGen, Cat# 144875-48-9) and 10ng/ml LPS (Sigma-Aldrich, Cat# L2630) for 16 h. Culturing supernatants were collected and stored at −20°C until the LEGENDplex™ assay execution day. LEGENDplex Human Macrophage/Microglia Panel (13-plex) with V-bottom Plate (Biolegend Cat# 740503) and Human Proinflammatory Chemokine Panel (13-plex) with V-bottom Plate (Biolegend Cat# 740003) was used according to manufacturer’s instructions. In short, samples and standards were thawed and plated with capture beads and incubated for 2 h. Plate was then washed, and Detection Antibodies were added. After 1 h incubation SA-PE was added and incubated for 30 min. Samples were acquired on BD FACSCanto II. Samples were analyzed using LEGENDplex™ Data Analysis Software.

75PBMCs and in vitro generated cells were stained and flow cytometry-sorted as described previously. 104 mononuclear phagocytes were cultured with TLR agonists cocktail containing 25 μg/ml PolyI:C, 1 μg/ml R848 and 10 ng/ml LPS for 16 h. T cells were isolated from fresh or frozen PBMCs using Naive Pan T Cell Isolation Kit (Miltenyi, Cat# 130-097-095) and 105 cells were plated on top of mononuclear phagocytes in presence of CytoStim (Miltenyi, Cat# 130-092-172), according to manufacturer’s instructions. For mixed leukocyte reaction (MLR) experiments, isolated Naive Pan T cells were labeled with Cell Tracer Violet (CTV) (Thermo Fischer, Cat# C34557) as per manufacturer’s instructions and cultured with flow cytometry-sorted DCs without CytoStim. At day 5, cells were collected and stained for extracellular and intracellular marker expression and analyzed using BD LSRFortessa. For intracellular staining, fixation and permeabilization were performed using BD Cytofix/Cytoperm solution (BD Biosciences, Cat# 554714) according to manufacturer’s instructions.

76Statistical analysis was performed using Prism 8.3 (GraphPad Software Inc., USA). When two experimental groups were compared, non-parametric Mann-Whitney test was used. When three or more groups were compared, statistically significant differences between means were determined using the one-way or two-way analysis of variance (ANOVA) method. A p value of less than 0.05 was considered as significant.