An Nfil3-Zeb2-Id2 pathway imposes Irf8 enhancer switching during cDC1 development.

Classical type 1 dendritic cells (cDC1s) are required for antiviral and antitumor immunity, which necessitates an understanding of their development. Development of the cDC1 progenitor requires an E-protein-dependent enhancer located 41 kilobases downstream of the transcription start site of the transcription factor Irf8 (+41-kb Irf8 enhancer), but its maturation instead requires the Batf3-dependent +32-kb Irf8 enhancer. To understand this switch, we performed single-cell RNA sequencing of the common dendritic cell progenitor (CDP) and identified a cluster of cells that expressed transcription factors that influence cDC1 development, such as Nfil3, Id2 and Zeb2. Genetic epistasis among these factors revealed that Nfil3 expression is required for the transition from Zeb2hi and Id2lo CDPs to Zeb2lo and Id2hi CDPs, which represent the earliest committed cDC1 progenitors. This genetic circuit blocks E-protein activity to exclude plasmacytoid dendritic cell potential and explains the switch in Irf8 enhancer usage during cDC1 development.

Introduction

Development of classical type 1 dendritic cells (cDC1s) has become a topic of interest because of the critical role this lineage plays in anti-tumor immunity and checkpoint blockade therapy[1]. DCs are an immune lineage encompassing classical DCs (cDCs) and plasmacytoid DCs (pDCs)[2,3]. cDCs comprise two branches, cDC1 and cDC2, that exert distinct functions in vivo and rely on different transcriptional programs[4]. pDCs and cDCs can both arise from the common DC progenitor (CDP)[5-7]. cDC progenitors (pre-cDCs) include clonogenic populations separately committed to cDC1 or cDC2 lineages[8,9]. Similar progenitors have been confirmed in human DC development[10-12]. However, the precise transcriptional programs underlying DC specification and commitment remain unclear.

The transcription factors Irf8 and Batf3 are required for cDC1 development[9,13,14], but cDC1 develop from CDP progenitors that express Irf8 independently of Batf3, yet later become dependent on Batf3 to maintain Irf8 expression. The basis for this switch from Batf3-independent to Batf3-dependent Irf8 expression is unclear. A clonogenic cDC1 progenitor, the pre-cDC1, develops normally in Batf3 bone marrow (BM) but fails to maintain Irf8 expression[9], causing it to divert into cells that are transcriptionally similar to cDC2 (XXXX). An enhancer located at +32 kb of the IRF8 transcription start site contained several AP1-IRF composite elements (AICEs) that bind IRF8 and BATF3 in cDC1s in vivo[9]. CRISPR-mediated deletion of the +32 kb Irf8 enhancer in mice (Irf8 +32) suggests that Batf3 supports Irf8 autoactivation using this enhancer (XXXX). Like Batf3 mice, Irf8 +32 mice lack mature cDC1 but maintain pre-cDC1 development in vivo. Development of this progenitor instead depends upon a +41 kb Irf8 enhancer, which binds E proteins and is active in mature pDCs and cDC1 progenitors, but not mature cDC1s. In vivo deletion of this enhancer eliminated Irf8 expression in pDCs and also completely eliminated development of the specified pre-cDC1. This enhancer activity requires E proteins to induce sufficient levels of IRF8 during specification of the pre-cDC1, but it is still unclear why mature cDC1s require BATF3 and the +32 kb Irf8 enhancer to maintain Irf8 expression.

Other transcription factors are known to influence cDC1 development, such as Nfil3, Id2, and Zeb2[15-19]. Nfil3, a basic leucine zipper (bZIP) transcriptional repressor[20], is expressed in cDC1s and is required for cDC1 development[15,21], but how it functions is unknown[4,15]. Id2 is a known inhibitor of E proteins, is expressed in both cDC1 and cDC2, and is required only for cDC1 development[16,17]. Id2 may exclude pDC fate by blocking activity of E proteins, particularly E2-2 (Tcf4), required for pDCs[22-24]. However, this model predicts that Id2 mice should lack both cDC1 and cDC2 lineages, since both lineages must exclude pDC fate. Finally, the transcriptional repressor Zeb2 is required for pDC development and suppresses cDC1 development, perhaps through inhibition of Id2 transcription[18,19]. How these factors precisely interact and at what stage they influence cDC1 specification is unknown.

Here, we used single-cell RNA-sequencing and genetic epistasis to determine the functional hierarchy of transcription factors involved in cDC1 specification. We organized a transcriptional circuit that explains the switch in Irf8 expression from being Batf3-independent to being Batf3-dependent. The CDP originates in a Zeb2hi and Id2lo state in which Irf8 expression is maintained by the +41 kb Irf8 enhancer. Single-cell RNA-sequencing identified a fraction of the CDP that exclusively possesses cDC1 fate potential. This fraction’s development arises when Nfil3 induces a transition into a Zeb2lo and Id2hi state. A circuit of mutual Zeb2-Id2 repression serves to stabilize states before and after this transition. Id2 expression in the specified pre-cDC1 inhibits E proteins, blocking activity of the +41 kb Irf8 enhancer, and thereby imposing a new requirement for Batf3 for maintaining Irf8 expression via the +32 kb Irf8 enhancer.

Results

The earliest committed cDC1 progenitor arises within the CDP

The CDP was originally defined as a Lin−CD117intCD135+CD115+ BM population and was observed to be, although not defined as, largely negative for MHC-II and CD11c expression[6]. Subsequently, pre-cDC1 and pre-cDC2 progenitors were identified to arise from the CDP but were not contained within the CDP[8,9]. Pre-cDC1s were defined as Lin−CD117intCD135+CD11c+MHC-IIlo-int and were largely CD115−. They can be defined using two methods, relying either on Zbtb46-GFP expression in Zbtb46 reporter mice, or on conventional surface markers (Fig. 1a)[9,25]. In each case, we noticed that approximately 10% of pre-cDC1s expressed CD115. The expression of CD115 in the pre-cDC1 suggested that cDC1 specification could occur at an earlier developmental stage in the CDP. In agreement, 5-10% of CDPs, defined on the strict exclusion of CD11c− and MHC-II− expressing cells, are Zbtb46-GFPpos (Fig. 1b). These Zbtb46-GFPpos CDPs had nearly exclusive cDC1 potential in vitro, comparable to pre-cDC1, and completely lacked pDC and cDC2 potential. This was in contrast to the Zbtb46-GFPneg CDPs, which produced cells from all three DC lineages (Fig. 1c, Supplementary Fig. 1a).

Fig. 1:

Zbtb46-GFP Expression in CDPs Identifies the Earliest Committed cDC1 Progenitor.

a, BM from Zbtb46 mice was analyzed by flow cytometry to identify pre-cDC1 as defined by Zbtb46-GFP or by CD24 expression. Lineage (Lin) included CD3, CD19, NK1.1, Ly-6G, TER-119, CD105, CD127 and Siglec-H. Numbers are the percent of cells in the indicated gates (representative of three independent experiments, n = 3 mice).b, BM from Zbtb46 mice was analyzed by flow cytometry to identify the percentage of Zbtb46-GFP expression within the CDP. Lineage was defined as in (a) (representative of three independent experiments, n = 3 mice).c, Zbtb46-GFPpos CDPs, Zbtb46-GFPneg CDPs, pre-cDC1 and pre-cDC2 were sort purified from Zbtb46 mice, cultured for 5 d in Flt3L, and analyzed by flow cytometry for development of pDCs and cDC1 (representative of three independent experiments, n = 4 for Zbtb46-GFPpos, Zbtb46-GFPneg CDPs, pre-cDC1 and n = 3 for pre-cDC2) d-e, Zbtb46-GFPpos CDPs, Zbtb46-GFPneg CDPs, pre-cDC1 and pre-cDC2 were purified as in (c) and analyzed using gene expression microarrays. Shown is expression of transcription factors with at least 4-fold differences between Zbtb46-GFPneg CDP and pre-cDC1s (d) or hierarchical clustering for genes with a least 8-fold differences between Zbtb46-GFPneg CDP and pre- cDC1s (e) (results averaged from biological triplicates for Zbtb46-GFPpos CDPs, Zbtb46-GFPneg CDPs, and pre-cDC1 or biological replicate for pre-cDC2).

The transcriptional profile of these Zbtb46-GFPpos CDPs suggests they represent an intermediate population between a non-specified CDP, the Zbtb46-GFPneg CDP, and the pre-cDC1 (Fig. 1d,e). For example, we considered genes whose expression changed more than 8-fold between the Zbtb46-GFPneg CDP and the pre-cDC1. For such genes, their expression in Zbtb46-GFPpos CDPs was consistently intermediate between their expression in Zbtb46-GFPneg CDPs and pre-cDC1s (Fig. 1d,e and Supplementary Tables 1,2). Id2 expression in Zbtb46-GFPneg CDPs was increased by 34-fold in pre-cDC1s, but only by 15-fold in Zbtb46-GFPpos CDPs. Likewise, Zeb2 expression in Zbtb46-GFPneg CDPs was reduced by 9-fold in pre-cDC1s, but only by 3.6-fold in Zbtb46-GFPpos CDPs. As expected, the Zbtb46-GFPpos CDPs were segregated away from the pre-cDC2 (Fig. 1e). Thus, these results indicate that Zbtb46-GFPpos CDPs are an earlier and distinct stage of cDC1 specification compared with the more abundant pre-cDC1 described previously.

Single-cell RNA-sequencing of the CDP identifies factors associated with cDC1 specification

The identification of Zbtb46-GFP expressing cells in the CDP that had nearly exclusive cDC1 potential suggested that the CDP might contain cells that have already specified to cDC1 fate. Single-cell RNA-sequencing (scRNA-seq) was performed on 9,554 CDPs defined as Lin−CD127−CD117intCD115+CD135+MHC-II−CD11c− (Fig. 2a) on the 10X Genomics platform to assay for unrecognized heterogeneity within this population. Uniform Manifold Approximation and Projection (UMAP) analysis[26-28] identified 8 closely connected clusters (Fig. 2b,c). Although we were able to identify genes that were specifically enriched in certain clusters, others such as Klf4 and Ly6d were not specifically enriched in one cluster (Supplementary Fig. 1b). However, scRNA-seq was able to identify a cluster that was enriched in Zbtb46 expression, corroborating our data above with the Zbtb46-GFP reporter mice. Zbtb46 was expressed in cluster 3, which also showed restricted expression of Id2 and Batf3, but excluded expression of Tcf4 (E2-2) and Zeb2 (Fig. 2d,e). Cluster 3 also showed reduced Csf1r expression (Fig. 2d), consistent with lower CD115 expression in pre-cDC1 and incongruent with the higher CD115 expression in the bulk CDP (Fig. 1a). As expected, Flt3 and Irf8 were uniformly and highly expressed (Fig. 2d, e). Cluster 7, the only other Tcf4 negative cluster, likely contained macrophage or neutrophil contamination as this cluster expressed Ccl6 and did not contain many cells (Fig. 2c,d). Other factors impacting DC development such as Bcl11a, Spi1, Klf4, and Notch2[29,30,31,32] were not differentially expressed across the CDP, perhaps suggesting that specification of cDC2s and pDCs occurs after the CDP (Fig. 2d, Supplementary Fig. 1b). In addition, the CDP appeared homogenous with respect to markers of proliferation (Fig. 2f). Thus, scRNA-seq identifies a cluster of cells within the CDP that coordinately induces Nfil3, Id2, Batf3 and Zbtb46, and reduces Tcf4 and Zeb2, suggesting these genes may regulate cDC1 specification at an earlier stage than previously recognized.

Fig. 2:

Single-cell RNA Transcriptome Analysis of CDPs.

a, CDPs gated as Live,[CD105, CD3, CD19, Ly6G, Ter119]−CD127−CD117intCD115+CD135+MHC-II−CD11c− cells were purified by sorting from C57BL/6J mice. Shown are pre-sort (top) and post-sort (bottom) for cells collected for single-cell RNA-sequencing. b, UMAP clustering of CDPs from Seurat analysis (data represents combined analysis of two independent sequencing runs)c, Heatmap of 9,954 cells for the top ten genes of each cluster from Seurat analysis. Shown are names of representative genes within each cluster.d, Violin plots depicting cluster identity and expression level for the indicated genes expressed in each cluster as described in (b).e, UMAP plots for the indicated genes as described in (b).f, Joy plots depicting expression level and cell cycle stage for genes involved in the cell cycle.

cDC1 specification is functionally characterized by low Zeb2 and high Id2 expression

To test the functional importance of these genes for cDC1 specification, we first analyzed two reporter mouse lines expressing a ZEB2-EGFP fusion protein (Zeb2)[33] or an Id2-IRES-GFP cassette (Id2)[34]. Both reporters exhibit a GFP expression pattern consistent with the level of Zeb2 and Id2 gene expression across many immune lineages (Supplementary Fig. 2a,b). In Zeb2 mice, 90% of CDPs expressed high levels of ZEB2-EGFP, but 10% expressed low levels of ZEB2-EGFP, similar to low levels of ZEB2-EGFP expressed by pre-cDC1s (Fig. 3a). In Id2 mice, 94% of CDPs expressed low Id2-GFP, but 6% expressed high levels of Id2-GFP similar to the high levels of Id2-GFP expressed by pre-cDC1s (Fig. 3b). Thus, both Zeb2 and Id2 reporter lines confirm the existence of ZEB2-EGFPlo and Id2-GFPhi cells within the CDP as predicted by scRNA-seq.

Fig. 3:

Zeb2 and Id2 Heterogeneity Identifies cDC1 Specification in CDPs.

BM from Zeb2 (a) and Id2 (b) mice were analyzed by flow cytometry to identify GFP expression in CDPs and pre-cDC1s. WT mice (Zeb2 and Id2) are shown as gray histograms. Numbers indicate the percentage of cells in the indicated gates. (representative of three independent experiments, n = 3 mice).c-d, ZEB2-EGFPlo and ZEB2-EGFPhi CDPs (c), and Id2-GFPhi and Id2-GFPlo CDPs (d) were purified by sorting, cultured for 5 d in Flt3L, and analyzed by flow cytometry for development of cDC1 (red) and cDC2 (blue) (representative of three independent experiments, n = 5 for ZEB2-EGFPlo and ZEB2-EGFPhi CDPs and n = 4 for Id2-GFPhi and Id2-GFPlo CDPs). e, The indicated cells purified as described in (c) and (d) or in Fig. 1c were cultured as in (c) and analyzed by flow cytometry for cDC1 development shown as a percentage of total cDCs (CD45R−CD317−MHC-II+CD11c+) (pooled from three independent experiments, n = 5 for ZEB2-EGFPlo and ZEB2-EGFPhi CDPs, n = 4 for Id2-GFPhi or Id2-GFPlo CDPs and Zbtb46-GFPpos or Zbtb46-GFPneg CDPs). Small horizontal lines indicate the mean. f, Hierarchical clustering of genes expressed at least 5-fold differently between pre-cDC1 and ZEB2-EGFPhi CDPs (results averaged from three independent experiments). g, Expression of the indicated genes described in (f). h, Hierarchical clustering of genes expressed at least 5-fold differently between pre-cDC1 and Id2-GFPlo CDPs (results averaged from two independent experiments). i, Expression of the indicated genes described in (h). Data are presented as mean and two-tailed unpaired Student’s t test was used to compare groups. *p < 0.05, **p < 0.01, ****p < 0.0001.

We next analyzed the developmental potential of CDPs expressing high or low levels of ZEB2-EGFP, Id2-GFP, and Zbtb46-GFP in an in vitro Flt3L culture system. CDPs expressing low levels of ZEB2-EGFP showed significantly increased cDC1 potential (66%) compared with CDPs expressing high levels of ZEB2-EGFP (26%) (Fig. 3c,e). Likewise, CDPs expressing high levels of Id2-GFP showed significantly increased cDC1 potential (77%) compared with CDPs expressing low levels of Id2-GFP (30%) at both days 5 and 7 of in vitro Flt3L culture (Fig. 3d,e, Supplementary Fig. 2c,d). Finally, CDPs expressing Zbtb46-GFP developed nearly exclusively into cDC1 (96%), while CDPs lacking Zbtb46-GFP developed into both cDC1 (30%) and cDC2 (70%) (Fig. 1c, 3e). In all three cases, pDCs developed exclusively from CDPs that were either Zbtb46-GFPneg, ZEB2-EGFPhi, or Id2-GFPlo (Supplementary Fig. 2e–j). These results suggest that CDPs expressing low levels of Zeb2-EGFP or high levels of Id2-GFP are biased toward cDC1 development, but not as completely as CDPs expressing Zbtb46-GFP.

The transcriptional profile of CDPs expressing low levels of ZEB2-EGFP or high levels of Id2-GFP suggests that these cells are an intermediate population between non-specified CDPs and the pre-cDC1 (Fig. 3f–i). We considered genes whose expression differed more than 5-fold between the pre-cDC1 and either ZEB2-EGFPhi CDPs (Fig. 3f,g) or Id2-GFPlo CDPs (Fig. 3h,i). The expression of such genes in ZEB2-EGFPlo CDPs was consistently intermediate between the expression in ZEB2-EGFPhi CDPs and pre-cDC1s (Fig. 3f,g and Supplementary Table 3). Likewise, the expression of such genes in Id2-GFPhi CDPs was consistently intermediate between the expression in Id2-GFPlo CDPs and pre-cDC1s (Fig. 3h,i and Supplementary Table 4). Additionally, the cells that are ZEB2-EGFPlo within the CDP have induced Id2, and cells that are Id2-GFPhi within the CDP have downregulated Zeb2 (Fig. 3f–i). Both of these populations also show increasing Zbtb46 expression compared to the non-specified CDPs. Although these three cDC1-specificed CDP populations differ in cDC1 potential, their transcriptional profiles suggest that they are highly overlapping. In summary, CDPs that express low ZEB2-EGFP or high Id2-GFP represent an earlier stage of cDC1 specification compared to the previously identified pre-cDC1.

Nfil3 is required for cDC1 specification within the CDP

Nfil3 is required for cDC1 development[15], but its mechanism and timing of action remain obscure. To determine the stage where Nfil3 acts in cDC1 development, we crossed Nfil3 mice with ZEB2-EGFP, Id2-GFP and Zbtb46-GFP reporter mice, and assayed whether cDC1-specified progenitors developed in BM. In Nfil3+/+Zbtb46 reporter mice, cDC1-specified cells can be identified as CD117intZbtb46-GFPpos cells that include pre-cDC1s and Zbtb46-GFPpos CDPs and comprise approximately 5% of Lin−CD135+ BM (Fig. 4a,b). However, these cells are absent in Nfil3Zbtb46 mice, but do develop normally in Batf3Zbtb46 mice as previously described (Supplementary Fig. 3a)[9]. Within the CDP, cDC1-specified cells can be identified as Zbtb46-GFPpos cells that comprise 5% of the CDP (Fig. 4a,b). However, these cells are also absent in Nfil3Zbtb46 mice.

Fig. 4:

Nfil3 is Required for cDC1 Specification.

a, BM from Nfil3Zbtb46 and Nfil3Zbtb46 mice was analyzed by flow cytometry for Lin−CD135+CD117int Zbtb46-GFPpos cells (left) or Zbtb46-GFPpos CDPs (right). Numbers indicate the percent of cells in the indicated gates (representative of five independent experiments, n = 5 mice). b, Cells from (a) are shown as a percentage of Lin−CD135+ (left) or CDPs (right). Small horizontal lines indicate the mean.c, BM from Nfil3Zeb2 and Nfil3Zeb2 mice was analyzed for Lin−CD135+CD117int ZEB2-EGFPlo cells (left) or ZEB2-EGFPlo CDPs (right) (representative of three independent experiments, n = 6 mice). d, Cells from (c) are shown as a percentage of Lin− CD135+ (left) or CDPs (right). Small horizontal lines indicate the mean.e, BM from Nfil3Id2 and Nfil3Id2 mice was analyzed for Lin−CD135+CD117intId2-GFPhi cells (left) or Id2-GFPhi CDPs (right) (representative of three independent experiments, n = 3 for Nfil3Id2 mice and n = 4 for Nfil3Id2 mice). f, Cells from (e) are shown as a percentage of Lin− CD135+ (left) or CDPs (right). Small horizontal lines indicate the mean .Data in b, d, and f are presented as mean and two-tailed unpaired Student’s t test was used to compare groups. ***p < 0.001; ****p < 0.0001.

In Nfil3+/+Zeb2 reporter mice, cDC1-specified cells are identified as CD117int ZEB2-EGFP lo cells that includes pre-cDC1s and ZEB2-EGFPlo CDPs and comprise approximately 6% of Lin−CD135+ BM (Fig. 4c,d). However, these cells are absent in Nfil3Zeb2 mice. In Nfil3+/+Zeb2 reporter mice, cDC1-specified CDPs can be identified as ZEB2-EGFPlo cells that comprise 7% of CDPs (Fig. 4c,d), which again are absent in Nfil3Zeb2 mice. Finally, in Nfil3+/+Id2 reporter mice, cDC1-specified cells can be identified as CD117int Id2-GFPhi cells that include pre-cDC1s and Id2-GFPhi CDPs and comprises approximately 2% of Lin−CD135+ BM (Fig. 4e,f). However, these cells are absent in Nfil3Id2 mice. Further, cDC1-specified CDPs can be identified as Id2-GFPhi cells that comprise 7% of the CDP (Fig. 4e,f), but which are absent in Nfil3Id2 mice. In summary, Nfil3 is required for the appearance of all cDC1-specified progenitors identified by Zbtb46-GFP, ZEB2-EGFP, or Id2-GFP.

Zeb2 functions downstream of Nfil3 in cDC1 specification

We next evaluated the interactions between Nfil3 and other factors using genetic mutants rather than GFP reporters. We first examined interactions between Nfil3 and Zeb2. We crossed Nfil3 mice to Zeb2Mx1-Cre mice in which ZEB2 can be inactivated by poly(I:C) treatment (Zeb2). We compared cDC1 development and the presence of cDC1-specified progenitors in Nfil3+/+Zeb2Mx1-cre− (wildtype), Nfil3, Zeb2, mice as well as Nfil3Zeb2 mice (Fig. 5). First, Zeb2 mice have more than a 2-fold increase in splenic cDC1s compared with wildtype mice (Fig. 5a,b), consistent with our previous study[19]. Further, Nfil3 mice lacked cDC1s in spleen, as previously reported[15]. However, Nfil3 Zeb2 DKO mice had a splenic cDC1 population that, like Zeb2 mice, is about 2-fold greater than WT mice. Similarly, in vitro cDC1 development was increased in Zeb2 BM and reduced in Nfil3 BM (Fig. 5c,d). However, in vitro cDC1 development from Nfil3 Zeb2 DKO BM was increased compared to Nfil3 BM. Finally, we directly examined pre-cDC1 development in these mice. Zeb2 mice have increased numbers of pre-cDC1 compared to wildtype mice, while Nfil3 mice have greatly reduced numbers of pre-cDC1 (Fig. 5e,f). However, Nfil3Zeb2 DKO mice have markedly restored pre-cDC1 development compared to Nfil3 mice. In summary, for both in vivo and in vitro cDC1 development and for in vivo cDC1 specification, the phenotype of Zeb2 deficiency dominates over that of Nfil3 deficiency, suggesting that Zeb2 genetically functions downstream of Nfil3. The repression of Zeb2 by Nfil3 is required in the early stages of cDC1 specification.

Fig. 5:

Zeb2 is Downstream of Nfil3 in cDC1 Development.

a, Splenic cDCs from Nfil3Zeb2 Mx1-cre (WT), Zeb2 Mx1-cre (Zeb2), Nfil3 (Nfil3), and Nfil3 Zeb2 Mx1-cre (Nfil3 Zeb2) mice, gated as in Fig. 3e, were analyzed for cDC1 (red) and cDC2 (blue) frequency. Numbers are the percent of cells in the indicated gates (data representative of three independent experiments, n = 7 for WT and Zeb2 mice, n = 8 for Nfil3 mice and n = 9 for Nfil3 Zeb2 mice).b, Analysis from (a) are presented as individual mice. Small horizontal lines indicate the mean. c, cDCs derived in vitro from Flt3L-treated BM cultures from mice in (a) were analyzed for cDC1 (red) and cDC2 (blue) frequency as in (a) (data representative of three independent experiments, n = 7 for WT and Zeb2 mice, n = 8 for Nfil3 mice, and n = 9 for Nfil3 Zeb2 mice). d, Analysis from (c) are presented for individual mice. Small horizontal lines indicate the mean. e, BM from mice in (a) was analyzed for the frequency of pre-cDC1 (red). BM cells are pre-gated as Lin SiglecHCD135+ (data representative of three independent experiments, n = 7 for WT and Zeb2 mice, n = 8 for Nfil3mice, and n = 9 for Nfil3 Zeb2 mice). f, Analysis from (e) are presented for individual mice. Small horizontal lines indicate the mean. Mean and two-tailed unpaired Student’s t test was used to compare groups. *p < 0.05; ****p < 0.0001; ns, not significant.

Zeb2 functions downstream of Id2 with respect to cDC1 specification

Some evidence suggests that Zeb2 may function genetically upstream of Id2 in cDC1 development[18,19], but no mechanism has been established. To evaluate the genetic interaction between Zeb2 and Id2, we crossed the Rosa26 strain with Zeb2, Id2, and Zeb2 Id2 mice to produce mice in which tamoxifen administration can conditionally inactivate ZEB2 (Zeb2), ID2 (Id2), or both (Zeb2Id2) , respectively. We first evaluated pre-cDC1 specification and cDC1 development in these mice (Fig. 6a–d). Zeb2 mice show a 2-fold increase in cDC1 and pre-cDC1 compared with wildtype mice, similar to mice with ZEB2 deficiency generated using poly(I:C) and Mx1-Cre (Fig. 5). Id2 mice lack splenic cDC1, as expected[17], and also lack pre-cDC1 in BM. However, Zeb2Id2 mice showed a restored development of splenic cDC1 and BM pre-cDC1 (Fig. 6a–d). Moreover, similar results were obtained from in vitro Flt3L cultures of BM cells from these mice (Supplementary Fig. 4a,b). In summary, for cDC1 development, Zeb2 deficiency dominates over Id2 deficiency in Zeb2Id2 DKO mice, suggesting that with respect to cDC1 specification, Zeb2 genetically functions downstream of Id2.

Fig. 6:

Expression of Id2 and Zeb2 is Mutually Repressive in the CDP.

a, Splenic cDCs harvested from WT, Zeb2 Rosa26 (Zeb2), Id2 Rosa26 (Id2), and Id2 Zeb2 Rosa26 (Zeb2 Id2 ) were analyzed for cDC1 (red) and cDC2 (blue) frequency, gated as in Fig. 3e. Numbers are the percent of cells in the indicated gates (data representative of two independent experiments, n = 2 for Id2 mice, n = 3 for Zeb2 Id2 mice, n = 4 for Zeb2 mice, and n = 5 for WT mice). b, Data from (a) are presented for individual mice. Small horizontal lines indicate the mean.c, BM from mice in (a) was analyzed for the frequency of pre-cDC1 (red). BM cells are pre-gated as Lin SiglecHCD135+ (data representative of two independent experiments, n = 2 for Id2 mice, n = 3 for Zeb2 Id2 mice, n = 4 for Zeb2 mice, and n = 5 for WT mice).d, Data from (c) are presented for individual mice. Small horizontal lines indicate the mean.e, Shown is the expression of Irf8, Nfil3, and Id2 in splenic cDC1 sorted from WT, Zeb2, Zeb2 Id2 , and Zeb2 Nfil3 mice (n = 3 for WT and Zeb2 mice, n = 2 for Zeb2 Id2 and Zeb2 Nfil3 mice). Small horizontal lines indicate the mean.f, BM from Zbtb46(WT), Id2Zbtb46 (Id2), and Batf3Zbtb46 (Batf3) mice was gated as Lin− cells, and the CD117intZbtb46-GFP−- (red) or CD117intZbtb46-GFP+ (blue) cells were separately analyzed for CD115 and CD135 expression (data representative of five independent experiments, n = 5 mice) g, CDPs and Zbtb46-GFPpos cells in (f) were sort purified and analyzed by gene expression microarray. Shown are gene expression levels for Zeb2, Nfil3, and Batf3 (data representative of three independent experiments, n = 2 for CDPs and n = 3 for Zbtb46-GFPpos cells). Small horizontal lines indicate the mean. Data are shown as mean and two-tailed unpaired Student’s t test was used to compare groups. *p < 0.05, ***p < 0.001, ns, not significant.

Zeb2 functions upstream of Id2 with respect to Id2 expression

We next compared the transcriptional profiles of splenic cDC1 in wildtype Zeb2, Zeb2 Id2, and Nfil3 Zeb2 mice using gene expression microarrays (Fig. 6e, Supplementary Fig. 4c, Supplementary Table 5). cDC1 from all genotypes expressed high Irf8 and Batf3, and low Irf4 and Tcf4, levels, as expected. Nfil3 was highly expressed in cDC1 isolated from wildtype, Zeb2, and Zeb2Id2 mice and was absent in cDC1 isolated from Nfil3 Zeb2 mice, consistent with Nfil3 genetically functioning upstream of both Zeb2 and Id2. Further, Id2 was expressed at the expected high levels in cDC1 from wildtype and Zeb2 mice, and absent in cDC1 from Zeb2Id2mice, in agreement with Id2 genetically functioning upstream of Zeb2. Unexpectedly, Id2 gene expression remained high in cDC1 from Nfil3 Zeb2 mice, despite the absence of Nfil3 normally required for cDC1 specification. These results indicate that, in the absence of Nfil3, loss of Zeb2 is sufficient for Id2 induction, suggesting Zeb2 acts upstream of Id2 with respect to Id2 expression.

Id2 and Zeb2 expression are mutually repressive

The above results indicate that Zeb2 functions downstream of Id2 with respect to cDC1 specification, as ZEB2 deficiency can restore cDC1 in Id2 mice, but acts upstream of Id2 with respect to Id2 gene expression. Thus, Id2 appears to repress Zeb2 expression, and Zeb2 appears to repress Id2 expression, to create a circuit of mutual repression in which Nfil3 seems to initiate cDC1 specification by repressing Zeb2.

This model predicts that cDC1 specification in the CDP could occur in the absence of Id2, and that Id2 pre-cDC1 would maintain Zeb2 expression, unlike Id2 pre-cDC1. To test this, we used chimeric mice reconstituted with Id2Zbtb46 BM (Id2Zbtb46). We first confirmed that splenic pDCs and cDC2s develop normally in Id2Zbtb46 chimeras (Supplementary Fig. 5a). We also showed that Id2Zbtb46 cDC2s are transcriptionally essentially identical to Id2Zbtb46cDC2s (Supplementary Fig. 5b and Supplementary Table 6). Further, unspecified CDPs, defined as Lin−CD117intZbtb46-GFP− CDPs, in Id2 mice are similar to Id2and Batf3 CDPs, both in frequency, expression of CD115 and CD135 (Fig. 6f), and transcriptional profile (Fig. 6g, Supplementary Fig. 5c, and Supplementary Table 7). However, in Id2Zbtb46 chimeras, cDC1-specified cells (Lin−CD117intZbtb46-GFPpos) were present but were reduced in frequency by 3-fold. The cDC1-specified cells in Id2Zbtb46 chimeras maintained CD135 expression but had higher expression of CD115 compared to Id2Zbtb46, implying a partial block in development of specified cDC1s. In addition, these cells failed to induce Batf3 but maintained expression of Zeb2 compared to Id2 Zbtb46-GFPpos cells (Fig. 6g and Supplementary Fig. 5c). These results confirm a role for Id2 in inducing Batf3 and repressing Zeb2 expression during cDC1 specification. Since Id2 inhibits E protein transcription factors, Id2 might indirectly repress Zeb2 if E proteins supported Zeb2 expression. In agreement, E2A is expressed in CDPs and binds to E-box motifs in the Zeb2 locus based on ChIP-seq analysis (Supplementary Fig. 5d,e,f)[35].

Id2 induction imposes a switch in Irf8 enhancer usage during cDC1 development

Data has revealed that E proteins may be necessary for the sufficient induction of Irf8 in the CDP by activating the +41 kb Irf8 enhancer (XXXX). This enhancer is transiently active during cDC1 progenitor development, but is required for the development of both pre-cDC1 and cDC1 in vivo[9,36](XXXX). This 454 bp region contains six E-box motifs that are conserved between human and murine Irf8 loci (Fig. 7a) and is known to bind E2-2 in human pDCs (Supplementary Fig. 6)[37]. Using the 454 bp region in a retroviral reporter system[9], we found robust activity that was specific for pDCs, but not cDC1s or cDC2 (Fig. 7b,c). We also examined the activity of three individual enhancer segments each containing 2 E-box motifs. Segments A and C showed reduced overall activity compared with the 454 bp enhancer, but retained pDC specificity, while the middle segment B retained overall activity, but reduced pDC specificity (Fig. 7b,c). Mutation of both E-boxes 1 and 2 in the 454 bp enhancer significantly reduced enhancer activity in pDCs (Supplementary Fig. 7a,b). Within segment A, mutation of either E-box alone reduced overall activity, while mutation of both E-boxes together completely extinguished activity (Fig. 7d, Supplementary Fig. 7c). The most active segment B was also E-box dependent, showing reduced overall activity upon mutation of E-boxes 3 and 4 (Fig. 7e, Supplementary Fig. 7d). These results indicate that the +41 kb Irf8 enhancer activity relies on the redundant activity of the six E-box motifs contained within this 454 bp region. In agreement with the role of Id2 in repressing E-box motifs, overexpression of retroviral ID2 diminished +41 kb Irf8 enhancer activity (Fig. 7f).

Fig. 7:

Id2 imposes a switch from the +41 kb Irf8 enhancer to the +32 kb Irf8 enhancer by Reducing E protein activity,

a, Conservation of E-box motifs between human (red) and mouse (blue) loci within the +41 kb Irf8 enhancer.b, GFP expression from RV reporters with (IRF8 +41) or without (empty) the 454 bp +41 kb enhancer, or with intact segment A (A), intact segment B (B), intact segment C (C), or intact segments A and B (A+B), or intact segments B and C (B+C), in pDCs, cDC1s, and cDC2s, shown as histograms (data pooled from >5 independent experiments, n > 5).c, Data shown in (b) shown as integrated MFI (data pooled from >5 independent experiments, n > 5). Small horizontal lines indicate the mean.d, GFP expression in pDCs of RV reporters without (empty) or with the 454 bp +41 kb enhancer (IRF8 +41), or with intact segment A (A), or with mutations in E-box 1 (A-m1), E-box 2 (A-m2) or both (A-m1/m2), shown as integrated MFI (data pooled from >5 independent experiments, n > 5). Small horizontal lines indicate the mean.e, GFP expression in pDCs of RV reporters without (empty) or with the 454 bp +41 kb enhancer (IRF8 +41), or with intact segment B (B), or with mutations in E-box 3 (B-m3), E-box 4 (B-m4) or both (B-m3/m4), shown as integrated MFI (data pooled from >5 independent experiments , n > 5). Small horizontal lines indicate the mean.f, GFP expression in WEHI-231 cells of RV reporters with (IRF8 +41) or without (empty) the 454 bp +41 kb enhancer, or with intact segment A (A), intact segment B (B), intact segment C (C) and co-transduced with either empty RV (gray) or ID2 RV (purple), shown as integrated MFI (data pooled from three independent experiments, n = 3). Small horizontal lines indicate the mean.g, ATAC-Seq was performed on the indicated progenitor or DC populations. Shown is the Irf8 locus, with the Irf8 +41 kb enhancer region (black box) and the +32 kb enhancer region (dotted box). (representative of three independent experiments and the Immunological Genome Project Open Chromatin Regions, n= 1 biological replicate per population). Data are presented as mean and one-way or two-way ANOVA was used to compare groups. *p < 0.05, **p < 0.01, ****p < 0.0001.

This suggests that Id2 induction in the CDP can extinguish E protein activity at the +41 kb Irf8 enhancer, thereby imposing a requirement for a new enhancer in the pre-cDC1 to maintain Irf8 expression necessary for cDC1 development. To identify a potential enhancer, we performed ATAC-seq on MDP, CDP, and pre-cDC1 progenitors and found a peak that indicated accessibility within the Irf8 region only in the pre-cDC1 and in mature cDC1, but not in the earlier MDP or CDP or mature cDC2 (Fig. 7g, red dashed line). This peak was located at +32 kb of the Irf8 TSS and was shown to be BATF3-dependent[9] (XXXX). The induction of Id2, and the subsequent repression of Zeb2, thus forces a new requirement for Batf3 in maintaining Irf8 expression during cDC1 development.

Discussion

This study resolves several long-standing puzzles regarding cDC1 development. First, Id2 was proposed to be required for cDC development by excluding pDC fate potential[22,23], but Id2 mice lacked only cDC1, and did not show the expected loss of all cDCs[16]. Second, cDC1 develop from CDP progenitors that express Irf8 independently of Batf3, yet later become dependent on Batf3 to maintain Irf8 expression. The basis for this switch from Batf3-independent to Batf3-dependent Irf8 expression was unclear. Third, mature cDC1 do not express E proteins or show +41 kb Irf8 enhancer activity, yet their development requires both. These apparent inconsistencies all result from a cryptic stage in cDC1 development in which Irf8 expression relies on the E protein-dependent +41 kb Irf8 enhancer. In this study, we examined this cryptic stage of development to reveal the hierarchy of transcription factors governing cDC1 specification.

Our results define a genetic hierarchy that unifies the actions of the known transcription factors required for cDC1 development. cDC1s were known to require Irf8, Batf3, Id2, and Nfil3, but how these factors interacted was unknown. We used Zbtb46-GFP to identify an earlier stage of cDC1 specification than previously described that occurs within the CDP itself[9]. Single-cell RNA-sequencing of the CDP identified a cluster of cells defined by the expression pattern of Nfil3, Id2, and Zeb2. Epistatic analysis revealed a genetic hierarchy in which Nfil3 induces a transition from CDPs that express high levels of Zeb2 and low levels of Id2, to CDPs that express high levels of Id2 and low levels of Zeb2. A circuit of mutual repression between Zeb2 and Id2 stabilizes these distinct states, such that repression of Zeb2 by Nfil3 is required to induce this transition. In Zeb2hi and Id2lo CDPs, Irf8 expression is maintained by the +41 kb Irf8 enhancer, which is dependent on E proteins for activity. Upon Id2 induction, E protein activity is lost and Irf8 expression becomes dependent on Batf3 acting at the +32 kb Irf8 enhancer. It is currently unclear whether Nfil3 directly represses Zeb2 and whether Zeb2 directly represses Id2, as there may be other factors in this proposed genetic circuit. Nfil3 acts largely as a repressor [20,38],but may activate transcription in contexts[39]. Likewise, Zeb2 has been suggested to directly repress Id2 expression[18,19], although this has not been rigorously tested. Nfil3, Zeb2, and Id2 have also been shown to regulate ILC development[40], but the mechanisms by which these transcription factors act in these cells has not been studied.

Although our study seems to clarify several outstanding questions in cDC1 development, it may raise the possible necessity of a revised DC development scheme. We identified a cDC1-specified stage that occurs before the development of the pre-cDC1. The cells in this stage express a high level of Irf8, consistent with the high level of Irf8 in the CDP. Early expression of Irf8 seems to correlate with commitment to the cDC1 lineage, as shown recently in a report in which IRF8 expression in human hematopoietic stem cells specifies to the DC1 lineage[41]. cDC1 specification may occur even earlier than our report suggests, but may rely on a minimum threshold of Irf8 expression, and not simply early expression in the BM. The requirement of the +41 kb Irf8 enhancer during the transition from the MDP to the CDP for subsequent cDC1 specification is consistent with this idea of a minimum threshold for Irf8 expression. A revised DC development model may require a deeper understanding of the relationship between IRF8 expression level and activity.

Our results also suggest that cDC1 development may be more closely related to pDC development than previously appreciated. The actions of the proposed genetic circuit on the +41 kb Irf8 enhancer suggest that Id2 extinguishes E protein activity at the +41 kb Irf8 enhancer and imposes a requirement for Batf3 at the +32 kb Irf8 enhancer. It is possible that pDCs and cDC1s share a common progenitor. The emergence of pDCs from myeloid or lymphoid BM progenitors is debated, as early studies suggested that pDCs can arise from both lymphoid and myeloid BM progenitors[42]. However, two recent studies indicated that late pDC progenitors emerge from the common lymphoid progenitor and a “pre-pDC” was described[43,44]. Since these studies did not perform lineage tracing for prior expression of myeloid markers, such as CD115-Cre, pDCs progenitors conceivably could emerge in a series of stages that include both myeloid and lymphoid features, as recently suggested[45]. Resolving whether pDC and cDC1 share a common progenitor that has segregated from the cDC2 lineage, or simply share molecular transcriptional requirements will require additional studies.

Methods

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Kenneth Murphy (kmurphy@wustl.edu)

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

WT C57BL6/J mice were obtained from The Jackson laboratory. Zbtb46 mice were described[25]. Nfil3−/− mice were from A. Look and Tak Mak[46]. Mx1-Cre [B6.Cg-Tg(Mx1-cre)1Cgn/J] mice (stock no. 003556), and Rosa26 [B6.129-Gt(ROSA)26Sor] mice (stock no. 008463) were obtained from The Jackson Laboratory. B6.SJL (B6.SJL-Ptprc Pepc /BoyJ) mice (strain code 564), were obtained from Charles River. ZEB2-EGFP fusion protein reporter (STOCK Zfhxlb) mice[33] were derived from biological material provided by the RIKEN BioResource Center through the National BioResource Project of the Ministry of Education, Culture, Sports, Science and Technology, Japan. SIP1 (Zeb2) were from Y. Higashi[47]. For experiments shown in Fig. 6f,g, Id2-CreER mice (JAX stock #016222)[48] were bred to Zbtb46 mice to generate Id2Zbtb46 mice. These mice were crossed to generate Id2 Zbtb46 mice. Livers from day 1 old Id2 pups were dispersed and cells injected into 4-6 week old lethally irradiated SJL WT mice (Charles Rivers) and chimeras used eight weeks after reconstitution. Id2-flox and Id2-IRES-GFP mice[34] were generously donated by G. Belz. Tcf3 were generated by crossing the Tcfe2afl allele (B6.129-Tcf3tm1Mbu/J JAX stock #028184) with Vav-iCre mice (JAX stock #008610).

All mice were generated, bred, and maintained on the C57BL/6 background in the Washington University in St. Louis School of Medicine specific pathogen-free animal facility. Animals were housed in individually ventilated cages covered with autoclaved bedding and provided with nesting material for environmental enrichment. Up to five mice were housed per cage. Cages were changed once a week, and irradiated food and water in autoclaved bottles were provided ad libitum. Animal manipulation was performed using standard protective procedures, including filtered air exchange systems, chlorine-based disinfection, and personnel protective equipment including gloves, gowns, shoe covers, face masks, and head caps. All animal studies followed institutional guidelines with protocols approved by the Animal Studies Committee at Washington University in St. Louis.

Unless otherwise specified, experiments were performed with mice between 6 and 10 weeks of age. No differences were observed between male and female mice in any assays performed and so mice of both genders were used interchangeably throughout this study. Within individual experiments, mice used were age- and sex-matched littermates whenever possible.

Antibodies and flow cytometry.

Cells were kept at 4ºC while being stained in PBS supplemented with 0.5% BSA and 2mM EDTA in the presence of antibody blocking CD16/32 (clone 2.4G2; BD 553142). All antibodies were used at a 1:200 dilution vol/vol (v/v), unless otherwise indicated.

The following antibodies were from BD: Brilliant Ultraviolet 395–anti-CD117 (clone 2B8, catalog number 564011, 1:100 v/v), PE-CF594–anti-CD135 (clone A2F10.1, catalog number 562537, 1:100 v/v), V500–anti-MHC-II (clone M5/114.15.2, catalog number 742893), Brilliant Violet 421–anti-CCR9 (clone CW-1.2, catalog number 565412, 1:100 v/v), Alexa Fluor 700–anti-Ly6C (clone AL-21, catalog number 561237), Brilliant Violet 421–anti-CD127 (clone SB/199, catalog number 562959, 1:100 v/v), biotin–anti-CD19 (clone 1D3, catalog number 553784), BV510–anti-CD45R (clone RA3–6B2, catalog number 563103), PE-anti-CD90.1 (clone OX-7, catalog number 554898). The following antibodies were from eBioscience: allophycocyanin–anti-CD317 (clone eBio927, catalog number 17–3172-82, 1:100 v/v), PE-Cy7–anti-CD24 (clone M1/69, catalog number 25–0242-82), peridinin chlorophyll protein (PerCP)–eFluor 710–anti-CD172a (clone P84, catalog number 46–1721-82), PerCP-Cy5.5–anti-SiglecH (clone eBio-440c, catalog number 46–0333-82), PE–anti-CD11c (clone N418, catalog number 12–0114-82).

The following antibodies were from BioLegend: Brilliant Violet 711–anti-CD115 (clone AFS98, catalog number 135515, 1:100 v/v), PE or Brilliant Violet 421–anti-XCR1 (clone ZET, catalog number 148204 or 148216), Alexa Flour 700 or APC/Cy7-–anti-F4/80 (clone BM8, catalog number 123130 or 123118, 1:100 v/v), PE–anti-CD45.2 (clone 104, catalog number 109808), biotin or PE/Dazzle 594–anti-CD45R (clone RA3–6B2, catalog number 103203 or 103258), biotin–anti-Ly6G (clone 1A8, catalog number 127603), biotin–anti-Ter119 (clone TER-119, catalog number 116204), biotin–anti-CD105 (clone MJ/718, catalog number 120404), biotin–anti-NK1.1 (clone PK136, catalog number 108704), biotin–anti-CD127 (clone A7R34, catalog number 135006, 1:100 v/v), biotin–anti-Ly-6A/E (clone D7, catalog number 108104), PE-anti-human-CD4 (clone RPA-T4, catalog number 300550, 1:50 v/v). The following antibodies were from Tonbo Bioscience: FITC–anti-CD45.1 (clone A20, catalog number 35–0453-U500), biotin or APC–anti-CD3e (clone 145–2c11, catalog number 30–0031-U500 or 20–0032-U100), violetFluor 450–anti-MHC Class II (I-A/I-E) (clone M5/114.15.2, catalog number 75–5321-U100). The following antibodies were from Invitrogen: allophycocyanin–eFluor 780–anti-CD11c (clone N418, catalog number 47–0114-82). Cells were analyzed on a FACSCanto II or FACSAria Fusion flow cytometer (BD), and data were analyzed with FlowJo v10 software (TreeStar).

Induced Gene Deletion.

Conditional gene deletion in Nfil3Zeb2Mx1-cre (Nfil3 Zeb2), Zeb2 Mx1-cre− Nfil3 (WT), Zeb2 Mx1-cre− Nfil3 (Nfil3) and Zeb2 Mx1-cre Nfil3 (Zeb2) mice was induced by i.p. injection of 150 μg poly(I:C) (SigmaAldrich; 1.0 mg/mL stock solution dissolved in saline) twice within 36–72 h. Gene deletion in WT, Zeb2 Rosa26 (Zeb2), Id2 Rosa26 (Id2) and Zeb2 Id2 Rosa26 (Zeb2 Id2) mice was induced by administration of tamoxifen citrate chow (Envigo) for 4–5 weeks. Mice were given up to 2 d of regular chow per week if significant weight loss was observed. After treatment, mice were rested on regular chow for one week before analysis.

Isolation and culture of BM progenitor cells and splenic DCs.

Bone marrow progenitors and DCs were isolated as described[9]. For BM sorting experiments, BM was isolated and depleted of CD3-, CD19-, CD105-, Ter119-, and in some instances Ly6G- and CD45R-expressing cells by staining with the corresponding biotinylated antibodies followed by depletion with MagniSort Streptavidin Negative Selection Beads (Thermo Fisher). All remaining BM cells were then stained with fluorescent antibodies prior to sorting. MDPs were identified as Lin−CD117hiCD135+CD115+ BM cells; CDPs were Lin−CD117intCD135+CD115+MHC-II−CD11c+; pre-cDC1s are Lin−CD117intCD135+CD115−MHC-IIlo-intCD11c+CD24+Siglec-H− or as Lin−CD117intCD135+MHC-IIlo-intCD11c+Siglec-H−Zbtb46-GFP+, and pre-cDC2s as Lin−CD117loCD135+CD115+MHC-II−CD11c+. For splenic sorting experiments, spleen was isolated and depleted of Ly6G-, B220-, and CD3-expressing cells. cDC2 were identified as Lin−CD45R−CD317−MHC-II+CD11c+CD172a+ cells. Cells were purified on a FACSAria Fusion into IMDM plus 10% FBS with 5% Flt3L conditioned media. Sort purity of >95% was confirmed by post-sort analysis before cells were used for further experiments. For experiments that included Flt3L cultures, sorted cells (1×103 to 10×103 cells per 200 µl complete IMDM) were cultured for 5 or 7 d at 37 °C with 5% Flt3L conditioned media.

Expression microarray analysis.

RNA was extracted with a RNAqueous-Micro Kit (Ambion) or a NucleoSpin RNA XS Kit (Machery-Nagel), then was amplified with Ovation Pico WTA System (NuGEN) or WT Pico System (Affymetrix) and hybridized to GeneChip Mouse Gene 1.0 ST microarrays (Affymetrix) for 18 h at 45 °C in a GeneChip Hybridization Oven 640. The data was analyzed with the Affymetrix GeneChip Command Console. Microarray expression data was processed using Command Console (Affymetrix, Inc) and the raw (.CEL) files generated were analyzed using Expression Console software with Affymetrix default RMA Gene analysis settings (Affymetrix, Inc). Probe summarization (Robust Multichip Analysis, RMA), quality control analysis, and probe annotation were performed according to recommended guidelines (Expression Console Software, Affymetrix, Inc.). Data were normalized by robust multiarray average summarization and underwent quartile normalization with ArrayStar software (DNASTAR). Unsupervised hierarchical clustering of differentially expressed genes was computed with ArrayStar (DNASTAR) with the Euclidean distance metric and centroid linkage method.

Single-cell RNA-sequencing.

100,000 CDPs were sort purified as Live,[CD105, CD3, CD19, Ly6G, Ter119]−CD127−CD117intCD115+CD135+MHC-II−CD11c− cells and single-cell gene measured with the Chromium system using Chromium Single Cell 3’ Library and Gel Bead Kit v2 (10X Genomics). Cell density and viability of sorted cells were determined by Vi-CELL XR cell counter (Beckman Coulter), and all processed samples had cell viability at >90%. The cell density was used to impute volume of single cell suspension needed in the reverse transcription (RT) master mix, to achieve ~6,000 cells per sample. After Gel Bead-in-Emulsion reverse transcription (GEM-RT) reaction and clean-up, a total of 12 cycles of PCR amplification was performed to obtain cDNAs. Libraries for RNA-seq were prepared following the manufacturer’s user guide (10x Genomics), profiled using Bioanalyzer High Sensitivity DNA kit (Agilent Technologies) and quantified with Kapa Library Quantification Kit (Kapa Biosystems). Each single-cell RNA-seq library was sequenced in one lane of HiSeq4000 (Illumina). Sequencing data were pooled from two runs of 4,796 and 4,758 individual cells. Run 1 had 2,354 median genes and 85,247 means reads per cell. Run 2 had 2,247 median genes and 85,265 mean reads per cell. Sequencing was filtered and processed using the Seurat R toolkit[49].

ATAC-Seq.

ATAC-Seq of DC progenitors was performed using the Omni-ATAC protocol as previously described with minor modifications[36]. 10,000 MDPs, CDPs, and pre-cDC1s were sorted from bone marrow as described above and lysed in ice-cold ATAC-RSB buffer containing 0.1% NP40, 0.1% Tween-20, and 0.01% digitonin. Cells were incubated at 4° C for 3 min, then washed with ATAC-RSB buffer containing only 0.1% Tween-20. Nuclei were spun down by centrifugation and then incubated in 50 µL of transposition buffer (25 µL 2X TD buffer, 22.5 µL dH2O, 2.5 µL Tn5 transposase (Nextera DNA Library Prep Kit, Illumina)) and incubated at 37° C for 30 min. If 10,000 cells could not be obtained for a certain population then the quantity of Tn5 transposase was titrated down proportionately to the number of cells obtained but cells were still incubated in 50 µL total. Transposed DNA was purified with a DNA Clean & Concentrator kit (Zymo Research), eluted in 21 µL of elution buffer, and stored at −20° C until amplification. Three biological replicates for each cell population were obtained and sequenced. ATAC-Seq libraries were prepared as previously described, barcoded and sequenced on an Illumina Nextseq.

Retroviral analysis of murine +41 kb Irf8 enhancer.

The 454 bp region of the +41 kb Irf8 enhancer was cloned into hCD4 pA GFP-RV[9]. Each E-box motif (CANNTG) in the enhancer was mutated to a binding site-free DNA sequence (AACTAC) determined by SiteOut[50].

The primer sequences for the entire enhancer and the associated mutations are as follows: for +41 kb Irf8 enhancer: aaaagatctGATCTGGGGTATGTGGGAAC and GAAAGAAGATCTGGGGTATGT; for segment A: aaaagatctGATCTGGGGTATGTGGGAAC and aaaaaagcttTGTGCTAATTAAAGCCAAGAGG; for segment B: aaaaggatccCTGTACCCCAGATCCCATC and aaaaaagcttGAGGAACCACCACTCAAGG; for segment C: aaaaggatccTCAGGTTTGGGGAAGAAG and aatcttttattttatcgatagcaagCTTGACACTCTGGGAATAG; for segment A+B: GCGACGGTCGCGCGAGCtagaaaagatctGATCTGGGGTATGTGGG and aatcttttattttatcgataaaaaaagcttGAGGAACCACCACT; for segment B+C: aaaaggatccCTGTACCCCAGATCCCATC and aatcttttattttatcgatagcaagCTTGACACTCTGGGAATAG; for mE1: GTGTCTCTCACaactacGGATCCCATATAAGGTTTATTTTTAC and CCTTATATGGGATCCgtagttGTGAGAGACACAAAGGGTTC; for mE2: GCCCAGGCCCaactacTTCCCCCCTGTACCCCAG and GTACAGGGGGGAAgtagttGGGCCTGGGCGATGTTCTG; for mE3: TCCTCCTCTGGTAGAGAAGAAGCTGCGGGCTGGGaactacCCGCACCCTCCCC and GGGGAGGGTGCGGgtagttCCCAGCCCGCAGCTTCTTCTCTACCAGAGGAGG; for mE4: GCACCCTCCCCGGaactacTCTTCACCGTGCGGTCAGG and CGCACGGTGAAGAgtagttCCGGGGAGGGTGCGGg; for mE5: GGCTGGAAGCCTTGAGTGGTGGTTCCTCaactacTCTTTGGGCACCTG and CAGGTGCCCAAAGAgtagttGAGGAACCACCACTCAAGGCTTCCAGCC; for mE6: ctacTCTTTGGGaactacGGATGCGTCCTGTTAGGACC and CCTAACAGGACGCATCCgtagttCCCAAAGAgtagttGAGG; and for mE¾: AGCTGCGGGCTGGGaactacCCGCACCCTCCCCGGaactacTCTTCACCGT and ACGGTGAAGAgtagttCCGGGGAGGGTGCGGgtagttCCCAGCCCGCAGCT.

Retroviral vectors were transfected into Plat-E cells with TransIT-LTI (Mirus Bio) and viral supernatants were collected two days later. For retroviral analysis in Flt3L cultures, Lin−CD117high BM cells were infected on day 1 after plating with the supernatants of transfected packaging cells and concentrated by centrifugation with 2 ug/ml polybrene by ‘spin infection’ at 2,250 r.p.m. for 60 min. Viral supernatant was replaced by complete IMDM + 5% Flt3L one day after transduction and the culture was read out on day 8. For analysis, the enhancer activity was quantitated using integrated MFI[51,52].

For retroviral analysis in WEHI-231 cultures, WEHI-231 cells were infected on day 1 after plating with supernatants of transfected packaging cells with the reporter constructs and either empty or ID2 retrovirus and concentrated by centrifugation with 2 ug/ml polybrene by ‘spin infection’ at 2,250 r.p.m. for 60 min. Viral supernatant was replaced by complete IMDM one day after transduction and the culture was read out on day 3.For analysis, the enhancer activity was quantitated using integrated MFI in cells that were co-infected with either empty or ID2 retrovirus[51,52].

Analysis of E-box motifs in human +58 kb IRF8 enhancer.

The occurrence of E-box motifs in the element +41 kb relative to the Irf8 TSS was found with FIMO[53] motif-identification program at a P-value threshold of 1 × 10−3 with the E-box position weight matrix obtained for the E2-2 peaks of human pDCs[37].

Human and mouse elements were aligned via Clustal Omega W.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis for single cell RNA-sequencing data is described above. Horizontal lines in figures indicate the mean. Results from independent experiments were pooled as indicated in figure legends. Data were analyzed using Prism (GraphPad), using unpaired two-tailed Student’s t tests when comparing two groups or ordinary one-way or two-way ANOVA when comparing multiple groups. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

LIFE SCIENCES REPORTING SUMMARY

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

DATA AND SOFTWARE AVAILABILITY

The data that support the findings of this study are available from the corresponding author upon request. Microarrays are available on the GEO database with the SuperSeries accession number GSE123800. Data from Fig. 1 is available with accession number GSE123747; from Fig. 3 are GSE123794 and GSE123796; from Fig. 6 are GSE123797 and GSE123799. Data from Supplementary Fig. 4 is GSE123797 and from Supplementary Fig. 5 are GSE123798 and GSE123799. The single-cell RNA-sequencing data is available on the GEO database with the following accession number: and is utilized in Fig. 2 and Supplementary Fig. 1b. The ATAC-seq data of DC progenitors is available on the GEO database with the following accession number: GSE132240 and is utilized in Fig. 7.

Supplementary Table 1: Expression of transcription factor-encoding genes with a difference in expression of at least four fold in the pre-cDC1 relative to their expression in Zbtb46-GFPneg CDP.

Supplementary Table 2: Expression of genes with a difference in expression of at least eight fold in the pre-cDC1 relative to their expression in Zbtb46-GFPneg CDP.

Supplementary Table 3: Expression of genes with a difference in expression of at least five fold in the pre-cDC1s relative to ZEB2-EGFPhi CDPs.

Supplementary Table 4: Expression of genes with a difference in expression of at least five fold in the pre-cDC1 relative to Id2-GFP+ CDPs.

Supplementary Table 5: Expression of lineage-defining transcription factors in WT, Zeb2 and Zeb2cDC1s.

Supplementary Table 6: Expression of genes with a difference in expression of at least two fold in Id2 cDCs compared to Id2 cDC2.

Supplementary Table 7: Expression of genes with a difference in expression of at least three fold in Id2CD117intZbtb46-GFPpos compared to Id2CD117intZbtb46-GFPpos.