Cohesin-mediated chromatin organization controls the differentiation and function of dendritic cells

Abstract

The cohesin complex extrudes chromatin loops, stopping at sites bound by CCCTC-binding factor (CTCF) and organizing chromosomes into topologically associated domains, yet biological implications of this process remain obscure. We show that cohesin controls in vivo differentiation and function of murine antigen-presenting dendritic cells (DCs), particularly antigen cross-presentation and IL-12 secretion by type 1 conventional DCs (cDC1s). The chromatin organization of DCs was shaped by cohesin and the transcription factor IRF8, which facilitated chromatin looping and chromosome compartmentalization, respectively. Optimal expression of IRF8 itself required CTCF/cohesin-binding sites demarcating the Irf8 gene. During DC activation, cohesin enabled the induction of a subset of genes that were preferentially located in Polycomb-repressed regions and enriched in more distal enhancers. Accordingly, deletion of CTCF sites flanking the Il12b gene in mice reduced IL-12 production by cDC1s. Our data reveal an essential role of cohesin-mediated chromatin folding in cell differentiation and function in vivo, and its bi-directional crosstalk with lineage-specifying transcription factors.

One-sentence summary:

Chromatin-organizing cohesin complex cooperates with transcription factor IRF8 to enable the functionality of dendritic cells in vivo.

Introduction

A fundamental level of epigenetic regulation in eukaryotic cells is genome organization, the hierarchical folding of chromatin in 3D space that is critical for genome stability and function. High-resolution genome-wide chromosome conformation capture (Hi-C)-based technologies have revealed sub-megabase topologically associating domains (TADs) as a hallmark of chromatin organization (1, 2). TADs are formed by the cohesin complex, which extrudes chromatin loops until halted by the protein CCCTC-binding factor (CTCF) bound in the proper orientation (3, 4). Cohesin-mediated TAD formation is thought to facilitate enhancer-promoter interactions, particularly those occurring over long distances and inside TADs, while insulating cross-TAD interactions (5, 6). It may also optimize temporal control of gene expression during development (7), in part by counteracting Polycomb-mediated repression (8). Independent of loop extrusion, the genome is spatially compartmentalized based on the transcriptional and histone modification status of chromatin, with transcriptionally active chromatin located centrally (A compartments) and inactive chromatin located peripherally (B compartments) within the nucleus (9).

Acute depletion of cohesin in post-mitotic cells in vivo demonstrated its central role in chromosome organization during interphase (10). The role of cohesin/CTCF-mediated chromatin organization has been demonstrated in special cases such as antigen receptor gene rearrangement in lymphocytes (11, 12), stochastic protocadherin expression (13) or organization of multilobular nuclei (14). Conversely, the effects of cohesin on the terminal differentiation and function of various cell types in vivo have not been established. In addition to cohesin, various lineage-specifying transcription factors (TFs) have been implicated in the control of 3D genome organization (15–18). However, the respective roles of and potential crosstalk between TFs and cohesin in shaping chromatin architecture remain to be explored.

Dendritic cells (DCs) bridge innate and adaptive immunity, linking the recognition of pathogens to the priming of antigen-specific T cells (19). Pathogen recognition by DCs leads to DC activation, production of immunostimulatory cytokines and antigen presentation to T cells. DC progenitors in the bone marrow (BM) undergo Flt3 ligand (Flt3L)-dependent differentiation into antigen-presenting conventional DCs (cDCs) and type I interferon (IFN-I)-producing plasmacytoid DCs (pDCs) (20, 21). BM-derived cDC progenitors exit into the periphery and undergo terminal differentiation driven by local signals such as Notch2 (22, 23). Differentiated DCs are largely non-proliferative and short-lived, with a lifespan of several days. cDCs themselves are composed of two developmentally and functionally distinct subsets, cDC1s and cDC2s (24). cDC1s efficiently cross-present exogenous antigens on MHC class I to CD8⁺ T cells (25) and secrete interleukin-12 (IL-12) during infection with intracellular pathogens, qualities that also underlie their key role in anti-tumor responses (26). Among the TFs that control DC differentiation, interferon regulatory factor 8 (IRF8) is induced early in DC specification and controls the development of committed DC progenitors, pDCs and cDC1s (27–30). Here, we explored the role of cohesin and its interactions with key DC-specifying TFs such as IRF8 in DC differentiation and function in vivo.

Results

Cohesin is required for the differentiation of conventional dendritic cells

To test the role of the cohesin complex in the DC lineage, we targeted its essential subunit Smc3, which is required for cohesin complex binding to DNA (31). We crossed mice with a LoxP-flanked (“floxed”) allele of Smc3 (Smc3^fl, (32)) to the CD11c-Cre deleter strain (33) to generate mice with DC-specific deletion of Smc3 (Smc3^ΔDC). Cre recombination in the CD11c-Cre strain occurs after the commitment of progenitors to the DC lineage, is highly efficient in mature cDCs and long-lived, tissue-resident CD11c⁺ cells (e.g. Langerhans cells and alveolar macrophages), but is less efficient in pDCs (33). Indeed, cDCs isolated from spleens of Smc3^ΔDC mice harbored excised Smc3 alleles (Fig. S1A), confirming the robust DC-specific deletion of Smc3. Splenic DCs from Smc3^ΔDC mice showed a ~50% reduction in absolute number (Fig. 1A) and frequency (Fig. S1B) of cDC1 and of the Notch2-dependent Esam⁺ subset of cDC2 (22). In contrast, Esam⁻ cDC2 and pDCs were unaffected in Smc3^ΔDC mice (Fig. 1A, S1B,C). We also observed a near-complete absence of lung cDC1s (Fig. 1B), and a reduction of skin-derived migratory cDCs in the skin-draining lymph nodes (sdLN) of Smc3^ΔDC mice (Fig. S1D). Finally, epidermal Langerhans cells and alveolar macrophages were nearly absent from the skin and lungs of Smc3^ΔDC mice, respectively (Fig. S1E,F). Because these CD11c⁺ populations are established before birth and maintained in tissues by local self-renewal (34), their depletion was likely due to the essential role of cohesin in cell division (35). To exclude a similar role in the observed DC phenotype, we measured DC turnover by administering the nucleoside analog EdU for 3 days. In this time frame, proliferating DC progenitors are expected to incorporate EdU and give rise to mature DCs, as evidenced by the detection of EdU in splenic cDCs (Fig. 1C). In the spleens of Smc3^ΔDC mice, EdU incorporation was marginally reduced in Esam⁺ cDC2s and normal in cDC1s, suggesting a normal cell turnover (Fig. 1C). Thus, the post-commitment loss of Smc3 impaired cDC1 homeostasis independently of proliferation.

Figure 1. Cohesin is required for cDC differentiation independently of proliferation.

(A) Representative flow plots (left) and total cDC subset numbers (right) of splenic cDC populations. cDC1 = Lin⁻ (TCR-β⁻ CD3⁻ CD19⁻ NKp46⁻ NK1.1⁻) CD11c⁺ MHC-II⁺ XCR1⁺ SIRPα⁻, cDC2 = Lin⁻ CD11c⁺ MHC-II⁺ XCR1⁻ SIRPα⁺. Pooled from two independent experiments.

(B) Representative flow plots (top) and frequencies (bottom) of lung cDC populations. Pooled from two independent experiments.

(C) Mice were injected on three consecutive days with EdU. Representative histograms (left) and frequencies (right) of EdU⁺ splenic cDCs one day after the last injection.

(D) Representative histograms (left) and frequencies (right) of CD8α expression on splenic cDC1s. Pooled from three independent experiments.

(E-F) RNA-seq was performed on purified splenic cDCs from CD11c-Cre Smc3^fl/fl (n=3), CD11c-Cre Smc3^fl/- (n=2) and control CD11c-Cre (n=3) mice. Shown are PCA (E) and GSEA (F) of Mat^ON (36) and Notch (23) signatures among differentially expressed genes (DEGs) between control and Smc3^Δ cDC1s. ES = enrichment score, NES = normalized enrichment score.

(G-H) Bone marrow cells from R26^CreER/+ (Control) or R26^CreER/+ Smc3^fl/fl (Smc3^Δ) mice were differentiated into DCs in vitro in the presence of Flt3L and OP9-DL1 stromal cells for 8 days. 4-hydroxytamoxifen (4-OHT) was added to cultures at day 3 to induce recombination of the floxed Smc3 allele. Shown are representative flow plots (left) and numbers (right) of indicated cDC subsets (G), and representative histograms of CD8α on cDC1s (H).

Symbols represent individual mice, and bars represent mean. For (A-D), Control = CD11c-Cre (grey circles) and Smc3^fl/fl (grey triangles), Smc3^ΔDC = CD11c-Cre Smc3^fl/fl (blue circles) and CD11c-Cre Smc3^fl/- (blue triangles).

Statistical differences were evaluated using Mann-Whitney test, except for one-way ANOVA followed by Tukey’s test (D). *p < 0.05, **p < 0.01, ****p < 0.0001; n.s. not significant.

Analysis of the remaining XCR1⁺ Smc3^ΔDC cDC1s revealed the loss of CD8α, which was more profound in mice with one germline null allele of Smc3 (CD11c-Cre Smc3^fl/-) (Fig. 1D). Furthermore, Smc3^ΔDC Esam⁺ cDC2s exhibited aberrant upregulation of the integrin CD11b (Fig. S1G). To further characterize DC differentiation, we performed bulk RNA sequencing of sorted splenic cDC1s and cDC2s from CD11c-Cre Smc3^fl/fl, CD11c-Cre Smc3^fl/- and control CD11c-Cre animals (Data File S1). By principal component analysis (PCA), PC1 resolved cDC subsets, whereas PC2 resolved a cohesin-dependent signature that scaled with Smc3 gene dosage (Fig. 1E). The differences between Smc3^ΔDC and control cDCs along PC2 were greater for cDC1s than cDC2s (Fig. 1E). Gene set enrichment analysis (GSEA) revealed an enrichment of the Mat^ON signature, a set of genes upregulated during cDC1 homeostatic maturation in several tissues (36), among differentially expressed genes (DEGs) downregulated in Smc3^ΔDC cDC1s (Fig. 1F and Data File S2). A similar enrichment was observed for target genes of Notch2, an important signal for terminal cDC maturation (22, 23), even after removing genes shared with the Mat^ON signature (Fig. 1F and S1H–I, and Data File S2). In contrast, proliferation signature genes were not enriched among the DEGs downregulated in Smc3^ΔDC cDCs (Data File S1), consistent with the low proliferation of cDCs that was unaffected by Smc3 deletion. Collectively, these data reveal a requirement for cohesin in the in vivo differentiation of cDCs, particularly cDC1s and Esam⁺ cDC2s.

To confirm the role of cohesin in cDC differentiation in vitro, we crossed the Smc3^fl mice to mice expressing a tamoxifen-inducible Cre recombinase-estrogen receptor fusion protein (CreER) from the ubiquitous Rosa26 locus (R26^CreER). The development of the major DC lineages (cDC1, cDC2, pDC) can be modeled by culturing total BM cells with Flt3L for 7–8 days (FL-BMDC cultures) (37). In FL-BMDC cultures from R26^CreER Smc3^fl/fl mice, the addition of 4-hydroxytamoxifen (4-OHT) at day 3 of culture, after the peak of progenitor proliferation has occurred (38), permitted DC development but depleted Smc3 protein by day 8 (Fig. S2A–C). Such Smc3-deleted (Smc3^Δ) FL-BMDC cultures showed a reduction of pDCs (~1.5-fold), cDC2s (~2-fold) and especially cDC1s (~4-fold) compared to cultures from control R26^CreER mice (Fig. S2D). We purified the residual Smc3^Δ DC subsets from FL-BMDC cultures and analyzed them by RNA-seq (Data File S3). By PCA, control and Smc3^Δ pDCs and cDC2s clustered together, whereas Smc3^Δ cDC1s clustered nearest to control cDC2s (Fig. S2E). Furthermore, DEGs upregulated in Smc3^Δ cDC1s demonstrated enrichment of the signature of primary splenic cDC2s (39), suggesting that cohesin optimizes cDC1 subset specification in this system (Fig. S2F,G and Data Files S4–S5). The induction of Notch2 signaling, via co-culture with OP9 stromal cells expressing the Notch ligand Delta-like 1 (OP9-DL1), facilitates the differentiation of bona fide cDC1s and Esam⁺ cDC2s (40). In these conditions, the output of both cDC1s and cDC2s was strongly reduced by Smc3 deletion (Fig. 1G), and Smc3^Δ cDC1s failed to upregulate CD8α (Fig. 1H). These data confirm the function of cohesin in cDC differentiation, particularly in the context of physiological signals such as Notch ligands.

Cohesin controls the function of cDCs in vivo

We next examined the function of cohesin-deficient cDCs, starting from their responses to pathogen-derived ligands of Toll-like receptors (TLR). RNA-Seq of Smc3^Δ cDCs ex vivo (Fig. 1E) and from cultures (Fig. S2E) showed reduced expression of Il12b, the gene encoding the main p40 isoform of IL-12 (Fig. 2A). Accordingly, Smc3^Δ FL-BMDC showed less IL-12p40 production when untreated or treated with the TLR9 ligand unmethylated CpG oligonucleotides (CpG) or with the TLR11 ligand profilin (Fig. 2B). Besides IL-12, only IL-6 and IFN-γ were severely reduced in the supernatants of profilin-stimulated Smc3^Δ DCs, whereas most other factors were unaffected (Fig. S3A). In response to TLR3 ligand poly(I:C) and TLR4 ligand lipopolysaccharide (LPS), Smc3^Δ DCs showed a selective impairment of IL-12 production, with either normal or increased levels of other secreted factors (Fig. S3A); these differences could not be explained solely by differences in cell numbers (Fig. S3B). Thus, the activation program of DCs displays stimulus- and cytokine-specific dependencies on cohesin, which appears dispensable for the production of many secreted factors.

Figure 2. Cohesin is required for cDC function in vivo.

(A) Normalized Il12b read counts from RNA-Seq of cDC1s purified from FL-BMDC cultures (left) and spleen (right).

(B) IL-12p40 concentration in supernatant of FL-BMDC cultures stimulated for 24 hours with profilin (middle), CpG-B (right) or left unstimulated (left).

(C-D) IL-12p40 (C) and IL-12p70 (D) concentration in the serum of mice prior to and 6 hours post challenge with profilin. Pooled from two independent experiments.

(E) IL-12p40 in the serum (left) and from peritoneal exudate cell (PEC) (middle) and splenocyte (right) supernatant 5 days after T. gondii infection.

(F) Mice received OVA-pulsed or control BALB/c splenocytes. Shown are representative flow plots (left) and fractions (right) of endogenous OVA-specific CD8⁺ T cells in spleen on day 8 post-transfer.

(G-I) Mice received bilateral subcutaneous implantation of MC38 tumor cells without immunotherapy (G) or with αPD-1 + αCD137 (H-I). Shown are tumor growth curves (G and H, left), Kaplan-Meier curves showing mice with tumor burden >250 mm³ (H, right), and largest volume tumors at day 21 endpoint (I).

(J-K) Mice were injected with sheep RBCs or PBS. Shown are representative flow plots (left), frequencies (middle) and numbers (right) of germinal center B cells (J) and T_FH cells (K) in spleen on day 8. Pooled from two independent experiments.

Symbols represent individual mice, and bars represent mean. In (A-F), (J-K), Control = CD11c-Cre (grey circles) and Smc3^fl/fl (grey triangles), Smc3^ΔDC = CD11c-Cre Smc3^fl/fl (blue circles) and CD11c-Cre Smc3^fl/- (blue triangles). In tumor experiments, symbols represent the mean of all tumors.

Statistical differences were evaluated using Mann-Whitney test, except for the log-rank (Mantel-Cox) test for evaluating the Kaplan-Meier curve (H). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s. not significant.

Given that IL-12p40 production was most markedly and consistently reduced in Smc3^Δ FL-BMDCs, we tested in vivo IL-12 production in Smc3^ΔDC mice. During acute infection with Toxoplasma gondii or its extract soluble tachyzoite antigen (STAg) containing profilin, cDC1s are the critical source of IL-12 required for host resistance (41–43). Smc3^ΔDC mice showed a dramatic loss of serum IL-12p40 in the steady state (Fig. 2C), and of both IL-12p40 and IL-12p70 in profilin-treated mice (Fig. 2C,D). Five days after infection with Toxoplasma gondii, Smc3^ΔDC mice showed reduced serum IL-12p40 and reduced production of IL-12p40 by peritoneal exudate cells and splenocytes (Fig. 2E). These data show that cohesin is required for both homeostatic and inducible production of IL-12 by cDC1s.

To test the ability of cohesin-deficient cDC1s to cross-present cell-associated antigens, we injected mice with allogeneic (H-2K^d) splenocytes loaded with ovalbumin (OVA) protein and measured by tetramer staining the expansion of OVA-specific CD8⁺ T cells (39). Smc3^ΔDC mice were defective in the cross-priming of OVA-specific CD8⁺ T cells despite expressing normal amounts of surface MHC-I (Fig. 2F and S3C). To analyze cDC1 function in anti-tumor immunity, we implanted MC38 adenocarcinoma cells into control and Smc3^ΔDC mice and treated them with a blocking anti-PD-1 antibody (αPD-1) and agonist antibody to CD137 (4-1BB), a combination that requires cDC1s for a productive anti-tumor CD8⁺ T cell response (44). Without immunotherapy, MC38 tumors grew at comparable rates in all control (CD11c-Cre or Smc3^fl/fl) and Smc3^ΔDC (CD11c-Cre Smc3^fl/fl or CD11c-Cre Smc3^fl/-) mice (Fig. 2G). Immunotherapy impaired tumor growth in control but not in Smc3^ΔDC mice (Fig. 2H), which also showed a increase in larger tumors (Fig. 2H,I). cDC2s are specialized in antigen presentation to CD4⁺ T cells, with Esam⁺ cDC2s facilitating follicular helper T (T_FH) cell differentiation and germinal center (GC) reactions after immunization (45). In line with the diminished Esam⁺ cDC2 population, Smc3^ΔDC mice showed a reduction of GC B cells and T_FH cells following immunization with sheep red blood cells (Fig. 2J,K). Thus, cohesin is required to execute the hallmark DC functions in vivo, including cross-presentation and anti-tumor response priming by cDC1s and T_FH cell priming by cDC2s.

Cohesin controls a specific subset of activation-induced genes in DCs

To characterize the differentiation and activation of primary cohesin-deficient cDCs at the single-cell level, we sorted total cDC populations from the spleen of control and Smc3^ΔDC mice, either untreated or treated with a combination of LPS and profilin, and analyzed by single-cell RNA sequencing (scRNA-Seq). Clustering and subsequent visualization with Uniform Manifold Approximation and Projection (UMAP) of the resulting scRNA-Seq dataset yielded 16 clusters in total (Fig. 3A). In addition to non-target monocytes (cl. 13), macrophages (cl. 14) and Rorγt⁺ DC (cl. 16), UMAP clearly separated naïve cDC1 (cl. 2, 7), Esam⁻ (cl. 4) and Esam⁺ (cl. 6, 10) cDC2 from untreated mice, homeostatically mature cDC1 (cl. 9) and cDC2 (cl. 12) primarily from untreated mice, and activated cDC1 (cl. 3, 8) and cDC2 (cl. 1, 5) from LPS+profilin-treated mice (Fig. 3A, S4A and Data File S6).

Figure 3. Single-cell analysis of cohesin-deficient cDC differentiation and activation.

Control (CD11c-Cre) and Smc3^ΔDC (CD11c-Cre Smc3^fl/fl) mice were either untreated (UT) or injected with LPS + profilin (cells from n=2 mice per genotype/condition were pooled into a single sample). After 6 hours, “hashtagged” cDC1 and cDC2 were sorted from the spleen and analyzed by CITE-Seq.

(A) UMAP plots colored by cell cluster (left) or by sample origin (right).

(B-C) Volcano plot showing pairwise analysis of Smc3^Δ vs control cDC1 (cl. 2+7) (B) and Esam⁺ cDC2 (cl. 6+10) (C) from untreated mice. Genes upregulated or downregulated (>2-fold difference with p_adj < 0.1) in Smc3^Δ cDC1 are colored red or blue respectively.

(D) GSEA of Notch (23) signature among rank ordered genes in the pairwise differential expression analysis between control and Smc3^Δ cDC1 (cl. 2+7) from untreated mice. NES = normalized enrichment score.

(E-F) Volcano plot showing pairwise analysis of Smc3^Δ vs control cDC1 (cl. 3+8) (E) and cDC2 (cl. 1+5) (F) from LPS+profilin-treated mice. Genes upregulated or downregulated in Smc3^Δ cDC1 are colored red or blue respectively.

(G) Violin plots of select inducible cytokines and chemokines within cDC1 and cDC2 clusters from UT or LPS+profilin-treated mice. Scale (log₂(expression)) is indicated to the right.

In untreated mice, naïve cDC1 and Esam⁺ cDC2 split into separate clusters of either control or Smc3^Δ cells; in contrast, all Esam⁻ cDC2 comprised a single cluster derived from both control and Smc3^Δ cells (Fig. 3A and S4B). Pairwise comparison revealed multiple DEGs between the corresponding subsets of control vs Smc3^Δ cDC that were highly concordant with DEGs from bulk RNA-Seq (Fig. 3B,C, S4C,D and Data File S6). Again, Smc3^Δ cDC1 demonstrated greater transcriptional differences versus control cells than did Smc3^Δ cDC2, illustrating the strong dependence of cDC1 differentiation on cohesin (Fig. 3B,C). Consistent with bulk RNA-Seq, GSEA revealed an enrichment of the Notch signature among DEGs downregulated in naïve Smc3^Δ cDC1s, exemplified by the reduction in Cd8a and Adam23 (Fig. 3B,D). Furthermore, mirroring the reduction of migratory cDCs in the sdLN of Smc3^ΔDC mice (Fig. S1D), the corresponding cluster of homeostatically mature splenic cDC1s (46) was under-represented by Smc3^Δ cells (Fig. 3A and S4B). Collectively, these data confirm at the single-cell level the requirement for cohesin in the in vivo differentiation and maturation of cDCs.

After LPS+profilin treatment, nearly all cDCs isolated from both control and Smc3^ΔDC mice were found within clusters expressing markers of activated cDCs (Ccr7, Ly6a, Cxcl10) (Fig. 3A and S4A,B), consistent with the largely intact response of Smc3^Δ cDCs to activation (Fig. S3A). Nevertheless, Smc3^Δ cDC1s and cDC2s occupied different activated clusters from controls, a distinction which was again more accentuated in cDC1 (Fig. 3E,F). Indeed, pairwise analysis and inspection of activation-inducible cytokines and chemokines confirmed that some were strongly cohesin-dependent (Il12b, Il27, Il6, Il15) whereas others were induced comparably in control and Smc3^Δ cDCs (Ccl5, Cxcl9, Cxcl10, Ccl22) (Fig. 3G). Thus, the in vivo activation program of cohesin-deficient cDCs is predominantly intact, except for a specific subset of inducible genes.

Cohesin-mediated genome organization in DCs

To test the role of cohesin in the genome organization of DCs, we purified cDCs (both subsets) and pDCs from 4-OHT-treated control and Smc3^Δ FL-BMDC cultures (Fig. S2) and performed Hi-C (Fig. 4A). PCA of pairwise Hi-C map similarities revealed the separation of all cDCs and pDCs along PC1 and the separation between genotypes along PC2 (Fig. S5A), suggesting that cohesin depletion preserves subset identities but induces changes in chromosome organization. The curves of the contact frequency P(s) as a function of genomic separation (s) are used to detect extruded loops, seen as a characteristic hump (elevated contact frequency, and a peak at the P(s) derivative) at the scale corresponding to the loop size (47, 48). While this curve for control DCs showed a hump corresponding to ~100 Kb loops, this hump was reduced and the curve of the P(s) derivative was flattened in Smc3^Δ cDCs (Fig. 4B) and Smc3^Δ pDCs (Fig. S5B). Using Hi-C data from mouse embryonic stem cells with a near-complete loss of cohesin (49) (Fig. S5C), we estimated that Smc3^Δ cDCs and pDCs retain ~54% and ~73% of residual cohesin activity, respectively (Fig. S5D,E). Such ~2-fold reduction would increase the separation between cohesin complexes on the genome, allowing larger loops to be extruded (50). Indeed, the hump in the P(s) curve showed a ~2-fold increase in the loop sizes in Smc3^Δ cells (Fig. 4B and S5B), emphasizing the sensitivity of DC differentiation to cohesin dosage.

Figure 4. Cohesin regulates genome organization in DCs.

(A) Experimental schematic. 4-OHT-treated FL-BMDC cultures from R26^CreER/+ (Control) or R26^CreER/+ Smc3^fl/fl (Smc3^Δ) mice were analyzed by RNA-Seq (n=3 per genotype/DC subset), or by Hi-C for cDCs and pDCs (n=2 for each genotype/DC subset, post-sequencing replicates combined for analysis).

(B) P(s) curves (top) and their slopes (bottom) for cDCs. Shaded regions represent 90% confidence intervals. Dashed horizontal lines (top plot) show mean interchromosomal (trans-) interactions.

(C) Average pileups around stripes (top row), dots (middle row) and fountains (bottom row) (left) and summary statistics of their strength (right) in cDCs.

(D) Contact frequency maps at a representative locus highlighting the above-mentioned chromatin structures in cDCs.

(E) Effect of Smc3^Δ on contact probability in cDCs for indicated region types (top), including CTCF in convergent orientation (P_{CTCF +-} (s)), active genomic regions (P_active (s)), and putative promoter-enhancer pairs (P_EP (s)). Insets represent the ratio between these specific P(s) and whole-genome P(s), displayed in Fig. 4B. Beneath shows the Smc3^Δ/control cDC ratio of P(s) for that genomic region type (color), alongside the same ratio for whole-genome P(s) (grey).

(F) Experimental schematic. FL-BMDCs were unstimulated or stimulated for 6 hours with CpG. cDC1s and cDC2s were purified and analyzed by RNA-seq (n=2 for each genotype/DC subset post-activation), and for control unstimulated cells only, by CUT&RUN for histone modifications (n=1 per genotype/DC subset/histone mark).

(G) Scatterplots of genes in control and Smc3^Δ cDC1s with or without CpG activation. Cohesin-dependent, cohesin-independent and non-inducible genes defined as described in the text are indicated.

(H) Median distance of cohesin-dependent, cohesin-independent and non-inducible genes to their nearest enhancer in cDC1s. Right shows proportions of gene subsets that have their nearest enhancer located within indicated distances from their transcription start site.

(I) H3K27me3 CUT&RUN was performed on unstimulated control cDCs as well as CpG-stimulated control and Smc3^Δ cDCs from FL-BMDC cultures (n=2 for each genotype/condition). Median Z-score of H3K27me3 signal within the insulated neighborhood of genes from the indicated gene categories in activated cDC1s. Statistical differences were calculated with the Mann-Whitney one-sided test.

Data are presented as box plots. *p < 0.05, ***p < 0.001, ****p < 0.0001; n.s. not significant.

While the overall compartmentalization of the genome as defined by the first eigenvector (E1) was unchanged in Smc3^Δ cDCs (Fig. S5F), extrusion-associated features were reduced. These included weakened insulation at CTCF sites and dissolution of TADs (Fig. S5G); reduction of “stripes” reflecting the stoppage of loop extrusion at CTCF sites (9) (Fig. 4C,D); and weakening of “dots”, which reflect the stopping of cohesin at two CTCF sites and a transient loop between them (49), particularly for the loops <200 Kb (Fig. 4C,D). We also examined the so-called “jets” or “fountains”, which reflect the loading of cohesin at active enhancers (51, 52). We identified an average “fountain” by aggregating Hi-C maps at active chromatin that show such a pattern (53), and observed a diminished fountain signature in the Smc3^Δ cDCs (Fig. 4C,D). Long-range contacts were either unaffected (all contacts with s>4 Mb, Fig. 4E) or became slightly enriched (interchromosomally, Fig. 4B, S5B, S5H), consistent with chromosome decompaction upon cohesin loss (47). Similar (albeit smaller) effects were observed in Smc3^Δ pDCs (Fig. S5I). Altogether, cohesin-deficient cDCs manifest genome-wide reductions in major features associated with loop extrusion. We also examined effects of cohesin depletion on interactions between functional loci at different genomic separations. Similar to the P(s) for all contacts, we made plots of contact frequency for pairs of convergent CTCFs, P_CTCF+-(s); between pairs of active elements (all enhancers and promoters), P_active(s); and for DC-specific enhancer-promoter pairs (called using activity-by-contact ABC model), P_EP(s) (Fig. 4E). All of these interactions were enriched relative to their backgrounds (Fig. 4E, insets) in DCs, and were reduced in Smc3^Δ cDCs at genomic separations s~30–500Kb. This is consistent with previous estimates {Kane, 2022 #32016;Rinzema, 2022 #32015;Hansen, 2025 #32167} and suggests that cohesin mediates long-range enhancer-promoter interactions in cDCs.

Cohesin regulates a subset of the cDC activation program

To further dissect the role of cohesin in activation-induced DC responses, we stimulated Smc3^Δ and control FL-BMDC cultures (Fig. S2) with CpG (both A- and B-types), which activate all DC subsets through TLR9. After 6 hours, unstimulated or CpG-stimulated cDC1 and cDC2 subsets were sorted and analyzed by RNA-seq (Fig. 4F and Data File S3). PCA of RNA-seq profiles showed that both control and Smc3^Δ cDCs separated equally from their unstimulated counterparts along PC1 (Fig. S6A), whereas the two genotypes separated along PC2. A small subset of activation-induced genes (e.g. Shisa3, Cxcl11, Il6 and Saa3) showed preferential expression in both subsets of control cDCs. In addition, several inducible genes showed higher expression in control activated cDC1s, including Il12b and Nos2 (Fig. S6B). Thus, consistent with the analysis of in vivo DC responses by scRNA-Seq (Fig. 3G), the loss of cohesin does not block the cDC activation program but impairs the induction of a few specific genes.

Consistent with PCA, pairwise comparison showed only a slight reduction of activation-induced transcription in Smc3^Δ cDCs, except for a small subset of inducible genes (Fig. 4G and S7A, and Data Files S7–S8). We defined cohesin-dependent inducible genes as those induced by CpG in control but not Smc3^Δ DCs (n=117 and 112 in cDC1s and cDC2s, respectively), versus cohesin-independent genes that were induced in both control and Smc3^Δ DCs (n=989 and 1348 in cDC1s and cDC2s, respectively), and non-inducible genes that were not upregulated upon stimulation (n=7501 and 7186 in cDC1s and cDC2s, respectively) (Fig. 4G and S7A). We then used CUT&RUN to profile histone modifications in wild-type quiescent cDC1s and cDC2s (Fig. 4F) and identified active enhancers as ATAC-Seq peaks with “enhancer state” called by ChromHMM. We then defined a putative enhancer for each gene as the closest enhancer in the gene’s insulated neighborhood (i.e. not insulated by a DNA-bound CTCF). Cohesin-dependent inducible genes were >1.7-fold more distant from their putative enhancers than cohesin-independent inducible genes, with a particular enrichment among genes located >200 Kb from the nearest enhancer (Fig. 4H and S7B). Similar results were obtained with enhancers defined using the activity-by-contact (ABC) model (57) (Fig. S7C). Contact frequencies between promoters of inducible genes and their ABC-defined putative enhancers was reduced in Smc3^Δ DCs for genomic separations up to 2 Mb, comparably for cohesin-dependent and independent genes (Fig. S7D–E).

We further compared the three groups of genes using histone profiling described above, as well as ATAC-Seq and ChIP-Seq for CTCF from in vitro-derived cDCs described below (in Fig. 5). Compared to cohesin-independent genes, cohesin-dependent genes showed an increased distance to the nearest CTCF binding site (Fig. S7F) and a reduction of open chromatin peaks (Fig. S7G). Among histone modifications, histone H3 tri-methylation at lysine 27 (H3K27me3) was enriched in cohesin-dependent vs cohesin-independent genes, and generally in inducible compared to non-inducible genes (Fig. 4I and S7H). H3K27me3 is a mark of repression by the Polycomb repressive complex, consistent with the role of cohesin in disrupting Polycomb-dependent chromatin interactions (8). Indeed, H3K27me3 signal decreased at cohesin-dependent inducible genes, as compared to cohesin-independent genes, in control but not Smc3^Δ cDCs upon activation (Fig. 4I and S7H). Other differences between cohesin-dependent vs -independent genes included increased histone H3 tri-methylation at lysine 9 (H3K9me3) in both cDC1 and cDC2 (Fig. S8A), a mark generally associated with heterochromatin. They also included decreased histone H3 tri-methylation at lysine 4 (H3K4me3) in cDC1 only (Fig. S8B), a mark associated with active gene promoters. Both of these differences, along with reduced ATAC-Seq signal, are consistent with increased Polycomb mark on cohesin-dependent genes. No differences were observed in H3 lysine 4 monomethylation or lysine 27 acetylation (Fig. S8C,D). Collectively, these data suggest that cohesin in DCs is required for the induction of a distinct group of genes that are located farther from their enhancers and CTCF-binding sites, and are subject to Polycomb-mediated repression.

Figure 5. Uni-IPG analysis reveals dynamics of genome organization during DC differentiation.

(A) Experimental schematic. RNA-Seq (n=3), ATAC-Seq (n=2), Hi-C (n=2, post-sequencing replicates combined for analysis) and ChIP-Seq for CTCF and Rad21 (n=1) were performed on HoxB8-FL progenitors (Undiff) and differentiated cDCs and pDCs, accompanying the datasets performed on FL-BMDCs in Fig. 4.

(B) Workflow to generate unified interaction profile groups (uni-IPGs). The simultaneous decomposition of all Hi-C subsets was performed on a consensus compartmental profile, which allowed for harmonization across samples and experimental systems.

(C) Heatmap of mean enrichment of CUT&RUN-derived histone marks (Z-score normalized log10 signal in 25 Kb bins) in uni-IPG states in HoxB8-FL-derived cDCs.

(D) Sankey plot of IPG transitions of genomic bins between progenitors and cDCs.

(E) Sankey plot of IPG transitions of genes between progenitors and cDCs for expressed genes, with the fractions of upregulated (red) and downregulated (blue) genes in each compartment indicated.

(F) The matrix showing numbers of genes in each uni-IPG in progenitors and cDCs for the genes that were upregulated during cDC differentiation.

(G) Hi-C maps in the vicinity of the activation-inducible Stat4/Stat1 locus in HoxB8-FL progenitors and cDCs. First projected spectral component (E1) and uni-IPG tracks are shown above the Hi-C maps.

(H) The matrix showing numbers of genes in each uni-IPG in progenitors and cDCs for the genes that were upregulated during the activation of cDC1s by CpG.

The role of genome compartmentalization in DC differentiation and activation

A recent analysis of Hi-C data from cell lines revealed genome compartmentalization into multiple active and inactive compartments with distinct properties, termed the interaction profile groups (IPG) (58). We sought to apply the IPG approach to the dynamics of genome compartments during DC differentiation and activation. To compare DCs to their progenitors, we used the Flt3L-dependent HoxB8-FL cells that can be grown as multipotent progenitors or differentiated into a mixture of pDCs and cDC2s (59, 60). We performed Hi-C on HoxB8-FL progenitors and their DC progeny, as well as epigenetic profiling of HoxB8-FL progenitors to match that of BM-derived DCs (Fig. 5A). The resulting Hi-C profiles corresponded to the differentiation trajectory of ex vivo DC progenitors (61) as determined by PCA (Fig. S9A), but at higher Hi-C resolution. First, we sought to integrate Hi-C profiles of HoxB8-FL progenitors, HoxB8-FL-derived pDCs and cDCs, and BM-derived DCs (Fig. 5A). Because the standard Hi-C decomposition (62) generated vastly different compartmental profiles of progenitors and DCs (Fig. S9B), we used simultaneous decomposition of all Hi-C subsets and their projection on a consensus compartmental profile, which allowed efficient harmonization of samples from both experimental systems (Fig. 5B and S9B). Subsequent clustering of projections yielded two active (A1-A2) and three inactive (B1-B3) unified IPG (uni-IPG) (Fig. 5B and Data File S9), whose identity was consistent with their epigenetic profiles (Fig. 5C). The analysis of uni-IPG in Smc3-deficient cDCs showed very few differences from controls (Fig. S9C), consistent with cohesin being dispensable for chromatin compartmentalization (10).

Progenitor differentiation into cDCs was accompanied by extensive reciprocal transitions between uni-IPG, only a small fraction of which represented “generic” A<>B compartment transitions (Fig. 5D). Most of the transitions were between “adjacent” uni-IPG (e.g. B2<>B1), although “jumps” across a uni-IPG (e.g. B2<>A2) could be observed. Transitions into a more active uni-IPG during differentiation (e.g. A2->A1 or B1->A2) were enriched for upregulated genes, and vice versa (Fig. 5E and S9D, and Data File S10). Conversely, genes that were upregulated in cDCs vs progenitors either remained in the same uni-IPG or transitioned to a more active uni-IPG (e.g. A2->A1 or B1->A2) (Fig. 5F). We then examined the compartmentalization of genes that were induced by activation in cDCs (Data File S7), as exemplified by the Stat4/Stat1 locus (Fig. 5G). The vast majority of them resided in active uni-IPG (A1 or A2) in differentiated cDCs even prior to activation (Fig. 5H and S9E,F). Most of these genes were in the same active uni-IPG in progenitors, whereas others transitioned to a more active compartment (A2->A1 or B1->A2) during differentiation (Fig. 5H). The transitions between compartments also correlated with changes in histone modifications (Fig. S9G). Finally, we examined the relationship between uni-IPG partitioning and cohesin dependence of inducible genes. Consistent with the enrichment of H3K27me3 in cohesin-dependent genes, a larger fraction of them were in B2 or B1 compartments (for cDC1 and cDC2, respectively), comprising a larger net fraction of the H3K27me3-marked B1/B2 compartment (Fig. S9H). Thus, DC differentiation is accompanied by cohesin-independent redistribution of genes between chromosomal compartments, with the majority of activation-inducible genes transitioning into the active compartments prior to activation.

IRF8 controls genome compartmentalization during DC differentiation

We hypothesized that the cohesin-independent compartment dynamics in DCs might be controlled by a key DC-specifying TF such as IRF8. To test this notion, we deleted IRF8 from HoxB8-FL cells using CRISPR/Cas9-mediated gene targeting (Fig. S10A). As expected, the resulting IRF8-deficient (Irf8^Δ) HoxB8-FL cells (Fig. 6A) generated only immature cDC-like cells lacking MHC class II expression (Fig. 6B). We analyzed the transcriptome of Irf8^Δ CD11c⁺ DCs by RNA-Seq and integrated it with RNA-Seq of differentiating HoxB8-FL cells ((60) and this study, Data File S11). By PCA, IRF8-deficient cells mapped close to immature DCs between differentiation days 2 and 4 (Fig. 6C). Clustering analysis confirmed the similarity of Irf8^Δ HoxB8-FL-derived cells and DCs on day 4 of differentiation (Fig. S10B and Data File S12). Thus, similar to its key role in vivo, IRF8 is required for terminal DC differentiation in the HoxB8-FL system.

Figure 6. IRF8 regulates the genome compartmentalization of differentiating DCs.

(A) Experimental schematic. In addition to datasets in Fig. 5, RNA-Seq was performed on HoxB8-FL cells at day 2 and day 4 along their differentiation trajectory (n=3, left). An sgRNA targeting Irf8 was used to generate IRF8-deficient (Irf8^Δ) HoxB8-FL cells using Cas9-RNPs (right). RNA-Seq (n=3) and Hi-C (n=2, post-sequencing replicates combined for analysis) was performed on differentiated Irf8^Δ DCs well as RNA-Seq of wild-type control bulk DCs (n=3).

(B) Representative flow plots of CD11c⁺ cells (left), DC subsets (middle) and MHC-II⁺ cDCs (right). Representative of three independent experiments.

(C-D) PCA of RNA-Seq (C) and Hi-C (D) showing position of Irf8^Δ DCs relative to the normal transcriptional and chromatin trajectory of differentiating DCs.

(E) Hi-C maps in the vicinity of the Ciita locus in HoxB8-FL progenitors as well as cDCs and Irf8^Δ DCs. First projected spectral component (E1) and uni-IPG tracks are shown above the Hi-C maps. Boxed regions highlight compartments with disrupted dynamics in Irf8^Δ DCs.

(F) Sankey plot of IPG transitions of genomic bins during normal differentiation of progenitors into cDCs, colored (in orange) by transitions altered in Irf8^Δ DCs.

(G) Sankey plot of IPG states of genes between cDCs and Irf8^Δ DCs for expressed genes, with the fractions of upregulated (red) and downregulated (blue) genes in each compartment indicated.

(H-I) Matrices showing numbers of upregulated (H) and downregulated (I) genes in Irf8^Δ DCs, organized by their IPG status in cDCs and Irf8^Δ DCs.

We performed Hi-C on IRF8-deficient HoxB8-FL-derived DCs and compared it to Hi-C profiles of wild-type HoxB8-FL progenitors and DCs (Fig. 6A). PCA showed that Irf8^Δ cells incompletely progressed along the DC differentiation trajectory (PC1) and remained unaffiliated with either pDC or cDC subsets (Fig. 6D). This was consistent with the impaired chromatin organization of primary IRF8-deficient monocyte-DC progenitors (MDP), which develop normally without IRF8 ((61), Fig. S10C). We then compared cohesin-dependent features such as Hi-C contact frequency and the strength of stripes, dots and fountains between undifferentiated progenitors and differentiated wild-type or IRF8-deficient DCs. All these features in IRF8-deficient cells were either similar to progenitors (for contact frequency, dots and stripes) or intermediate between progenitors and DCs (for fountains enriched specifically at enhancers) (Fig. S10D–G). Thus, IRF8, directly or indirectly, facilitates optimal cohesin-mediated chromatin organization in DCs.

We then analyzed the compartmentalization of Irf8^Δ cells using the uni-IPG approach (Data File S9). Representative genes that are upregulated (Ciita, Fig. 6E) or downregulated (Cd34, Fig. S10H) during DC differentiation showed disrupted compartments in Irf8^Δ cells. Accordingly, many transitions between uni-IPG in progenitors vs DCs were altered, including the majority of “jumps” over one or two uni-IPG (Fig. 6F and S10I,J). Direct comparison between wild-type cDCs and Irf8^Δ cells revealed major differences in their uni-IPG structure, which correlated with their gene expression profiles (Fig. 6G and S10K). Indeed, many genes that were up- or downregulated in Irf8^Δ cells were also localized in a different compartment than in wild-type cDCs (Fig. 6H,I). Collectively, these results show that IRF8 controls the genome organization of differentiated DCs through two mechanisms: by facilitating cohesin-mediated features, and by enforcing cohesin-independent compartmentalization.

Local genome organization optimizes IRF8 expression in DCs

Having established the role of IRF8 in the control of chromatin dynamics, we asked the opposite question, namely: how does chromatin affect the expression of Irf8 itself? We reasoned that as an upstream regulator, the Irf8 locus may have a distinct chromatin arrangement that ensures its proper expression. Indeed, we observed that Irf8 is located as a sole gene within a compact TAD present in both progenitors and mature DCs (Fig. 7A). This Irf8 TAD was demarcated by two TAD boundaries (TB1 and TB2), each containing at least three CTCF binding sites (Fig. 7B). All enhancers that regulate Irf8 expression in DCs (63, 64) and have been shown to physically interact (65) are located within this TAD. To test whether the enhancer-containing TAD may regulate Irf8 expression, we generated mice lacking TB1 and TB2 (Irf8^ΔTB) through two successive rounds of CRISPR/Cas9-mediated targeting in zygotes (Fig. 7C). In the resulting Irf8^ΔTB mice compared to controls (Irf8^WT), the DC-enriched fraction of the spleen and BM showed a ~50% reduction in Irf8 expression (Fig. 7D). Accordingly, the levels of IRF8 protein in Irf8^ΔTB cDC1s and pDCs were similar to those in heterozygous Irf8^+/− mice (Fig. 7E). IRF8 expression in Ly6C⁺ monocytes and B cells, both of which express intermediate levels of IRF8, was similarly reduced in Irf8^ΔTB mice (Fig. S11A). IRF8 is strictly required for cDC1 development but is dispensable for pDC development in vivo due to compensation by IRF4 (28). Despite the observed reduction of IRF8 expression, cDC1 and pDC numbers were unaffected in Irf8^ΔTB mice (Fig. S11B,C). We therefore turned to FL-BMDC cultures, which force DC development without potential in vivo compensatory mechanisms. Similar to primary DCs, cDC1s and pDCs from Irf8^ΔTB FL-BMDC cultures demonstrated an ~50% reduction in IRF8 expression (Fig. 7F). In this setting, however, Irf8^ΔTB cDC1 development was specifically and severely crippled (Fig. 7G). These results reveal a role for Irf8-flanking TAD boundaries in the optimal IRF8 expression and the development of IRF8-dependent cDC1s.

Figure 7. Local architectural elements optimize IRF8 expression in developing DCs.

(A) Hi-C maps near the Irf8 locus (highlighted by dashed lines) in HoxB8-FL progenitors and differentiated DCs. First projected spectral component (E1) and uni-IPG tracks are shown above the Hi-C maps.

(B) ATAC-seq as well as Rad21 and CTCF ChIP-seq in HoxB8-FL progenitors and differentiated DCs within the region outlined in (A). Irf8 gene = purple; known myeloid/progenitor cell-type specific Irf8 enhancers = blue; TB1 and TB2 TAD boundary regions composed of clusters of CTCF/Rad21 binding sites = orange. Arrowheads indicate the directionality of CTCF binding sites. The scale of each track is indicated on its right.

(C) Strategy used to excise TB1 and TB2 through sequential rounds of targeting in zygotes to generate Irf8^ΔTB mice (top). Red lightning bolts indicate sgRNA target sites. Deletion of the TB1 and TB2 regions was screened by PCR (bottom) with primer sets indicated by arrows and confirmed by Sanger sequencing.

(D) RT-qPCR for Irf8 (expressed as a ratio relative to Actb) in the DC-enriched fractions of the spleen and BM. Pooled from two independent experiments.

(E) Representative histograms (left) and quantification of IRF8 median fluorescent intensity (MFI) (right) in cDC1s (spleen) and pDCs (spleen and BM) from mice of indicated genotypes. Irf8^+/− mice were used as controls with ~50% levels of IRF8.

(F) Representative histograms (left) and quantification of IRF8 MFI (right) in cDC1s and pDCs from FL-BMDC cultures derived from the mice of indicated genotypes.

(G) Representative flow plots (left) and numbers (right) of DC subsets from FL-BMDC cultures.

Symbols represent individual mice, and bars represent mean. Statistical differences were evaluated using Mann-Whitney test, except for one-way ANOVA followed by Tukey’s test (E). *p < 0.05, **p < 0.01, ****p < 0.0001; n.s. not significant.

CTCF sites control basal and inducible IL-12 production in cDC1s

We directly tested the role of cohesin-mediated control of gene expression in DCs in vivo focusing on its identified key target, IL-12 (Fig. 2A–E). Hi-C analysis revealed that the Il12b locus is located within a “private” ~123 Kb sub-TAD that transitioned from B to A genomic compartments during the differentiation of HoxB8-FL progenitors to mature DCs (Fig. 8A). This sub-TAD contained the upstream HSS1 enhancer of Il12b (66, 67), and single CTCF/cohesin binding sites constituted its upstream and downstream boundaries (Fig. 8B). To test the role of these CTCF sites in Il12b regulation, we concomitantly targeted the CTCF binding motifs at each TAD boundary with a single sgRNA in zygotes (Fig. 8C). By PCR and Sanger sequencing, we confirmed 11 bp deletions interrupting the majority of the CTCF motifs at both TAD boundaries (Fig. 8C). The resulting Il12b^ΔCTCF mice manifested lower levels of IL-12p40 in their serum in the steady state (Fig. 8D). Profilin-induced production of IL-12p40 was similarly impaired; furthermore, a substantial fraction of Il12b^ΔCTCF mice showed very little to no IL-12p40 response (Fig. 8D). The latter observation suggested that CTCF binding sites increase the probability (rather than the magnitude) of the activation-induced IL-12 response in vivo. Il12b^ΔCTCF cDC1s enriched from spleens or from FL-BMDC cultures displayed a similar impairment in the steady-state and profilin-induced Il12b expression (Fig. 8E,F). This impairment was specific to Il12b compared to other inducible cytokines (e.g. Il6, Tnf), suggesting that the development and activation programs of Il12b^ΔCTCF cDC1s are otherwise normal (Fig. S11D,E). Collectively, these data reveal a role for CTCF/cohesin sites in the hallmark function of cDC1s, namely IL-12 production.

Figure 8. Il12b CTCF sites control basal and inducible Il12b expression.

(A) Hi-C maps in the vicinity of the Il12b locus (highlighed by dashed lines) in HoxB8-FL progenitors and differentiated cDCs and pDCs. First projected spectral component (E1) and uni-IPG tracks are shown above the Hi-C maps.

(B) The chromatin profile (ATAC-seq) as well as Rad21 and CTCF binding profiles (ChIP-seq) in HoxB8-FL progenitors and differentiated cDCs and pDCs within the region outlined in (A). The Il12b gene is shaded purple. The HSS1 enhancer is highlighted in blue. The TAD boundaries composed of a single CTCF/Rad21 binding site on either side of the locus are highlighted in orange, with arrowheads indicating the directionality of CTCF binding sites. The scale of each track is indicated on its right.

(C) Schematic of the strategy used to target the CTCF binding sites upstream and downstream of Il12b to generate Il12b^ΔCTCF mice (top). Green lightning bolts indicate sgRNA target sites. Significant deletion of the CTCF binding motif was screened by PCR (bottom) with primer sets indicated by arrows and confirmed by Sanger sequencing (middle). The sequence of the wild-type CTCF binding site is underlined, and the deletion observed in founders is highlighted in green.

(D) Quantification of IL-12p40 concentration in serum of indicated mice prior to (left) and 6 hours post challenge with profilin (right). Pooled from two independent experiments.

(E-F) RT-qPCR for Il12b (expressed as a ratio relative to Actb) in spleen enriched for cDC1 (E) and cDC1s purified from FL-BMDC cultures (F) of indicated mice that were either unstimulated (left) or stimulated with profilin (right) for 4 hours in vitro.

Symbols represent individual mice, and bars represent mean. Statistical differences were evaluated using Mann-Whitney test. **p < 0.01, ***p < 0.001, ****p < 0.0001.

Discussion

The causal role of cohesin-mediated genome folding in gene expression and the resulting tissue physiology is not well understood. Deletion of the cohesin subunit Nipbl in hepatocytes ablated TADs yet only modestly affected transcription, with the effect on cell differentiation unknown (10). Conversely, major effects of cohesin loss in the developing brain (68) or in B lymphocyte responses (69) may reflect the role of cohesin in cell division. Here, we interrogated the role of cohesin in the differentiation and function of DCs independently of its effect on proliferation. We found that even an incomplete (<50%) deletion of Smc3 from DCs impaired all cohesin-dependent features including cohesin stopping at CTCF sites, CTCF-facilitated interactions (stripes) and cohesin loading at enhancers (fountains) (52). The results revealed a critical role for cohesin in the terminal differentiation of cDCs, particularly in hallmark cDC1 functions such as cross-presentation and antitumor responses in vivo. The preferential sensitivity of cDCs vs pDCs to cohesin levels may be attributable to continuous cDC differentiation in peripheral tissues. The observed preferential cohesin dependence of cDC1s may reflect their nature as a distinct lineage with unique progenitors, driven by multiple TFs (i.e. IRF8, NFIL3, ID2, BATF3) (70); in contrast, cDC2s represent a convergent state of differentiation from multiple progenitors (71, 72).

Whereas inducible gene expression in macrophages was found to be partially impaired by cohesin loss (73), the activation program of cohesin-deficient cDCs was predominantly intact, except for a specific subset of inducible genes. This may be because the majority of inducible genes have already repositioned into active compartments (IPG) during the DC differentiation process. Furthermore, during DC activation, which proceeds rapidly and without cell division, genome structure is only minimally reorganized (61, 74). Consistent with this anticipatory effect, we observed that steady-state cohesin-deficient cDCs exhibited a dysregulated maturation (Mat^ON) signature, which contains many genes that are superinduced during cDC activation. In that respect, “priming” via differentiation-dependent chromatin reorganization should be clearly distinguished from true inducibility. On the other hand, the few cohesin-dependent inducible genes were situated farther from active enhancers and were enriched in Polycomb-dependent repressive chromatin marks. This is consistent with recent studies in synthetic systems and/or cell lines, which showed that cohesin helps establish distal enhancer-promoter interactions (54, 55) and counteracts polycomb repression (8). Thus, cohesin-mediated genome organization facilitates inducible responses both by “priming” inducible loci during differentiation for their subsequent induction, and by facilitating enhancer-promoter interactions and de-repression during activation.

Lineage-specifying TFs can contribute to the control of genome organization by modulating chromatin looping in lymphocytes (15, 16, 18). Myeloid progenitors deficient in IRF8 were shown to have abnormal Hi-C profiles (61), whereas subsequent differentiation stages could not be examined. Our genetic analysis of IRF8 revealed its critical contribution to cohesin-mediated genome organization, which could be direct (via cooperation with cohesin or its regulators on the chromatin) and/or indirect (via the enforcement of terminal differentiation). In addition, generalization of the IPG approach (58) for developmental trajectories of Hi-C revealed that IRF8 mediated transitions between compartments that underlie gene expression changes during DC differentiation. Thus, both cohesin and specific TFs have essential roles in differentiation: while cohesin is responsible for local chromosome folding that mediates long-range enhancer-promoter communication, TFs drive differentiation and associated chromosome reorganization at all levels, including compartments. While the contribution of TFs to cohesin-mediated gene regulation is well established, this is not true for the opposite process, i.e. the regulation of TFs by cohesin. Here, we found that cohesin optimized the expression of IRF8 via the TAD boundary CTCF elements around the Irf8 gene itself. Such insulation of the gene encoding an essential TF is likely to represent a common mechanism that ensures robustness of TF-driven cell differentiation programs.

The role of CTCF sites has been studied largely within well-characterized networks of cis-regulatory elements in various developmental contexts, including lymphocytes (75, 76). Here, we deleted in the germline CTCF binding sites within loci that are critical for cDC1 development (Irf8) and function (Il12b). In the case of Irf8, the removal of its flanking TAD boundaries reduced Irf8 transcription and IRF8 protein expression ~2-fold, limiting cDC1 differentiation in vitro. Mutagenesis of Il12b-flanking CTCF sites decreased both basal and inducible Il12b expression ~2-fold, leading to a defective and heterogenous in vivo cDC1-driven IL-12 response to profilin. These reductions in expression upon CTCF site loss are also consistent with effects observed in cell lines (77) and beget several conclusions about the function of CTCF/cohesin. First, CTCF goes beyond its role as an insulator, facilitating enhancer-promoter communication and optimal gene expression. Yet even in the absence of their nearest CTCF site, enhancers apparently can reach their promoters, perhaps via cohesin-mediated extrusion not facilitated by CTCF. Second, in addition to optimizing the magnitude of gene expression, CTCF/cohesin-mediated regulation also controls its robustness, particularly under inducible conditions. In that respect, the inducible in vivo IL-12 response was not only diminished but also rendered more variable. The coupling of enhancer and promoter bursting was proposed as the key cohesin-dependent parameter during inducible responses in macrophages (66). Given that coupling is an intrinsically probabilistic event at the level of individual alleles, this may create a thresholding effect wherein a certain amount of coupling is required for a productive inducible response.

In conclusion, our studies demonstrate the essential role of cohesin-mediated chromatin organization in cell differentiation, inducible gene expression and ultimately in efficient immune responses in vivo. They also reveal both distinct and cooperative pathways of chromatin regulation by cohesin and transcriptional master regulators such as IRF8, as well as an unexpected role of the former in the control of the latter. These mechanisms appear to underlie an anticipatory state of chromatin that facilitates rapid DC responses.

Materials and Methods

Study design.

The aim of the study was to elucidate the role of cohesin-mediated genome organization, and its crosstalk with lineage-specifying TFs, in dendritic cell differentiation and function. The study used a combination of high-throughput chromatin and transcriptional profiling assays in orthogonal genetic models, including: genetic deletion of cohesin subunit Smc3 in murine DCs; deletion of TF IRF8; and mouse models of TAD boundary deletion and CTCF site disruption at Irf8 and Il12b loci respectively. Animals were assigned to groups based on genotype, without randomization or exclusion; experimenters were not blinded to animal identity except when measuring tumor growth.

Additional details of experimental methods, as well as all computational analysis methods, are described in Supplemental Text and Fig. S12. All experimental models, reagents, antibodies, oligonucleotides, datasets and software are listed in Data File S13.

Mice.

Mice were housed and bred under specific pathogen-free conditions at New York University Grossman School of Medicine (NYUGSoM). All animal studies were performed according to the investigator’s protocol approved by the Institutional Animal Care and Use Committee of NYUGSoM. Mice with a conditional Smc3 allele were crossed with DC-specific Itgax-Cre (33) (Smc3^ΔDC) or with tamoxifen-inducible Rosa26^CreER/+ (78). Mice with deletion of Irf8-flanking TAD boundaries (Irf8^ΔTB) or disruption of Il12b-flanking CTCF binding sites (Il12b^ΔCTCF) were generated on C57BL/6N background by microinjecting day 0.5 single-cell embryos with Cas9 protein and relevant sgRNAs at the Rodent Genetic Engineering Laboratory; the resulting pups were screened by PCR, confirmed by Sanger sequencing and back-crossed at least once to C57BL/6 prior to inter-crossing heterozygotes. The Il12b-flanking CTCF binding sites were disrupted concomitantly with a pair of sgRNAs; the TAD boundaries flanking Irf8 were removed sequentially. Experiments were performed with adult mice 6–10 weeks of age. No differences were observed between male and female mice in any experimental system, therefore mice of both sexes were used throughout this study, with age- and sex-matched mice used within each experiment.

Cell Lines.

Murine Flt3L-secreting B16 melanoma cell line (B16-Flt3L), DL1-expressing OP9 cell line (OP9-DL1) and immortalized progenitor HoxB8-FL cell line were cultured and used as described before (40, 60) and detailed in Supplementary Text.

Mouse procedures.

For EdU labeling, mice were injected i.p. on three consecutive days with 1 mg EdU and harvested for analysis one day after the last injection. For SRBC immunization, mice were injected i.p. with 2×10⁸ sheep RBCs and harvested for analysis eight days post-immunization. For in vivo cross-presentation, cells from spleen and skin-draining lymph nodes of BALB/c (H-2^d) mice were pulsed with a hypertonic solution containing 10 mg/ml chicken ovalbumin (OVA) protein in RPMI, flushed with a hypotonic solution of 40% water in RPMI, washed with PBS, and irradiated with 2000 rad using a Cell-Rad X-Ray irradiator (Precision) as described (39). Recipients were injected i.v. with 4×10⁷ cells, and the OVA-specific CD8⁺ T cell response was analyzed in the spleen on day seven post-injection using OVA/H-2K^b tetramer staining. As a control, recipient mice received irradiated BALB/c splenocytes pulsed with a hypertonic solution lacking OVA protein.

For IL-12 responses, mice were injected i.p. with 100 ng profilin. Blood was collected from mice both before and 6 hours after challenge, and serum separated from whole blood. For analysis of in vivo inducible responses by CITE-Seq, mice were injected i.v. with 100 ng profilin and 2.5 μg LPS. Spleens were collected from mice 6 hours after challenge. T. gondii (ME49 cysts) were obtained by homogenizing brain tissue from chronically infected mice, and 20 cysts were injected i.p. as described (79). Serum, spleen and peritoneal wash were obtained 5 days later, and IL-12p40 concentrations were measured by ELISA.

For antitumor responses, MC38 cells were washed once and resuspended in endotoxin-free PBS containing 2.5 mM EDTA. Mice were injected s.c. on both dorsal flanks with 5×10⁵ tumor cells in 100 ul volume. To analyze the response to immunotherapy, on day 6 post-tumor inoculation, mice of each genotype were randomized and injected i.p. with a combination immunotherapy regimen of 200 μg anti-PD-1 blocking antibody and 200 μg anti-4–1BB agonist antibody or rat IgG2a isotype control antibody in 200 ul PBS every third day. Tumor growth was measured using a caliper by a researcher who was blinded to genotype and treatment group. Tumor volume was estimated using the formula: Tumor volume ~ Length × Width² × π/6.

Mouse Tissue Primary Cell Preparation.

Spleens were minced and digested in 1 mg/ml collagenase D and 20 μg/ml DNaseI in RPMI supplemented with 10% FCS for 30 min at 37°C. Cleaned femur and tibia bones were crushed with mortar and pestle. Capsules of skin-draining lymph nodes (bilateral inguinal and brachial nodes) were pierced with sharp forceps, and lymph nodes were digested in 1 mg/ml collagenase D in HBSS with calcium and magnesium for 30 min at 37°C. Lungs were minced and digested in 1 mg/ml collagenase IV in RPMI supplemented with 10% FCS for 30 min at 37°C. Ears were collected and separated into dorsal and ventral halves. Each half was placed with the dermis layer down in digestion media (RPMI supplemented with 250 μg/ml Liberase TL and 125 μg/ml DNase I) for 90 min at 37°C. Halfway through the digestion, the ear skin was minced with scissors. Digested spleen, bone marrow and lung tissues were subjected to red blood cell (RBC) lysis.

Flow Cytometry and Cell Sorting.

Single-cell suspensions were blocked with TruStain FcX anti-mouse CD16/CD32 (BioLegend) and subsequently stained with fluorophore-conjugated or biotinylated antibodies (Data File S13). Intracellular staining was performed by fixing and permeabilizing with the eBioscience Foxp3/Transcription Factor Staining Set. EdU positive cells were stained with Click-iT Plus EdU AlexaFluor 488 Flow Cytometry Assay Kit according to the manufacturer protocol. eBioscience Fixable Viability Dye eFluor 506 was used to exclude dead cells in some experiments. Samples were acquired on Attune NxT using Attune NxT software and data analyzed with FlowJo software v10 (Tree Star). Cell sorting was performed on FACSAria II (BD Biosciences) at the Cytometry & Cell Sorting Laboratory at NYUGSoM. Flow cytometry was performed on sorted samples to confirm purity > 95% for populations of interest.

FL-BMDC cultures.

Primary mouse bone marrow cells were plated at a density of 2×10⁶ cells per well in 2 ml of DMEM supplemented with 10% FCS, 10% B16-Flt3L supernatant, 1% L-glutamine, 1% sodium pyruvate, 1% MEM-NEAA, 1% penicillin-streptomycin, and 55 μM 2-mercaptoethanol (DC medium) in 24-well plates. Cells were cultured at 37°C for 8 days without replating and collected by scraping the bottom of the well. To induce recombination of the floxed Smc3 allele in Rosa26^CreER/+ Smc3^fl/fl cells, 4-hydroxytamoxifen (4-OHT) was spiked into cultures at day 3 of differentiation to a final concentration of 500 nM.

For Notch-driven FL-BMDC cultures, on differentiation day 2, OP9-DL1 cells were treated with 10 μg/ml mitomycin C for 2 h, harvested, washed three times, and 9×10⁴ cells plated in 1 ml OP9 medium in 24-well plates. On differentiation day 3, differentiating bone marrow cells were harvested and 1 ml added per well of OP9-DL1 stromal cells, from which OP9 media was aspirated just prior. Subsequently, 500 nM 4-OHT was spiked into the cultures. On differentiation day 8, harvested cultures were passed through 70 μm filters to exclude OP9-DL1 stromal cells.

In Vitro Stimulation.

For assays testing Smc3^Δ or Il12b^ΔCTCF DC function, FL-BMDC cultures or splenic DCs (+/− enrichment/sort purification for cDC1s) were stimulated with the following ligands: 100 ng/ml profilin, 1 μg/ml LPS, 10 μg/ml poly(I:C) (Invivogen), 1 μM CpG-A/ODN2216 (Invivogen), 1 μM CpG-B (Invivogen), or left unstimulated as a control in DC medium. As detailed in the corresponding figure legends, supernatants or cells were collected at various timepoints after stimulation for analysis of cytokine production, RNA-Seq, RT-qPCR or CUT&RUN.

ELISA.

Serum or cell supernatants were collected for analysis of IL-12p40, IL-12p70 and IL-6 production by uncoated ELISA kits (Thermo Fisher Scientific) or for simultaneous analysis of multiple cytokines by LEGENDplex Anti-Virus Response panel according to the manufacturer protocol.

Generation of Irf8^Δ HoxB8-FL Cells.

Cas9 ribonucleoproteins (RNPs) were prepared by incubating 250 pmol IRF8 sgRNA, 250 pmol CD45 sgRNA, 75 pmol S. pyogenes Cas9 Nuclease V3, and 100 pmol Alt-R Cas9 electroporation enhancer (all from IDT) together for 15 min at room-temperature. Control Cas9-RNPs were prepared containing 250 pmol non-targeting control crRNA and 250 pmol tracrRNA instead of IRF8 sgRNA. HoxB8-FL cells (1×10⁶) were incubated in 100 ul Opti-MEM I reduced serum medium containing Cas9-RNPs and electroporated with 4D-Nucleofector X Unit (Lonza) using program CM-137. At day 6 post transfection and passaging, HoxB8-FL progenitors were differentiated into mature DCs. Successful transfection was confirmed by loss of CD45 expression.

Hi-C.

Sort-purified cells (~1–5×10⁶) were cross-linked for 10 min in 1 ml of 2% methanol-free formaldehyde in PBS + 3% BSA. Crosslinking was reversed by addition of glycine to 0.125 M for 5 min, followed by incubation on ice for 15 min. Libraries were prepared using the Arima Hi-C kit according to manufacturer protocol and paired-end 100 bp sequencing of libraries was performed on Illumina NovaSeq 6000.

ChIP-seq.

Sort-purified cells (~3.5–4.5×10⁶) were cross-linked with 1% methanol-free formaldehyde in PBS for 8 min followed by quenching with 0.125 M glycine for 5 min. Cells were lysed and nuclei isolated using the Covaris TruCHIP Kit, and chromatin was sheared using a Covaris M220 ultrasonicator. ChIP was performed using 4 μg of rabbit anti-CTCF polyclonal antibody (Millipore) or rabbit anti-mouse/human Rad21 polyclonal antibody (Abcam) and Dynabeads Protein A. Libraries were prepared using the KAPA HyperPrep Kit (Roche). As a control, libraries were also prepared from non-immunoprecipitated “input” DNA from each cell type. Paired-end 50 bp sequencing was performed on Illumina NovaSeq 6000 at the Genome Technology Center at NYUGSoM.

CUT&RUN.

CUT&RUN was performed on ~1.4–3×10⁵ sort-purified cells using the CUTANA ChIC/CUT&RUN Kit (Epicypher) according to manufacturer protocol. Briefly, cells were permeabilized with buffer containing 0.05% digitonin. After cells were bound to activated Concanavalin A (ConA)-coated beads, cells were bound with 4 μg of the following antibodies (or 0.5 μg for H3K4me3), which was used to guide cleavage by pAG-MNase: anti-histone H3K4me3 mixed monoclonal (Epicypher), anti-histone H3K4me1 polyclonal (Abcam), anti-histone H3K9me3 polyclonal (Abcam), anti-histone H3K27Ac polyclonal (Abcam), anti-histone H3K27me3 polyclonal (Millipore) and rabbit IgG isotype control polyclonal (Abcam). Libraries were prepared with NEBNext Ultra II Library Prep Kit. Paired-end 50 bp sequencing was performed on Illumina NovaSeq 6000 at the Genome Technology Center at NYUGSoM.

CITE-Seq.

The “hashtagging” functionality of the CITE-Seq protocol was used as described (80). Spleens were isolated from control and Smc3^ΔDC mice, either untreated or 6 hours after treatment with 100 ng profilin and 2.5 μg LPS. Spleens from each genotype and treatment group (n=2 per condition) were pooled, and single cell suspensions were stained with fluorophore-conjugated antibodies as well as TotalSeqB hashtag antibodies specific against mouse CD45 and MHC-I (BioLegend) to maintain sample identity (B0301 = Control untreated, B0302 = Control LPS+profilin, B0303 = Smc3^ΔDC untreated, B0304 = Smc3^ΔDC LPS+profilin. Successful cDC activation was confirmed by flow cytometric analysis of CD86 upregulation. 6,500 (B0301–0303) or 4,000 (B0304) cDC1s (Live singlets IgD⁻ TCR-β⁻ NK1.1⁻ CD11c⁺ MHC-II⁺ XCR1⁺ SIRPα⁻) and 8,500 cDC2s (Live singlets IgD⁻ TCR-β⁻ NK1.1⁻ CD11c⁺ MHC-II⁺ XCR1⁻ SIRPα⁺) from each sample were sorted and pooled together in 50 μl PBS containing 10% FBS. The resulting cell mixture was prepared and sequenced according to the 10x Genomics protocol.

Statistical analysis.

For graphs, data are shown as mean overlaid with values from individual replicates. Statistical differences were evaluated using a non-parametric Student’s t test with Mann-Whitney analysis. Three groups were compared with one-way ANOVA followed by Tukey’s test. Statistical differences in mice with < 250 mm³ tumor volume were determined with the log-rank (Mantel-Cox) test. P values less than or equal to 0.05 were considered significant, and significance was assigned according to the following breakdown: *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. GraphPad Prism software v8 and v10 were used for all graphing and statistical calculations except for gene expression and chromatin analysis, for which R was used. The number of experiment repetitions are indicated in the respective figure legends.

Supplementary Material

Supplementary Text

Figs. S1 to S13

References (82–97)

Data file S1 to S14

MDAR Reproducibility Checklist

Data, code, and materials availability:

All sequencing datasets have been deposited in the Gene Expression Omnibus repository GSE271298. Tabulated data underlying the figures is provided in Data File S14. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. The code used for data analysis is publicly available (81). Irf8^ΔTB and Il12b^ΔCTCF mice are available from B.R under a material transfer agreement with NYUGSoM. All other materials used in the manuscript are available from the sources listed in the Materials and Methods and data file S13.