DNMT1 deficiency impacts on plasmacytoid dendritic cells in homeostasis and autoimmune disease

Abstract

Dendritic cells (DCs) are heterogeneous immune regulators involved in autoimmune diseases. Epigenomic mechanisms orchestrating DC development and DC subset diversification remain insufficiently understood but could be important to modulate DC fate for clinical purposes. By combining whole-genome methylation assessment with the analysis of mice expressing reduced DNA methyltransferase 1 levels, we show that distinct DNA methylation levels and patterns are required for the development of plasmacytoid (pDC) and conventional DC subsets. We provide clonal in vivo evidence for DC lineage establishment at the stem cell level, and we show that a high DNA methylation threshold level is essential for Flt3-dependent survival of DC precursors. Importantly, reducing methylation predominantly depletes pDC and ameliorates systemic lupus erythematosus in an autoimmunity mouse model. This study shows how DNA methylation regulates the production of DC subsets and provides a potential rationale for targeting autoimmune disease using hypomethylating agents.

Introduction

Dendritic cells (DCs) have important functions in antigen presentation and cytokine secretion, and exist in different subsets of which conventional DCs (cDCs) and plasmacytoid DCs (pDC) predominate within lymphoid tissue.(1–3) While cDCs develop via the myeloid branch of hematopoiesis,(4–7) pDC are generated via the lymphoid system, but can also develop together with myeloid cells via macrophage-DC progenitors (MDP) and common DC progenitors (CDPs).(8–15)

DC differentiation is orchestrated by a network of transcription factors,(16) which generally establish gene expression in concert with the epigenome. One major layer of epigenomic regulation is DNA methylation, which is catalyzed by DNA methyltransferases (DNMTs).(17) DC maturation has been reported to be associated with demethylation at genomic regions that are predicted binding sites of DC-affiliated transcription factors.(18) However, this and other previous studies were restricted to only a fraction of the genome, were performed using an in vitro system, did not discriminate between DC subsets and did not include methylation changes in early DC precursors, in which DC fate is established.(19) Moreover, they did not address causality between methylation alterations and DC fate.

DCs are involved in immunological diseases, one of which is systemic lupus erythematosus (SLE), an autoimmune disorder in which patients present with a broad range of clinical features, ranging from mild symptoms up to life-threatening organ damage.(20) Changed pDC frequencies and pDC overactivation characterize SLE patients, and ablation of pDC could impair SLE in mice.(21–23) Hence, pDC represent an attractive target in SLE therapy, but therapeutic options for targeting pDC remain largely unknown.

Herein, we have investigated a possible causal role of DNA methylation in DC development during homeostasis and in SLE.

Materials & Methods

Mice

Dnmt1 ^c, Dnmt1 ^chip, Dnmt1 ^lox, Cx3cr1 ^Cre and Vav ^Cre mice were described.(24–28) All animal experiments were approved by the local authorities according to the German Federal Animal Protection Act. B6.SJL-PtprcaPepcb/BoyCrl and 129S1/SvImJ mice were crossed to obtain 129/SJL F1 offspring.

Flow cytometry

Flow cytometry was performed with 8-12 week old mice unless stated otherwise. Unspecific Fc binding was blocked by CD16/32 antibody. Detailed information on fluorochrome-conjugated antibodies are availbale on request. Lin^- cells were identified with a lineage ‘cocktail’ of antibodies to B220, CD4, CD8α, CD3ε, CD19, CD11c, Ter119, NK1.1 and Ly-6G. Discrimination of dead cells was performed by addition of 7-aminoactinomycin D (7-AAD; BioLegend). Fluorescence intensity was measured with Canto II or Aria III (BD) FACS cytometers equipped with FACSDiva. Data analysis was performed with FlowJo.

Generation and analysis of BM chimeras

5x10⁶ BM cells from 8-12 week old mice (CD45.1^-CD45.2⁺ 129S1/SvlmJ background) were transplanted into lethally irradiated (2x7 Gy) 129/SJL F1 recipient mice (CD45.1⁺/CD45.2⁺). 10-14 weeks after transplantation, donor cells were recovered and analyzed by flow cytometry.

Transplantation of single HSCs

Transplantation of single HSCs was performed as described.(29) In brief, single CD45.2⁺Lin^-Sca-1⁺Kit⁺(LSK)CD34^-CD150⁺CD48^- Vwf ^- tdTomato⁺ HSCs were flow-sorted from BM cell suspensions of Vwf-tdTomato/Gata1-eGFP (CD45.1^-CD45.2⁺) B6 adult mice and were mixed with 200,000 wildtype CD45.1⁺CD45.2^- SJL BM competitor cells before intravenous injection into lethally irradiated (10 Gy) CD45.1⁺CD45.2^- SJL recipient mice.

Retroviral constructs, viral supernatant production and cell transduction

MSCV retroviral constructs expressing IRES-GFP or murine Flt3-IRES-GFP (gift from Stephen L. Nutt) and production of retroviral supernatants were described.(30, 31) Transduced GFP⁺ cells were flow-sorted and cultured in pDC medium (RPMI 1640, 10% FCS, Pen/Str, 1mM Sodium Pyruvate, β-Mercaptoethanol (1:1000), 100ng/ml FLT3 ligand) for 8 days.(32)

Transcriptome analysis

Total RNA was extracted using the RNeasy Micro kit (Qiagen) and gDNA was digested with RNase-free DNAse. Sample processing and hybridization on Affymetrix GeneChIPs MG 430 PM was carried out according to standard procedures. Data was analyzed with the Bioconductor package version v.3.5.1 along with R version 3.4.0. Raw expression intensity values were normalized using the ‘Robust Multi-Array Average (RMA)’ function. Differentially expressed genes were identified by LIMMA. The gene expression raw-data are accessible in Gene Expression Omnibus (GSE163295, password: ijareeqczvihrgl).

Tagmentation-based whole-genome bisulfite sequencing (TWGBS)

For DNA methylation analysis of the indicated progenitor populations, cells were isolated from 8-12 week old C57BL/6J mice by fluorescence-activated cell sorting. The following cell populations were isolated from bone marrow cells: MDP (Lin^-cKit^hiCD115⁺Flt3⁺), CDPs (Lin^-cKit^loCD115⁺Flt3⁺), cMoP (Lin^-cKit^hiCD115⁺Flt3^-CD11b^-Ly6C⁺) and monocytes (CD11b⁺Ly6C⁺B220^-CD5^-CD8α^-Ter119^-SiglecF^-Ly6G^-FceRI^-). Monocytes were flow-sorted from total bone marrow. MDP, CDPs and cMoP were sorted from lineage-negative cells. Lineage depletion of mature hematopoietic cell lineages was performed using Dynabeads (Dynabeads Sheep anti-Rat IgG magnetic beads, Thermo Fischer) and rat IgG isotype antibodies to B220, CD4, CD8α, CD19 and Ter119, expanded by antibodies to CD11b and Gr-1 for isolation of MDP and CDP or Ly-6G for cMoP isolation. pDC (PDCA⁺CD11c^int), cDC1 (CD11c^hiMHCII⁺CD8α⁺CD11b^-) and cDC2 (CD11c^hiMHCII⁺CD8α⁺CD11b^-) were isolated from spleen suspensions. Per biological replicate, we sorted 20,000 monocytes (n=3, each probe was pooled from BM of 2 or 3 independent animals), 10,000 MDP (n=2, each probe was pooled from BM of 7 and 8 independent animals), 20,000-25,000 CDP (n=4, each probe was pooled from BM of 2 or 3 independent animals), 15,000-20,000 cMoP cells (n=2, and each probe is pooled from BM of 2 or 3 independent animals), 10,000 cDC1, 20,000 cCD2 and 20,000 pDC (n=3, and each probe was pooled from spleens of 3 independent animals). All sorted cell populations were snap-frozen as dry pellets and stored at -80 °C.

Genomic DNA (10-30 ng) was isolated using the QIAamp DNA Micro Kit (Qiagen) and used as input to generate sequencing libraries by tagmentation-based whole-genome bisulfite sequencing (TWGBS) as described previously.(33) To reduce PCR duplicates, four independent sequencing libraries were prepared per replicate and library amplification was monitored by real-time quantitative PCR. Pools of four libraries were sequenced on three separate lanes using the 125 bp paired-end mode on the HiSeq 2000 platform (Illumina). Sequencing was performed by the Genomics and Proteomics Core Facility at the German Cancer Research Center. TWGBS raw data as well as methylation calls in BED format can be accessed in Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE164124) password: yfqzwiyuhnkjtmt). TWGBS data from murine HSCs were taken from a previous study (GSE52709).(34)

TWGBS sequence alignment and methylation calling

Read alignment was performed by the Omics IT and Data Management Core Facility (ODCF) at the German Cancer Research Center using an updated version of the pipeline published by (33), which was implemented as a Roddy Workflow (https://github.com/DKFZ-ODCF/AlignmentAndQCWorkflows) in the automated ‘One Touch Pipeline’ (OTP).(35) Briefly, adaptor sequences of raw reads were trimmed using ‘Trimmomatic’.(36) Sequencing reads were then in silico bisulfite-converted (C>T for the first read in the pair, G>A for the second). The software package ‘BWA-MEM’ (https://arxiv.org/abs/1303.3997) was used with default parameters to align the converted reads to the in silico bisulfite-converted reference mm10 genome extended with the PhiX and lambda phage sequences. After alignment, reads were converted back to their original state. PCR duplicate removal was performed per library using ‘Picard MarkDuplicates’ before merging reads from all libraries per replicate. The alignment quality was validated by computing the mapping rates using ‘samtools flagstat’, (37) insert size distributions, and genome coverage statistics.

Methylation calling and M-bias trimming were performed using the bistro software package (version 0.2.0) (https://github.com/stephenkraemer/bistro). Automatic M-bias removal was performed using the ‘binomp’ algorithm. This approach removes the gap repair nucleotides introduced by the tagmentation reaction from both reads (first 9 bp following sequencing primer 2) and automatically detects and removes additional read positions with M-bias for each individual sample. Reads with a mapping quality ≥25 and nucleotides with a Phred-scaled quality score ≥25 were considered for further analysis. Bisulfite conversion rates were estimated using the autosomal CHH methylation levels.

Detection of differentially methylated regions

Differentially methylated regions (DMRs) were called pairwise between HSCs and each of the other cell populations using DSS. (38) Differential methylation at each CpG site was tested without smoothing. DMRs were defined by i) a minimal DMR length of 50 bp ii) a minimal number of 3 CpGs iii) a minimum fraction of differentially methylated CpGs of 50% (posterior p-value for a methylation delta greater or equal than 10% < 0.01). Within each pairwise comparison, DMRs separated by less than 50 bp were joined. For integrated analyses, DMRs from all pairwise comparisons were combined (union of the genomic intervals).

PCA and clustering

Principal component analysis (PCA) was performed on all DMRs based on the average DNA methylation levels. DMRs were clustered using hierarchical clustering (Ward’s method) of the z-score normalized average DNA methylation levels. Partitioning was performed with the cutreeHybrid algorithm of the dynamicTreeCut package (39) using R version 3.6.2 and the following parameters: deepSplit=1, minClusterSize=0.005 * number of DMRs, pamStage=False and minGap=0.1.

DMR gene annotation and gene set enrichment analysis

DMR gene annotation was performed by the gtfanno software package (https://github.com/stephenkraemer/gtfanno). Gene set enrichment analysis in DMRs was performed using regionset_profiler (https://github.com/stephenkraemer/regionset_profiler). Only promoter DMRs were considered and gene-set membership for each DMR was determined based on its gene annotation. Promoters were defined as regions 5000 bp upstream and 1000 bp downstream of the TSS. Enrichment of gene sets in individual DMR clusters against the background of all other DMRs was tested using Fisher’s exact test. Gene-sets were obtained from Supplemental Table 4.

Induction of systemic lupus erythematosus

Systemic lupus erythematosus (SLE) was induced by treating mice (at balanced gender ratio) with pristane (2,6,10,14-tetramethylpentadecane; Sigma-Aldrich) as described.(22) In brief, the animals received a single intraperitoneal pristane injection (0.5ml each) and were monitored by regular weight control and urine as well as blood sampling. The experiment was terminated after 4 months.

Histology

Tissue samples were fixed in 4% formalin, embedded in paraffin and cut into 3μm thick sections, which were stained with hematoxylin and eosin (HE) or periodic acid-Schiff reaction (PAS). Blinded histopathological scoring was done by a pathologist to evaluate the morphological changes according to ref. (40).

EM

Kidney tissue was sliced into 2³ mm blocks, were fixed in 2.5% glutaraldehyde overnight and were washed in water. Samples were treated with osmium tetroxide for 1h, dehydrated in an ascending alcohol series, and infiltrated with epon using a mixture of propylene oxide and epon (1:2) for 1.5h. After embedding in pure epon the samples were kept at 60°C for 36h, ultrathin slices of 60nm were cut and were contrasted with uranyl acetate and lead citrate. Ultrastructural images were taken with a Philips EM 208S transmission electron microscope at various magnifications.

Immunohistochemistry

Tissue was fixed in 4% formalin and dehydrated in a tissue processing machine (Leica ASP300, Leica) overnight. After paraffin embedding, tissue was cut into 3μm sections and mounted onto Superfrost objective slides. For initial deparaffinization, the slides were incubated at 80°C for 1h. The slides were then deparaffinized in xylene and cooked in citrate buffer for 40min (Target Retrieval Solution; DAKO). 5% bovine serum albumin (Albumin Fraction V; Roth) plus 0.5% Triton X 100 (Roth) in PBS was used for blocking for 1h. Goat anti-mouse IgM (Invitrogen) and donkey anti-mouse IgG (Invitrogen) antibodies were incubated in 5% bovine serum albumin plus 0.5% Triton X 100 overnight at +4°C, followed by three washes with PBS. DAPI (1:10,000; Boehringer) was added for 30min. After washing with PBS, coverslips were mounted using Mowiol solution. Images were taken with a BZ-9000 fluorescence microscope (BioRevo, Keyence). FIJI, a distribution of ImageJ, v1.52p, was used for the automated measurement of the signal intensity of the immunohistochemical reactions.

B-cell proliferation assay

B-cell proliferation was determined according to ref. (41). Briefly, 1×10⁶ cells/ml splenocytes from pristane-treated mice were loaded with 1μM CFSE (Life Technologies) in PBS/1% FCS for 20min at 37°C and were blocked by washing with IMDM/20%FCS. 2x10⁶ cells/ml were then cultured in IMDM/20%FCS under stimulation with 0.35μM CpG-B ODN2006 (Hycult Biotech) for 5 days. B-cell proliferation was evaluated by tracking dilution of CFSE within the B220⁺CD19⁺ flow-gate.

Statistical analysis

Statistical analysis was conducted using Prism software (GraphPad) or custom python scripts as indicated. The two-tailed Student’s t-test was used for statistical analysis of two-group comparisons. P<0.05 was considered significant.

Results

DNMT1 levels determine DC differentiation fates

Taking advantage of mice with hypomorphic DNMT1 expression (Dnmt1 ^c/chip mice),²⁴ we showed previously that high DNA methylation levels are essential for the generation of lymphocytes but not myeloid cells.(42) In the current study, we confirmed reduced B-cell frequencies and revealed normal monocyte/macrophage and slightly increased granulocyte frequencies in Dnmt1 ^c/chip mice (Fig. S1a-c). Dnmt1 ^c/chip mice are smaller than Dnmt1 ^+/+ controls and thus have overall reduced cell numbers.(24) As in ref. (42), we therefore display cell differences between both genotypes as frequencies of total BM or (when noted) LSK cells. Because it was unknown whether DNA methylation plays a causal role in DC development, we investigated DC subsets in Dnmt1 ^c/chip mice. Strikingly, pDC were almost completely absent in Dnmt1 ^c/chip animals (Fig. 1a and Fig. S1d-f). In contrast, cDCs were reduced to a lesser degree, caused by lower cDC2 frequencies, whereas cDC1 frequencies were slightly increased (Fig. 1b,c).

Figure 1. DC development in Dnmt1 ^c/chip mice.

a-d, Flow cytometry analysis of (a) pDC (PDCA⁺CD11c^int) in BM, of (b) cDCs (MHCII⁺CD11c^hi) and (c) cDC subsets (MHCII⁺CD11c^hiCD11b⁺CD8^- and MHCII⁺CD11c^hiCD11b^-CD8⁺) in spleens, and of (d) macrophage-DC progenitors (MDP: Lin^-CD117^hiCD115⁺Ly-6C^-CD11b^-) and common monocyte progenitors (cMoP: Lin^-CD117^hiCD115⁺Ly-6C^hiCD11b^-) in BM of Dnmt1 ^c/chip mice and Dnmt1 ^+/+ control littermates. Representative density plots are shown on the left (numbers indicate percentage of cells within the gates) and summaries of the analyzed cohorts are shown on the right, representing data from at least 3 to 4 independent experiments. Each symbol represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). Cell frequencies are indicated as percentage of living cells in a-c, or as percentage of Lin^- cells in d. n was 8 mice per genotype in a, 7 in b and c, and 15 in d. Normal MHCII levels on CD11c^- cells and normal CD11c levels on residual cDCs excluded that these markers were generally downregulated in Dnmt1 ^c/chip mice.

At the progenitor level, Dnmt1 ^c/chip mice demonstrated much lower MDP frequencies, but less affected common monocyte progenitors (cMoP), which directly develop from MDP (Fig. 1d).(43) CDPs could not be analyzed as Dnmt1 ^c/chip mice lacked Flt3 expression (see below). For this reason, we have also omitted FLT3 from gating MDP as in ref. (44).

To determine if the alterations in DC development were caused by lineage-intrinsic hypomethylation, we generated BM chimeric animals. Although Dnmt1 ^c/chip BM cells have reduced engraftment potential, they can fully repopulate recipients when transplanted in high numbers (Fig. S1g).(42) Dnmt1 ^c/chip donor cells failed almost completely to produce pDC and MDP (Fig. 2a,b). They produced also much fewer cDCs than the controls, which was in contrast to the relatively mild cDC phenotype in primary Dnmt1 ^c/chip mice (Fig. 2c). This difference suggested that under the enhanced pressure to maintain methylation during cell cycle, transplanted Dnmt1 ^c/chip progenitors had a more pronounced deficiency to replenish cDCs. In contrast, Dnmt1 ^c/chip donor cells reconstituted normal or even slightly more granulocytes (Fig. S1h), confirming their competence to replenish mature hematopietic lineages.

Figure 2. Impaired DC development in Dnmt1 ^c/chip mice is cell intrinsic.

a-c, Flow cytometry analysis of donor cell-derived (a) BM pDC (CD45.1^-CD45.2⁺PDCA⁺CD11c^int), (b) BM macrophage-DC progenitors (MDP: CD45.1^-CD45.2⁺Lin^-CD117^hiCD115⁺Ly-6C^-CD11b^-) and common monocyte progenitors (cMoP: CD45.1^-CD45.2⁺Lin^-CD117^hiCD115⁺Ly-6C^hiCD11b^-) and (c) splenic cDCs (CD45.1^-CD45.2⁺MHCII⁺CD11c^hi) of BM chimeras that had received 5x10⁶ BM cells of Dnmt1 ^+/+ (129S1/SvlmJ) or Dnmt1 ^c/chip donor mice 10-14 weeks before. Representative density plots of CD45.1^-CD45.2⁺ donor population gated cells are shown on the left (numbers indicate percentage of cells within the gates) and summaries of the analyzed cohorts are shown on the right, representing data from 2 independent experiments. Each symbol represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). Cell frequencies are indicated as percentage of CD45.1^-CD45.2⁺ donor-derived living cells in a and b, or as percentage of CD45.1^-CD45.2⁺ donor-derived Lin^- cells in c. n was 8 mice per genotype in a and b, and 7 in c.

Collectively, Dnmt1 ^c/chip mice revealed that development of pDC and cDCs requires different cell-intrinsic DNA methylation levels.

High DNA methylation levels are required for early pDC development

To assess at which developmental phase methylation is involved in DC fate control, we crossed Dnmt1 ^lox/chip mice (combining a loxP-flanked with the hypomorphic chip Dnmt1 allele) (24, 28) with mice expressing CRE either from Vav (Vav ^Cre+) or Cx3cr1 (Cx3cr1 ^Cre+) genes. Vav ^Cre+ mice express CRE in hematopoietic stem cells (HSCs) and their progeny, generating hematopoiesis-specific DNMT1 stem cell mutants. In contrast, CRE expression in Cx3cr1 ^Cre+ mice is restricted to mononuclear phagocytes including DCs and DC progenitors (refs (26, 27) and data shown below). Although Cx3cr1 ^Cre+ Dnmt1 ^lox/chip mice demonstrated efficient excision of the floxed Dnmt1 allele in pDC, cDCs and CDPs (Fig. S2a), pDC and cDC frequencies were normal, suggesting that both subsets were not affected by hypomethylation in lineage-committed progenitors (Fig. 3a).

Figure 3. Differentiation of HSCs/early progenitors into DC-committed progenitors requires high DNA methylation levels.

(a) Cx3cr1 ^Cre+ Dnmt1 ^lox/chip and Cx3cr1 ^Cre- Dnmt1 ^lox/chip or (b,c) Vav ^Cre+ Dnmt1 ^+/+, Vav ^Cre- Dnmt1 ^lox/chip and Vav ^Cre+ Dnmt1 ^lox/chip mice as determined by flow cytometry. pDC (PDCA⁺CD11c^int), monocytes (Lin^-CD117^-CD115⁺), MDP (Lin^- CD117^hiCD115⁺Ly-6C^-CD11b^-) and cMoP (Lin^-CD117^hiCD115⁺Ly-6C^hiCD11b^-) were measured in BM, cDCs (MHCII⁺CD11c^hi), B cells (B220⁺CD19⁺), T cells (CD3⁺) and granulocytes (CD11b⁺Gr1⁺) in spleens. (a-c) Each symbol represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). Cell frequencies are indicated as percentage of living cells in a and b, or as percentage of Lin^- cells in c. The figures represent data from 2 to 3 independent experiments.

In contrast, Vav ^Cre+ Dnmt1 ^lox/chip mice (which had an excised floxed Dnmt1 allele in HSCs) showed almost complete loss of pDC, strongly reduced MDP and a slighter reduction in cDCs (Fig. 3b,c and S2b,c). They had normal granulocyte, monocyte and cMoP frequencies and did not show significant reductions in T and B cells, which was in contrast to Dnmt1 ^c/chip mice. The reason for this difference is still unclear, but a potential explanation could be that in contrast to Dnmt1 ^c/chip HSCs, Vav ^Cre+ Dnmt1 ^lox/chip HSCs developed from normally methylated ancestors and as such underwent delayed hypomethylation.

Collectively, conditional DNMT1 hypomorphic mice confirmed that pDC/cDC subset establishment requires different cell-intrinsic methylation levels, and demonstrated that the critical phase for this requirement is in early hematopoiesis.

DCs develop from HSCs with different lineage restrictions

To obtain deeper insight into DC generation in early hematopoiesis, we tracked DC replenishment by single HSCs with different lineage outputs in vivo. We could recently classify HSCs into four different categories: multilineage HSCs or HSCs with platelet (P), platelet-erythrocyte-myeloid (PEM) or platelet-erythrocyte-myeloid-B-lymphoid (PEM-B) lineage restrictions.(29) We purified CD45.2⁺Lin^-Sca-1⁺Kit⁺(LSK)CD34^-CD150⁺CD48^- Vwf ^- tdTomato⁺ cells (HSCs) from the BM of mice expressing tdTomato from the von Willebrand factor gene (Vwf) promoter and eGFP from the Gata1 gene promoter to detect, in addition to other lineages, donor-derived erythrocytes (eGFP+) and platelets (tdTomato+ eGFP+) lacking the CD45 antigen. We transplanted single cells of these along with CD45.1⁺ BM cells for radioprotection into BM-ablated congenic CD45.1⁺ mice, controlled engraftment and identified the lineage potential of the transplanted single HSCs by regular peripheral blood analysis over a period of up to 42 weeks (data not shown). As in ref. (29), we defined lineage potential by a minimum donor contribution of 0.01% to the respective hematopoietic lineage in the recipient animals. After 32-42 weeks, we determined the contribution of the single HSCs to pDC and cDC development in spleens by a minimum of 0.01% donor-derived cells, as with the other lineages (Fig. S2d,e). Multilineage HSCs and PEM-B HSCs contributed to both pDC and cDC development, while P HSCs lacked DC potential (Fig. 4a-c). Notably, PEM HSCs gave rise to both pDC and cDCs as well, at frequencies similar to myeloid progeny, suggesting that DC progeny of both subsets can develop independently of the lymphoid pathway (Fig. 4d). This notion was further supported by the observation that the frequencies by which PEM-B HSCs produced pDC and cDCs correlated better with their myeloid than with B-cell potential (mean frequencies: pDC 11%, cDC 14%, myeloid 19%, B-cell 0.04%). A similar relationship between DC and myeloid in vivo output was noted with multilineage HSCs (Fig. S2f).

Figure 4. DC development from single HSCs.

In vivo contribution of single CD45.2⁺Lin^- Sca-1⁺Kit⁺(LSK)CD34^-CD150⁺CD48^- Vwf ^- tdTomato⁺ donor HSCs with either multipotent reconstitution capacity (a) or with different lineage restrictions (b-d) to DC progeny in spleens of BM-ablated CD45.1⁺ recipient mice. The plots summarize the donor-derived frequencies of n = 4 mice with multilineage HSCs, n = 4 with platelet-restricted HSCs, n = 3 with PEM-restricted HSCs and n = 5 with PEM-B-restricted HSCs generated in 4 independent experiments and presented in logarithmic scale. Purple and red dots indicate pDC or cDCs, respectively.

These results provide clonal evidence that DCs can be replenished via different lineages. Importantly, they suggest that the decision by which lineage individual DCs are produced is made at the HSC level.

DNA methylation analysis reveals DC subset specific patterns and supports a high methylation threshold for pDC

To investigate how DNA methylation impacts DC development, we measured genome-wide methylation in pDC, cDC1, cDC2, CDPs, MDP, monocytes and cMoP by tagmentation-based whole-genome bisulfite sequencing (TWGBS).(33) We analyzed 2 to 4 biological replicates each, resulting on average in 1.14x10⁹ total mapped paired reads per group (Supplemental Table 4). Mean genomic CpG-coverages of all populations are in Supplemental Table 4. We also included TWGBS data of HSCs, which we had previously generated by the same protocol.(34) Unsupervised analysis of DNA methylation in ENSEMBL regulatory regions confirmed that all samples clustered according to cell type (Fig. S3a).

We identified differentially methylated regions (DMRs) by pairwise comparison of HSCs and each of the downstream cell populations (cutoff ≥10% methylation difference). We pooled DMRs from all comparisons and merged overlapping DMRs, leading to a catalogue of 49,588 unique DMRs (Supplemental Table 4). Principal component analysis (PCA) based on these DMRs revealed that ontogenetic DC/myeloid trajectories were linked to continuously diverging methylation patterns (Fig. 5a). Unsupervised hierarchical clustering identified 9 unique DMR clusters across all cell populations (Fig. 5b and Fig. S3b). For example, cluster 9 consisted of DMRs with lower methylation in HSCs than in any of the other populations and, thus, represented a ‘HSC DMR cluster’. In contrast, cluster 3 (which was among the largest clusters, Fig. S3c) had lower methylation in DCs than in the other populations and, thus, constituted a ‘panDC DMR cluster’. TWGBS also revealed DC-subset-specific DMR clusters (clusters 5, 7 and 8) and a ‘monocyte DMR cluster’ (cluster 6) which had lower methylation in monocytes and cMoP than in any other cell population. Importantly, for each of the clusters activated (i.e. hypomethylated) DMRs could be annotated to genes known to be functionally linked with the respective cell types (examples are given in Fig. 5b,c and S3d,e).

Figure 5. Whole-genome DNA methylation data of DC and monocyte lineages.

(a) Principal component analysis of DMR methylation in the indicated populations. PC1 and PC2 depict principal components 1 and 2. Percentage values at axes indicate sample variation explained by the respective principal component. Each condition was analyzed from 2 to 4 independent experiments and each dot represents a biological replicate. b) Heatmap of the union of all DMRs (n = 49,588) detected between HSCs and each of the other indicated populations based on unsupervised hierarchical clustering of z-scores. Depicted are 500 randomly sampled DMRs per cluster. Each horizontal dash represents a DMR. Data show the average methylation of aggregated biological replicates. (c) Locus plot depicting the genomic region surrounding the transcription start site of Tlr9. DNA methylation tracks are displayed for HSC (grey), MDP (brown), CDP (yellow), cMoP (orange) monocytes (red) cDC2 (light green), cDC1 (green) and for pDC (purple) where each bar represents a CpG dinucleotide and the height of the bar represents the DNA methylation level. The numbers in the DMR track identify the clusters as labelled in panel (b). (d) Violin plot showing the distribution of DNA methylation levels across all DMRs in pDC, cDC1 and cDC2. Dots within the violins represent medians, the box heights represent the interquartile range (IQR) and the whiskers indicate 1.5x IQR. One way ANOVA was used to compare the mean methylation levels per population on a replicate basis (F=826.005, p=4.739*10-8). Post-hoc pairwise comparisons using Tukey’s method were conducted and p-values are shown.

The DC subsets revealed significant differences in DMR methylation levels, with pDC, cDC2 and cDC1 harboring the highest, intermediate and lowest methylation levels, respectively (Fig. 5d). Hence, these differences reflected the differential DC subset phenotype of Dnmt1 ^c/chip mice, and thus provided a possible explanation for the preferential ablation of pDC upon hypomorphic DNMT1 levels.

DNA methylation dynamics correlate with gene expression alterations in DC progenitors

We next flow-sorted MDP and cMoP for genome-wide gene expression profiling (Fig. S4a). In comparison to Dnmt1^+/+ MDP, Dnmt1 ^c/chip MDP showed 471 and 765 genes with increased or decreased expression, respectively (Fig. 6a, Supplemental Table 4). Dnmt1 ^c/chip cMoP showed fewer differentially expressed genes when compared to Dnmt1 ^+/+ cMoP: 118 and 179 genes with increased or decreased expression, respectively (Fig. S4b, Supplemental Table 4). To identify genes with known functions in DCs or myeloid cells, we compared the differentially expressed genes with DC, monocyte and granulocyte signature gene lists.(45) This comparison revealed that Dnmt1 ^c/chip MDP did not undergo global dysregulation of DC genes, but demonstrated a selective decrease in expression of several important DC genes, including those encoding DC transcription factors (e.g., Irf8, Runx2) and growth factor receptors (e.g., Flt3, Il7r, CD27) (Fig. 6b, Supplemental Table 4) and data not shown). Of note, the progenitors expressed DC signature genes much stronger than total BM cells did (examples shown in Fig. S4c), indicating their robust transcription in Dnmt1 ^+/+ progenitors and specific decrease in the Dnmt1 ^c/chip cells. In contrast, we noted that 90 genes with functions in granulocytes were higher expressed in Dnmt1 ^c/chip MDP (e.g., Gfi1, Elane, Ltf, Ccr1).

Figure 6. DNA methylation and gene expression in DC progenitors.

(a) Volcano plot showing a comparison between the transcriptomes of MDP (CD45.1^-CD45.2⁺Lin^- CD117^hiCD115⁺Ly-6C^-CD11b^-) isolated from Dnmt1 ^+/+ or Dnmt1 ^c/chip mice (n = 4 biologically independent samples each from at least 2 different experiments). Differentially expressed genes (cutoff criteria: log2 fold change difference ≥ 0.58 or -0.58 and p-value ≤ 0.05) are represented by black dots. A positive fold change indicates transcripts with decreased expression in Dnmt1 ^c/chip MDP and a negative fold change indicates genes with increased expression. Dnmt1 and Flt3 transcripts are highlighted. (b) Heatmap of DC signature genes (genes marked by blue bar on the left) with decreased expression or of granulocyte signature genes (genes marked by red bar on the left) with increased expression in Dnmt1 ^c/chip versus Dnmt1 ^+/+ MDP, respectively. Gene signatures were taken from (45). Color code on the right represents RMA normalized log2 expression levels. (c) Heatmap summarizing enrichment (red) or depletion (green) of up- or downregulated genes in Dnmt1 ^c/chip MDP or cMoP compared to their control counterparts in the DMR clusters shown in Fig. 3b. For enrichment analysis, only DMRs overlapping with gene promoter regions were considered. The color scheme represents ‘sign(log odds ratio) * log10 (p-values)’. (d) Summary of percentage of LSK stem/progenitor cells from Dnmt1 ^+/+ or Dnmt1 ^c/chip mice (n = 7 each) expressing FLT3 protein on their surface as determined by flow cytometry. Data are from 2 independent experiments. (e, f) Total cells (e) or pDC (f) that developed from FLT3 or empty vector reconstituted c-kit-enriched BM cells of Dnmt1 ^c/chip orDnmt1 ^+/+ mice cultured with FLT3 ligand. The graph summarizes 3 independent experiments with a total of n = 9 Dnmt1^+/+ and n = 10 Dnmt1 ^c/chip mice. Each symbol in d-f represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). (g) Representative flow cytometry images showing pDC (PDCA⁺CD11c⁺) frequencies developing from FLT3 (left) or empty vector (right) reconstituted c-kit-enriched Dnmt1 ^c/chip BM cells cultured with FLT3 ligand.

We mapped the promoter regions of the differently expressed genes to the clustered DMRs. This revealed that genes downregulated in Dnmt1 ^c/chip MDP were enriched in the ‘panDC DMR cluster 3’ and to a lesser extent in DC-specific clusters 5 (‘pDC DMR cluster’) and 7 (‘cDC DMR cluster’), describing a link between lineage-affiliated methylation changes and altered gene expression in Dnmt1 ^c/chip progenitors (Fig. 6c). These genes comprised those with known functions in DCs and their precursors, such as Flt3, Cxcr2, IL1r1 and Vcam1 (Supplemental Table 4). In contrast, genes upregulated in Dnmt1 ^c/chip cells were enriched in the ‘monocyte DMR cluster 6’ and in the ‘HSC cluster 9’.

Taken together, these data indicate that Dnmt1 ^c/chip progenitors failed to express specific DC genes, and were likely unable to methylate and shut down HSC and myeloid genes.

Restoring FLT3 signaling rescues Dnmt1 ^c/chip progenitor cell survival

Because Dnmt1 ^c/chip cells failed to express Flt3, we assessed the contribution of FLT3 signaling to the DC phenotype (Fig. 6d and Fig. S4d). To this end, we retrovirally re-expressed Flt3 in c-kit-enriched Dnmt1 ^c/chip BM progenitors and cultured the cells with FLT3 ligand for in vitro pDC production.(32) Indeed, restoration of FLT3 signaling rescued Dnmt1 ^c/chip progenitor survival to levels comparable to those of Dnmt1 ^+/+ cells (Fig. 6e, S4e). However, it failed to robustly rescue pDC maturation (Fig. 6f,g). Importantly, this result fitted well to our data from the conditional DNMT1 hypomorphic mice, showing that hypomethylation disrupted early DC production from stem/progenitor cells but was dispensable for subsequent maturation.

DNA hypomethylation attenuates systemic lupus erythematosus

Systemic lupus erythematosus (SLE) is an autoimmune disease for which pDC are important.(46) Therefore, we explored a potential therapeutic option for SLE by hypomethylation induction in a SLE mouse model based on pristane administration.(22, 47) We induced SLE in Vav ^Cre+ Dnmt1 ^lox/chip and Vav ^Cre+ Dnmt1 ^+/+ mice by injecting pristane and monitored them regularly during an incubation time of 4 months, and at the end of the experiment probed them for clinical and pathological signs of SLE as in ref. (22). We confirmed that Vav ^Cre+ Dnmt1 ^lox/chip mice lacked pDC but maintained normal B-cell frequencies after the pristane stimulus (Fig. 7a,b). We calculated the combined clinical and pathological SLE scores (called ‘SLE score’) as a summary of both clinical and pathological symptoms (described in Supplemental Table 1). To exclude mice with ordinary infections, we identified SLE by a score of ≥2. All Vav ^Cre+ Dnmt1 ^+/+ mice (12 of 12, 100%) but only 4 out of 12 of the Vav ^Cre+ Dnmt1 ^lox/chip mice (33%) developed SLE (mean total score of 5 for Vav ^Cre+ Dnmt1 ^+/+ versus 1.5 for Vav ^Cre+ Dnmt1 ^lox/chip; Fig. 7c, Supplemental Table 1).

Figure 7. Impaired SLE development by pDC elimination upon induced hypomethylation.

(a,b) Frequency of (a) pDC (PDCA⁺CD11c^int) and (b) B cells (B220⁺CD19⁺) in spleens of pristane-treated Vav ^Cre+ Dnmt1 ^+/+ and Vav ^Cre+ Dnmt1 ^lox/chip mice. Each symbol represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). Data are from 2 to 3 independent experiments. (c,d) Total clinical and pathological (c) and renal (d) scores demonstrating significantly impaired SLE development in pristane-treated Vav ^Cre+ Dnmt1 ^lox/chip mice as compared to Vav ^Cre+ Dnmt1 ^+/+ controls (n = 12 each, split over three independent experiments). Each symbol represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). (e) Light microscopic images of hematoxylin and eosin (HE) stained and periodic acid-Schiff (PAS) reacted (upper two panels) and electron-microscopic (EM) images of kidney sections from pristane-treated Vav ^Cre+ Dnmt1 ^+/+ and Vav ^Cre+ Dnmt1 ^lox/chip mice. Vav ^Cre+ Dnmt1 ^+/+ mice demonstrated increased glomerular inflammation, mesangial matrix and cellularity (black arrow), increased presentation of crescent formation (blue arrow) and small necrosis (yellow arrow). In EM, all types of deposits (subepithelial (green arrow), subendothelial (red arrow), paramesangial and mesangial) could be identified in Vav ^Cre+ Dnmt1 ^+/+ animals, while in the Vav ^Cre+ Dnmt1 ^lox/chip group neither subepithelial nor subendothelial and only few mesangial and paramesangial deposits were detected. The images are of single animals from 2 independent experiments. (f) Left: Representative immunofluorescence images of immune complexes by staining for IgG (top images) and IgM (middle images) of kidney sections from pristane-treated Vav ^Cre+ Dnmt1 ^+/+ and Vav ^Cre+ Dnmt1 ^lox/chip mice. The bottom images show IgG (red) and IgM (green) combined, together with DAPI (blue) nuclear staining. Per kidney, five randomly selected windows with a comparable distance to the tissue border (1000 x 1000 pixels, equivalent to 1065.5 x 1065.5 μm) were assessed. Background and noise were reduced by excluding extremely weak signal (threshold 26). Right: Quantification of IgG and IgM immunofluorescence intensities (as immunohistochemistry (IHC) positivity per mm²). Each symbol represents an individual mouse from 2 independent experiments; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). (g) Quantification of total IgG, IgM and IgA concentrations in the sera of pristane-treated Vav ^Cre+ Dnmt1 ^+/+ and Vav ^Cre+ Dnmt1 ^lox/chip mice. Each symbol represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). (h) Total splenocytes of pristane-treated Vav ^Cre+ Dnmt1 ^+/+ and Vav ^Cre+ Dnmt1 ^lox/chip mice were labeled with CFSE and then stimulated with CpG for 5 days in vitro. Proliferation of flow-gated B cells (CD19⁺B220⁺) was evaluated by tracking the dilution of the fluorescent dye CFSE. Each symbol represents an individual mouse; small horizontal lines indicate the mean ± s.d. (unpaired t-test, two-tailed). Data represent 2 independent experiments. (i) Representative flow histogram illustrating the dilution of CFSE in B-cell-gated (CD19⁺B220⁺) splenocytes of pristane-treated Vav ^Cre+ Dnmt1 ^+/+ and Vav ^Cre+ Dnmt1 ^lox/chip mice after CpG-stimulation for 5 days in vitro.

Renal involvement in SLE is associated with increased morbidity and mortality.(48) We therefore calculated a renal score by histopathological analysis of the kidneys as described,⁽⁴⁰⁾ also demonstrating significant differences between both genotypes (mean total score of 3.5 for Vav ^Cre+ Dnmt1 ^+/+ versus 1.5 for Vav ^Cre+ Dnmt1 ^lox/chip) (Fig. 7d, Supplemental Table 2). Figure 7e displays representative examples showing differences in glomerular inflammation, proliferation, crescent formation and necrosis (HE and PAS) and in the immune deposits (electron microscopic analysis) between both genotypes. We found immune deposits in 7 out of 8 (87%) Vav ^Cre+ Dnmt1 ^+/+ but only in 3 out of 6 (50%) Vav ^Cre+ Dnmt1 ^lox/chip mice. Moreover, those Vav ^Cre+ Dnmt1 ^lox/chip animals that succumbed to SLE developed a weaker disease than the controls: While the controls showed all types of deposits (i.e., subepithelial, subendothelial, paramesangial and mesangial), no subepithelial or subendothelial deposits and only few paramesangial and mesangial deposits were detected in the Vav ^Cre+ Dnmt1 ^lox/chip mutants (Supplemental Table 3).

We next quantified IgG and IgM immune complexes in the kidneys of the pristane-treated mice, detecting more complexes of both antibody classes in Vav ^Cre+ Dnmt1 ^+/+ than Vav ^Cre+ Dnmt1 ^lox/chip mice (Fig. 7f). Moreover, the pristane-treated Vav ^Cre+ Dnmt1 ^lox/chip animals had lower concentrations of total IgG, IgM and IgA than the equally treated controls (Fig. 7g). Finally, B-cells from pristane-treated Vav ^Cre+ Dnmt1 ^lox/chip mice proliferated slower upon in vitro CpG stimulation than the controls (Fig. 7h,i). Together, these results showed that although B-cells of Vav ^Cre+ Dnmt1 ^lox/chip mice were capable of antibody production and class switching, their autoimmune activation was impaired.

In summary, disease modeling showed that hypomethylation can ameliorate SLE by pDC ablation and attenuation of B-cell functions.

Discussion

Here, we have demonstrated that Dnmt1 ^c/chip progenitors have an almost complete disability to produce pDC, but their ability to produce cDCs was less affected. In contrast to previous data,(18) this finding established a causal link between methylation dynamics in progenitors and the programming of DC subset fates. Indeed, the differences in the hypomethylation-driven effects were also evident by graded differences in the genomic methylation levels between both DC subsets at differentially methylated regions, which was significantly higher in pDC than in cDCs and which involved regulatory elements of DC-relevant genes. Notably, the present study together with our previous data revealed that pDC and lymphoid cells share a tight requirement for high methylation levels.(42) However, Vav ^Cre+ Dnmt1 ^lox/chip mice highlighted also differences between pDC and lymphocytes, as they, like Dnmt1 ^c/chip mice, lacked pDC but unlike Dnmt1 ^c/chip mice, maintained normal lymphocyte numbers. Differences in pDC and lymphocytes were also suggested by clonal lineage tracing of individual HSCs in vivo, revealing that pDC can develop via different lineage trajectories including a trajectory lacking lymphoid replenishment. Interestingly, a previous barcoding study demonstrated that DCs separate from other hematopoietic lineages at lymphoid-primed multipotent progenitors stage.(19) Our study suggests that a split between myeloid and lymphoid DC trajectories at the stem cell level precedes this separation from other lineages.

DNMT1 strongly prefers hemimethylated DNA as substrate, and therefore is considered as the canonical maintenance methylating enzyme. Indeed, recent results with embryonal stem cells have confirmed that DNMT1 has little de novo methylating activity in vivo, with the exception of retrotransposable elements.(49) Congruently, Dnmt1 ^c/chip HSCs are likely capable to maintain their methylome landscape after self-renewal divisions, but are unable to faithfully maintain these patterns upon increased proliferation during differentiation. As a result, differentiation routes that are highly dependent on silencing alternative differentiation programs, like e.g. those towards B cells and pDCs, may be affected the most.(50) Similar to DNMT1, the de novo methyltransferases DNMT3a and DNMT3b were also shown to control differentiation.(51) Hence, although there are no specific data on DC development, it is likely that these enzymes affect the generation of DCs as well. However, the underlying mechanism and affected genes may be different from those in Dnmt1 ^c/chip HSCs, because DNMT3a was found to control specific features of the stem cell methylome (namely the preservation of borders separating highly from lowly methylated chromosomal regions) rather than having gross genome-spanning effects.(52)

Epigenetic modifications are reversible and, as such, can potentially be modulated for disease treatment. Indeed, DNA hypomethylating agents such as azacitidine and decitabine (both acting as DNMT inhibitors) are currently applied clinically in cancer therapy.(53) However, methylation changes described in SLE patients mainly comprised methylation loss, in particular at genes encoding cytokines and interleukins whose enhanced expression is thought to contribute to SLE pathophysiology.(54–57) Hence, the effect of hypomethylation on SLE has been unclear, and, in the light of prior data, it was assumed that hypomethylation-based therapy might be unsuitable to cure SLE. However, previous studies have measured methylation mainly at few gene loci. In contrast, our data considered genome-wide hypomethylation effects, demonstrating that reducing DNMT1 ablated pDC and impaired B-cell autoimmune activity, which ameliorated SLE in mice. Hypomethylating agents may have similar effects in patients and, as such, could represent a hitherto unappreciated option for SLE therapy.

Key points.

1. Myeloid-restricted hematopoietic stem cells can replenish pDC and cDC in vivo

2. DC diversification is associated with differences in genomic methylation levels

3. Reducing Dnmt1 activity ameliorates systemic lupus erythematosus autoimmunity