Activation of Notch and Myc Signaling via B-cell–Restricted Depletion of Dnmt3a Generates a Consistent Murine Model of Chronic Lymphocytic Leukemia

Recent large-scale cancer sequencing studies performed over the past decade have revealed a myriad of candidate genetic lesions that drive malignancies, however, the functional effects of these newly somatic mutations have remained largely unknown (1, 2). For chronic lymphocytic leukemia (CLL), growing efforts to address this challenge have yielded the recent generation of in vivo models engineered to recapitulate the B-cell–restricted expression of novel putative genetic drivers. These studies have confirmed the CLL-driving roles of alterations such as mutated SF3B1 in conjunction with ATM deletion, and hotspot mutations in IKZF3, and altogether highlight the common and diverse mechanisms underlying CLL pathogenesis [i.e., dysregulation of B-cell receptor (BCR) signaling; refs. 3, 4].

In addition to genetic lesions, recent studies have further highlighted the role of epigenetic aberrations in cancer development (5, 6). DNA methylation is one of the most well studied epigenetic abnormalities in human cancers and dysregulation of DNA methylation can lead to genome instability and the silencing of tumor suppressor genes (7, 8). In line with this notion, previous studies have revealed that CLL is characterized by prominent locally disordered methylation disorder (9, 10). These characteristic changes in DNA methylation suggest a critical role in the onset and progression of CLL (10–12). In mammalian cells, DNA methylation is dynamically established by the DNA methyltransferase 3 (DNMT3) family of de novo methyltransferases DNMT3A and DNMT3B and sustained by the maintenance methyltransferase DNMT1 (13). Previous studies suggested that alterations in DNA methylation/demethylation intermediates were closely linked to clinical outcomes in CLL (14).

Although mutations of DNMT3A are rarely observed in CLL, we demonstrate herein that low DNMT3A expression is associated with more aggressive disease through the analysis of two independent CLL cohorts. To explore the hypothesis that loss of DNMT3A expression functionally contributes to CLL pathogenesis, we engineered a conditional knockout (KO) mouse model with B-cell–restricted deletion of Dnmt3a through intercrossing with the CD19-Cre recombinase (Cre) system. Strikingly, homozygous Dnmt3a depletion in B cells resulted in the development of CLL with 100% penetrance at a median age of onset of 5.3 months, and heterozygous Dnmt3a depletion yielded a disease penetrance of 89% with a median onset at 18.5 months, confirming its role as a haplo‐insufficient tumor suppressor. Distinct from other existing CLL models tested (i.e., Sf3b1-Atm, Ikzf3, and MDR; refs. 3, 4, 15), we observed selective activation of Notch and Myc signaling, but not BCR signaling, and sensitivity to Notch inhibition, supporting this line as a Notch-driven mouse model of CLL.

All animals were housed at Dana-Farber Cancer Institute (DFCI). All animal procedures were completed in accordance with the Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committees at DFCI (protocol no.: 07–068; 12–004). Dnmt3a-floxed mice were gifted from Dr. Ross Levine [Memorial Sloan Kettering Cancer Center (MSKCC); ref. 16]. Cd19-Cre^+/− mice were obtained from The Jackson Laboratory. Dnmt3a-floxed mice were intercrossed with Cd19-Cre mice to generate Cd19-Cre Dnmt3a-floxed C57BL/6J mice.

Heparinized blood samples were obtained from healthy donors (HD) and patients enrolled on clinical research protocols with informed consent. The study was conducted in accordance with the ethical guidelines approved by the Human Subjects Protection Committees of DFCI; by the University of California San Diego (UCSD); by the NIH National Heart, Lung, and Blood Institute; and by the Department of Internal Medicine III at Ulm University, Germany. Peripheral blood mononuclear cells (PBMC) from normal donors and patients with CLL were isolated by Ficoll/Hypaque density gradient centrifugation. CD19⁺ B cells from normal volunteers were isolated by immunomagnetic selection (Miltenyi Biotec). PBMCs were used fresh or cryopreserved in FBS with 10% DMSO and stored in vapor-phase liquid nitrogen until the time of analysis.

To monitor disease progression, 50 μL of blood was collected from mice via submandibular bleeds into EDTA-containing tubes. Erythrocyte lysis was achieved by incubating the blood in 1 mL of ACK buffer for 5 minutes and then washing with PBS supplemented with 2% FCS and 2 mmol/L EDTA. The cells were then stained with a cocktail of antibodies as described in Supplementary Materials and Methods for 20 minutes at 4°C. Cells were then washed and analyzed (BD LSRFortessa cell analyzer). For flow cytometric evaluation of specific cell surface markers, erythrocyte lysed single-cell suspensions prepared from spleen, bone marrow (BM), and peritoneal cavity were washed with PBS supplemented with 2% FCS and 2 mmol/L EDTA and then incubated with antibodies for 20 minutes. Splenic marginal and follicular B-cell populations were stained with a panel of antibodies as described in Supplementary Materials and Methods. All analyses of splenic and BM B cells were performed using mice of 8 to 12 weeks old, while that of B cells from the peritoneal cavity were from mice of 4 to 6 weeks of age. To analyze germinal center generation, mice of 8 to 12 weeks of age were immunized intraperitoneally with 10⁹ sheep red blood cells (Immunoresearch) in 150 μL of PBS. Spleens were collected 10 days after immunization and analyzed by flow cytometry to quantify germinal center B cells using a panel of antibodies as described in Supplementary Materials and Methods. To assess proliferation of B1a cells, cells isolated from the peritoneal cavity of 8- to 12-week-old Cd19-Cre^+/− or Cd19-Cre^+/−Dnmt3a^fl/fl animals, were stained with the CellTrace CFSE Cell Proliferation Kit (Thermo Fisher Scientific) according to the manufacturer's protocol and injected intraperitoneally into irradiated immunodeficient recipients [10-week-old NOD/SCID gamma (NSG) mice]. Eleven days following injection, recipient mice were euthanized, and peritoneal fluid was collected and a panel of antibodies as described in Supplementary Materials and Methods. CD11b⁺ B1a cells were analyzed for carboxyfluorescein diacetate succinimidyl ester (CFSE) intensity by flow cytometry.

For IHC staining, spleens were fixed in 10% buffered neutral formalin overnight followed by 70% ethanol until the tissues were processed. Spleens were then paraffin-embedded and sectioned into 20 μm for staining. Bones were fixed in 10% buffered neutral formalin overnight followed by 2 hours of 70% ethanol until the tissues were processed. Hematoxylin and eosin staining, as well as PAX5 and CD5 immunostaining were performed on the sections using standard procedures. MYC IHC was performed on the Leica Bond III automated staining platform using the Leica Biosystems.

Transfer of HSC

Seven- to twelve-week-old female or male CD45.2⁺ Cd19-Cre^+/−Dnmt3a^fl/fl mice were used as donors. Femurs, tibias, hips, and spines were isolated from donor mice, crushed, and ACK-lysed. Cells were enriched with anti-CD117 microbeads (Miltenyi Biotec) and then sorted to collect LSK (lineage⁻ Sca-1⁺ c-Kit⁺) cells using a panel of antibodies as described in Supplementary Materials and Methods.

Transfer of B1a cells

Six- to eight-week-old female or male Cd19-Cre^+/− or Cd19-Cre^+/−Dnmt3a^fl/fl mice were used as donors. B1a cells from the peritoneal cavity were sorted and injected intraperitoneally into irradiated Cd45.1⁺ recipients, cells from of 1 donor mouse (50,000–100,000 cells) per 1 recipient mouse for total of 5 donors per genotype.

Transplantation of CLL cells

Five million splenocytes from Cd19-Cre^+/−Dnmt3a^fl/fl were resuspended in 100 microliters PBS and injected intravenously into 8- to 15-week-old NSG mice. Overall, CLL cells from 5 different donors were transplanted into recipient mice. Disease burden in the peripheral blood, spleen, and BM was evaluated by flow cytometry and IHC.

For analysis of BCR signaling transduction, purified splenic B cells (3 × 10⁶) or splenic CLL cells were stimulated in 1 mL RPMI 1640 medium (Life Technologies) supplemented with 10% FCS, and 1% penicillin/streptomycin in the presence of 10 μg/mL goat anti-mouse IgM (Southern Biotech) for 5 or 15 minutes at 37°C. Cells were harvested, washed twice with cold PBS, and lysed for 30 minutes on ice with RIPA buffer (Boston Bioproducts) supplemented with a protease inhibitor mix (1 tablet in 10 mL RIPA buffer; Roche) and PhosSTOP (Sigma Aldrich) according to manufacturer's protocol. Cell debris was removed by centrifugation at 10,000 per g for 15 minutes, followed by protein quantification using Pierce BCA protein assay kit (Thermo Scientific). Cell extracts were fractionated by SDS-PAGE and transferred to Immun-Blot PVDF or nitrocellulose membranes (Biorad). Membranes were stained overnight at 4°C with primary antibodies and appropriate secondary antibodies, e.g., horseradish peroxidase–coupled goat anti-rabbit and goat anti-mouse antibodies (Rockland Immunochemicals Inc). The bands were visualized by using a Biorad ChemiDoc touch imaging system. Antibodies are described in Supplementary Materials and Methods. Western blot quantification was performed using imageJ (RRID:SCR_003070).

BH3 profiling was performed as described previously (17). Experiments were performed using B cells from the spleen and peritoneal cavity of 2- to 3-month-old Cd19-Cre^+/− or Cd19-Cre^+/−Dnmt3a^fl/fl, immunomagnetically isolated (B-cell Isolation Kit, Miltenyi Biotec). B cells from the peritoneal cavity were pooled together to increase total cell number. Cells were suspended in MEB2 buffer (150 mmol/L mannitol, 10 mmol/L HEPES-KOH pH 7.5, 150 mmol/L KCI, 1 mmol/L EGTA, 1 mmol/L EDTA, 0.1% BSA, 5 mmol/L succinate). Cells were added to a 384-well plate and incubated at 25°C for 60 minutes combined with different BH3 peptides in 0.001% digitonin for permeabilization. Cell were fixed with 4% formaldehyde for 10 minutes at room temperature followed by neutralization by adding N2 buffer (1.7 M Tris base, 1.25 M glycine, pH 9.1). Cells were stained overnight at 4°C with Hoechst 33342 (H3570, Invitrogen) and anti–cytochrome c–Alexa Fluor 488 (6H2.B4/612308) and analyzed using an Intellicyt iQue flow cytometer to determine the rate of loss of cytochrome c in response to each BH3 peptide. Splenic and peritoneal cavity cells were also stained with CD19, CD11b, and CD5 antibodies and gated for B2 cells (CD19⁺) or CD11b⁺ B1a cells (CD19⁺;CD5⁺;CD11b⁺). Assays were generally conducted in triplicate. DMSO and alamethicin were used as negative and positive controls for cytochrome c release.

RNA extraction was performed on sorted splenic B cells or peritoneal cavity B1a cells from young, disease-free 4- to 6-week-old mice and from CLL cells isolated from the spleens of diseased mice and sorted to collect B220⁺CD5⁺IgK⁺ cells. Total RNA was extracted using Qiagen RNAeasy kit or Qiagen RNAeasy micro kit with on-column DNase treatment (Qiagen).

CLL cells from different genetic CLL mouse models (3, 4, 15) were resuspended in RPMI 1640 medium supplemented with 10% FBS and cultured in 96-well tissue culture plates (50,000 cells/100 μL). Cells were treated with indicated concentrations of the BCL2 inhibitor ABT-199, BTK inhibitor ibrutinib, Notch signaling inhibitor daptomycin (DAPT), and Wnt signaling inhibitor LGK-974. After incubation for 24 hours at 37°C with 5% CO₂, cell viability was measured by CellTiter-Glo Luminescent Assay and normalized by cells treated with the DMSO vehicle control. Human primary CLL cells were plated at 2 million/mL in complete RPMI and exposed to 50 mmol/L DAPT, 1 nmol/L venetoclax or 1 μmol/L ibrutinib overnight, followed by CellTiter-Glo assessment of cellular viability, assessed in technical triplicates. OP9 and OP9-DLL1 were kindly provided by J.C. Aster (Brigham and Women's Hospital), were seeded in 48-well plates (250K stromal cells per well) 16 hours prior to incubation with samples, and then drug treatments were performed as above.

Transplantation studies were performed on 8- to 12-week-old immunodeficient recipient NSG mice using viably cryopreserved splenocytes from Dnmt3a^fl/fl CLL animals. Five million CLL cells/recipient were resuspended in 100 μL PBS and injected intravenously into NSG. Two weeks after transplant, CLL burden in the peripheral blood of the transplanted recipients (NSGs) was analyzed by flow cytometry and the treatment was initiated. Vehicle control (4% DMSO in corn oil) or 30 mg/kg DAPT was administered daily by i.p. injections for 14 days on a 5 days on–2 days off schedule. Disease burden in the peripheral blood was evaluated by flow cytometry.

RNA was assessed for quality and quantity (RNA 6000 Nano LabChip kit on a 2100 Bioanalyzer; Agilent). RNA libraries were generated using NEBNext Single Cell/Low Input RNA Library Prep Kit for Illumina (New England BioLabs) according to manufacturer's protocol and checked for quality and quantified using the DNA-1000 kit (Agilent) on a 2100 Bioanalyzer. Libraries were sequenced with the Illumina Sequencing SP kit on two lanes to obtain 150-base paired-end (PE) reads. RNA sequencing (RNA-seq) reads were aligned to the mouse reference genome mm10 using STAR (v2.4.0.1; RRID:SCR_004463; ref. 18) and Transcripts Per Kilobase Million (TPM) value was used to measure gene expression. Differentially expressed genes were assessed using DESeq2 (RRID:SCR_000154; ref. 19). Transcription factor and pathway analyses were performed using Enrichr (RRID:SCR_001575) and the TRRUST transcription factor database and the MsigDB hallmark gene sets, respectively.

Total RNA was transcribed into cDNA using the SuperscriptII reverse transcription kit (Life Technologies). The expression of mRNAs was analyzed using the TaqMan Gene Expression Assays probes as described in Supplementary Materials and Methods. Target gene expression was normalized to the mean Ct values of the housekeeping gene Gapdh.

Genomic DNA was isolated using the QIAGEN DNeasy Blood & Tissue Kit (Qiagen). For immunoglobulin heavy-chain gene variable region analysis, DNA from CLL cells was subjected to PCR as previously described (20).

Initial DNA sample quality assessment, DNA library preparation, sequencing, and bioinformatics analysis were conducted at GENEWIZ, Inc. Genomic DNA samples were quantified using Qubit 2.0 Fluorometer (Life Technologies) and DNA integrity was checked with 0.6% agarose gel with 50- to 60-ng sample loaded in each well. SureSelect^XT Exome Enrichment System for Illumina Paired-End Multiplexed Sequencing Library and SureSelect Human All Exon V5 bait library were used for target enrichment DNA library preparation following the manufacturer's recommendations (Agilent) and the standard low-input protocol, which starts with 200 ng. Briefly, the genomic DNA was fragmented by acoustic shearing with a Covaris LE200 Focused Ultra-sonicator instrument. Fragmented DNAs were cleaned up and end-repaired, as well as adenylated at the 3′ends. Adapters were ligated to the DNA fragments, and adapter-ligated DNA fragments were enriched with limited cycle PCR. Adapter-ligated DNA fragments were validated using Agilent TapeStation (Agilent Technologies), and quantified using Qubit 2.0 Fluorometer. Seven hundred and fifty nanograms of adapter-ligated DNA fragments were hybridized with biotinylated RNA baits at 65°C for 24 hours. The hybrid DNAs were captured by streptavidin-coated magnetic beads. After extensive wash, the captured DNAs were amplified and indexed with Illumina indexing primers. Postcaptured DNA libraries were validated using Agilent TapeStation and quantified using Qubit 2.0 Fluorometer and Real-Time PCR (Applied Biosystems). Illumina reagents and kits for DNA library sequencing cluster generation and sequencing were used for enrichment DNA sequencing. Postcaptured DNA libraries were multiplexed in equal molar mass, and pooled DNA libraries were clustered on one lane of a flow cell, using the cBOT from Illumina. After clustering, the flow cell was loaded on the Illumina HiSeq instrument according to manufacturer's instructions. The samples were sequenced using a 2 × 150 PE configuration. Image analysis and base calling was conducted by the HiSeq Control Software (HCS 2.0) on the HiSeq instrument. Whole-exome sequencing raw FASTQ files were aligned to Genome Reference Consortium Mouse Build 38 (mm10) and the resulting BAM files were further processed by marking duplicated reads and recalibrating base qualities. To perform the copy number analysis, read coverage counts were calculated across the exome intervals for all samples. A copy-number variation (CNV) panel of normal (PoN) was created using all of the samples. The tumor read coverage data was normalized using the PoN. All of the analysis was performed following the recommendation of the GATK Best Practices Workflows using GATK4 toolkit (RRID:SCR_001876).

Reduced representation bisulfite sequencing (RRBS) libraries were generated from 50 ng input DNA using the Tecan Ovation RRBS Methyl-Seq System following the manufacturer's recommendation except that we performed scaled-down half reactions from restriction digest through final repair after adapter ligation. Libraries were PCR amplified for 12 cycles. PE 100-base reads were generated on Illumina HiSeq2500 rapid flow cells using custom primers (Tecan) without adding a PhiX spike-in.

Illumina sequencing reads were quality-controlled and adapter clipped using fastQC (RRID:SCR_014583) and cutadapt (21). These reads were aligned to the mouse mm9 reference genome using BSmap (22). In order to determine the methylation state of all CpGs captured and assess the bisulfite conversion rate, we used the mcall module in the MOABS software suite (RRID:SCR_012071) with standard parameter settings (23). Differentially methylated regions (DMR) were called using metilene version 0.2 to 8 (24). DMRs were defined to have an absolute minimum difference in methylation of 0.2 with a maximum distance of 300 nt between CpGs within a DMR and a minimum of 5 CpGs per DMR. Missing values were allowed per group. Adjusted P values were calculated using the -c 2 parameters, invoking Benjamini–Hochberg correction (FDR). For heatmap visualization, methylation levels of DMRs were calculated as the mean methylation of all CpGs with a DMR for each sample and plotted using the heatmap function of the R package pheatmap (RRID:SCR_016418). A DMR was associated with a gene if it overlapped or was in the promoter region [(−2500 nt to +500 nt of transcription start site (TSS)] of the respective gene.

Discordant read methylation was calculated from each read based on the fraction of transitions in methylation status (0 or 1) between adjacent CpGs. PE reads overlapping each other were cut in the middle of the overlap to avoid interpretation of the same molecule twice and reads covering less than 3 CpGs were excluded from this analysis. For each read the methylation status of the covered CpGs was determined as unmethylated (converted base) or methylated (unconverted base). Based on this, the number of transitions in methylation status along the read were calculated and normalized to the number of possible transitions.

The raw data described in this publication have been deposited in NCBI's Gene Expression Omnibus (GEO) under the accession numbers GSE169245, GSE143711, and GSE122668.

We reanalyzed transcriptome (RNA-seq) data generated from 107 human CLL samples and CD19⁺ B cells from 8 HDs (9) to evaluate the expression of DNMT3A, DNMT3B, and DNMT1. We observed robust expression of DNMT3A and DNMT1 in CLLs and HDs, whereas DNMT3B was poorly expressed (Supplementary Fig. S1A). Notably, DNMT3A expression was highly variable amongst these patients with CLL, and lower DNMT3A expression was associated with more aggressive disease with shorter failure-free survival (P = 0.013; Fig. 1A). We confirmed the high variability in DNMT3A expression in CLL in a separate set of 6 HDs and 17 CLLs (Supplementary Fig. S1B). We also validated the correlation between more aggressive disease and low DNMT3A expression in another independent cohort consisting of 127 patients with treatment-naïve CLL (Supplementary Fig. S1C; P = 0.02). Complementing this analysis, we examined DNMT3A protein expression in 25 CLLs and 6 HDs by Western blot and again found a high degree of expression variability but overall lower expression of DNMT3A in CLLs (Supplementary Fig. S1D–S1F; Supplementary Table S1; P < 0.001). Given the known key role of DNMT3A across various hematologic malignancies (25–27) and its association with more aggressive CLL that we observed, we sought to evaluate the potential impact for this lesion as a CLL driving event.

To investigate the role of DNMT3A depletion in vivo, we generated a B-cell–restricted Dnmt3a KO mouse model by intercrossing Cd19-Cre and Dnmt3a-LoxP mice (Fig. 1B), resulting in cohorts of wild-type (Cd19-Cre^+/−, WT), heterozygous KO (Cd19-Cre^+/− Dnmt3a^fl/+, referred as Dnmt3a^fl/+), and homozygous KO (Cd19-Cre^+/−Dnmt3a^fl/fl, referred as Dnmt3a^fl/fl) mice. Successful Dnmt3a KO was confirmed by Western blot (Fig. 1C), and different groups of mice were monitored for the development of CLL through monthly flow cytometric analysis of peripheral blood for the appearance of B220⁺CD5⁺ cells up to 24 months of age. Dnmt3a^fl/fl mice consistently developed CLL between 4 to 8 months of age (median onset of 5.3; n = 26) with 100% penetrance (Fig. 1D). In contrast, Dnmt3a^fl/+ mice developed CLL/MBL between 15 to 21 months of age (median of 18.5; n = 16) with 89% penetrance, while none of the Cd19-Cre^+/− control mice developed CLL. No malignancies other than CLL were observed in Dnmt3a KO mice over this time period, as confirmed by IHC. A subset of mice was further monitored for disease progression and survival, showing that CLL progressed rapidly in the Dnmt3a^fl/fl mice (n = 14), resulting in poorer survival than the Dnmt3a^fl/+ mice (n = 12; Fig. 1E and F).

Mice with CLL consistently displayed enlarged spleens, with infiltration of CLL cells in the peripheral blood, spleen, BM, and peritoneal cavity of affected mice (Fig. 1G). The presence of CLL cells in the BM and spleen was confirmed by colocalization of CD5 (red) and PAX5 (brown) expression within the tissue sections (Fig. 1H). We confirmed that the CLL cells were clonal based on Igκ expression and immunoglobulin heavy-chain gene variable region (IGHV) mutational status (20), which was found to be unmutated (i.e., 100% homology with germline) in all 5 samples tested (Supplementary Table S2). We further confirmed the transplantability of the CLL cells. Following intravenous injection of CLL cells into immunodeficient mice (NSG; n = 5), we consistently detected CLL cells in the peripheral blood of recipient mice 4 weeks following transplant. In each instance, upon euthanasia at 6 weeks following injection, these CLL cells were also detected in the spleen, BM, and peritoneal cavity by flow cytometry (Supplementary Fig. S1G). Altogether, we demonstrate that CD19-restricted deletion of Dnmt3a reliably results in CLL, confirming this lesion as a CLL driving event.

We examined the impact of Dnmt3a depletion on normal B-cell development in 2- to 3-month old preleukemic mice using previously established flow cytometry conditions (Supplementary Fig. S2A and S2B; refs. 3, 4). Splenocytes from Dnmt3a KO mice were reduced in a proportion of marginal zone cells (MZ; B220⁺;CD23⁻;CD21⁺; P < 0.001, Student t test) and increased in a proportion of follicular B cells (FO; B220⁺;CD23⁺;CD21⁻; P < 0.02, Student t test) compared with WT mice (Fig. 2A). In contrast, no proportion differences in BM B-cell subpopulations were observed (Supplementary Fig. S2C). Of note, splenic B cells from Dnmt3a KO mice exhibited no alteration in BCR signaling activity (Supplementary Fig. S2D) or immune response to sheep red blood cells (Supplementary Fig. S2E) compared with WT B cells, which differs from our previously reported CLL mouse models (3, 4).

We detected a substantially increased proportion of CD11b⁺ B1a cells (CD19⁺;CD43⁺;CD5⁺;CD11b⁺) in the peritoneal cavity of Dnmt3a^fl/fl mice (Fig. 2B), suggesting B1a cells as the cell of origin of CLL in this model system. Testing this hypothesis, we transplanted marrow hematopoietic stem cells (LSK cells; n = 5) or B1a cells (n = 5) isolated from Dnmt3a^fl/fl mice into irradiated CD45.1⁺ recipient mice. While no recipient mice transplanted with LSKs developed CLL up to 12 months following transplantation, clonal CD45.2⁺B220⁺CD5⁺Igk⁺ CLL cells were detectable in peripheral blood of recipient mice transplanted with B1a cells starting 10 weeks after transplantation (Fig. 2C). IHC staining of splenic and BM tissue sections confirmed the presence of CD5 and PAX5 double-positive CLL cells among the engrafted cells (Fig. 2D). Infiltration of CD45.2 CLL cells was evident in the peripheral blood, spleen, BM, and peritoneal cavity of affected mice (Supplementary Fig. S2F and S2G). Consistent with the parental model, CLL cells exhibited no obvious alteration in BCR signaling activity (Supplementary Fig. S2H). Together, these data support the notion that CLL cells in this model arise from the B1a population.

To determine the mechanism of Dnmt3a^fl/fl B1a cell expansion, we analyzed cell division of B1a cells from WT or Dnmt3a^fl/fl animals using CFSE staining. Peritoneal cells were isolated, labeled with CFSE, and injected intraperitoneally into irradiated immunodeficient recipients (NSG; n = 5). Cell division was assessed by measuring the corresponding decrease in cell fluorescence via flow cytometry. Since no difference (P = 0.72) in cellular proliferation between the 2 groups was observed (Fig. 2E), we asked whether Dnmt3a^fl/fl B1a expansion was caused by reduced cell death. Indeed, both splenic CD19⁺ B cells and peritoneal B1a cells from Dnmt3a^fl/fl mice showed reduced apoptotic priming, induced by BH3 peptides (i.e., BIM and PUMA), compared with WT cells (Fig. 2F). Dnmt3a depletion thus appears to affect B1a expansion through a reduction in cell death, rather than increased proliferation.

Given the known role of Dnmt3a as a de novo methyltransferase, we evaluated the impact of Dnmt3a depletion on global DNA methylation of preleukemic B1a cells (i.e., absence of circulating CLL cells at the time of analysis) and of B2 cells (as a comparator population), using RRBS. Although no difference in the average CGI methylation levels was observed between Dnmt3a^fl/fl versus WT B cells (n = 3 per group; Fig. 3A), increased local methylation disorder [measured by percent discordant reads (PDR)] was observed in CGI shelves and background regions (i.e., CpG sites that are not in CpG islands, shores, or shelves) in Dnmt3a^fl/fl B1a cells compared with WT B1a cells (Fig. 3B). We identified a set of DMRs (difference > 0.2; Supplementary Table S3), mostly hypomethylated, in Dnmt3a^fl/fl versus WT B1a cells (Fig. 3C and D). As representative examples, hypomethylated DMRs in the promoter regions of the Cxcr7, Wnt10b, and Hes5 genes are shown in Fig. 3E. In contrast, only minor changes in methylation were observed in Dnmt3a^fl/fl B2 cells. Notably, complete depletion of Dnmt3a was confirmed by RNA-seq in both Dnmt3a^fl/fl B1a and B2 cells (Supplementary Fig. S3A), suggesting that the methylation differences between the cell types could not be attributed to incomplete KO of Dnmt3a in B2 cells. As DNMTs might have overlapping functions to regulate DNA methylation, we quantified the expression of Dnmt3a, Dnmt3b, and Dnmt1 by real-time PCR in these cells. Our results revealed higher levels of Dnmt3b (P = 0.005) in B2 cells compared with B1a cells (Supplementary Fig. S3B), suggesting that Dnmt3b might functionally compensate for the loss of Dnmt3a in Dnmt3a^fl/fl B2 cells. We further confirmed that Dnmt3b1, the active isoform of Dnmt3b gene, was stably expressed in Dnmt3a^fl/fl B2 cells (Supplementary Fig. S3C). Pathway analysis revealed that genes associated with those DMRs were highly enriched in Notch, TNFα, TGF-β, and IL2/STAT5 signaling, and apoptosis (Supplementary Table S4).

Given the prominent hypomethylation changes observed in Dnmt3a depleted B1a cells, we investigated whether these would lead to altered gene transcript expression. Indeed, principal component analysis (PCA) of bulk transcriptomes generated from Dnmt3a^fl/fl and WT B1a, and B2 cells (n = 3 per group) revealed clear differences between Dnmt3a^fl/fl and WT B1a cells, but not between Dnmt3a^fl/fl and WT B2 cells, in line with the methylation changes (Fig. 3F). However, in Dnmt3a-depleted B1a cells, hypomethylation events were not necessarily associated with upregulation of gene expression (Supplementary Fig. S3D), suggesting that the methylation status per se was insufficient to determine gene expression, in line with prior findings (28). Nevertheless, we detected 460 downregulated and 168 upregulated genes in the Dnmt3a^fl/fl B1a cells compared with WT B1a cells [Fig. 3G, left; FDR < 0.05, fold change (FC) > 2; Supplementary Table S5], while the changes in B2 cells were negligible (Fig. 3G, right). Differentially expressed genes in Dnmt3a^fl/fl B1a were highly enriched for several oncogenic signaling pathways, including Notch, TNFα, JAK/STAT3, and Wnt-β catenin signaling (Fig. 3H; Supplementary Table S6).

Upon investigating the methylomes of the CLL cells arising in Dnmt3a^fl/fl animals compared with WT and Dnmt3a^fl/fl B1a cells, we observed a marked increase in CGI methylation rate in CLL cells (Fig. 4A). This was accompanied by increased PDR levels in background regions and CpG shelves, and decreased PDR levels in CpG islands (Fig. 4B). Notably, we identified multiple DMRs in CLL cells compared with either WT or Dnmt3a^fl/fl B1a cells (difference > 0.2; Fig. 4C–E; Supplementary Tables S7 and S8), a large fraction of which were hypermethylated. In the promoter regions of the CLL cells, we identified 1,514 hypermethylated and 247 hypomethylated DMRs compared with WT B1a cells, and 1,574 hypermethylated and 103 hypomethylated DMRs compared with Dnmt3a^fl/fl B1a cells. In gene body regions, we identified 855 hypermethylated and 1,088 hypomethylated DMRs compared with WT B1a cells, and 892 hypermethylated and 503 hypomethylated DMRs compared with Dnmt3a^fl/fl B1a cells (examples of the differentially methylated DMRs in the promoters shown in Fig. 4F). Of the 466 hypomethylated DMRs observed in the preleukemia Dnmt3a^fl/fl B1a cells (Supplementary Table S3), only 45 remained hypomethylated in Dnmt3a^fl/fl CLL cells compared with WT B1a (including Hes5), and the others remained unchanged (n = 383) or even hypermethylated (n = 38), suggesting a dramatic reprogramming of the methylation profile associated with progression to CLL. Genes with promoter and/or gene body DMRs in CLL compared with WT B1a cells were highly enriched for several oncogenic signaling pathways including Notch and Hedgehog signaling, epithelial–mesenchymal transition, and Wnt signaling (Supplementary Table S9).

RNA-seq analysis of Dnmt3a^fl/+ and Dnmt3a^fl/fl CLLs (n = 3 each) revealed that expression changes in Dnmt3a^fl/+ and Dnmt3a^fl/fl CLLs compared with WT B1a cells were highly concordant (Supplementary Fig. S3E; R = 0.72, P < 0.0001), suggesting a consistent transcriptional reprogramming during CLL development. We thus combined the 2 groups to identify commonly altered genes in CLL cells. We identified more upregulated (n = 2801) than downregulated (n = 1,244) genes in CLL cells compared with WT B1a cells (FDR < 0.05, FC > 2; Supplementary Table S10). Moreover, we identified 3,030 upregulated and 1,652 downregulated genes in CLL versus Dnmt3a^fl/fl B1a cells (Supplementary Table S11). To determine the signaling pathways affected in CLL cells, we performed an unbiased pathway enrichment analysis of these differentially expressed genes using MSigDB hallmark gene sets. This allowed us to identify several pathways that were commonly enriched in both comparisons; again, we observed enrichment of Notch signaling genes (Fig. 4G; Supplementary Table S12), pointing to its putative important role in CLL development.

Compared with methylome data of human CLLs (11), we identified 20 hypermethylated genes and 210 hypomethylated genes that are shared between human and mouse CLLs. The 444 DMRs of these genes in human CLLs are shown in Supplementary Table S13. These genes were highly enriched for TGF-β signaling, UV response, and mitotic spindle (all FDR < 0.005), supporting a role of DNA damage repair pathway in CLL development. To determine the gene signatures associated with poor DNMT3A expression in human CLL, we reanalyzed transcriptome profiles of 8 samples with the lowest DNMT3A expression in the cohort (Fig. 1A). Compared with normal B cells (n = 8), altogether there were 715 upregulated and 428 downregulated genes in these CLLs (FC > 2, FDR < 0.05). Of these genes, we identified 104 upregulated and 18 downregulated genes that overlapped with the Dnmt3a mouse CLL model (Supplementary Table S14). Pathway analysis revealed that these genes were highly enriched for apical junction, IL2/STAT5 signaling and complement (all FDR < 0.05).

Closer inspection of the role of the Notch signaling pathway in the Dnmt3a-depleted CLL models revealed a general upregulation of Notch signaling genes in CLL cells compared with either WT or Dnmt3a^fl/fl B1a cells (Fig. 5A). To investigate whether upregulation of Notch signaling genes was associated with abnormal activation of the signaling pathway, we examined expression changes of known Notch targets in these CLL cells (29). As a comparison, we also examined Notch target expression in three other CLL genetically engineered mouse models, namely those driven by MDR, Ikzf3, or Sf3b1-Atm lesions (Fig. 5B; refs. 3, 4, 15). The distinct features of these mouse models are summarized in Supplementary Fig. S3F. Indeed, we observed high variability of Notch target expression across the different CLL models, with selective upregulation of Notch targets exclusively observed in the Dnmt3a-depleted CLL model but not in others (Fig. 5C). The results remained robust even after we separated Dnmt3a^fl/+ CLLs from Dnmt3a^fl/fl CLLs, supporting a unique role of Notch activation in this model (Supplementary Fig. S3G). We were able to validate upregulation of several well established Notch targets (i.e., Hes5, Dtx1, and St3gal6) in CLL cells versus normal B cells (Fig. 5D).

To examine the therapeutic vulnerabilities of the diverse genetic CLL models to various drugs, we exposed CLL cells to the BCL2 inhibitor ABT-199 (venetoclax), BTK inhibitor ibrutinib, the Notch signaling inhibitor DAPT, the Wnt signaling inhibitor LGK-974, and DMSO. The optimal working concentration of different drugs had been determined by treating tumor cells with multiple doses (Supplementary Fig. S4A). CLL cells from the four distinct CLL models were treated with two concentrations of each drug for 24 hours, and percent survival was calculated as the ratio between the cellular viability following test drug exposure and DMSO. Dnmt3a-depleted CLLs were more sensitive to ABT than the other CLLs, consistent with the BH3 results suggesting that Dnmt3a-depleted CLLs were more dependent on BCL2 for survival (Supplementary Fig. S4B). Moreover, Dnmt3a-depleted CLL cells were highly sensitive to Notch inhibition by DAPT (Fig. 5E; Supplementary Fig. S4C), supporting hyperactivation of Notch signaling in these CLL cells and high dependency on this pathway for survival. In vivo, NSG mice transplanted with Dnmt3a^fl/fl CLLs and subsequently treated with DAPT (30 mg/kg; n = 5) or DMSO (n = 5) for 14 days likewise showed highly sensitivity to treatment with Notch inhibition (Fig. 5F; P = 0.008).

To determine if Dnmt3a^fl/fl CLLs had a distinct genetic profile, we performed whole-genome sequencing of the CLL cells. Notably, our analysis revealed amplification of chromosome 15 in these CLL cells (Fig. 5G), consistent with our other genetically engineered CLL models (3, 4), and thereby suggesting a common oncogenic mechanism contributing to CLL generation. A putative oncogenic driver localized in chromosome 15 is Myc, a direct downstream target activated by Notch signaling. Consistent with Myc copy number gain, IHC staining of spleen sections also revealed strong upregulation of MYC protein expression in Dnmt3a^fl/fl CLLs compared with WT B cells (Fig. 5H). In line with this notion, we observed that a large fraction of known Myc targets in B cells (30) were preferentially upregulated in CLL cells versus normal B cells (Fig. 5I). In contrast, very few somatic mutations (26 and 32, respectively) were identified in these CLL samples (Supplementary Table S15).

To determine if the aforementioned insights identified in our mouse model were reflected in human CLL, we evaluated the relationship between baseline expression of DNMT3A and sensitivity to Notch inhibition in primary untreated CLL samples (Supplementary Table S16). Nine samples with differing levels of DNMT3A were exposed to 50 μmol/L DAPT, 1 nmol/L ABT-100, or 1 μmol/L ibrutinib overnight followed by CellTiter-Glo assessment of cellular viability (31–33). Primary CLLs with lower expression of DNMT3A were more sensitive to DAPT, but not to the other two drugs, supporting a specific dependence on Notch signaling for survival (Fig. 5J). Moreover, activation of Notch signaling through coculturing with Notch ligand–expressing OP9 stroma could sensitize DNMT3A-high CLLs (n = 5) to DAPT (Supplementary Fig. S4D), again supporting a direct link between Notch activity and the observed drug response.

DNMT3A is essential for de novo DNA methylation and its inactivation in embryos results in lethally of mice at 4 weeks of age (34). In blood malignancies, dysregulation of Dnmt3a, whether by mutation or altered expression, is increasingly appreciated to be crucial to the pathogenesis of both myeloid and lymphoid malignancies. Dnmt3a is essential for hematopoietic stem cell differentiation, and hematopoietic lineage skewing was observed in its absence (35, 36). Previous murine model studies have only occasionally identified CLL resulting from disruption of Dnmt3a expression, while developing aggressive myeloid or lymphoid neoplasms with early (<12 month) onset (Supplementary Table S17). Conditional KO of Dnmt3a in Mx1-Cre;Dnmt3a^fl/fl mice by pIpC treatment, followed by transplantation of whole BM cells predominantly resulted in myeloid neoplasms, including myelodysplastic syndromes (MDS; 76%), acute myelogenous leukemia (AML; 8%), and MDS/myeloproliferative neoplasms (MPN; 16%; ref. 37). Interestingly, deletion of Dnmt3a together with expression of mutant c-Kit lead to the development of B-ALL in 25% of the mice, suggesting cooperation between the two mutations is necessary for acute B-cell lymphomagenesis (37). In parallel studies, irradiated mice transplanted with Mx1-Cre;Dnmt3a^fl/fl HSCs develop wide spectrum of blood malignancies, including T-cell and B-cell acute lymphocytic leukemia, myelodysplastic syndrome, acute myeloid leukemia, and primary myelofibrosis (16, 38, 39). In contrast, when homozygous Dnmt3a KO in HSCs through the EmSR-tTA;Teto-Cre system, peripheral T-cell lymphomas (PTCL) and CLL were observed, whereas heterozygous deletion of Dnmt3a resulted in MBL/CLL (65% of the cases) and myeloproliferative disease (MPD, 15%; refs. 40–42). In another study, deletion of Dnmt3a specifically in B cells resulted in expansion of B1a cells and development of monoclonal leukemia (43). Finally, deletion of Dnmt3a using the Vav-Cre;Dnmt3a^fl/fl model resulted in MDS (53%), AML (7%), MPN (7%), and mixed phenotype acute leukemia (13%; ref. 44). The heterogeneity of disease phenotypes observed in these studies may be explained by the distinct conditional strategies, as Mx1-cre and Vav1-cre are known to impact transgene expression also in non-HSC populations, including endothelial and BM mesenchymal cells (45). Overall, all these studies support the role of Dnmt3a as a tumor suppressor across cell lineages.

In this study, by restricting Dnmt3a loss to B cells through intercrossing with CD19-Cre mice, we detected only the generation of CLL (without occurrence of myeloid disease or PTCL). Alterations in methylation produced by lack of DNMT3A present within the distinct cellular context and transcriptional regulatory milieu of B1a cells consistently led to disease. Notably, we identified that DNMT3A depletion resulted in the activation of the key CLL-associated pathway of Notch signaling as well as Myc overexpression. Thus, distinct from other existing genetically engineered mouse models of CLL, which prominently feature dysregulation of BCR signaling, this DNMT3A-depleted model most critically dysregulates Notch. As NOTCH mutations and altered expression have been highly linked to CLL generation, our results provide a novel mechanism linking epigenetic alterations in CLL to NOTCH dysregulation.

Our analysis of the changes in the methylation profiles of our heterozygous and homozygous Dnmt3a KO mouse models and the mechanism of Notch activation within these models provided us with insight into the process by which Dnmt3a depletion exerts its pro-oncogenic activity. To determine the influence of Dnmt3a depletion on methylation we performed RRBS analysis. The principle of RRBS is to focus coverage on CpG dense regions of the genome, but not the whole genome, in order to reduce sequencing requirements and cost. Using RRBS, about 50% of covered CpGs are located in CpG islands (CGI) and the remaining CpGs span other CpG dense regions including gene bodies and repeats. Of note, CGIs are often located close to gene promoters, and methylation of CGIs is one of the major mechanisms of gene-expression regulation (46). We observed two levels of change in the DNA methylation profiles within our models. In the preleukemic setting, we found that loss of Dnmt3a induced limited but significant hypomethylation whose impact was reduced apoptotic priming and facilitated accumulation of B1a cells. Following CLL transformation, we observed the CLL cells to become hypermethylated. This process was seemingly independent of Dnmt3a function as this is also observed in other CLL mouse models with WT Dnmt3a, suggesting that the hypermethylation is a response to, rather than a cause of, CLL progression, in line with prior findings (28). In combination with Dnmt3a-associated methylation changes, these result in transcriptional dysregulation of several oncogenic pathways that are important for CLL development, particularly Notch signaling.

Altogether, we identified an association between low DNMT3A expression and poor failure-free survival in human datasets, and have confirmed a causal role of Dnmt3a depletion in CLL generation in mice. Our results support the interaction between Dnmt3a-dependent methylation change and activation of Notch and Myc signaling as a mechanism by which Dnmt3a depletion induces CLL. Moreover, the Dnmt3a models provide a unique opportunity for the study of nonmutational Notch activation, and a useful platform for the study of Notch-signaling targeted therapeutics.

C.J. Wu is an equity holder of BioNtech, Inc. and C.J. Wu and D.S. Neuberg receive research funding from Pharmacyclics. D.S. Neuberg has been a consultant for H3 Biomedicine and received research funding from Celgene. A. Letai serves as an equity holding SAB member for Zentalis Pharmaceuticals, Flash Therapeutics, and Dialectic Therapeutics. His laboratory receives research support from Novartis. T.J. Kipps has received research funding and/or has served as an advisor to Ascerta/AstraZeneca, Celgene, Genentech/Roche, Gilead, Janssen, Loxo Oncology, Octernal Therapeutics, Pharmacyclics/AbbVie, TG Therapeutics, VelosBio, and Verastem. Cirmtuzumab was developed by T.J. Kipps and licensed by the University of California to Oncternal Therapeutics, Inc., which has provided stock/options to the university and T.J. Kipps. C. Sun received research funding from Genmab. J.R. Brown has served as a consultant for Abbvie, Acerta/Astra-Zeneca, Beigene, Bristol-Myers Squibb/Juno/Celgene, Catapult, Eli Lilly, Genentech/Roche, Janssen, MEI Pharma, Morphosys AG, Nextcea, Novartis, Pfizer, Rigel; received research funding from Gilead, Loxo/Lilly, SecuraBio, Sun, TG Therapeutics; and served on the data safety monitoring committee for Invectys. No disclosures were reported by the other authors.

A. Biran: Conceptualization, writing–review and editing, formal analysis, data curation. S. Yin: Conceptualization, data curation, formal analysis, writing–review and editing. H. Kretzmer: Resources, formal analysis, data curation, validation. E. ten Hacken: Resources, data curation, formal analysis. S. Parvin: Validation, data curation, formal analysis. F. Lucas: Data curation, formal analysis. M. Uduman: Data curation, formal analysis. C. Gutierrez: Data curation. N. Dangle: Data curation. L. Billington: Data curation. F.F. Regis: Data curation. L.Z. Rassenti: Resources. A. Mohammad: Data curation. G.B. Hoffmann: Data curation. K. Stevenson: Formal analysis. M. Zheng: Resources. E. Witten: Data curation. S.M. Fernandes: Resources. E. Tausch: Resources. C. Sun: Resources. S. Stilgenbauer: Resources. J.R. Brown: Resources. T.J. Kipps: Resources. J.C. Aster: Resources. A. Gnirke: Resources. D.S. Neuberg: Formal analysis. A. Letai: Resources. L. Wang: Resources. R.D. Carrasco: Resources, data curation, formal analysis. A. Meissner: Resources. C.J. Wu: Conceptualization, resources, data curation, supervision, funding acquisition.

The authors thank the members of the Wu lab, in particular Drs. Erin Parry and Satyen Gohil for their valuable feedback and critical insights. The authors are also grateful for the DFCI Animal Research Facility technical team and Flow Cytometry Core for truly excellent technical support, and the Dana-Farber/Harvard Cancer Center in Boston, MA, for the use of the Specialized Histopathology Core. Dana-Farber/Harvard Cancer Center is supported in part by an NCI Cancer Center Support Grant (NIH 5 P30 CA06516). The authors are also grateful to Binyamin Knisbacher, Cynthia Hahn, J.C. Aster, Oriol Olive, Adrian Wiestner, and Tuan Tran for their generous support. This study was supported by grants from the NIH/NCI (P01 CA206978, P01-CA081534, R01CA216273, and UG1CA233338). This work was further supported in part by the Lymphoma Research Foundation (A. Biran) and the Leukemia & Lymphoma Society (A. Biran). A. Biran is an Awardee of the Weizmann Institute of Science—Israel National Postdoctoral Award Program for Advancing Women in Science. E. ten Hacken is a Scholar of the American Society of Hematology. A. Mohammad is supported by the Max Planck Society. C. Sun is supported by the Intramural Research Program of the National Heart, Lung, and Blood Institute, NIH. E. Tausch and S. Stilgenbauer received research support from DFG SFB1074 subproject B1 and B2. J.R. Brown is supported by NIH R01 CA 213442, P01 CA206978, and P01-CA081534.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.