Decitabine suppresses triple-negative breast cancer by disrupting DNMT1 liquid-liquid phase separation and synergizes with G9a inhibition

Abstract

Triple-negative breast cancer (TNBC) remains a substantial clinical challenge due to the lack of well-defined therapeutic targets. Previous studies have demonstrated that epigenetic modifiers play crucial roles in the progression of various cancers. In this study, we identify DNA methyltransferase 1 (DNMT1) as a critical epigenetic driver of TNBC proliferation and metastasis. Our findings show that DNMT1 is highly expressed in TNBC and forms liquid-liquid phase separation (LLPS) condensates. Knockout of DNMT1 sensitizes TNBC cells to apoptosis and lipid peroxidation, downregulates oncogenic pathways, and upregulates apoptosis-related genes. The DNMT1 inhibitor decitabine (DAC) disrupts DNMT1 condensates, consequently inducing apoptosis and lipid peroxidation. Mechanistically, DAC directly binds to the disordered domain of DNMT1, thereby dissolving LLPS condensates and impairing epigenetic silencing. Remarkably, DAC synergizes with histone methyltransferase G9a inhibitor BIX-01294 (BIX) to suppress TNBC proliferation, migration, and invasion. Combined treatment with DAC and BIX significantly inhibits lung metastasis in murine and human TNBC mouse models, reducing proliferative cancer cells and metastatic burden. This study reveals DNMT1 LLPS as a druggable vulnerability in TNBC and establishes LLPS disruption as a promising therapeutic strategy, providing a rationale for combining epigenetic inhibitors to treat TNBC and other cancers with high DNMT1 expression.

Introduction

Breast cancer is the most prevalent malignant tumor among women, accounting for approximately one-quarter of all female cancer cases [1]. According to the expression status of molecular markers such as estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and Ki67, breast cancer can be classified into four subtypes, which include luminal A-like, luminal B-like, HER2-positive, and triple-negative breast cancer (TNBC) [2]. The systemic treatment strategies for breast cancers are based on the characteristics of their subtypes. Patients with ER- or PR-positive breast cancer primarily receive endocrine therapy, while patients with HER2-positive breast cancer mainly undergo HER2-targeted therapy. The primary treatment for TNBC patients is chemotherapy [3]. Endocrine therapy has significantly reduced the recurrence and mortality rates of hormone receptor-positive breast cancers [4]. HER2-targeted therapy is one of the most significant breakthroughs in breast cancer treatment over the past two decades. Patients with HER2-positive breast cancers have shown substantial benefits from monotherapy against HER2, dual anti-HER2 therapy, or combination chemotherapy [5–7].

TNBC is the subtype of breast cancer with the poorest prognosis, comprising 15 to 20% of all breast cancer cases [8, 9]. Due to the absence of relevant receptor markers, TNBC patients cannot benefit from existing endocrine therapies or HER2-targeted drugs. Consequently, for inoperable TNBC patients, non-specific chemotherapy remains the current standard treatment. Although many TNBC patients respond well to conventional anthracycline and taxane-based chemotherapy agents, the recurrence rate remains high, and long-term survival rates are still low [10]. Therefore, researchers have developed various targeted therapeutic strategies for TNBC, including poly (ADP-ribose) polymerase inhibitors (PARPi), cyclin-dependent kinase (CDK) inhibitors, PI3K/AKT/mTOR inhibitors, Notch signaling pathway inhibitors, STAT3 inhibitors, transforming growth factor-β (TGF-β) inhibitors, and antibody-drug conjugates (ADCs) [11, 12]. Although these candidates have shown promise in preclinical studies and early clinical trials, their widespread clinical application still faces challenges such as suboptimal efficacy, side effects, and resistance. Thus, identifying new therapeutic targets and drugs for TNBC patients remains an urgent priority.

The pathogenesis and progression of TNBC are closely related to epigenetic dysregulation. DNA methylation is one of the most extensively studied epigenetic processes in human cancer [13]. Studies have shown that DNA methyltransferase 1 (DNMT1) is implicated in various cancers, including TNBC, lung cancer, leukemia, lymphoma, liver cancer, gastric cancer, and bone cancer [14–21]. Histone modifications, such as methylation, acetylation, and phosphorylation, are also critical mechanisms regulating chromatin structure and gene expression [22]. The histone methyltransferase G9a catalyzes the methylation of H3K9 (H3K9me), thereby repressing gene expression, particularly tumor suppressor genes (TSGs) [5]. Studies have shown that high G9a expression is associated with poor prognosis in breast cancer patients, and G9a inhibitors (G9ai) effectively suppress the migration and invasion of TNBC cells [23, 24]. The inhibitory histone mark H3K27me3 is more prevalent in luminal A tumors compared to HER2-positive and TNBC tumors, while HER2-positive and TNBC tumors exhibit aberrant acetylation of H3K9 (H3K9ac) [25–28]. Phosphorylation of histone H1 in TNBC cells has been identified as a potential biomarker for distinguishing different cancer subtypes and grades [29]. Additionally, targeting chromatin remodeling mechanisms may offer promising strategies to address TNBC drug resistance [30, 31].

Unlike irreversible genetic abnormalities, epigenetic alterations such as DNA methylation and histone modifications are reversible, rendering them attractive therapeutic targets. DNMT inhibitors (DNMTi) encompass first-generation nucleoside analogs like 5-azacytidine (AZA) and 5-aza-2’-deoxycytidine (decitabine, DAC), as well as second-generation inhibitors, which include nucleoside analogs such as guadecitabine and FdCyd, and non-nucleoside inhibitors like GSK3685032. Notably, AZA and DAC have been approved for the treatment of myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML) [32, 33]. In preclinical TNBC models, both DAC and AZA have exhibited tumor-suppressive effects [34–36]. Historically, high-dose DAC administration in solid tumor studies has resulted in significant toxic side effects [37]. However, recent research has shown that low concentrations of DAC can effectively inhibit tumor growth in patient-derived xenograft (PDX) models of TNBC, including both chemotherapy-sensitive and chemotherapy-resistant cases [36].

DAC exerts its effects by incorporating into DNA strands and forming covalent bonds with DNMTs, thereby inhibiting their enzymatic activity and reducing DNA methylation levels in cancer cells. DNA demethylation can reactivate TSGs, promote apoptosis, induce differentiation or senescence, inhibit cell growth, and diminish clonogenic potential [38, 39]. Current research is also exploring the synergistic effects of DNMTi with other therapies, such as cytotoxic drugs and molecularly targeted agents. Studies have shown that the combination of DAC and Cisplatin enhances pro-apoptotic responses in TNBC cells that are less sensitive to Cisplatin [40]. Moreover, combining DAC with PARPi potentiates the anti-tumor efficacy of PARPi in TNBC cells [41]. Additionally, the combination of DNMTi and histone deacetylase inhibitors (HDACi) has been shown to reverse epithelial-mesenchymal transition (EMT) in TNBC, thereby exhibiting enhanced anti-tumor activity [42].

G9a is a pivotal epigenetic regulatory protein that catalyzes H3K9 methylation, leading to chromatin condensation and the transition from an active to a repressed state. The interaction between G9a and DNMT1 facilitates DNA methylation during cell replication, with both DNA methylation and H3K9 methylation being implicated in the silencing of TSGs [43, 44]. BIX-01294 (BIX) is a specific inhibitor of G9a, which induces H3K9 demethylation [45]. Previous studies have found that simultaneously targeting G9a and DNMT1 synergistically inhibits the growth of ovarian and liver cancer cells [46, 47]. However, no studies have yet evaluated the inhibitory effects of combining the DNMT inhibitor DAC with the G9a inhibitor BIX on TNBC and many other cancers, highlighting the critical need to explore this novel combination as a potential epigenetic therapy for TNBC.

Recent studies have shown that cellular proteins can form liquid-liquid phase-separated (LLPS) droplets to safeguard the efficiency of biological activities in both viral infection and cancer cells. Specifically, the polycomb group (PcG) protein CBX4 and chromatin assembly factor 1 (CAF-1) form LLPS condensates that promote the tri-methylation of H3K27 and H3K9, respectively, thereby maintaining HIV-1 latency in infected individuals [48, 49]. The nucleocapsid (N) protein of SARS-CoV-2 forms LLPS condensates and regulates viral RNA transcription through phase separation [50]. Yes-associated protein (YAP) forms LLPS condensates in the nucleus, promoting the transcription of its target genes and driving the proliferation of TNBC cells [51]. Recent studies have identified that phosphorylated histone deacetylase 6 (phospho-HDAC6) forms LLPS condensates in the nuclei of TNBC cells, which drive the formation of aberrant chromatin structures, highlighting the potential of targeting phospho-HDAC6 or its phase separation properties as a therapeutic strategy for TNBC [52]. These findings underscore the close relationship between the LLPS of epigenetic regulatory proteins and disease progression.

Although substantial progress has been made in understanding DNA methylation, the role of LLPS condensates in the DNA methylation process and metastasis of TNBC remains unknown. In this study, we demonstrated that DNMT1 formed LLPS condensates in the nucleus of TNBC cells, the process of which was inhibited by the DNMT1 inhibitor DAC. DAC is directly bound to the intrinsically disordered domain of DNMT1 to dissolve LLPS condensates, thereby suppressing TNBC proliferation, migration, and invasion. Additionally, the combination of DAC and the G9a inhibitor BIX synergistically inhibited TNBC growth and metastasis by inducing lipid peroxidation and apoptosis. Both DAC monotherapy and its combination with BIX reduced cancer cell proliferation and metastatic burden. This study provided a rationale for developing therapeutic strategies targeting LLPS of epigenetic proteins in TNBC.

Materials and methods

Cell lines and compounds

The HEK293T (CRL-3216), 4T1 (CRL-2539), MDA-MB-231 (HTB-26), MDA-MB-468 (HTB-132), MCF10A (CRL-10317), and HC11 (CRL-3062) cell lines were obtained from the American Type Culture Collection (ATCC). HEK293T, MDA-MB-231, MDA-MB-468, and MCF10A cells were maintained in DMEM (11965118, Gibco, USA), while 4T1 and HC11 cells were cultured in RPMI 1640 medium (11875119, Gibco). The 4T1-luc and MDA-MB-231-luc cell lines were constructed by infecting wild-type cells with luciferase-expressing lentiviruses (LV-NC009, Guangzhou IGE Biotechnology, Guangzhou, China). All the culture media were supplemented with 10% fetal bovine serum (FBS) (A5670701, Gibco, California, USA) and 1% penicillin–streptomycin (15140163, Gibco). Cells were maintained in a humidified incubator at 37 °C with 5% CO₂. All the cell lines have been confirmed to be mycoplasma-free via PCR-based detection (Forward primer: 5’-GGGAGCAAACAGGATTAGTATCCCT-3’; Reverse primer: 5’-TGCACCATCTGTCACTCTGTTACCCTC-3’).

Decitabine (DAC) (T1508), erastin (T1765), RSL3 (T3646), Ferrostatin-1 (Fer-1) (T6500), LCL161 (T2080), and Cisplatin (T1564) were purchased from TargetMol Chemicals. BIX-01294 (BIX) (S8006) and mitomycin C (S8146) were purchased from Selleck Chemicals. All compounds were dissolved in DMSO, except that Cisplatin was dissolved in saline. The stock solutions of these compounds were prepared as follows: DAC, 100 mM; BIX, 100 mM; erastin, 10 mM; Fer-1, 50 mM; RSL3, 10 mM; LCL161, 50 mM; mitomycin C, 1 mg/mL; Cisplatin, 3 mM (in saline).

Microbial strains

Escherichia coli (E. coli) strains DH5α (KTSM101L, AlpaLifeBio, Shenzhen, China) and BL21 (DE3) (KTSM104L) were cultured at 37 °C in Luria-Bertani (LB) medium containing specified antibiotics. DH5α was utilized for propagating constructed protein-coding clones, while BL21 (DE3) was used for the expression and purification of prokaryotically expressed proteins, including GFP-tagged DNMT1-DS1 and DNMT1-DS1 mutants.

Cell viability assay

To evaluate the dose-dependent response of targeted compounds, about 10,000 4T1 cells, 5000 MDA-MB-231 cells, and 10,000 MDA-MB-468 cells were seeded into each well of the 96-well plates, respectively. Triplicate wells were prepared for all three independent experiments. Cells were incubated overnight at 37 °C with 5% CO₂. On the next day, the culture medium was replaced with freshly prepared 100 μL of DMEM or 1640 culture medium containing different concentrations of DAC or BIX. The control wells were replaced with fresh medium containing DMSO. 4T1 and MDA-MB-468 cells were incubated for another 3 days, while the MDA-MB-231 cells were incubated for 7 days. Four hours (h) before detection, about 10 μL of Cell Counting Kit-8 (CCK-8) reagent (C0005, TargetMol) was added to each well, followed by incubaton for 4 h. The optical density (OD) value was measured at a wavelength of 450 nm using a multifunctional microplate reader (Tecan Spark, Männedorf, Switzerland). OD450 values were then curve-fitted using GraphPad Prism 9 software to calculate the half-maximal inhibitory concentration (IC₅₀) of each compound in each cell line.

Evaluation of combination effects

To calculate the combination index (CI) value, the inhibitory effects on cell growth were assessed utilizing combinations of different concentrations of the DNMT1 inhibitor DAC and the G9a inhibitor BIX. For each treatment condition, the cell viability data were normalized to represent the fraction affected (Fa) value as a percentage relative to the untreated control. The CI values corresponding to different conditions were calculated using CompuSyn software. A CI value < 1 was defined as a synergistic effect. A CI value = 1 was defined as an additive effect, while a CI value > 1 was defined as an antagonistic effect.

Colony formation assay

4T1, MDA-MB-231, and MDA-MB-468 cells undergoing exponential growth were digested with 0.25% trypsin (25200114, Gibco) to obtain the single-cell suspension. Approximately 1000 cells were seeded in 6-well plates in triplicate. Three independent experiments were performed. On the next day, 4T1 cells were treated with 100 nM DAC and/or 1 μM BIX for 7 days. MDA-MB-231 cells were treated with 50 nM DAC and/or 100 nM BIX for 14 days. MDA-MB-468 cells were treated with 10 nM DAC and/or 10 nM BIX for 14 days. The control group was treated with equal volume of DMSO. Each treatment was conducted in triplicate. During the period of cultivation, the medium was replaced with freshly prepared compound-containing medium every 3 days. After the cultivation, cells were fixed with 4% paraformaldehyde for 20 min (min) and stained with 0.2% crystal violet for an additional 20 min. After washing with PBS, the plates were dried at room temperature. Images of each well were captured utilizing an inverted optical microscope (Sunny Optical Technology, Hangzhou, China). Cellular colonies within each well were subsequently quantified using ImageJ software.

Wound healing scratch assay

4T1, MDA-MB-231, and MDA-MB-468 cells in the exponential growth phase were seeded in 6-well plates at a density of 1 × 10⁶ cells per well, in triplicate. On the next day, a 10 μL pipette tip was used to generate a scratch in the cell monolayer, followed by washing the cells three times with PBS to remove scratched cells. To exclude effects of cell proliferation-mediated wound healing, cells were treated with 1 μg/mL mitomycin C for 1 h. Cells within each well were then treated with 1 μM DAC and/or 1 μM BIX, while the control group was treated with an equal volume of DMSO. Images of the scratched area were captured at 0 h and 48 h post-scratch using an inverted optical microscope (Sunny Optical Technology). The experiment was repeated three times. The formula for calculating the wound healing rate (%) was: [(0 h scratch width - 48 h scratch width) / 0 h scratch width] × 100%.

Transwell migration assay

TNBC cells were seeded at a density of 3 × 10⁵ cells per well in 6-well plates containing DMEM or RPMI 1640 medium supplemented with 10% FBS. Triplicate wells were prepared for all three independent experiments. Cells were then treated with 1 μM DAC and/or 1 μM BIX on the next day. After 12-h treatment, cells were digested with 0.25% trypsin, washed once with PBS, and resuspended in serum-free DMEM or RPMI 1640 medium. Subsequently, about 100 μL of the cell suspension (1 × 10⁵ cells) was added to the upper chamber of a transwell insert with 6.5 mm diameter and 8.0 µm pore size polycarbonate membrane (3422, Corning, New York, USA). The lower chamber was filled with 0.6 mL of culture medium containing 10% FBS. After incubating for 48 h, the top of the membrane was wiped with a cotton swab to remove non-migrated cells. Migrated cells on the bottom of the membrane were then fixed with 4% paraformaldehyde for 20 min and stained with 0.2% crystal violet for an additional 20 min. Finally, representative images were captured using an inverted optical microscope (Sunny Optical Technology). Transmigrated cells were manually counted using ImageJ software in randomly selected fields of view, and results were expressed as cell counts per field. At least three random fields were analyzed per group, and all experiments were independently repeated three times.

Transwell invasion assay

The procedure for the transwell invasion assay was similar to that of the transwell migration assay, with the exception that the membrane of the transwell insert was pre-coated with 1 mg/mL Matrigel matrix (356234, Corning) for 3 h at 37 °C. Subsequent steps involving cell seeding, incubation, fixation, staining, and quantification were identical to those in the transwell migration assay.

Apoptosis assay

TNBC cells were seeded at a density of 2 × 10⁵ cells in 12-well plates containing DMEM or 1640 medium supplemented with 10% FBS. Triplicate wells were prepared for subsequent experiments. After 24 h, cells were treated with 2 μM DAC and/or 2 μM BIX. Following a 48-h treatment period, cells were subjected to apoptosis detection using the Annexin V-FITC/PI apoptosis kit (AP101-30, LianKe Biotech, Hangzhou, China) according to the manufacturer’s instructions. Briefly, cells were resuspended in 500 μL of 1× Binding Buffer containing 5 μL of Annexin V-FITC and 10 μL of PI, followed by incubation in the dark for 5 min. Subsequently, quantitative analysis of Annexin V-positive and/or PI-positive cells was performed using a flow cytometer (CytoFLEX S, Beckman Coulter, California, USA) with detection parameters set at 488 nm and 561 nm lasers. The experiment was independently repeated three times. The total apoptotic rate was calculated as the sum of the proportions of Annexin V single-positive (early apoptosis) and Annexin V/PI double-positive (later apoptosis) cells. Simultaneously, cleaved caspase-3 was detected by western blot with antibodies against caspase-3 (14220 T, CST, Massachusetts, USA) or cleaved caspase-3 (9664 T, CST, Massachusetts, USA). Anti-GAPDH (60004-1-Ig, Proteintech, Tianjin, China) was used as a loading control. All samples were taken from the same batch of cells and subjected to identical treatment. DNMT1 WT and KO cells were treated for 48 h with the apoptotic inducers LCL161 (0.1 μM) and Cisplatin (50 μM). Apoptotic cell populations were quantified by flow cytometry. In parallel, levels of cleaved caspase-3 were assessed by western blot analysis. To further validate the role of DNMT1, DNMT1 was re-expressed in KO cells via transient transfection (rescue experiment). The rescued DNMT1-expressing KO cells were then treated in parallel with LCL161 (0.1 μM) or cisplatin (50 μM) for 48 h, and apoptotic cell populations were quantified by flow cytometry.

Lipid ROS analysis

TNBC cells were seeded at a density of 3 × 10⁵ cells in 6-well plates containing DMEM or 1640 medium supplemented with 10% FBS. Cells were treated in triplicate with 1 μM DAC and/or 1 μM BIX after 24 h. Erastin (1 μM) was used as a positive control. Another 48 h later, intracellular lipid reactive oxygen species (ROS) levels were assessed with 2 μM C11-BODIPY dye (D3861, Invitrogen) for 30 min in the dark, followed by washing with PBS 3 times. Cells were then resuspended in 200 μL of PBS. The oxidation of the polyunsaturated butadienyl moiety of the dye caused a shift in the fluorescence emission from approximately 590 nm (PE-positive) to about 510 nm (FITC-positive). The percentage of FITC-positive cells was quantified to represent lipid ROS levels. The mean fluorescence intensity (MFI) of FITC-positive cells was also measured. Additionally, TNBC cells were treated with 1 μM DAC and/or BIX for 48 h in the presence or absence of Ferrostatin-1 (Fer-1, 10 μM), and relative lipid ROS levels were assessed using the C11-BODIPY dye. For genetic validation, DNMT1 WT and KO cells were treated with erastin (1 μM) for 48 h, and lipid ROS levels were measured as described above. For rescue experiments, DNMT1 was re-expressed in KO cells via transient transfection prior to identical erastin treatment and analysis. Three independent experiments were conducted.

FerroOrange assay

TNBC cells were seeded in confocal dishes at a density of 3 × 10⁵ cells per dish. After 24-h attachment, cells were treated in triplicate with 1 μM DAC and/or 1 μM BIX. Subsequently, following an additional 48-h incubation, intracellular labile ferrous iron (Fe²⁺) levels were detected using BioTracker FerroOrange Live Cell Dye (SCT210, Sigma-Aldrich) according to the manufacturer’s instructions. Specifically, cells were incubated with 1 μM FerroOrange working solution in serum-free medium at 37 °C for 30 min in the dark. After washing with PBS, fluorescence images were immediately captured using a confocal fluorescence microscope (Ex/Em = 548/561 nm), and fluorescence intensity was quantified using ImageJ software. The fluorescence intensity of FerroOrange was regarded as an indicator of intracellular labile Fe²⁺ levels. The fluorescence intensity in the control group was normalized to 1 for relative quantification.

RSL3 inhibition assay

To evaluate the dose-dependent response of TNBC cells to RSL3, cells were seeded at 10,000 cells per well in 96-well plates, with three technical replicates per group and three independent biological replicates. After attachment, cells were cultured at 37 °C in a humidified incubator with 5% CO₂ for 24 h. The medium was then replaced with fresh culture medium containing different concentrations of RSL3, and an equivalent volume of DMSO was added to the control group. After an additional 48 h of incubation, 10 μL of CCK-8 reagent was added to each well, followed by incubation at 37 °C for 4 h. Absorbance was measured at 450 nm. Cell viability and IC₅₀ values were analyzed using GraphPad Prism 9 software.

Lipid ROS imaging

4T1, MDA-MB-231, and MDA-MB-468 cells were seeded in confocal dishes (D35-20-1-N, Cellvis, California, USA) at a density of 3 × 10⁵ cells per dish, in triplicate. Cells were treated with 1 μM DAC at 24 h post-seeding, while DMSO was used as the vehicle control. After 48-h treatment, cells were stained with C11-BODIPY and Hoechst (62249, ThermoFisher, Massachusetts, USA) dyes for 30 min at 37 °C and 5% CO₂. Subsequently, these cells were washed with PBS and imaged utilizing a confocal microscope (LSM 980, Zeiss, Oberkochen, Germany). The intensity of the FITC-positive component indicated lipid ROS levels. Three independent imaging assays were performed.

DNMT1 knockout cell line construction

The sgRNA expression vector was generated by cloning the target sequence (sgDNMT1: 5’-TCCTCCCAGGCTCAAAGATT-3’) into the lentiCRISPRv2.0 vector (52961, Addgene, Massachusetts, USA). Pseudotyped viruses were produced by co-transfecting HEK293T cells with VSV-G glycoprotein-expressing vector pMD2.G (12259, Addgene), lentiviral packaging construct psPAX2 (12260, Addgene), and sgRNA-expressing construct lentiCRISPRv2.0. Subsequently, MDA-MB-231 cells were infected with the sgDNMT1 lentiviruses, followed by treatment with 1 μg/mL puromycin (A1113803, Gibco) to eliminate uninfected cells at 48 h post-infection. Three days post-puromycin selection, surviving cells were transferred to fresh DMEM medium for further cultivation. The efficiency of DNMT1 knockout was verified by western blot analysis using a rabbit polyclonal antibody against DNMT1 (5032 T, CST).

RNA extraction and sequencing

Total RNAs were extracted from wild-type (DNMT1-WT) and DNMT1-knockout (DNMT1-KO) MDA-MB-231 cells using TRIzol reagent (15596018, Invitrogen, California, USA). RNA samples from each cell type were prepared in duplicate to construct cDNA libraries. RNA purity and quantification were evaluated using a NanoDrop 2000 spectrophotometer (ThermoFisher), while RNA integrity was assessed using an Agilent 4200 Bioanalyzer (Agilent Technologies, California, USA). The RNA libraries required for RNA-sequencing (RNA-Seq) were prepared as follows. Poly(A)-enriched mRNAs were purified from total RNAs and fragmented into 300 to 350 bp pieces. The fragmented RNAs were reverse transcribed into cDNAs, followed by end repair, adenylation, and adapter ligation. The cDNA templates underwent PCR amplification, followed by purification of PCR products to obtain final libraries. The constructed RNA libraries were sequenced on the Illumina NovaSeq 6000 sequencing system by Berry Genomics. RNA-seq data have been deposited in the NCBI database with the following accession numbers: B231-WT-1 (SRR31392303), B231-WT-2 (SRR31392302), B231-KO-1 (SRR31392301), B231-KO-2 (SRR31392300).

For reverse transcription-quantitative polymerase chain reaction (RT-qPCR), total RNA from each cell type was extracted using an EZ-press RNA Purification Kit (B004D, EZBioscience, Minnesota, USA) according to the manufacturer’s instructions. cDNAs were synthesized from total RNAs via StarScript Pro All-in-one RT Mix Kit (A240-10, GenStar, Beijing, China), followed by qPCR with Taq Pro Universal SYBR qPCR Master Mix Kit (Q712-03, Vazyme, Nanjing, China) on a QuantStudio 1 Plus Real-Time PCR System (Applied Biosystems, California, USA) in triplicate. The relative expression levels of each gene were calculated by first normalizing to internal control genes, followed by normalizing to the control groups. Primers used for RT-qPCR in this study have been listed in Table S1 (Supplementary Material). We validated the DNMT1 knockout-induced changes in protein expression of c-Fos, GPX4, DAPK1, and KISS1 by western blotting. Western blot analyses were performed using antibodies against c-Fos (2250 T, CST), GPX4 (67763-1-Ig, Proteintech), DAPK1 (67815-1-Ig, Proteintech), and KISS1 (85068 T, CST), with GAPDH serving as the loading control.

SIM imaging

To prepare samples for super-resolution structured illumination microscopy (SIM) imaging, 4T1, MDA-MB-231, and MDA-MB-468 cells were seeded in a 6-well plate in triplicate with coverslips pre-treated with poly-L-lysine (polyK) (P1399, Sigma-Aldrich). These cells were transfected with GFP-tagged constructs at 24 h post-seeding. Another 24 h later, cells were fixed with 4% paraformaldehyde (16005, Sigma-Aldrich) and permeabilized with 0.2% Triton X-100 (A600198, Sangon Biotech, Shanghai, China). Subsequently, cells were incubated sequentially with specific primary antibodies and fluorescently labeled secondary antibodies. DNA was stained with 4’,6-diamidino-2-phenylindole (DAPI) (62248, ThermoFisher). For cells transfected with GFP-tagged constructs, only fixation, permeabilization, and DAPI staining were required. Coverslips with cells were mounted with ProLong Diamond Antifade Mountant (P36961, Invitrogen) on slides for imaging.

Cell samples were imaged using a Nikon A1 confocal microscope equipped with a CFI Apo TIRF objective (1.49 NA, oil immersion), an sCMOS camera (Hamamatsu Flash 4.0, pixel size 6.5 μm × 6.5 μm), four lasers (SIM 405, SIM 488, SIM 561, and SIM 647), and NIS-Elements AR software. During 3D-Stack SIM imaging, the z-axis step size was set to 0.20 μm per frame. For each focal plane, 15 images were captured in 3D-SIM mode (5 phases and 3 angles). The reconstruction and analysis of SIM images were performed using the N-SIM module of NIS-Elements AR software. The size, number, distribution, and sphericity of DNMT1 bodies were measured and analyzed using Imaris software (Imaris x64 9.0.1). Briefly, raw images were imported and background correction was applied to the FITC channel to remove noise while preserving target signals. The “Spots” function was used for detection, with the option “different spot sizes” enabled, and the FITC channel selected for analysis. The estimated spot diameter was set according to condensate size (approximately 0.1 μm for endogenous condensates and 0.2 μm for exogenous condensates). After automatic initial thresholding, manual adjustment was performed to ensure accurate identification of all discernible condensates while minimizing background signals. Parameters including spot number and diameter were exported for quantitative analysis of condensates.

FRAP experiment

The internal diffusion characteristics of DNMT1 bodies were analyzed through fluorescence recovery after photobleaching (FRAP) experiments. GFP-tagged DNMT1 constructs were transfected into MDA-MB-468 or HEK293T cells. Live cells were then imaged on a Zeiss Axio Observer Z1 confocal microscope equipped with a 63×/1.40 NA oil immersion objective. GFP fluorescence was excited using a 488 nm laser, with the pinhole set to 1.0 AU and signals detected using a GaAsP-PMT detector. Three condensates were selected as regions of interest (ROIs) for photobleaching (approximately 0.5 μm in diameter), which was performed using a 488 nm laser at 90% power with 100 scanning iterations. For HEK293T cells, image acquisition was conducted at 0.2% laser power with a time interval of 1.26 s per frame. For MDA-MB-468 cells, image acquisition was performed at 0.5% laser power with a time interval of 1.89 s per frame. Another three ROIs served as the unbleached negative controls. A total of 45 frames were acquired for each experiment. FRAP data were analyzed using ZEN 3.9 software (Zeiss), with background subtraction, normalization, and correction for overall photobleaching. Recovery curves were fitted using a single-exponential model, and the half-time of recovery (T_1/2) was calculated. The FRAP assays were performed for three independent experiments.

Protein purification

The GFP-tagged DNMT1-DS1 and DNMT1-DS1 mutants were cloned into the pET28a vector containing a 6×His tag. The constructed clones were transformed into BL21 competent cells and cultured in LB medium containing kanamycin until the OD600 reached 0.4 to 0.6. The isopropyl β-D-1-thiogalactopyranoside (IPTG) (A600168, Sangon Biotech) solution was added to induce the expression of target genes for 18 h. Subsequently, the bacterial pellet was collected by centrifugation, followed by resuspension in Buffer A (50 mM Tris-HCl, pH 7.5, 500 mM NaCl). The bacteria were then lysed by sonication in an ice bath via VCX-500 Ultrasonic Microprocessor (Sonics Materials, Connecticut, USA) (20 min, 70% power, 5 sec on / 5 sec off). Cell debris and insoluble materials were removed by centrifugation. The cleared supernatant was incubated with Ni Sepharose Excel beads (17371202, Cytiva, New York, USA) to enrich the 6×His-tagged target proteins. The protein-bound beads were then washed sequentially with 30 mL of Buffer B (50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 10 mM imidazole), Buffer C (50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 20 mM imidazole), and Buffer D (50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 50 mM imidazole). Target proteins were eluted stepwise using Buffer E (50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 100 mM imidazole), Buffer F (50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 200 mM imidazole), and Buffer G (50 mM Tris-HCl, pH 7.5, 500 mM NaCl, 500 mM imidazole). Elutions from Buffer E to G were concentrated and buffer-replaced with Buffer A using a 30 kDa ultrafiltration device.

In vitro droplet formation assay

Protein solutions were diluted into buffers with varying NaCl concentrations. The NaCl concentration gradients were prepared by serially diluting protein solutions with Buffer H (50 mM Tris-HCl, pH 7.5, 10% glycerol, 1 mM DTT). The droplet formation buffers included Buffer I (50 mM Tris-HCl, pH 7.5, 250 mM NaCl, 10% glycerol, 1 mM DTT), Buffer J (50 mM Tris-HCl, pH 7.5, 125 mM NaCl, 10% glycerol, 1 mM DTT), Buffer K (50 mM Tris-HCl, pH 7.5, 62.5 mM NaCl, 10% glycerol, 1 mM DTT), Buffer L (50 mM Tris-HCl, pH 7.5, 31.25 mM NaCl, 10% glycerol, 1 mM DTT), and Buffer M (50 mM Tris-HCl, pH 7.5, 15.625 mM NaCl, 10% glycerol, 1 mM DTT). Under different conditions of NaCl and protein concentrations, droplets were rapidly imaged using a Nikon A1 confocal microscope with a 488 nm laser. The areas of droplets in three preparations were calculated via Imaris software (Imaris x64 9.0.1) and analyzed by GraphPad Prism 9.

Molecular docking

For the binding mode of DAC to the catalytic domain of DNMT1, the crystal structure of human DNMT1 (351-1600) (PDB ID: 4WXX) was used as the receptor. DAC was docked to the catalytic pocket of DNMT1 using Schrödinger Maestro software in covalent docking mode with default parameters. For the binding mode of DAC to DS1 region of DNMT1, the structure of DS1 was predicted by AlphaFold 3 (AF3). The model with the highest-ranking score of 0.68 from the prediction results was chosen as the receptor model for subsequent molecular docking. The ranking score was calculated based on the composite score of normalized pLDDT, pTM, ipTM, fraction disordered, and clash score, according to the equation of 0.8×ipTM + 0.2×pTM + 0.5×disorder - 100×has_clash. The pTM value of the selected model was 0.28, while the value of fraction disordered was 0.8, and the clash score was 0.0. The ipTM provided the score of the predicted interface, which was not applicable to DS1 monomer prediction. The weights of both ipTM and pTM were assigned to pTM alone. The highest ranking score of 0.68 was calculated as follows: 1×0.28 + 0.5×0.8-100×0.0 = 0.68. DAC was docked to DS1 using AutoDock 4.2 software in blind docking mode. The model generated by AF3 only contains the protein backbone, without any heteroatoms or water molecules. Therefore, during docking preparation, no water molecules were added to either the DS1 receptor model or the DAC ligand model. Eight clusters with binding energies below -3.00 kcal/mol were the possible binding sites and selected for further analysis. The docking results were analyzed and visualized via PyMOL 3.1 software.

SPR assay

Surface plasmon resonance (SPR) measurements of the binding of DAC to recombinant GFP-tagged DNMT1-DS1 were performed using a Biacore 8 K+ instrument (GE Healthcare). A Biacore CM7 sensor chip and an amine coupling kit were employed according to the manufacturer’s protocol. The optimal pH for the immobilization of DS1 was determined to be 5.5. The surface of CM7 sensor chip was activated, followed by covalent coupling of DS1 (20 μg/mL in 10 mM acetate buffer, pH 5.5) for 10 min. Unreacted active esters were blocked by injecting 1 M ethanolamine-HCl for 7 min. All immobilization and binding analyses were conducted at 25 °C using PBS-P running buffer (10 mM phosphate, 137 mM NaCl, 2.7 mM KCl, and 0.05% P20, pH 7.4) supplemented with 5% DMSO. DAC solutions were prepared as a two-fold serial dilution series (ranging from 1.56 to 25 μM) in running buffer containing 5% DMSO. To minimize bulk refractive index effects, solvent correction was applied. A reference flow cell without immobilized DS1 was employed to account for non-specific binding. DAC at increasing concentrations was injected over the DS1-coated surface at a flow rate of 30 μL/min. The association phase was monitored for 120 s, followed by a 300-s dissociation phase in running buffer for each cycle. Response units were recorded at the end of the injection phase. Sensorgrams were fitted using the “Affinity” model implemented in the Biacore 8 K+ evaluation software to obtain binding parameters.

MST assay

Microscale thermophoresis (MST) assays were carried out using a Monolith NT.115 Pico instrument (NanoTemper Technologies) following established protocols. All measurements were performed in an assay buffer containing 50 mM Tris, 500 mM NaCl, and 0.02% (v/v) Tween-20, pH 7.5. To assess the binding affinity between DAC and GFP-tagged DNMT1-DS1, DAC ligand solutions were prepared through serial 2-fold dilutions in quadruplicate, with concentrations ranging from 1.5 nM to 50 μM. Each ligand dilution was incubated with a fixed concentration of GFP-DNMT1-DS1 (150 nM) at room temperature for 10 min. Following incubation, the samples were loaded into Monolith NT.115 Standard Treated Capillaries (MO-K002, NanoTemper Technologies), and MST measurements were performed using 20% LED power and medium MST power settings. All experiments were independently repeated with separate protein preparations to ensure reproducibility. Dissociation constants (K_D) were determined based on the law of mass action using NanoTemper MO. Affinity Analysis (NanoTemper Technologies).

In vivo animal model

To establish an in vivo 4T1 TNBC lung metastasis model, 5-week-old female BALB/cJGpt mice (N000020, GemPharmatech, California, USA) were housed at Guangzhou National Laboratory Animal Center under specific pathogen-free (SPF) conditions. A saline suspension containing 1 × 10⁶ 4T1-luc cells (approximately 100 μL) was injected into BALB/c mice via the tail vein. Mice were randomly allocated to each group (n = 4 per group) with similar initial average body weights. Subsequently, mice were intraperitoneally injected with saline, DAC (2 mg/kg), BIX (2 mg/kg), and the combination of two compounds (2 mg/kg DAC + 2 mg/kg BIX), respectively, five times per week for two weeks. The body weight of each mouse was monitored every two days. Bioluminescence imaging was conducted weekly to monitor lung metastasis. The bioluminescence quantification was performed based on total photon flux. All mice were euthanized after two weeks. The metastatic nodule quantification was conducted based on manual microscopic examination of lung tissues by two investigators.

For the in vivo MDA-MB-231 TNBC orthotopic tumor model, approximately 1 × 10⁶ MDA-MB-231 cells were orthotopically implanted into the mammary fat pad of 5-week-old female immunodeficient NCG mice (NOD/ShiLtJGpt-Prkdc^em26Cd52Il2rg^em26Cd22/Gpt) (T001475, GemPharmatech). One week after inoculation, mice were randomly assigned to four groups (n = 4 per group) and intraperitoneally injected with saline, DAC (2 mg/kg), BIX (2 mg/kg), and the combination of two compounds (2 mg/kg DAC + 2 mg/kg BIX), respectively, twice a week for three weeks. Tumor volume was measured three times per week using calipers and calculated as (length × width²)/2. Body weight was recorded every two days. Mice were euthanized on day 25 and tumors were harvested.

For the in vivo MDA-MB-231-luc TNBC lung metastasis model, the procedure was identical to that in the 4T1 lung metastasis model, except that compound treatments began on day 7 post-tumor inoculation in NCG mice (n = 4 per group), continuing for two weeks with two doses per week. Bioluminescence imaging was conducted twice a week to evaluate therapeutic effects. All animal experiments were approved by the Ethics Committees of the Guangzhou Medical University Animal Center and the Guangzhou National Laboratory Animal Center.

Ki67 immunostaining of mouse lung or tumor tissues was quantitatively analyzed using ImageJ software. The IHC images were transformed into 8-bit grayscales, and a consistent threshold was applied to identify positive signals. The integrated density (IntDen) was measured, and the IntDen/Area ratio was calculated using the entire image area serving as the normalization baseline to represent Ki67 expression. The control group was set to a value of 1, and the relative Ki67 expression levels of the experimental groups were calculated accordingly. Statistical analysis was performed by comparing the relative values among groups.

Statistical analysis

Statistical analysis was performed using GraphPad Prism 9 or Microsoft Excel. Comparisons between two groups of data were conducted using Student’s t-tests, while comparisons between multiple groups were performed utilizing one-way or two-way ANOVA. All results were presented as mean ± SD. Network analysis and clustering analysis were performed using the STRING database. Statistical significance was defined as follows: P values < 0.05 (*), < 0.01 (**), < 0.001 (***), and < 0.0001 (****).

Results

Decitabine and BIX-01294 synergistically suppressed TNBC cell proliferation

Previous reports have shown that both DNA methyltransferase 1 (DNMT1) and histone H3K9 methyltransferase G9a can epigenetically contribute to the progression of triple-negative breast cancer (TNBC) [15, 23, 24]. Treatment with DNMT1 inhibitor decitabine (DAC) or G9a inhibitor BIX-01294 (BIX) alone impairs TNBC growth in specific cell lines or patient-derived xenograft (PDX) organoids [24, 36]. To systematically evaluate the anti-TNBC effects of DAC, BIX, or their combination on TNBC cells, we treated high-metastatic murine 4T1 and human MDA-MB-231 and MDA-MB-468 TNBC cell lines with different concentrations of DAC and BIX. Our results showed that both DAC and BIX significantly inhibited the proliferation of TNBC cells (Fig. 1A). Half-maximal inhibitory concentrations (IC₅₀) of DAC on these cells ranged from approximately 40 to 400 nM, while IC₅₀ values of BIX varied from 5 to 6 μM. Compared with individual treatments, the combination of DAC and BIX significantly enhanced the inhibitory effect on TNBC cells (Fig. 1B). Further Chou-Talalay analysis revealed that DAC and BIX exhibited synergistic inhibitory effects on 4T1, MDA-MB-231, and MDA-MB-468 cells, as indicated by combination index (CI) values less than 1 (Fig. 1C). Previous reports have shown that DAC induces the degradation of DNMT1 upon long-term treatments, whereas BIX promotes the demethylation of H3K9me2 [45, 53]. Consistently, our western blot assays revealed that the endogenous DNMT1 protein levels in 4T1, MDA-MB-231, and MDA-MB-468 cells were decreased upon DAC treatment at concentrations of IC₅₀ and IC₉₀ for 3-7 days (Fig. S1A–C). BIX also dose-dependently decreased endogenous H3K9me2 levels within 4T1, MDA-MB-231, and MDA-MB-468 cells, supporting the mechanism of BIX on G9a inhibition (Fig. S1D–F). To further determine whether DAC and BIX engage their respective targets simultaneously under combination treatment, we assessed both DNMT1 and H3K9me2 levels in the same set of TNBC cells treated with DMSO, BIX alone, DAC alone, or BIX plus DAC at their respective IC₅₀ concentrations. Western blot analysis revealed that BIX alone or DAC alone modestly reduced DNMT1 levels (Fig. S2A–C). However, the combination of DAC and BIX resulted in a more pronounced decrease in DNMT1 levels across all three TNBC cell lines than either agent alone. Similarly, BIX alone or DAC alone slightly reduced H3K9me2 levels, whereas the combination treatment further enhanced the reduction of H3K9me2 compared with their separate treatment (Fig. S2D–F). These results demonstrated that DAC and BIX simultaneously engaged DNMT1 and G9a, respectively, under combination conditions, thereby providing direct evidence for the synergistic epigenetic inhibition underlying their combined anti-TNBC activity.

Fig. 1. Effects of DAC and BIX on the proliferation, migration, and invasion of TNBC cells.

A The inhibitory effects of decitabine (DAC) and BIX-01294 (BIX) monotherapies on the proliferation of 4T1, MDA-MB-231, and MDA-MB-468 cells were evaluated using the CCK-8 assay (n = 3). The absorbance values measured by the CCK-8 assay were used to determine the inhibitory effects of the compounds on cell proliferation, and the IC₅₀ values for each compound were calculated and presented in the respective panels. B The percentage changes in cell proliferation relative to the control group under various concentrations of DAC and BIX treatments were quantified (n = 3). C The Chou-Talalay analysis was used to assess the effects of DAC and BIX combination treatments at different concentrations. The fraction affected (Fa) values were determined using the CCK-8 assay, and the combination index (CI) values across different Fa ranges were calculated using CompuSyn software. D 4T1 cells were treated with 100 nM DAC and/or 1 μM BIX for 7 days. MDA-MB-231 cells were treated with 50 nM DAC and/or 100 nM BIX for 14 days. MDA-MB-468 cells were treated with 10 nM DAC and/or 10 nM BIX for 14 days. Crystal violet staining was conducted to visualize and quantify cellular colonies (n = 3). E TNBC cells seeded in 6-well plates were treated with 10 μL pipette tips to generate scratches, followed by treating with mitomycin C. Cells within each well were then treated with 1 μM DAC and/or 1 μM BIX for 48 h (n = 3). The wound healing rates were calculated as follows: [(0 h scratch width - 48 h scratch width)/0 h scratch width] × 100%. F TNBC cells were seeded into transwell inserts, followed by treating with 1 μM DAC and/or 1 μM BIX for 48 h. The number of migrated cells per field on the bottom of transwell inserts were quantified (n = 3). G TNBC cells were seeded into transwell inserts which were pre-coated with Matrigel matrix, followed by treating with 1 μM DAC and/or 1 μM BIX for 48 h (n = 3). The number of invaded cells per field on the bottom of transwell inserts were determined. Data in (A, B) and D–G were presented as mean ± SD in triplicate. Scale bars in (E) represented 500 μm, while scale bars in (F, G) represented 100 μm. All data were obtained from three independent experiments. P values were calculated using one-way ANOVA. ns, p > 0.05; *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001.

To comprehensively characterize anti-tumor effects of DAC, BIX, and their combination, we performed a series of functional assays to evaluate their effects on TNBC cells. In colony formation assays, different concentration combinations were employed to facilitate long-term monitoring. 4T1 cells were treated with 100 nM DAC and/or 1 μM BIX for 7 days. MDA-MB-231 cells were treated with 50 nM DAC and/or 100 nM BIX for 14 days. MDA-MB-468 cells were treated with 10 nM DAC and/or 10 nM BIX for 14 days. Both DAC and BIX significantly suppressed the colony-forming ability of 4T1, MDA-MB-231, and MDA-MB-468 cells, while the inhibitory effect of the combination treatment was more pronounced (Fig. 1D). Furthermore, mono-treatment with either 1 μM DAC or 1 μM BIX for 48 hours (h) significantly delayed the migration of TNBC cells, as demonstrated by wound healing scratch assays (Fig. 1E). The co-treatment of DAC and BIX further enhanced the inhibition of TNBC cell migration. Additionally, the inhibitory effects of 1 μM DAC and 1 μM BIX treatments for 48 h, as well as their synergistic inhibitory effects on TNBC cell migration were confirmed by transwell migration assays (Fig. 1F). Given that metastatic TNBCs are characterized by high invasion ability, our transwell invasion assays demonstrated that the mono-treatment with either 1 μM DAC or 1 μM BIX for 48 h significantly reduced the invasion capabilities of TNBC cells, while the combination treatment exhibited a significantly superior inhibitory effect compared to individual treatments (Fig. 1G). These results indicated that both the DNMT1 inhibitor DAC and the G9a inhibitor BIX effectively suppressed the proliferation, migration, and invasion of TNBC cells. Importantly, the combination treatment with DAC and BIX exhibited a synergistic effect that significantly enhanced anti-TNBC efficacy, demonstrating that both DNA and histone methylations epigenetically contributed to TNBC progression.

Decitabine and BIX-01294 induced apoptosis and lipid peroxidation in TNBC cells

DAC and BIX have been found to induce apoptosis of human leukemia cells [54, 55]. To investigate whether apoptosis was involved in the anti-TNBC activity of DAC and BIX, we treated 4T1, MDA-MB-231, and MDA-MB-468 cells with DAC and BIX individually or in combination and assessed the percentages of apoptotic cells via Annexin V and PI staining. Cells positive for Annexin V alone indicated early apoptosis, while cells positive for both Annexin V and PI were indicative of late apoptosis. We calculated the apoptotic rates by quantifying the proportions of cells in both early and late apoptosis. Both 2 μM DAC and 2 μM BIX significantly induced apoptosis in all three TNBC cell lines upon treatment for 48 h, with a more pronounced apoptotic effect observed in the combination treatment (Fig. 2A, B). Additionally, the treatment of these compounds in these cells led to an elevation in cleaved caspase-3 fragments, indicating the activation of caspase-3 and apoptosis (Fig. S3A–C).

Fig. 2. DAC and BIX induced apoptosis and promoted lipid ROS accumulation in TNBC cells.

A The effects of 2 μM DAC, 2 μM BIX, or their combination on apoptosis induction in 4T1, MDA-MB-231, and MDA-MB-468 cells were evaluated by flow cytometry after 48-h treatment. B The percentage of apoptotic cells was determined as the sum of Annexin V single-positive and Annexin V/PI double-positive cells (n = 3). C–E TNBC cells were treated with 1 μM DAC and/or 1 μM BIX for 48 h, followed by the treatment with 2 μM C11-BODIPY dye for 30 min. The mean fluorescence intensity (MFI) and percentage of C11-BODIPY (FITC)-positive cells were quantified by flow cytometry at 48 h post-treatment. Cells treated with 1 μM erastin as a positive control to induce lipid peroxidation. Lipid ROS levels were represented as percentages of FITC-positive cells (n = 3). F–I TNBC cells were treated with DAC, BIX, and C11-BODIPY as in (C–E), followed by imaging with a confocal microscope. Hoechst dye was used for DNA staining. The PE-positive C11-BODIPY indicated non-oxidized lipids, while the FITC-positive C11-BODIPY represented oxidized lipids (F, G). The MFIs of both oxidized (FITC) and non-oxidized (PE) C11-BODIPY were quantified for DMSO-and DAC-treated cells (H, I). J, L TNBC cells were treated with 1 μM DAC, 1 μM BIX, or their combination for 48 h, in the absence or presence of 10 μM Ferrostatin-1 (Fer-1). Relative lipid ROS levels were measured using C11-BODIPY-based flow cytometry (n = 3). Data in (B–E) and (H–L) were presented as mean ± SD in triplicate from three independent experiments. Scale bars in (F, G) represented 20 μm. P values in (B–E) and H–L were calculated using one-way ANOVA and two-way ANOVA, respectively. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001.

Previous studies have indicated that DAC can exert its effects by inducing the accumulation of reactive oxygen species (ROS) in leukemia cells [56]. The damage to both the plasma membrane and mitochondria leads to an increase in ROS, thus enhancing apoptosis-related lipid peroxidation. Therefore, lipid ROS directly reflects the damage induced by apoptosis, which can be commonly detected using the lipid peroxidation probe C11-BODIPY. To verify whether DAC and BIX also inducee lipid peroxidatioin in TNBC cells, we measured the accumulation of lipid ROS in DAC- and/or BIX-treated cells via C11-BODIPY staining. The C11-BODIPY is a redox-sensitive dye that integrates into the plasma membrane and changes its fluorescence from red (PE-positive) to green (FITC-positive) upon oxidation. The results showed that both 1 μM DAC and 1 μM BIX treatments for 48 h significantly increased lipid ROS levels in TNBC cells, as evidenced by enhanced mean fluorescence intensity (MFI) and higher percentages of C11-BODIPY (FITC)-positive cells (Fig. 2C–E). The combination treatment with DAC and BIX demonstrated significantly enhanced effects on lipid peroxidation in TNBC cells, comparable to those induced by 1 μM erastin, a well-established inducer of lipid peroxidation. These results were also confirmed by lipid ROS-based immunofluorescence. DAC treatment led to a significant accumulation of lipid ROS in TNBC cells, demonstrated by higher levels of oxidized C11-BODIPY (FITC) compared to dimethyl sulfoxide (DMSO)-treated groups (Fig. 2F–I).

Given that labile Fe²⁺ accumulation and lipid peroxidation are key indicators of oxidative stress-mediated cell damage, we performed FerroOrange assays to detect intracellular Fe²⁺ upon compound treatments. FerroOrange is a fluorescent probe widely used to detect Fe²⁺. When binding to Fe²⁺, FerroOrange exhibits an increased orange-red fluorescence signal. The results showed that both BIX and DAC treatments led to significant accumulation of intracellular Fe²⁺, as evidenced by high levels of FerroOrange intensity (Fig. S4A–C). Furthermore, the combined treatment with DAC and BIX demonstrated significantly enhanced Fe²⁺-positive FerroOrange intensity compared to the individual treatments. These results indicated that both DAC and BIX treatments elevated intracellular labile Fe²⁺ levels, consequently driving Fenton chemistry to produce ROS and promoting lipid peroxidation. To further elucidate the functional role of lipid peroxidation in TNBC inhibition, we utilized RSL3 assays to directly induce lipid peroxidation. RSL3 directly inhibits GPX4, resulting in the accumulation of lipid peroxides. Consistently, RSL3 dose-dependently inhibited TNBC cell growth with IC₅₀ values ranging from 0.03 to 0.56 μM (Fig. S4D–F). Furthermore, the co-treatment with the lipid peroxidation inhibitor Ferrostatin-1 (Fer-1) effectively inhibited DAC- and/or BIX-induced lipid peroxidation (Fig. 2J–L). These results demonstrated that DAC and BIX promoted lipid peroxidation in TNBC cells by elevating intracellular Fe²⁺ levels and ROS production.

DNMT1 knockout enhanced the sensitivity of TNBC cells to apoptosis and lipid peroxidation

The primary targets of DAC are DNMT enzymes, specifically DNMT1, which maintains methylation patterns during DNA replication. To evaluate the prognostic significance of DNMT1 in TNBC patients, we conducted survival analysis using the Kaplan-Meier Plotter (http://kmplot.com/analysis/) for TNBC and several other cancer types. The results indicated that elevated DNMT1 expression levels were significantly correlated with poorer overall survival (OS) in TNBC patients, particularly in those with high-grade tumors (Fig. 3A). Furthermore, higher DNMT1 expression was also associated with poorer prognosis across multiple cancer types, including lung cancer, liver cancer, acute myeloid leukemia (AML), multiple myeloma, kidney renal papillary cell carcinoma (KIRP), and various types of sarcoma (Fig. S5A). These findings suggested that DNMT1 could serve as a potential biomarker for poor prognosis in TNBC patients. Additionally, DNMT1 was highly expressed in mouse-derived 4T1 cells compared to HC11 normal-like epithelial cells (Fig. S5B). Similarly, human MDA-MB-231 and MDA-MB-468 cells exhibited higher DNMT1 expression levels compared to normal breast MCF10A epithelial cells (Fig. S5C, D).

Fig. 3. DNMT1 knockout sensitized TNBC cells to apoptosis and lipid peroxidation.

A Kaplan-Meier survival analysis was conducted using the breast cancer database to evaluate overall survival (OS) in total and Grade 3 TNBC patients based on DNMT1 mRNA expression levels. The hazard ratios (HR) with 95% confidence intervals and logrank P values were also calculated. B Wild-type (WT) and DNMT1-knockout (KO) MDA-MB-231 cells were treated with DAC and/or BIX in various concentrations. Percentages of inhibition and IC₅₀ values for DAC and/or BIX were calculated for both cell types (n = 3). C, D Apoptosis in WT and DNMT1-KO MDA-MB-231 cells was analyzed by flow cytometry after treatment with LCL161 (0.1 μM) or Cisplatin (50 μM) for 48 h (n = 3). Apoptotic cells were defined as the sum of Annexin V single-positive cells and Annexin V/PI double-positive cells. Representative flow cytometry plots (left), quantification (middle), as well as western blot analyses of both full-length and cleaved caspase-3 (right) were shown. E, F Lipid ROS levels in erastin (1 μM)-treated WT and DNMT1-KO MDA-MB-231 cells were determined by analyzing FITC-positive C11-BODIPY portions. The PE-positive C11-BODIPY represented non-oxidized lipids (n = 3). The MFIs of both oxidized and non-oxidized lipids were quantified. G RNA-Seq analysis was performed for WT and DNMT1-KO MDA-MB-231 cells in biological duplicates. The heatmap illustrated the differentially expressed genes between two groups, with a color scale representing the fold change in gene expression levels. A table adjacent to the heatmap listed representative upregulated and downregulated genes. H, I Pathway analysis of downregulated and upregulated genes upon DNMT1 KO in MDA-MB-231 cells was performed using the PANTHER classification system, highlighting 10 representative signaling pathways. J, K RT-qPCR validation confirmed the differential expression of representative genes identified by RNA-Seq (n = 3). L Western blot analyses of c-Fos, GPX4, DAPK1, and KISS1 within both WT and KO MDA-MB-231 cells. Data in (B–F) and J, K were presented as mean ± SD in triplicate from three independent experiments. Scale bars in (F) represented 20 μm. P values were calculated using Student’s t-test. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001.

To explore the impact of DNMT1 on gene expression and compound sensitivity in TNBC cells, we utilized CRISPR/Cas9 gene editing technology to knock out the endogenous DNMT1 gene in MDA-MB-231 cells. MDA-MB-231 cells are inherently heterogeneous [57–59]. Therefore, to avoid the random selection of monoclonal cell lines, we generated DNMT1-KO cell lines from pooled populations. Both western blot and immunofluorescence assays with antibodies against endogenous DNMT1 confirmed the successful and complete deletion of DNMT1 (Fig. S5E, F). Subsequently, we investigated the effect of DNMT1 deletion on DAC and/or BIX sensitivity. The results showed that the absence of DNMT1 markedly reduced the sensitivity of MDA-MB-231 cells to DAC, with IC₅₀ values increasing from 144.21 nM in wild-type (WT) cells to 963.83 nM in DNMT1 knockout (KO) cells (Fig. 3B). The IC₅₀ values of BIX increased from 1.95 μM to 2.23 μM upon DNMT1 KO, while the IC₅₀ values of DAC/BIX combination increased from 1.82 μM to 3.53 μM. Notably, the sensitivity of DNMT1-KO MDA-MB-231 cells to DAC decreased by almost 7-fold, while the sensitivity of KO cells to BIX and the DAC/BIX combination decreased by only 1- to 2-fold. These results further supported the notion that DNMT1 was the main target of DAC. DAC and BIX synergistically suppressed TNB cell growth by inhibiting DNMT1 and G9a, respectively.

Moreover, we investigated the changes in apoptosis sensitivity and lipid peroxidation levels in MDA-MB-231 cells following DNMT1 KO. The results showed that DNMT1 deficiency significantly increased the sensitivity to apoptosis, as evidenced by higher proportions of apoptotic cells and elevated lipid ROS levels in DNMT1 KO cells upon treatment with apoptosis inducers LCL161 (0.1 μM) and Cisplatin (50 μM) as well as the lipid peroxidation inducer erastin (1 μM) for 48 h (Fig. 3C–F). The treatment of LCL161 and Cisplatin in DNMT1-KO MDA-MB-231 cells significantly increased the levels of cleaved caspase-3 (Fig. 3C, D). Additionally, the re-expression of DNMT1 in KO cells significantly decreased the sensitivity of MDA-MB-231 cells to LCL161, Cisplatin, and erastin, as evidenced by the lower percentages of apoptotic cells and lower levels of lipid peroxidation in DNMT1-re-expressed cells (Fig. S5G–I). These findings suggested that DNMT1 could influence the survival of TNBC cells by modulating cell death pathways such as apoptosis and lipid peroxidation.

To further reveal the global impact of DNMT1 deletion on gene expression in TNBC cells, we performed RNA sequencing (RNA-Seq) on both WT and DNMT1-KO MDA-MB-231 cells. Using a significance threshold of FDR < 0.05, we identified 935 differentially expressed genes across two independent experiments, with 366 genes upregulated and 569 genes downregulated (Fig. 3G). Specifically, among the downregulated genes, several key proto-oncogenes associated with TNBC were identified, including FOS and PIM1, as well as several antioxidant stress regulatory genes, such as glutathione peroxidase 4 (GPX4) and NFE2-like BZIP transcription factor 2 (NFE2L2). Among the upregulated genes, significant changes were observed in multiple tumor suppressor genes (TSGs) relevant to TNBC, including death-associated protein kinase 1 (DAPK1), dual-specificity phosphatase 4 (DUSP4), and KiSS-1 metastasis suppressor (KISS1).

The unbiased analysis of differentially expressed genes revealed that negative regulation of cell killing, developmental programmed cell death, positive regulation of senescence, and additional pathways were significantly upregulated upon DNMT1 KO in MDA-MB-231 cells. Conversely, the cGMP metabolic process, cyclic nucleotide metabolic process, mitochondrial transport along microtubule, and many other functional pathways were significantly downregulated (Fig. S6A). Notably, the biased analysis revealed that the downregulated genes were enriched in 23 functional categories, predominantly related to signaling pathways promoting tumor growth and metastasis, including the Ras pathway, VEGF signaling pathway, and Wnt signaling pathway (Fig. 3H). Conversely, the upregulated genes were associated with 22 functional categories, primarily involving pathways that inhibited tumor growth, such as apoptosis, oxidative stress response, interferon-γ (IFN-γ) signaling pathway, and ubiquitin-proteasome pathway (Fig. 3I). Both downregulated and upregulated genes formed extensive STRING networks with an interaction confidence score of 0.7 (Fig. S6B, C). Protein Class (PC) analysis revealed that both protein modifying and metabolite interconversion enzymes, as well as gene-specific transcriptional regulators and modulators, were influenced upon DNMT1 KO (Fig. S6D, E). To validate the differential expression levels of these genes upon DNMT1 KO, we performed reverse transcription-quantitative polymerase chain reaction (RT-qPCR) assays to compare mRNA expression levels between WT and DNMT1-KO MDA-MB-231 cells. The results were highly consistent with the RNA-seq data (Fig. 3J, K). Additionally, the western blot assays confirmed that the protein levels of both c-Fos and GPX4 decreased upon DNMT1 KO, while the protein levels of DAPK1 and KISS1 increased when DNMT1 was knocked out in MDA-MB-231 cells (Fig. 3L). These findings further supported the role of DNMT1 as a critical epigenetic regulator in TNBC cells.

DNMT1 formed distinct nuclear bodies in TNBC cells

DNMT1 is the largest DNA methyltransferase, containing an N-terminal regulatory domain and a C-terminal catalytic domain (Fig. S7A) [60]. The N-terminal domain harbors two intrinsically disordered regions (IDRs) designated as disordered segment 1 (DS1) and DS2 (Fig. S7A, B). Both mouse and human DNMT1 proteins exhibit high conservation, with more than 75% sequence identity, particularly within the N-terminal and C-terminal domains (Fig. S7C) [61]. Previous studies have shown that proteins containing IDRs can form liquid-liquid phase-separated (LLPS) condensates through weak multivalent interactions mediated by these regions [62]. Notably, endogenous DNMT1 formed hundreds of nuclear condensates in TNBC cells as well as common kidney cell-derived HEK293T cells, whereas exogenously expressed DNMT1 formed tens to hundreds of nuclear condensates, with diameters slightly larger than endogenous condensates (Figs. 4A, B and S8A–H). We also examined endogenous DNMT1 in the normal breast epithelial cell line MCF10A by immunofluorescence staining (Fig. S8I). Although MCF10A cells did form DNMT1 nuclear condensates, with an average diameter of approximately 200 nm, the number of condensates per cell was markedly lower than that observed in TNBC cells (Fig. S8J, K). This disparity was likely attributable to the substantially lower expression levels of DNMT1 in normal breast epithelial cells compared with TNBC cells (Fig. S5B–D). The reduced DNMT1 protein abundance in normal cells would limit the local concentration required to drive multivalent interactions and condensation, thereby resulting in fewer nuclear bodies. Conversely, the elevated DNMT1 expression in TNBC cells facilitated robust condensate formation, which could create a cancer-specific dependency on DNMT1 for maintaining oncogenic transcriptional programs. Utilizing 3D-stack super-resolution structured illumination microscopy (SIM), we characterized the size and sphericity distributions of these condensates (Fig. 4C). The results demonstrated that the average diameter of DNMT1 bodies was approximately 400 nm, ranging from 200 to 600 nm (Fig. 4D). Additionally, these DNMT1 bodies exhibited high sphericity, with a median sphericity value of 0.95 (Fig. 4D).

Fig. 4. DNMT1 formed nuclear condensates in TNBC cells.

A, B Super-resolution SIM images of endogenous and exogenously-expressed mouse or human DNMT1 proteins in 4T1, MDA-MB-231, MDA-MB-468, and HEK293T cells. DAPI was used for DNA staining. C GFP-tagged DNMT1 was overexpressed in HEK293T cells, followed by 3D-stack SIM imaging to visualize the distribution of GFP-DNMT1 bodies within the nucleus. D Rendered 3D shapes of GFP-DNMT1 bodies were shown. The three panels represented the XY, XZ, and YZ planes of a representative GFP-DNMT1 body. The diameter and sphericity distributions of GFP-DNMT1 bodies were quantified via Imaris software. E The GFP-tagged truncated DNMT1 IDR mutant (GFP-DNMT1-IDR) was visualized via SIM imaging. F The DNMT1 mutant harboring DS1 only was conjugated with a GFP tag and imaged via SIM imaging. G, H The GFP-tagged DNMT1-DS1 and full-length DNMT1 were introduced with a Q-to-A mutation at position 180 (Q180A), respectively, followed by SIM imaging to indicate their distributions in HEK293T cells. Data in (D) were collected from three independent experiments. Scale bars in (A–C) and E–H represented 5 μm, while scale bars in (D) represented 0.2 μm.

Subsequently, we attempted to identify the key domain or amino acid motif responsible for DNMT1 condensate formation. SIM imaging showed that the truncated DNMT1 IDR mutant, which harbored both DS1 and DS2, still formed nuclear bodies, albeit with slight differences in number and size compared to WT DNMT1 (Fig. 4E). Notably, the DS1-only mutant also formed nuclear bodies, exhibiting nearly identical number and size distributions to those of WT DNMT1 (Fig. 4F). DS1 harbors several functional domains, encompassing the DNMT1-associated protein 1 binding domain (DMAP1BD), PCNA-interacting protein motif (PIP), and nuclear localization signal (NLS) (Fig. S7A). Previously, we have demonstrated that PIP within IDR of CHAF1A contributes to CHAF1A condensation [49]. Mutation of the glutamine residue (Q) within PIP to an alanine residue (A) disrupted IDR-mediated body formation. We speculated that the PIP motif within DS1 might serve as the driving force for DNMT1 condensation. Therefore, we introduced a Q-to-A mutation at position 180 (Q180A) within PIP and assessed its effect on body formation (Fig. S8L, M). The results showed that the Q180A mutation dissolved DS1-mediated condensation (Fig. 4G). Additionally, we constructed a full-length DNMT1 with the Q180A mutation, which also led to the dissolution of DNMT1 bodies (Figs. 4H and S8L, M). Collectively, our findings revealed that DNMT1 formed distinct nuclear bodies in TNBC cells, driven by multivalent interactions mediated by the PIP motif.

DNMT1 bodies were liquid-liquid phase-separated condensates

Previously, we have shown that the PIP motif promotes the LLPS of CHAF1A [49]. Based on our above results, we speculated that DNMT1 bodies might also exhibit LLPS properties. The basic characteristics of LLPS proteins include spherical and membraneless structure with high sphericity, internal and boundary diffusion, as well as sodium salt concentration- and protein concentration-dependent droplet formation. The properties of internal and boundary diffusion can be verified by the fluorescence recovery after photobleaching (FRAP) assay. To this end, we conducted live-cell FRAP assays on GFP-tagged DNMT1 bodies in three randomly selected MDA-MB-468 cells. Two condensates, including bleached and unbleached ones, were designated for analysis within each cell. The results showed that upon 488 nm laser-induced photobleaching to quench the fluorescence of GFP-DNMT1 bodies, these bleached bodies rapidly recovered fluorescence intensities within approximately 5 to 10 seconds (sec), with a half-time recovery (T_1/2) of 1.996 sec (Fig. 5A–C). Similarly, the FRAP assay in HEK293T cells revealed that bleached GFP-DNMT1 bodies quickly recovered fluorescence intensities with a T_1/2 of 1.024 sec (Fig. S8N–P). Furthermore, we utilized the in vitro purified recombinant proteins to validate their concentration-dependent properties. Our results demonstrated that as the sodium chloride (NaCl) concentrations decreased, GFP-tagged DNMT1-DS1 formed more and larger liquid-like droplets (Fig. 5D, E). Additionally, these recombinant proteins also exhibited an increase in both the number and size of droplets, as the protein concentrations increased in a NaCl concentration of 31.25 mM (Fig. 5F, G).

Fig. 5. DNMT1 condensates exhibited LLPS properties.

FRAP imaging demonstrated the internal diffusion characteristics of DNMT1 bodies. GFP-tagged DNMT1 was overexpressed in MDA-MB-468 cells, and the live-cell time-lapse imaging was performed at 24 h post-transfection. Three cells were randomly selected, and two GFP-DNMT1 bodies within each cell were designated as regions of interest (ROIs). ROI 1 bodies were photobleached using a strong 488 nm laser pulse, while ROI 2 bodies were left unbleached as controls. Images were captured at 1.89-sec intervals for 60 sec (A). Fluorescence intensity at each time point was recorded and analyzed via ZEN 3.9 software (B). The half-time of recovery (T_1/2) was calculated based on the time required for photobleached ROIs to recover halfway between initial and steady-state fluorescence intensities (n = 3) (C). Approximately 10 μM of purified recombinant GFP-DNMT1-DS1 proteins were incubated with droplet formation buffers containing various NaCl concentrations which included 15.625, 31.25, 62.5, 125, 250, and 500 mM, followed by imaging with SIM (D). The scatter plot on the right panel presented statistical analyses of droplet areas and numbers under different NaCl concentrations (E). Purified GFP-DNMT1-DS1 proteins at different concentrations, which included 0.78125, 1.5625, 3.125, 6.25, 12.5, and 25 μM, were incubated with droplet formation buffer containing 31.25 mM NaCl (F). The scatter plot on the right represented distributions of droplet areas and numbers based on SIM imaging (G). H–K GFP-tagged recombinant DNMT1-DS1-Q180A and DNMT1-DS1-8MUT proteins were visualized and analyzed as in (D, E). L–M GFP-DNMT1-DS1 proteins (12.5 μM) in 31.25 mM NaCl were co-treated with 5% 1,6-hexanediol (1,6-Hex) for 30 min. SIM images of droplets were visualized and analyzed as in (E). N GFP-tagged DNMT1-8MUT proteins were overexpressed in HEK293T cells and imaged with SIM to reveal their distributions. DAPI was used to stain DNA. All images were collected from three independent experiments. Representative images were presented for each experiment. Scale bars in (A, N) represented 5 μm, while scale bars in (D, F, H, J, L) represented 10 μm. ‘ND’ in (E, I, K) indicates ‘not detected’.

Our results showed that Q180 within the PIP motif of DNMT1-DS1 contributed to the condensate formation of DNMT1 in vivo (Fig. 4). Therefore, we also purified Q180A-mutated DS1 mutant (DNMT1-DS1-Q180A) to evaluate their ability of droplet formation. The result revealed that the introduction of Q180A significantly decreased both the number and size of liquid-like droplets (Fig. 5H, I). Specifically, only a few droplets formed at NaCl concentrations of 15.625, 31.25, and 62.5 mM. Additionally, the entire PIP motif mutants with eight mutations from positions 180 to 187 (QTTITSHF to AAAAAAAA) (DNMT1-DS1-8MUT) completely abolished droplet formation (Fig. 5J, K). The 1,6-hexanediol (1,6-Hex) compound in sufficient concentration has been widely used to disrupt phase-separated condensates. Therefore, we evaluated the sensitivity of DNMT1 bodies to 1,6-Hex. We co-treated 12.5 μM of GFP-DNMT1-DS1 proteins in 31.25 mM NaCl with 5% 1,6-Hex for 30 min. The immunofluorescent imaging revealed that GFP-tagged DS1 droplets were dissolved upon 1,6-Hex treatment, further confirming their LLPS characteristic (Fig. 5L, M). Consistent with these findings, the full-length GFP-tagged DNMT1-8MUT also failed to form nuclear condensates in HEK293T cells, similar to the result observed for DNMT1-Q180A (Fig. 5N). We also performed 1,6-Hex experiments in live cells. Surprisingly, we did not observe the complete disappearance of GFP-DNMT1 bodies after 1,6-Hex treatment (Fig. S8Q). During long-term imaging, the numbers and sizes of GFP-DNMT1 bodies gradually decreased rather than disappeared. Although 1,6-Hex has been extensively used in LLPS research, its effects have been reported to be context-dependent and vary between in vitro and in vivo systems [63–65]. Furthermore, DNMT1 condensates could be stabilized by additional interactions beyond the homotypic DS1-mediated forces that drive droplet formation in vitro. DNMT1 binds DNA through its CXXC domain and catalytic region, associates with chromatin-related proteins such as PCNA and UHRF1, and resides within a crowded nuclear environment. These factors could collectively reinforce condensate integrity and render DNMT1 bodies less susceptible to 1,6-Hex than isolated DS1 droplets in a simplified buffer system. This discrepancy therefore suggests that DNMT1 LLPS in cells is governed not only by the intrinsic phase separation propensity of its disordered regions but also by extrinsic stabilizing factors, a scenario increasingly recognized in the condensate field. Taken together, our results demonstrated that DNMT1 bodies were LLPS condensates. Mutations on the PIP motif of DS1 disrupted intermolecular multivalent interactions, consequently leading to the dissolution of DNMT1 bodies.

Decitabine inhibited the formation of DNMT1 condensates

Previous studies have revealed that DAC irreversibly binds to the C-terminal catalytic domain of DNMT1, leading to the trapping and degradation of DNMT1 on DNA [53]. Our above results have indicated that DAC at concentrations of IC₅₀ and IC₉₀ also moderately decreased DNMT1 levels in 4T1, MDA-MB-231, and MDA-MB-468 cells after treatment for 3 or 7 days (Figs. S1 and S2). To evaluate the impact of DAC on DNMT1 LLPS, we treated 4T1, MDA-MB-231, MDA-MB-468, and HEK293T cells with 2 μM DAC for 24 h. The results revealed that DAC significantly suppressed both endogenous and exogenous DNMT1 condensates (Fig. 6A, B). Nevertheless, DAC only reduced endogenous DNMT1 levels instead of achieving complete elimination (Fig. S8R–U). The protein expression levels of exogenous DNMT1 did not exhibit a significant decline after a 1-day DAC treatment (Fig. S8V–Y). Our previous results have demonstrated that the DS1 region within IDR drives the LLPS of DNMT1 (Fig. 5). To evaluate whether DAC inhibited DNMT1 LLPS by directly interacting with DS1, we co-incubated 10 μM of GFP-tagged recombinant DNMT1-DS1 proteins with 10 μM DAC for 10 minutes (min). The in vitro LLPS assay showed that the addition of DAC completely abolished the phase separation of DNMT1-DS1 at NaCl concentrations of 125, 250, and 500 mM (Fig. 6C, D). Only a small number of DNMT1-DS1 droplets with reduced volumes formed at low NaCl concentrations of 15.625, 31.25, and 62.5 mM. To further distinguish whether DAC disrupted DNMT1 condensates through DS1 binding rather than DNA entrapment with the C-terminal catalytic domain, we expressed GFP-DNMT1-DS1 in HEK293T cells. Treatment with 2 μM DAC for 24 h markedly dissolved GFP-DNMT1-DS1 condensates, supporting that DAC disrupted DNMT1 LLPS through direct DS1 engagement, independent of its DNA-demethylating activity (Fig. 6E).

Fig. 6. DAC inhibited the LLPS of DNMT1 condensates.

4T1, MDA-MB-231, MDA-MB-468, and HEK293T cells were treated with 2 μM DAC for 24 h. Endogenous DNMT1 was labeled with FITC-conjugated anti-DNMT1 antibody and visualized by SIM imaging. 4T1, MDA-MB-231, MDA-MB-468, and HEK293T cells were transfected with GFP-DNMT1 for 24 h, followed by treatment with 2 μM DAC for an additional 24 h. GFP-DNMT1 was visualized by SIM imaging. DAPI was used for nuclear counterstaining in (A) and (B). Approximately 10 μM of in vitro purified GFP-DNMT1-DS1 proteins were co-incubated with 10 μM DAC in droplet formation buffers containing various NaCl concentrations for 10 min (C). Areas of GFP-DNMT1-DS1 droplets within each buffer condition were quantified (D). E HEK293T cells were transfected with GFP-DNMT1-DS1 constructs, followed by the treatment with 2 μM DAC for 24 h. The subcellular distribution and aggregation of GFP-DNMT1-DS1 were visualized. DAPI was used for nuclear counterstaining. F The crystal structure of human DNMT1 (residues 351-1600, PDB ID: 4WXX), which harbored the catalytic domain, was used as the receptor. DAC was covalently docked to the catalytic pocket utilizing Schrödinger software. The structure of DNMT1-DS1 region was predicted by AlphaFold 3 (AF3). DAC was blindly docked to DS1 via AutoDock 4.2 software. Eight clusters with binding energies lower than -3.00 kcal/mol were selected and indicated in (G). Values of binding energies and inhibition constants of these potential binding sites were shown in (H). I DAC was docked to cluster 1, which overlapped with the PIP motif of DNMT1. The molecular docking mode showed that two hydrogen bonds were formed between hydroxyl groups on the deoxyribose ring of DAC and residues S177 and T178 of DNMT1. Two additional hydrogen bonds were formed between the carbonyl group and amino group on the pyrimidine ring of DAC and the side chains of R179 and Q180 of DNMT1. J Surface plasmon resonance (SPR) analysis of the direct binding interaction between DAC and GFP-tagged DNMT1-DS1 protein. Representative sensorgram and kinetic fitting revealed a dissociation constant (K_D) of 6.64 μM. K Microscale thermophoresis (MST) analysis of the binding interaction between DAC and GFP-tagged DNMT1-DS1 protein. The dissociation constant (K_D) was determined to be 4.07 ± 2.08 μM. Data in (K) were presented as mean ± SD (n = 4). Scale bars in (A, B) and E represented 5 μm, while scale bars in (C) represented 10 μm. ‘ND’ in (D) indicated ‘not detected’. All images in (A–E) were representative of three independent experiments.

DAC primarily binds to the C-terminal catalytic domain of DNMT1, which was also confirmed by our molecular docking analysis (Fig. 6F). As the N-terminal domain structure of DNMT1 has not yet been resolved, we utilized AlphaFold 3 (AF3) to predict the possible structure of the DS1 region. The model with the highest-ranking score of 0.68 from the prediction results was chosen as the receptor model. The ranking score was calculated based on the composite score of normalized pLDDT, pTM, ipTM, fraction disordered, and clash score. Molecular docking analysis of the predicted DS1 structure with DAC revealed that eight clusters could be the potential binding sites of DAC, with low binding energies ranging from –7.07 to –3.38 kcal/mol and inhibition constants ranging from 6.610 to 3350 μM (Fig. 6G, H). Among these sites, cluster 1, which overlapped with the PIP motif of DNMT1, exhibited the lowest binding energy (–7.07 kcal/mol) and inhibition constant (6.610 μM). The molecular docking analysis indicated that two hydroxyl groups on the deoxyribose ring of DAC formed hydrogen bonds with S177 and T178 of DNMT1 respectively, while the carbonyl group and amino group on the pyrimidine ring formed two hydrogen bonds with the side chains of R179 and Q180 respectively (Fig. 6I). To experimentally validate the direct interaction between DAC and the DS1 region, we performed both surface plasmon resonance (SPR) and microscale thermophoresis (MST) assays. The results showed that DAC bound to DNMT1-DS1 proteins with K_D values of 6.64 μM in the SPR assay and 4.07 μM in the MST assay, both confirming binding in the low micromolar range (Fig. 6J, K). The slight difference in K_D values likely reflects the inherent differences between the two assay formats, as SPR measures binding on an immobilized surface whereas MST detects interactions in solution. Altogether, our above results revealed that DAC inhibited DNMT1 LLPS through direct binding to the DS1 region, independent of the DAC-induced degradation of DNMT1 proteins.

Decitabine and BIX-01294 synergistically inhibited TNBC lung metastasis

Our above results have demonstrated the synergistic inhibitory effects of DAC and BIX on TNBC cells in vitro. To further investigate the therapeutic effects of DAC and BIX in advanced TNBC in vivo, we established a 4T1 TNBC lung metastasis model in BALB/c mice. The mouse model was constructed by the injection of luciferase-transduced 4T1 (4T1-luc) cells via the tail vein. Successful establishment of lung metastasis was confirmed on the day of injection using bioluminescence imaging (Fig. 7A). Subsequently, mice were randomly allocated to four groups with similar initial average body weights (n = 4 per group). Mice in different groups were treated with saline, 2 mg/kg of DAC, 2 mg/kg of BIX, or the combination of DAC and BIX via intraperitoneal injection, five times per week for two weeks. The results showed that 4T1-luc cells rapidly established lung metastasis upon tumor inoculation, while the treatment with DAC significantly inhibited tumor growth compared with the saline control group or BIX treatment (Fig. 7B–D). The administration of DAC and/or BIX did not induce potential toxicity, as evidenced by the consistent body weight change among different groups (Fig. 7B). The combination treatment of DAC and BIX further enhanced the inhibition of 4T1-luc tumor growth compared with DAC or BIX treatment only. Mice's lungs were isolated and proceeded to pathological evaluation. The results showed that the number of lung metastatic nodules in the combined treatment group was significantly reduced compared to the control or mono-compound treatment group (Fig. S9A). Notably, lung weight in DAC-treated and combined treatment groups was significantly lower than that in control and BIX-treated groups, suggesting a reduced tumor burden in mice subjected to DAC or combined treatment (Fig. S9B). The bioluminescence imaging of isolated lungs also showed that pulmonary metastatic lesions in DAC-treated and combined treatment groups were alleviated compared to those in saline- or BIX-treated groups (Fig. S9C, D). Further histopathological and immunohistochemical analyses confirmed that treatment with DAC alone and in combination with BIX reduced lung metastasis burden, characterized by fewer metastatic foci and decreased Ki67-positive proliferating cells (Figs. 7E, F and S9E).

Fig. 7. DAC and BIX inhibited the lung metastasis of TNBC.

A Schematic of 4T1 TNBC lung metastasis model in BALB/c mice. 4T1-luc cells were injected into each female mouse via the tail vein, followed by in vivo bioluminescence imaging to confirm the establishment of lung metastasis on the same day. Mice from each group (n = 4 per group) were intraperitoneally administered with saline, 2 mg/kg of DAC, 2 mg/kg BIX, or a combination of DAC and BIX, respectively, five times per week for two weeks. In vivo imaging was conducted on day 6 and day 14 post-inoculation to evaluate therapeutic effect. All mice were euthanized on day 14 and proceeded to pathological analysis. B Body weight monitoring of BALB/c mice after 4T1-luc cell inoculation. The body weight of each mouse was measured every two days. C Representative images of bioluminescence imaging on day 1, 6, and 14 post-inoculation, which depicted tumor growth in mice from each group. D Radiance quantification of bioluminescence images for each mouse at each time point (n = 4). E Histopathological analysis of mouse lungs was evaluated by hematoxylin and eosin (HE) staining. Dashed circles indicated metastatic foci. F Immunohistochemical (IHC) analysis of mouse lungs was conducted with antibodies against Ki67. G Schematic of MDA-MB-231 TNBC lung metastasis model in NCG mice. MDA-MB-231-luc cells were injected into mice via the tail vein. 2 mg/kg of DAC, 2 mg/kg BIX, and their combinations were intraperitoneally administered mice on day 7, 10, 14, and 17 post-inoculation. In vivo bioluminescence imaging was conducted on day 7, 12, 14, 19, and 21. Mice were euthanized on day 21. H Body weight monitoring of NCG mice after inoculation with MDA-MB-231-luc cells. I Representative bioluminescence images of mice at each time point. J Statistical analysis of mouse radiance at 5 time points (n = 4). K HE staining was conducted to indicate histopathological changes of mouse lung tissues. Metastatic foci were marked with dashed circles. L IHC staining of mouse lung tissues with antibodies against Ki67. Data in (B, D, H, J) were presented as mean ± SD in quadruplicate. Scale bars in (E, F, K, L) represented 100 μm. P values in (D, J) were calculated using one-way ANOVA. *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001.

To evaluate the inhibitory effect of DAC on human TNBC cell-derived tumors, we established the in vivo models by inoculating MDA-MB-231 cells into female NCG mice. Specifically, MDA-MB-231 cells were orthotopically implanted into the mammary fat pad of NCG mice to construct orthotopic tumor models. Seven days post-inoculation, mice were treated with DAC (2 mg/kg) and/or BIX (2 mg/kg) twice a week for three consecutive weeks. On day 25, mice were euthanized, and tumors were isolated for further analysis (Fig. S9F). The results showed that mono-treatment with DAC or in combination with BIX significantly inhibited orthotopic tumor growth without significant toxicity, as indicated by consistent body weight change, reduced tumor volume, delayed tumor progression, and a decrease in Ki67-positive proliferating cells within tumor tissues (Fig. S9G–K). To evaluate whether DAC treatment could also inhibit lung metastasis of human TNBC in vivo, we established MDA-MB-231 cell-derived TNBC lung metastasis models in female NCG mice. Similarly, NCG mice were inoculated with luciferase-expressing MDA-MB-231 (MDA-MB-231-luc) cells via the tail vein. The successful establishment of lung metastasis was confirmed on day 7 post-inoculation via in vivo bioluminescence imaging, followed by treatment with DAC (2 mg/kg) and/or BIX (2 mg/kg) twice a week for two consecutive weeks (Fig. 7G). The results showed that treatment with DAC and/or BIX did not significantly alter the body weight change in lung metastasis model either (Fig. 7H). DAC significantly delayed tumor progression in mouse lungs compared to both control and BIX-treated groups (Fig. 7I, J). Moreover, the combination of DAC and BIX treatment exhibited an enhanced suppression of tumor growth. Treatment with DAC or in combination with BIX significantly reduced the number of metastatic foci and Ki67-positive cells in lung tissues (Figs. 7K, L and S9L). Taken together, our results demonstrated that DAC effectively inhibited lung metastasis in advanced TNBC, while the combination of DAC and BIX exhibited synergistic inhibitory effects on tumor progression.

Discussion

Epigenetic dysregulation is highly correlated with the pathogenesis and progression of cancers. Previous studies have revealed that both DNMT1 and G9a modulate the homeostasis of DNA and H3K9 methylation in TNBC, respectively [15, 23, 24, 36]. The DNMT1 inhibitor decitabine (DAC) and G9a inhibitor BIX-01294 (BIX) can suppress the proliferation of TNBC and many other cancer types. DAC primarily acts on DNMT1 by covalently binding to the C-terminal catalytic domain of DNMT1, resulting in the constraint and degradation of DNMT1 on replicating DNA [53]. However, in our study, we found that DAC was also non-covalently bound to the N-terminal disordered segment 1 (DS1). Specifically, DAC bound to the PCNA-interacting protein motif (PIP), which may disrupt the interaction with PCNA. Furthermore, we revealed that DNMT1 formed LLPS condensates within TNBC cells, while DAC disrupted these condensates, consequently suppressing DNMT1 catalytic efficiency. Of note, while 1,6-hexanediol efficiently dissolved DNMT1-DS1 droplets in vitro, it only partially reduced full-length DNMT1 condensates in live cells. This differential sensitivity has been observed for other phase-separated compartments and likely reflects the contribution of cellular factors, including DNA binding, protein-protein interactions, and macromolecular crowding, that reinforce condensate stability beyond the intrinsic forces driving LLPS in simplified buffer systems [63–65]. Rather than arguing against the LLPS nature of DNMT1 bodies, this observation highlights that condensate integrity in the native nuclear environment is governed by both intrinsic phase separation propensity and extrinsic stabilizing interactions, a dual regulation increasingly appreciated in the field. Based on our findings, we proposed a novel mechanism by which DAC exerted its effects on TNBC by inhibiting DNMT1 LLPS (Fig. 8). In high-metastatic TNBC cells, DNMT1 proteins were highly expressed and formed nuclear condensates, leading to the epigenetic silencing of multiple tumor suppressor genes (TSGs). Consequently, the LLPS of DNMT1 promoted the proliferation, invasion, and metastasis of TNBC. However, treatment with DAC disrupted DNMT1 LLPS, leading to the reactivation of TSGs, thereby delaying TNBC progression.

Fig. 8. Schematic of DAC-mediated inhibition of DNMT1 LLPS.

A Inoculation of TNBC cells in mice induced rapid lung metastasis, leading to lethal progression. While the treatment of DAC alleviated lung metastasis and delayed TNBC progression. Mechanistically, DNMT1 formed LLPS condensates which simultaneously suppressed the expression of multiple TSGs. Consequently, DNMT1 facilitated the proliferation, invasion, and metastasis of TNBC. However, DAC disrupted DNMT1 LLPS condensates and reactivated TSGs, thereby inhibiting the proliferation, invasion, and metastasis of TNBC.

Currently, non-specific chemotherapy remains the predominant treatment for TNBC patients due to the lack of specific biomarkers for this subtype. Therefore, identifying new targets and corresponding therapeutic drugs are urgently needed to improve the prognosis of TNBC. Our study revealed that DNMT1 was significantly overexpressed in TNBC cells. The elevated expression levels of DNMT1 were markedly correlated with shorter overall survival (OS) in TNBC patients. Both low-dose DAC treatment and DNMT1 knockout (KO) significantly enhanced the sensitivity of TNBC cells to apoptosis and lipid peroxidation, thereby inhibiting the proliferation and metastasis of TNBC. These findings suggest that DNMT1 could be a potential biomarker for TNBC, particularly in high-grade tumors. Both DNA methylation and H3K9 methylation, primarily maintained by DNMT1 and G9a respectively, are closely correlated with poor TNBC prognosis [13, 23]. Consistent with these observations, our pharmacological experiments showed that the G9a inhibitor BIX also efficiently inhibited the proliferation, migration, and invasion of TNBC. Furthermore, co-treatment with DAC and BIX synergistically potentiated the anti-TNBC activities, highlighting the combined influences of DNA and histone methylations on TNBC progression.

Numerous studies have demonstrated that the co-treatment of DAC with Cisplatin and PARPi can increase the sensitivities of TNBC cells to these drugs, thereby potentiating the anti-tumor efficacy [40–42]. During DNA replication, DNMT1 is recruited by PCNA or UHRF1 to TSGs. Subsequently, histone modifiers like G9a and HDACs are recruited by DNMT1 to establish and maintain suppressive epigenetic memory [66, 67]. Given that our RNA-Seq data revealed the upregulation of multiple TSGs and tumor-suppressive pathways upon DNMT1 depletion, we speculate that DNMT1 could be one of the pivotal epigenetic modifiers that orchestrate the assembly of multiple modifiers to maintain suppressive microenvironments around TSGs. Targeting DNMT1 could broadly disrupt suppressive epigenetic modifications and reactivate multiple TSGs, thereby effectively inhibiting TNBC proliferation and metastasis.

Cellular proteins have been found to form liquid-liquid phase-separated (LLPS) condensates to promote cancer progression, while most studies focus on transcription factors. Targeting LLPS-associated transcription factors with small molecular compounds has demonstrated significant potential in treating malignant tumors [68]. Recently, the phosphorylated HDAC6 (phospho-HDAC6) has been found to drive aberrant chromatin architecture via LLPS in TNBC [52]. Disrupting phospho-HDAC6 LLPS by Nexturastat A effectively suppresses tumor growth. In this study, we comprehensively elucidated that elevated DNMT1 formed LLPS condensates within TNBC cells, leading to the simultaneous suppression of multiple TSGs, apoptosis-associated genes, and peroxidation regulatory genes. Additionally, our studies revealed that DAC was bound to the PIP motif besides the conventional C-terminal catalytic pocket, leading to the disruption of DNMT1 LLPS and the dissociation from PCNA on replicating DNA. Without the safeguard of DNMT1 condensates in a suppressive microenvironment, sequestered TSGs, apoptosis-associated genes, and peroxidation regulatory genes would eventually be reactivated. Therefore, our study demonstrates the druggable potential of epigenetic modifiers involving LLPS. Developing next-generation drugs targeting the LLPS of epigenetic proteins without degrading them could offer low toxicity and enhanced efficiency.

Searching for universal biomarkers or neoantigens for specific cancer types represents one of the critical challenges in preventing, predicting, and treating cancer progression. Specifically, TNBC patients cannot benefit from well-established endocrine- or HER2-targeted breast cancer therapies due to the absence of relevant targets. However, our study demonstrated that DNMT1 expression levels in TNBC were significantly higher than those in normal breast epithelial cells. Moreover, higher DNMT1 expression in patient-derived TNBC cells was correlated with poorer prognosis. Importantly, the Kaplan-Meier Plotter analysis revealed that elevated DNMT1 expression was also significantly associated with poorer prognosis within multiple cancer types, including lung cancer, liver cancer, AML, KIRP, as well as various types of myeloma and sarcoma. These findings suggest that high DNMT1 expression may serve as a potential universal biomarker for diverse cancers and represent an attractive target for the development of broad-spectrum anti-tumor drugs. Furthermore, given our finding that DAC inhibited DNMT1 activity via an alternative mechanism involving disruption of its LLPS, targeting DNMT1 LLPS could represent a universal therapeutic strategy for treating heterogeneous TNBC as well as other cancer types characterized by high DNMT1 expression.

Although this study has established DNMT1 LLPS as a druggable vulnerability in TNBC and demonstrated the therapeutic potential of combining DAC with BIX, several limitations should be acknowledged. First, the apoptosis assay was conducted at a DAC concentration of 2 μM, which substantially exceeded the IC₅₀ values of 40-400 nM determined for TNBC cell proliferation. At this concentration, DAC alone induced considerable cell death, which could confound the assessment of synergistic effects with BIX. Future studies employing lower and sub-cytotoxic concentrations would help clarify the extent of synergy. Second, while 1,6-hexanediol efficiently dissolved DNMT1-DS1 droplets in vitro, it failed to fully disperse DNMT1 condensates in live cells. This discrepancy suggests that cellular factors beyond simple LLPS, such as protein-protein or protein-DNA interactions, may contribute to condensate stabilization and warrant further investigation. Third, the in vivo experiments were performed with four animals per group without a formal power calculation, and some endpoints exhibited considerable variability. Studies with larger cohorts are needed to confirm the therapeutic efficacy of the DAC and BIX combination. Finally, although our data demonstrated that DAC and BIX promoted lipid peroxidation and Fe²⁺ accumulation, the evidence remains insufficient to establish ferroptosis as the definitive mode of cell death, given that lipid peroxidation can also occur during apoptosis. Future studies incorporating additional ferroptosis-specific endpoints, such as GPX4 dependence or cell viability rescue by iron chelation, will be required to resolve this question.

Conclusion

In summary, our study elucidated a novel mechanism by which DNMT1 contributed to TNBC progression through forming LLPS condensates. DAC disrupted DNMT1 condensates, leading to increased apoptosis and lipid peroxidation in TNBC cells. G9a inhibitor BIX synergistically enhanced the effects of DAC, significantly suppressing TNBC proliferation, migration, invasion, and lung metastasis. Developing therapeutic strategies targeting LLPS of epigenetic modifiers could be a safe and efficient intervention to treat TNBC and other cancers.

Author contributions

XM and XC designed and supervised the study. MH, HP, GQ, SL, QL, TL, FL, YZ, XL, LX, YL, TC, BZ, HR, and SX performed experiments and acquired data. MH, HP, JL, and PL analyzed data. MH, HP, GQ, and XM designed and evaluated the methodology. JL, SX, and XM validated the results of experiments. MH, XM, and XC wrote the manuscript. XM and XC reviewed and revised the manuscript. All authors have read and approved the final version of the manuscript.

Funding

XC discloses support for the research of this work from the National Natural Science Foundation of China (NSFC) [grant number 81802817] and the Guangzhou Basic Research Program Municipal School (Hospital) Joint Funded Foundation and Application Basic Research Project [grant number 202201020115]. XM discloses support for the research of this work from the Guangdong Basic and Applied Basic Research Foundation [grant number 2024B1515020068], the NSFC [grant number 82572540], and Major Projects of Guangzhou National Laboratory [grant numbers GZNL2023A01009 and GZNL2024A01017]. All other authors declare no relevant funding.

Data availability

The RNA-Seq data in this article have been deposited in the Sequence Read Archive (SRA) database and assigned the identifier (PRJNA1187680).

Competing interests

The authors declare no competing interests.

Ethics approval

The Animal Ethics Committees of Guangzhou Medical University Animal Center and Guangzhou National Laboratory Animal Center approved the mouse experiments in this study (Approval Number: 20240675 and GZLAB-AUCP-2023-03-A05).

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41388-026-03971-x.