The mRNA expression of DNMT1 was remarkably higher in TNBC/basal-like subtype (BLS; n=135) than in normal breast tissue (NBT; n=112; p=3.00×10–36) based on Gene Expression Profiling Interactive Analysis 2 (GEPIA2) in Figure 1A . To investigate the DNA demethylation effect of DSF in TNBC cell lines, we tested the protein levels of DNMT1 in human BT-549 and MDA-MB-231 cell lines exposed to DSF at different does for 48 h by western blot. As shown in Figures 1B, C , the DNMT1 expression was significantly reduced by DSF. At transcription level, a consistent phenomenon was observed in human MDA-MB-231 cells and BT-549 cells exposed to the indicated concentrations of DSF for 48 h in Figure 1D . Further experiments showed that the activity of DNMT was also inhibited by DSF in BT-549 and MDA-MB-231 cells exposed to DSF for 48 h ( Figure 1E ).

We explored whether DSF can affect the methylation level of IRF7 by affecting DNMT1, thereby regulating its expression. First, we analyzed the hypomethylation status of IRF7 promoter regions assessed by bisulfite sequencing analysis in the human MDA-MB-231 cells and BT-549 cells exposed to DSF ( Figure 2A ). To unravel the underlying relationship between the expression and the DNA methylation level of IRF7 in BC, we analyzed data from 1,217 patients in TCGA. The tumor samples are divided into high methylation groups (n=24) and low methylation groups (n= 206). In Figure 2B , the violin-box plot revealed that IRF7 mRNA expression was negatively correlated with its DNA methylation level. Gene expression analysis of IRF7 using GEPIA2 database based on the TCGA and GTEx database. Box plots represent the gene expression level in terms of log2 (TPM+1) in the tumor (TNBC/basal-like subtype, red, n=135) and normal breast tissue (gray, n=291) samples, respectively ( Figure 2C ). As presented in Figure 2D , the results showed that there was a negative correlation between IRF7 mRNA expression and DNA-methylated CpG islands in the gene body region, especially the promoter region. The immunoblot results revealed that the protein levels of IRF7 were upregulated in human BT549 cells and MDA-MB-231 cells exposed to DSF at different doses for 48 h ( Figures 2E, F ). Next, real-time PCR (RT-PCR) was conducted to determine the mRNA expression of IRF7. Notably, the expression level of IRF7 was remarkably upregulated in human MDA-MB-231 cells and BT-549 cells exposed to the indicated concentration of DSF for 48 h ( Figure 2G ).

Type I interferon regulated by IRFs can induce PD-L1 expression (34). We observed that the mRNA expression of PD-L1 was positively associated with IRF7 in BC patients from TCGA database (R=0.21, p<0.001) in Figure 3A . Human BT-549 and MDA -MB-231 cells were exposed to DSF for 48 h at doses 50 and 80 μM. DSF treatments significantly increased surface PD-L1 level and induced a twofold increase in PD-L1 levels. Flow cytometry and immunoblotting showed that PD-L1 protein expression was upregulated in the human MDA-MB-231 and BT549 cells after DSF treatment in Figures 3B–E . In human BT-549 cells, RT-PCR analysis further indicated that DSF treatment upregulated the mRNA expression of PD-L1 ( Figure 3F ), suggesting that DSF controls PD-L1 expression at the transcriptional levels. Upregulation of PD-L1 surface expression was found in human BT-549 cell line treated with DSF. Collectively, these findings show that DSF can upregulate the transcription and surface expression of PD-L1 in TNBC cell lines.

To investigate if DSF regulated PD-L1 through IRF7 in TNBC cell lines, we then detected whether the human BT-549 and MDA-MB-231 cells transfected with siRNA-IRF7 can significantly attenuate the upregulation of PD-L1 caused by DSF treatment in membrane protein level ( Figure 4A ). Next, the mRNA expression of PD-L1 was quantified by RT-PCR after DSF/si-RNA-IRF7 treatment. The results showed that when combined with si-RNA-IRF7, the upregulating effects of DSF on PD-L1 were significantly reduced. The level of PD-L1 transcription was significantly inhibited after knocking down IRF7 ( Figure 4B ). Western blot was applied to verify that si-RNA-IRF7 can cancel PD-L1 upregulation by DSF at the total protein level ( Figure 4C ). These results implicated that the regulation effects of DSF on PD-L1 was mediated by IRF7.

Increasing PD-L1 expression could improve the response to PD-1 blockade therapy. Thus, we explored combined efficiency of anti-PD-1 Ab and DSF in vivo. The schedule of animal experiments was shown in a flowchart ( Figure 5A ). The results showed that DSF reduced the tumor burden moderately, anti-PD-1 Ab inhibited tumor growth slightly, and the co-treatment strategy exhibited higher antitumor efficacy than each treatment alone ( Figures 5B, C ). In addition, mice can tolerate this combination treatment well. Immunohistochemistry (IHC) analysis demonstrated that both PD-L1 expression was upregulated in the DSF group as well as the DSF and anti-PD-1 Ab co-treatment group, while the population of tumor-infiltrating CD8+ T cells was downregulated in the tumor tissues of mice treated with DSF alone. Furthermore, the population of granzyme B+, CD8+ in the tumor tissue was significantly increased in the combination group, suggesting that the combined treatment could improve T cell activities in mice. Taken together, these findings illustrated that DSF combined with anti-PD-1 Ab had potential therapeutic benefits ( Figures 5D, F ). Finally, we detected the expression of DNMT1 and IRF7 in mice tissues. DNMT1 was downregulated and IRF7 upregulated in the DSF group and the DSF and anti-PD-1 Ab co-treatment group, which was consistent with the results in vitro ( Figures 5E, F ).

To further explore the effects of DSF and anti-PD-1 blocking Ab co-treatment on the immune microenvironment in 4T1 mouse xenograft tumor model, the expression profiles of immune microenvironment-related genes in the four groups were analyzed by RNA-seq. Notably, the gene expression profile of DSF and anti-PD-1 Ab co-treatment group was significantly different from other three groups ( Figure 6A ). For further validation of DSF and anti-PD-1 blocking Ab co-treatment as a potential immune regulator, we measured the tendency of residential immune cells in different treatment groups by calculating the degree of immune infiltration through MCP counter in R. We found the tumors in DSF group were in a low immune activation state ( Figure 6B ). This might be due to the upregulated expression of PD-L1 by DSF in tumor cells resulted in an impaired T cell function. To further explore the potential molecular mechanism of DSF regulating tumor immune function, the GO and KEGG pathway enrichment analysis was done to reveal the inactive state of immune-associated pathways ( Figure 6C ), such as Th1 and Th2 cell differentiation (Q<0.01), as well as antigen processing and presentation (Q<0.01). We speculated that the co-treatment of DSF and anti-PD-1 Ab can overcome the immunological side effects caused by DSF. To reveal the differences in biological function between the anti-PD-1 Ab group and the DSF and anti-PD-1 Ab co-treatment group, gene set enrichment analysis (GSEA) was performed ( Figure 6D ). The results indicated that four pathways (i.e., Th1 and Th2 cell differentiation, antigen processing and presentation, natural killer cell-mediated cytotoxicity, and T cell receptor signal transduction) were significantly enriched in the combination groups, whereas no pathway was markedly enriched in anti-PD-1 blocking Ab groups (P<0.05).

Human BC cell lines BT-549, MDA-MB-231 and mouse BC cell line 4T1 were supplied by ATCC. BT549 was cultured in RPMI 1640 (Thermo Scientific HyClone, USA), while MDA-MB-231 and 4T1 were cultured in DMEM (Thermo Scientific HyClone) containing 1% penicillin-streptomycin and 10% fetal bovine serum (FBS), followed by incubation in an atmosphere of 5% CO2 at 37°C.

DSF was obtained from Sigma (USA) and then dissolved in DMSO. siRNA of IRF7 (5′-TCGAGTGCTTCCTTATGGA- 3′) (RiboBio, Guangzhou, China). Anti-DNMT1, IRF7, PD-L1, granzyme B, CD3, CD4, CD8, and GAPDH Abs were obtains from ABclonal (Wuhan, China). PE-conjugated PD-L1 Ab were obtained from BioLegend (San Diego, CA, USA). The DNMT Activity/Inhibition Assay Ultra Kit (Colorimetric) was purchased from EpiGentek (USA). The anti-PD-1 blocking Ab were obtained from Innovent Biologics (China).

The cells were rinsed twice with pre-cold PBS and lysed in RIPA Buffer for 15 min, followed by centrifugation (15,000 rpm, 15 min, 4°C). Total protein content was measured using the BCA kit (Beyotime). After separation on 10% SDSPAGE, the samples were transferred onto a PVDF membrane. Subsequently, the membrane was inhibited with 5% skimmed milk in 0.1% TBST for 1 h, and then exposed to primary Abs at 4°C. On the next day, the membranes were rinsed thrice with TBST for 15 min and exposed to the corresponding secondary Ab (1:4,000 dilution, Servicebio, Wuhan, China) for 1 h. After washing, the protein blots were detected using a SuperSignal West Pico Chemiluminescent Substrate (Pierce, USA). GAPDH was employed as an internal control for normalization.

Total RNA was extracted with Total RNA Kit I (Omega, USA). cDNA synthesis was conducted with SYBR^® Premix Ex TaqTM II (Takara Bio, Japan) by following the manufacturer’s protocols. RT-PCR was performed on a Step One Plus RT-PCR system (Applied Biosystems) using SYBR Green Mastermix (Takara Bio). The sequences of the primers used are as follows: DNMT1 F: 5ʹ 5-CGGCAGACCATCAGGCATTCTAC-3ʹ and R: 5ʹ-CACACCTCACAGACGCCACATC-3ʹ; IRF7 F: 5ʹ-CTCCTTGGAGAGATCAGCAG-3ʹ, and R: 5ʹ-CAGCGG-AAGTTGGTTTTCC-3ʹ; PD-L1 F: 5ʹ-GCTGCACTAATTGTCTATTG-GG-3ʹ and R: 5ʹ-CACAGTAATTCGCTTGTAGTCG-3ʹ; GAPDH F: 5ʹ-ACCACAGTCCATGCCATCAC-3ʹ and R: 5ʹ-TCCACCACCCTGTTGCTGTA-3ʹ.

The harvested cells were incubated with PE-conjugated PD-L1 Ab (BioLegend) for 30 min at 4°C, rinsed twice with 4°C PBS, evaluated with flow cytometry, and analyzed by FACSDiva Software (BD Bioscience, NJ, USA).

MDA-MB-231 and BT-549 cells (1×10⁵/wells) were grown on 6 cm dishes for 24 h, and then exposed to vehicle or the corresponding drugs for 48 h. Then, genomic DNA was extracted, followed by bisulfite treatment using the Qiagen EpiTect kit. PCR cycle (98°C for 30 s; followed by 35 cycles of 98°C for 10 s, 60°C for 30 s, and 72°C for 10 s; 72°C for 2 min; 4°C hold) was conducted with Q5 Hot Start High-Fidelity Master Mix (NEB). Methyl Primer Express™ v1.0 (ThermoFisher) was used to design primer sequences for IRF7 promoter region (F: 5′-TTGGGTTGTAGTGGAGTGGTTTTATT-3′; R: 5′-CATCTCTCAAACTCCCCCAACTCTT-3′). The PCR products were detected through electrophoresis and purified by Gel and PCR Clean-up System (Promega). After purification, Zero Blunt™ TOPO™ PCR Cloning Kit (Invitrogen) was used to insert the products into pCR™4Blunt-TOPO^® Vector. Lastly, Sanger sequencing (Sangon) was carried out.

The gene expression RNAseq data (1,104 tumor tissues and 113 normal tissues) of the cohort: Genomic Data Commons Cancer Genome Atlas-Breast Cancer (GDC TCGA-BRCA) were obtained from the https://xenabrowser.net/datapages/. Total BC specimens were ordered according to methylation score levels. The beta-value cutoff ranges for hypermethylation and hypomethylation were 0.7–0.5 and 0.3–0.25, respectively (35, 36). R 3.6.0 was used to analyze the relationship between the level of IRF7 methylation and transcription. MEXPRESS is an easy-to-use tool for visualizing gene expression, DNA methylation, clinical TCGA data, and the relationship among them (37). Given the important effects of methylation and CNVs on gene expression, MEXPRESS was used to explore the association between IRF7 expression and methylation/CNVs.

GEPIA database (http://gepia.cancer-pku.cn/index.html) was employed to verify the significant relation between IRF7 and PD-L1, and the Pearson correlation was used to analyze the correlation between the two genes. The gene transcript expression in TNBC/basal-like compared with normal breast cases in TCGA and GTEx data cohorts by GEPIA2 (http://gepia2.cancer-pku.cn/). The p-value were set as 0.01.

siRNA targeting IRF7 was supplied by RiboBio (China). After transfection with siRNA (50 nM), the cells were analyzed by GenMute transfection reagent (SL100568; SignaGen Laboratories, China) as per the manufacturer’s protocol.

BALB/c mice (4–6 weeks old, female) were procured from Beijing HFK Bioscience (China) and maintained under SPF conditions in accordance with the animal care guidelines of Huazhong University of Science and Technology (HUST). The experimental protocol was approved by the Ethical Committee of HUST. 4T1 cells (1 × 10⁶) were subcutaneously transplanted into the right flank of mice. The tumor-bearing mice were randomly categorized into four groups (n = 5/group): control; anti-PD-1 blocking Ab i.p.; DSF p.o; anti-PD-1 blocking Ab + DSF. Tumor dimension was assessed by vernier caliper, and tumor volume was calculated as follows: 0.5×width²×length. After 28 days, the mice were sacrificed, and the subcutaneous tumors was isolated, recorded, and subsequently analyzed. DSF (100 mg/kg) was administered daily via the p.o. route starting from day 7 after tumor implantation and continuously for 3 weeks. Anti-PD-1 blocking Ab (200 µg; BE101, BioXCell) was administered i.p. every other day for four times starting from day 7 after tumor implantation. The mice in control group were treated with solvent (5% Tween 80, 30% PEG300, and 2% DMSO). Tumor size was recorded using a digital caliper every 3 days and calculated as follows: 0.5×width²×length.

4T1 xenografted tumors in BALB/c mice were formalin-fixed and embedded in paraffin. Briefly, after deparaffinization, rehydration, and antigen retrieval, tissue sections were exposed to primary antibody (DNMT1, IRF7, PD-L1, granzyme B, CD3, CD4, or CD8). Then, the sections were exposed to biotinylated goat-anti-mouse IgG secondary antibody and streptavidin-conjugated HRP, and finally developed with 3,3′-diaminobenzidine (DAB). Images were collected using a high-resolution slide scanning system (3DHISTECH Ltd, Pannoramic MIDI). Image-pro Plus 6.0 software (Media Cybernetics, Inc., Rockville, MD, USA) was used to select the same brown color as the unified standard for judging the positivity of all photos, and the positive area of each photo was obtained by analyzing each photo.

Total RNA was isolated from the 4T1 xenografted tumors tissue sample using TRIzol Reagent (Invitrogen, USA), and then subjected to RNA deep sequencing with MGISEQ2000 platform at Beijing Genomics Institute (BGI, China). The obtained sequencing reads were expressed as the FPKM (fragments per kilobase of exon per million reads) for each transcript. KEGG and single sample GSEA (ssGSEA) analyses were described in the previous study (38). The R-3.4.3 software tools were used to perform bioinformatics analysis and generate figures.

The data were shown as mean ± S.E. Statistical tests were performed by GraphPad Prism v8.0 and R language v3.4.3. Unpaired Student’s t-test (n<30) and Wilcoxon test (n>30) were used to compare the difference between two groups. Level of significance was set as p<0.05. The Pearson correlation was used to analyze the correlation between the two genes.