The chromatin architectural regulator SND1 mediates metastasis in triple-negative breast cancer by promoting CDH1 gene methylation

The human tissues of 58 TNBC and 10 glioblastomas (GBM) as positive control were collected from the Tianjin Medical University General Hospital (TMUGH). All the data involving patients were handled under the guidance of the Declaration of Helsinki and the principles of the TMUGH Ethics Committee. The tissues were formalin fixed and paraffin embedded (FFPE). The clinical features of the 58 TNBC from TMUGH and 142 TNBC from The Cancer Genome Atlas (TCGA) (https://​cancergenome.​nih.​gov/​) are shown in Additional file 5: Table S1.

IHC was performed with mouse anti-human SND1 antibody (Santa Cruz sc-166676, dilution ratio 1:25, overnight at 4℃), biotinylated goat anti-mouse secondary antibody (Beyotime A0286, dilution ratio 1:50, incubation for 1 h at room temperature) and HRP-labelled streptavidin (Beyotime A0303, dilution ratio 1:200, incubation for 1 h at room temperature). Finally, the proteins were visualized by diaminobenzidine (DAB) colour development kit (Beyotime P0203). The per cent (labelling index, LI) of SND1 positive cells in 10 fields (× 400) was count under a Leica DM6000B microscope with Image Pro Plus 5.0 software (Media Cybernetics).

The HEK293T (ATCC CRL-3216), MDA-MB-231 (ATCC HTB-26) and BT549 (ATCC HTB-122) cell were purchased from the American Type Culture Collection. MDA-MB-231 and BT549 were used for transwell and wound healing assay. The details are listed in the Additional file 6: Methods.

The details of plasmid construction, lentivirus package and stable cell line transfection were listed in the Additional file 6: Methods.

Additional file 5: Table S4 lists all the probes used in this study. Details are shown in the Additional file 6: Methods.

The transcription profiles of MDA-MB-231 cell with different treatments were detected by cDNA microarray (n = 3/group). See details in Additional file 6: Methods.

The qRT-PCR primers were listed in Additional file 5: Table S4; the antibodies used for Western blot were listed in Additional file 5: Table S5. See details in Additional file 6: Methods.

Scramble and SND1 knockdown TNBCs were used for ChIP and 3C assay; the experimental details are outlined in the Additional file 6: Methods.

The DNMT3A promoter reporter plasmid was established by inserting full length or different fragments of DNMT3A promoter (DNMT3A pro full length/ΔTSS/ΔR1/ΔR2/ΔR1 + R2) into pGL3 vector (Promega, E1751). Then, the aforementioned reporter plasmids were cotransfected with pSG5-SND1 into BT549 and MDA-MB-231 cells via Lipofectamine 3000 (Thermo Fisher L3000015). After 48 h of transfection, the luciferase activities among two breast cancer cells were detected with Dual-Luciferase Reporter Assay Kit (Promega E1910).

The expression of CDH1 and DNMT3A in BT549 and MDA-MB-231 was detected by immunofluorescence assay, and the experimental details were outlined in the Additional file 6: Methods.

The bisulfite sequencing PCR assay was used to detected the methylation on the CDH1 gene promoter, and the details were shown in the Additional file 6: Methods.

Genomic DNA was extracted as described above, and 300–500 bp DNA fragments were obtained after sonication. Immunoprecipitation was performed with a mouse monoclonal antibody against 5-methylcytosine (5-mC, Abcam ab10805) and Protein A/G Mix magnetic beads (Millipore LSKMAGAG02). Then, methylated DNA was extracted with DNA extraction reagent (Solarbio D1700-100 T) and analysed by real-time RT‒PCR. All primers were listed at Additional file 5: Table S4.

MDA-MB-231 expressing firefly luciferase stably (Xenogen, Alameda, CA, USA) was transfected by viruses containing scramble shRNA or either of two different shRNAs targeting SND1. These cells (5 × 10⁶ cells per mouse) were used for tail lateral vein injection pulmonary metastasis model by female SCID mice at six weeks of age. The number of mice in each group was determined by Mead's resource Eq. 200 mg/g D-luciferin (MEM HY-12591A) was injected to the enterocoelia of each mouse for luminescence assay. The autofluorescence images of metastasis tumour were taken by IVIS bioluminescence CCD device (Xenogen), and the image-forming parameters were view field as fifteen cm, binning factor of eight, f-stop as one and exposure time as 120 s. The data were presented by normalized bioluminescence photon flux, which were calculated by background mice without luciferin injection. The Tianjin Medical University Institutional Animal Care Committee approved the entire animal feeding, processing and experimental handing procedures, and confirmed they were in compliance with the ethical regulations.

All data calculation was processed via SPSS version 21.0 (IBM, Chicago, USA). Data were presented as mean ± SD. The significance among means was compared by one-way ANOVA or Student t test. The survival time was analysed by Kaplan–Meier (KM) method and compared with log-rank test. The median of protein LI or mRNA log₂ (fpkm fold) of SND1 was used as the cutoff in survival analyses of the TNBC cohort from TMUGH and TCGA database. The Cox’s proportional hazards regression model was applied for multivariate survival analyses. Statistical significance was assigned at *P < 0.05, **P < 0.01 or ***P < 0.001. All the experiments of cell lines were performed at least in triplicate.

The SND1 protein level in TNBC tissues from 38 patients with lymph node metastasis (Met) and 20 patients without metastasis (non-Met) were evaluated by IHC, and 10 glioblastomas (GBM) high-expressing SND1 were used as positive control. The results indicated that the SND1 expression in Met TNBC was significantly higher than that in non-Met TNBC (P < 0.01), but the similar as positive control GBM group (P > 0.05; Fig. 1A and B). The KM analysis verified that TNBC patients high-expressing SND1 suffered poor outcome (Fig. 1C) and TNBC cohort from TCGA showed similar pattern (Additional file 1: Fig. S1). The multivariate Cox proportional-hazards regression analysis confirmed that SND1 expressive level and TMN stage was the independent predictor for disease-free survival(DFS) and overall survival(OS) of patients with TNBC from both TMUGH and TCGA (Additional file 5: Tables S2 and S3). We also compared the SND1 level among breast cancer cell lines, and the oestrogen receptor-negative MDA-MB-231 and BT549 cells had stronger SND1 expression than luminal-like cell lines (Fig. 1D, E).

To detect the role of SND1 on TNBC migration and invasion, we used BT549 and MDA-MB-231 cell lines as models for wound healing, Transwell and establishment of a tail vein injection inducing pulmonary metastasis model in mice. Both qPCR and Western blot were performed to double check the knockdown efficiency of SND1 (with SND1-sh1 and SND1-sh2) in two breast cancer cell lines (Fig. 2A, B). The wound healing assay indicated the migration of MDA-MB-231 and BT549 cells was diminished after knockdown of SND1 (Fig. 2C). The Transwell results also confirmed that SND1 knockdown blocked the invasion of aforementioned cells (Fig. 2D). Such results indicated that SND1 might promote the metastatic phenotype of TNBC cells in vitro.

To confirm the importance of SND1 to in vivo TNBC cell metastasis, MDA-MB-231 cells with stable luciferase expression were infected by lentiviruses with control RNA (scramble) or an SND1 shRNA (sh1 or sh2). The aforementioned cells were intravenously injected to the tail veins of SCID mice (n = 8). Bioluminescence imaging was generated via an IVIS luminescence CCD device (Xenogen) to quantify the lung metastasis of ectopically implanted TNBC cells every week. The results demonstrated that, compared to the scramble subclones, the SND1-sh1 and SND1-sh2 cells had a significantly decreased metastatic capacity (Fig. 2E). The Kaplan–Meier analysis data indicated that survival times of mice from SND1 knockdown group (SND1-sh1/sh2) were extended comparing to the scramble group (Fig. 2F). Western blot was perform with the pulmonary metastatic tissues in order to detect SND1 expression and guarantee the implementation of follow-up experiments (Additional file 2: Fig. S2). The above results indicate that SND1 may play an important role in breast cancer metastasis and predict worse prognosis of TNBC.

To clarify the detail mechanism by which SND1 promotes TNBC metastasis, we identified SND1 downstream genes via cDNA microarray assay in MDA-MB-231 cells. After SND1 knockdown, the expressions of 17 genes were decreased and 22 genes were increased (Additional file 5: Table S6). Among the 39 potential SND1 downstream genes, DNMT3A, DNMT3B, SMAD2 and SMAD3 were the most downregulated, and CDH1, CLDN3 and CLDN7 were the most upregulated (Fig. 3A, data are presented as the means of each group, n = 3, two way fold > 2, FDR < 0.05). Gene Ontology analysis showed that the potential downstream genes were involved in DNA methyltransferase activity, unmethylated CpG binding, TGF-β receptor signalling and the regulation of epithelial-to-mesenchymal transition (Fig. 3B). Through Pearson correlation analysis of gene expression data in the TCGA database, 1132 genes coexpressed with SND1 were selected (P < 0.05, R > 0.3). By overlapping these genes with the 39 genes identified in the above cDNA chip analysis, it was found that there were 4 direct target candidate genes (green) and 2 indirect target candidates (red). Across all the candidates, DNMT3A had the strongest positive relevance to SND1 expression, and CDH1 level was the most negatively related to that of SND1 (Fig. 3C). Then, the Cytoscape GeneMANIA plug-in was used to analyse and create the SND1 regulatory schematic diagram (Fig. 3D). These results suggested SND1 may regulate CDH1 expression by promoting DNMT3A transcription.

Since bioinformatics analysis showed that DNMT3A was the putative candidate SND1 downstream gene, we further examined the relationship between DNMT3A and SND1. The qRT‒PCR and Western blot results showed expression levels of DNMT3A were dramatically decreased in MDA-MB-231 SND1-sh1/2 and BT-549 SND1-sh1/2 cells (Fig. 4A, B). Overexpression of SND1 up-regulated the expression of DNMT3A in both TNBC cells (Fig. 4C). The above results indicated that SND1 might promote DNMT3A expression via transcriptional activation. Then, a luciferase reporter assay was adopted to explore whether DNMT3A transcription is directly regulated by SND1. The promoter sequence of DNMT3A was inserted into pGL3 reporter plasmid to construct the pGL3-DNMT3A pro reporter vector. Then, the reporter vector was cotransfected into BT549 and MDA-MB-231 cells with pSG5-SND1 or pSG5-con. The relative luciferase activity was significantly increased in all SND1 overexpression groups, and the luciferase signals were positively correlated with the SND1 expression level (Fig. 4D).

Bioinformatics analysis indicated three individual SND1 interaction positions on the DNMT3A promoter (Fig. 4E, upper part); one was located at the transcription start site (TSS), and the other two were located upstream of the TSS (R1 and R2). The ChIP assay showed SND1 protein could bind at R1, R2 and TSS in both tested TNBC cell lines and that knockdown of SND1 attenuated the interaction of SND1 with all three conserved sites, further confirming the specificity for the recruitment of SND1 on DNMT3A promoter (Fig. 4E). The role of SND1 enrichment at conserved sites in the transcription of DNMT3A was further determined by dual-luciferase reporter assay. Deletion of TSS completely blocked the inducing effect of SND1 on DNMT3A transcription, and deletion of R1, R2 or both R1 and R2 only partially inhibited the transcriptional activation of DNMT3A induced by SND1 (Fig. 4F). The aforementioned data showed that the recruitment of SND1 on TSS is essential for transcriptional activation of DNMT3A and that the interaction between SND1 and R1 or R2 can enhance DNMT3A transcription. R1, R2 and TSS have a universal conserved motif, 5’-TGATCTCGGCTCACT-3’, which might be the SND1 recognition site. To confirm the detailed interaction mechanism of SND1 with these different sites, in vitro translated SND1 protein was purified and subjected to EMSA (Fig. 4G). SND1 formed complexes with labelled probes containing the conserved motif (Fig. 4G, Lane 2), mutation of the 5’ end, the 3’ end, or both the 5’ and 3’ ends of the conserved motif inhibited the binding of SND1 (Fig. 4G, Lanes 3, 4 and 5), and the unlabelled cold probe serving as the negative control also competitively inhibited the formation of the band corresponding to the complex (Fig. 4G, Lane 6). Such results indicated that SND1 interacts with DNMT3A promoter by recognizing the conserved motif in R1, R2 and TSS.

Chromatin histone acetylation is an essential event in initiating long-range chromatin architectural alterations and inducing gene transcription. Our previous results showed that histone acetyltransferases can be recruited by SND1 to certain gene loci and lead to chromatin conformation rearrangement and gene transcription. ChIP‒qPCR targeting histone H3K9Ac and H3K27Ac showed that the histone acetylation occurred at the same position of SND1 binding regions in TNBC cells (Fig. 5A, B; scramble; R1, R2 and TSS) and that silencing of endogenous SND1 decreased the H3K9Ac and H3K27Ac levels in both TNBC cell lines (Fig. 5A, B; SND1-sh1/sh2; R1, R2 and TSS). The histone acetylation induced by SND1 unlocks the potential for chromatin to be remodelled to another conformation.

To assemble the transcription factors and the distant cis-transcription elements that occupied by them, topological alterations in chromatin are necessary. 3C analysis was applied to determine the stereoscopic proximity of the conserved motif in R1, R2 and TSS. When the R1 site was set for the anchor of PCR, strong chromatin cross-linking signals were detected at R1, R2 and TSS in TNBC cells, while SND1 knockdown blocked only the long-range association between R1 and TSS (Fig. 5C). When R2 region was set as anchor, a similar cross-linking signal between R2 and TSS was detected (Fig. 5D). When TSS was set as anchor, TSS region could interact with both R1 and R2 (Fig. 5E). SND1 could recruits to the histone acetylation-modified R1, R2 and TSS and induce chromatin conformation remodelling, which finally promote DNMT3A transcription (Fig. 5F).

The microarray data showed that SND1 expression was also negatively correlated with that of the key metastasis factor CDH1 in TNBC cells (Fig. 3A). CDH1 is a potential DNMT3A target. Therefore, we speculated that SND1 could inhibit CDH1 expression by upregulating DNMT3A and then, in turn promote TNBC cell invasion and migration. To verify the above-mentioned hypothesis, we first measured the expression of CHD1 and DNMT3A in MDA-MB-231 and BT549 cells. As expected, we found that SND1 knockdown caused a decrease in DNMT3A mRNA level and an increase in CDH1 (Fig. 6A; SND1-sh1/sh2), while restoration of SND1 expression reversed these changes in DNMT3A and CDH1 in SND1 knockdown cells (Fig. 6A; SND1-sh1/sh2 + SND1 res). The universal DNA methylation inhibitor 5-azacytidine (5-aza) dose-dependently repressed the expression of CDH1, indicating CDH1 should be one of the downstream targets of DNMT3A in TNBC cells (Fig. 6B). The methylated DNA immunoprecipitation (MeDIP) results showed that the CpG island at the promoter of CDH1 was highly methylated in scramble control TNBC cells (Fig. 6C; scramble). SND1 knockdown inhibited the methylation of the CDH1 CpG island (Fig. 6C; SND1-sh1/sh2), while SND1 reconstitution by transfection restored this methylation (Fig. 6C; SND1-sh1/sh2 + SND1 res). DNMT3A needs to be recruited to certain loci of CpG islands to catalyse DNA methylation. The ChIP assay results showed that DNMT3A was recruited at the CpG Island on promoter of CDH1 to maintain its methylation level (Fig. 6D; scramble). SND1 knockdown led to dissociation of DNMT3A from the CDH1 promoter (Fig. 6D; SND1-sh1/sh2), while SND1 reconstitution by transfection restored the recruitment of DNMT3A to the CDH1 promoter (Fig. 6D; SND1-sh1/sh2 + SND1 res). The bisulfite sequencing PCR results showed a similar pattern as the MeDIP and ChIP results. Of the 14 potential CpG methylation sites, most were demethylated after SND1 knockdown, and those sites were remethylated after SND1 reconstitution by transfection (Fig. 6E). The Western blotting results further demonstrated SND1 knocking down led to the depletion of DNMT3A and appearance of CDH1, while reconstitution by transfection of SND1 reversed these changes in DNMT3A and CDH1 expression in SND1 knockdown cell lines (Fig. 6F). These above results confirmed that SND1 can promote DNMT3A expression, inducing CDH1 promoter methylation and finally inhibiting CDH1 expression.

Then, we used DNMT3A reconstitution by transfection to further confirm that DNMT3A is the key mediator of SND1-induced CDH1 promoter methylation. Subsequently, BT549 and MDA-MB-231 cell lines were constitutively transduced with scrambled shRNA (scramble; control), constitutively transduced with SND1-shRNA (SND1-sh1/sh2; for endogenous SND1 knockdown), or subjected to SND1-shRNA/DNMT3A co-expression (SND1-sh1/sh2 + DNMT3A; rescue of DNMT3A expression). As expected, we found that rescue of DNMT3A expression completely blocked the CDH1 mRNA transcription induced by SND1 knockdown (Fig. 7A). Compared to the demethylated CDH1 promoter in SND1 knockdown cells, reconstitution of DNMT3A also catalysed the remethylation of the CDH1 promoter to the basal levels measured in scramble control TNBC cells (Fig. 7B). Western blotting further confirmed that reconstitution of DNMT3A in SND1 knockdown cells reduced CDH1 expression to the same level as that in scramble control TNBC cells (Fig. 7C). The immunofluorescence (IF) results revealed a similar pattern as the above Western blot results (Fig. 7D); SND1 knockdown inhibited DNMT3A expression and increased CDH1 expression. After reconstitution of DNMT3A in SND1-sh1/sh2 cells, the CDH1 protein level was decreased back to basal level measured in scramble group (Fig. 7E).

To identify the effect of DNMT3A to SND1-induced TNBC malignancy in vitro, we evaluated the metastatic phenotype of the above-mentioned TNBC cells used to generate the data in Fig. 7 by transwell and wound healing assays. The migration ability of MDA-MB-231 and BT459 cells was significantly decreased in the SND1-sh1/sh2 groups, while cell migration in the SND1-sh1/sh2 + DNMT3A groups was maintained at the normal level observed in scramble control cells (Fig. 8A). The Transwell assay results revealed a similar pattern as the wound healing assay results, and DNMT3A reconstitution by transfection abolished the inhibition of TNBC cell invasion, consistent with SND1 knockdown effects (Fig. 8B).

We established an experimental model of pulmonary metastasis via tail vein injection to further investigate whether DNMT3A is the key mediator of SND1-induced in vivo metastasis of TNBC cells. MDA-MB-231 cells with SND1 knockdown or DNMT3A reconstitution (scramble, SND1-sh1/sh2, SND1-sh1/sh2 + DNMT3A) were infected with lentivirus expressing firefly luciferase. Bioluminescence imaging of MDA-MB-231 cells confirmed that SND1 knocking down decreased the pulmonary tumour signal significantly, whereas restoration of the expression of DNMT3A would neutralized the aforementioned effect of SND1 knockdown (Fig. 8C). The Kaplan–Meier analysis results indicated that mice in the SND1 knocking down groups (SND1-sh1, sh2) had longer survival time than those in the scramble control group (scramble), while mice in the DNMT3A rescue group (SND1-sh1/sh2 + DNMT3A) had similar life span as those of mice in the scramble control groups (Fig. 8D). In summary, SND1 promotes DNMT3A expression via inducing chromatin architecture rearrangement, DNMT3A supresses CDH1 and promotes invasion and migration of TNBC cells. DNMT3A is core mediator through the SND1/DNMT3A/CDH1 axis which SND1 promotes TNBC cell metastasis (Fig. 8E).