KM plotter (http://kmplot.com/analysis/) is a website used for on-line analysis, which is capable of assessing 54k genes on the survival of 21 cancer types, including breast cancer. The association between the distance-metastasis-free survival (DMFS) and the expression of m6A modulators (METTL3, METTL14, WTAP, FTO, and ALKBH5), COL3A1 was analyzed using KM plotter, respectively. GEPIA (http://gepia.cancer-pku.cn) is an on-line database including the RNA sequencing expression data of 9,736 tumors and 8,587 normal tissue samples from the Cancer Genome Atlas (TCGA) and the Genotype Tissue Expression (GTEx). The transcriptional levels of five m6A modulators above in breast cancer tissues and normal breast tissues were obtained from GEPIA. The mRNA expression data of 91 patients with TNBC were downloaded from TCGA (https://www.cancer.gov/) for analysis of the correlation between the m6A modulators and target genes.

The human breast cancer cell lines, MDA-MB-231, and MDA-MB-468, were acquired from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). All the cells were incubated in L15 culture medium (Gibco, NY, USA) supplemented with 10% FBS at 37°C under 5% CO₂ and saturated humidity.

The primary antibodies for western blot, anti-METTL3 (#96391) were sourced from Cell Signaling Technology (MA, USA), anti-COL3A1 (sc-514601) was sourced from Santa Cruz (CA, USA), anti-α-tubulin (ab7291) was sourced from Abcam (CA, USA). HRP-conjugated goat anti-mouse/rabbit secondary antibodies (ZDR-5306/5307) were sourced from ZSBIO (Beijing, China). The antibodies for immunohistochemistry, anti-METTL3 (ab195352) were sourced from Abcam (CA, USA), and anti-COL3A1 (sc-166316) was sourced from Santa Cruz (CA, USA).

For the migration assay, ~2 × 10⁴ cells were suspended in 200 μl serum-free L15 medium and added into the upper chamber of a trans-well plate (Corning, USA) with an 8-μm pore size polycarbonate filter, and 500 μl L15 medium with 10% FBS were dispensed into the lower chambers, and incubated for 24 h. Then the upper chambers were fixed in 75% ethanol, and the cells on the upper surface of the filter were removed manually with a cotton swab. Then the migrating cells were stained with Wright–Giemsa stain.

The invasion assay was similar to the migration assay except that 3% matrigel was dispensed into the upper chamber before seeding 3 × 10⁴ cells into the culture system. Migrating and invading cells were observed under an optical microscope. The cells from three fields were counted with Image J (https://imagej.nih.gov/ij/download.html).

The cells were seeded at 3 × 10⁴ per well into 96-well-plates pre-coated with 10% matrigel overnight. After 30 min-incubation at 37 °C, the non-adherent cells were removed by PBS washing. Then the cells were fixed with 75% ethanol for 10 min and stained with Wright–Giemsa stain. Images were acquired by microscope and the quantities of cells were counted with Image J.

RNA was isolated with TRIzol reagent (Invitrogen, CA, USA) and identified using a NanoDrop spectrophotometer (Thermo Scientific, Rockford, IL, USA). The Reverse Transcription Kit (Promega, WI, USA) was used for mRNA reverse transcription. Quantitative real-time PCR (qRT-PCR) assay was then performed using the SYBR Green kit (Promega, WI, USA) on the ABI7500 instrument (Thermofisher, IL, USA). All the reactions were conducted for triplicates. 18S RNA was used as the internal control. Primers of METTL3, COL3A1, MYH11, and 18S were used as follows:

The methylated RNA immunoprecipitation-qRT-PCR (MeRIP-qRT-PCR) assay was conducted according to the standard protocol of Magna MeRIP m6A Kit (Millipore, MA, USA, 17-10499) with a slight modification. The total RNA (300 μg) was isolated with TRIzol reagent and fragmented. Except for 3 μg of the total RNA as input, the remaining RNA was used for m6A-immunoprecipitation with m6A antibody. The Protein A/G Magnetic Beads were prepared by 30 min-incubation with m6A-specific antibody in immunoprecipitation buffer at room temperature, then incubated with the MeRIP reaction mixture and RNA for 2h at 4°C. Finally, the MeRIPed-RNA was cleaned up and concentrated with RNeasy MinElute Clean-up Kit (Qiagen, Hilden, Germany). The enriched RNA level of COL3A1 was analyzed by qRT-PCR. The primers were as same as the primers used in real-time PCR assay.

For the knockdown of METTL3 and COL3A1, the cells were seeded at 10⁵ cells/well in a 6-well-plate and the siRNAs targeted to METTL3 and COL3A1 (GENEWIZ, Beijing, China) at a final concentration of 50 nM were transfected using JetPRIME Transfection Reagent (Polyplus, Illkirch, France) according to the manufacturer's instructions. Negative control (NC) siRNA was used as a control. For the overexpression of METLL3, the cells were seeded at 10⁵ per 6-well-plate and transfected with plasmids pcDNA3.1-FLAG and pcDNA3.1-METTL3 (Obio, Shanghai, China) at a final concentration of 1 μg/ml using JetPRIME Transfection Reagent. The siRNA sequences of NC, METTL3, and COL3A1 were used as follows:

The harvested cells were lysed in a lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 10 mM EDTA, 50 mM NaF, 1 mM NaVanadate, 1% Triton X-100, 1 mM PMSF, and 0.5% aprotin) and the protein concentration was quantified according to the Coomassie blue G250 staining technique. Equivalent protein was electrophoresed on 8–10% SDS-PAGE gels and transferred to polyvinylidene difluoride (PVDF) membranes (Perkin-Elmer, Waltham, MA, USA). Then the membranes were blocked with 5% skimmed milk in TBST for 1 h and incubated with the primary antibodies overnight (more than 6 h). The concentration of anti-METTL3 (#96391) was 1:1,000, the concentration of anti-COL3A1 (sc-514601) was 1:200, the concentration of anti-α-tubulin (ab7291) was 1:1,000. After being immunoblotted with HRP-conjugated goat anti-mouse/rabbit antibody (1:2,000) for 40 min, the signal strength of revealed protein bands could be detected with enhanced chemiluminescence reagent (SuperSignal Western Pico Chemiluminescent Substrate; Pierce, Rockford, IL, USA) and visualized with the Electrophoresis Gel Imaging Analysis System (DNR Bio-Imaging Systems, Jerusalem, Israel). The blots were scanned and the abundance assessed quantitatively using ImageJ.

Cells at a density of 1 × 10⁵/well in a 6-well-plate were incubated for 72 h, and Collagen α1(III) levels in cell culture medium and cell lysate were measured, respectively, using a commercially available Collagen α1(III) ELISA kit at 450 nm by a microplate spectrophotometer (CSB-E13446h, Cusabio, Wuhan, China). The experiment process was carried out according to the instructions of the kit. The samples were added into wells and incubated 2 h in 37°C. After removing the liquid of each well, Biotin-antibody was added into each well. After incubated 1 h in 37°C, HRP-avidin and TMB Substrate were used for color rendering. The standard curve was constructed by the mean absorbance of each standard and the concentration. The concentration of each sample was determined according to standard curve. To acquire the total amount of Collagen α1(III) in each sample, the concentration was multiplied by the total volume. Then the amount of Collagen α1(III) in each sample was normalized to the amount of Collagen α1(III) in 10⁷ cells. All samples and standards were detected in duplicate.

The tissue microarray (TMA) sections (HBreD140Su06) and the relevant clinical data were obtained from Shanghai Outdo Biotechnology Company. This study was approved by the Ethics Committee of Shanghai Outdo Biotechnology Company (YB M-05-02), and all patients have given their informed consent. The TMA sections were deparaffinized and rehydrated with an ethanol gradient. Then antigen retrieval was performed with citrate buffer (MXB, Fuzhou, China, MVS-0066) and the TMA sections were blocked with endogenous peroxidases in UltraSensitive™ SP (Mouse/Rabbit) IHC Kit (MXB, Fuzhou, China, KIT-9730-A&B). The concentration of anti-METTL3 (ab195352) was 1:500, the concentration of anti-COL3A1 (sc-166316) was 1:50. After overnight incubation with primary antibody, the TMA sections were incubated with biotinylated secondary antibody for 10 min in UltraSensitive™ SP (Mouse/Rabbit) IHC Kit (MXB, Fuzhou, China, KIT-9730-C&D) and developed with DAB Kit (MXB, China, DAB-0031). Finally, the TMA sections were counterstained with hematoxylin (Solarbio, Beijing, China), and dehydrated with an ethanol gradient and mounted with neutral balsam (Solarbio, Beijing, China). METTL3 expression was evaluated by two independent reviewers by calculating the average positively stained tumor cells at 400×magnification. The positive signal of COL3A1 was quantified as integrate optical density (IOD) value using ImageJ software.

Statistical analyses were carried out using SPSS (version 16.0) and R (V.3.2.5). Limma package analysis was conducted to explore the correlations between METTL3 and other genes. One-way ANOVA and Student's t-test were used to determine statistical significance. Statistical significance was identified as P-values of <0.05.

To investigate which m6A modulator plays an important role in breast cancer, especially in TNBC, the mRNA expressions of m6A methyltransferases (METTL3, METTL14, and WTAP) and demethylases (FTO and ALKBH5) were first compared between breast cancer tissues and normal breast tissues using the RNA sequencing expression data in GEPIA (http://gepia.cancer-pku.cn) on-line database. The result showed that the expression of METTL3 (T = 4.8, N = 5.4) and FTO (T = 4.9, N = 5.3) was significantly lower, whereas that of METTL14 (T = 4.2, N = 3.8) was higher in breast cancer tissues than that in normal tissues, but the difference in the expression of WTAP (T = 5.9, N = 6.2) and ALKBH5 (T = 5.7, N = 5.5) was not very significant (Figure 1A). Next, the effects of METTL3, METTL14, and FTO on distant metastasis free survival (DMFS) of total breast cancer patients and TNBC patients were analyzed using KM-plotter on-line database, respectively. For overall patients, the analysis result indicated that no significant difference of DMFS was obtained between the patients with all three modulators in high-expression groups and low-expression groups (Figure 1B); however, for the TNBC patients, although no significant difference was found between the DMFS of the METTL14 high-expression group and low-expression group, the DMFS of METTL3 high expression group was shown to be longer than that in the METTL3 low-expression group, whereas the DMFS of the FTO high-expression group was shorter than that of the FTO low-expression group, indicating that METTL3 is a protective factor, but FTO is a risk factor for DMFS of TNBC (Figure 1C). As the result, the low expression of FTO in TNBC tissues was contradictory to its role as a risk factor (Figures 1A,E) and only METTL3 was shown to play an important inhibitory role in the metastasis of TNBC (Figures 1A,D), indicating that METTL3 might contribute to the metastasis of TNBC. Therefore, the role of METTL3 in TNBC metastasis was focused on in the following investigation.

For the knockdown of METTL3, three siRNAs targeted to METTL3 were transfected according to the manufacturer's instructions (Figures 2A,B). Two sequences, si-METTL3-1604 (KD2) and si-METTL3-1126 (KD3), were selected for subsequent experiments. To investigate whether METTL3 could inhibit TNBC metastasis, the effect of METTL3 on migration, invasion, and adhesion to cell-extracellular matrix (ECM) was detected by trans-well assay or adhesion assay in TNBC cell lines, MDA-MB-231, and MDA-MB-468. The results showed that METTL3 knockdown (KD) significantly increased the ability of migration and invasion, as well as the adhesion capability to ECM in both MDA-MB-231 and MDA-MB-468 cells (Figures 2C–E), indicating that METTL3 could inhibit the potential of cell mobility of TNBC cells. Then, to determine whether METTL3-inhibited potential of cell mobility was related to m6A modification or not, METTL3 was transiently overexpressed in MDA-MB-231 and MDA-MB-468 cells followed by the treatment with cycloleucine, a small molecule inhibitor of m6A modification. The results of trans-well assays demonstrated that METTL3 overexpression (OE)-suppressed migration, invasion, and adhesion were significantly recovered by cycloleucine (Figures 3A–D). These results strongly suggested that METTL3 inhibited the potential of cell mobility of TNBC cells by enhancing m6A modification.

It was known that m6A modification could down-regulate gene expression by accelerating RNA degradation (12–14). Therefore, to identify the target gene of METTL3 involved in METTL3-inhibited metastasis, multi-step screening was performed as summarized in Figure 4A. Firstly, mRNA expression profiles of 91 TNBC patients in the TCGA dataset were downloaded, and the differentially expressed genes (DEG) were screened to identify those with P < 0.05 and log fold change (FC) using the “limma” package in R. The Log FC of DEG genes more than 0 was identified as representing positive related genes, whereas that <0 represented negative related genes. Among the genes for which mRNA expression was negatively related to METTL3, the top 100 genes according to the correlation coefficient were selected for further m6A methylation analysis using the m6Avar database (http://m6avar.renlab.org). All these genes and related information were listed in Table S1. As a result, 51 genes, which were verified to be able to be modified by m6A, were screened. Then, with the KEGG pathway enrichment analysis by DAVID (https://david.ncifcrf.gov), 18 genes were shown to be associated with the focal adhesion pathway, metabolism pathway, and so on, suggesting close involvement with metastasis in breast cancer. Subsequently, the association of 18 genes with the DMFS of TNBC patients was further analyzed using KM-plotter, and six alternative genes were found to have shorter DMFS at high levels in TNBC (Table 1). In particular, COL3A1 and MYH11 aroused our attention, because, according to the Gene Ontology (GO) analysis in DAVID, it was shown that COL3A1 was involved in the biological process of skeletal system development and cell-matrix adhesion, and MYH11 was involved in the biological process of elastic fiber assembly, which were similar to the findings of previous studies that they could promote metastasis in breast cancer. Therefore, the two genes were chosen as target gene candidates for METTL3. The further verification result of qRT-PCR detection showed that METTL3-KD only up-regulated the mRNA expression of COL3A1, but not MYH11 in MDA-MB-231 and MDA-MB-468 cells (Figures 4B,C). Similarly, the result of western blot assay also confirmed that METTL3-KD increased the protein level of COL3A1 in TNBC cells (Figure 4D). All of the above data strongly suggested that COL3A1 might be the target gene of METTL3.

To investigate whether COL3A1 was regulated by METTL3-mediated m6A methylation or not, the relative m6A enrichment level change of COL3A1 before and after METTL3-KD was detected by MeRIP-qRT-PCR in MDA-MB-231 and MDA-MB-468 cells. As shown in Figure 5A, METTL3-KD significantly reduced m6A-methylated COL3A1 mRNA expression. Furthermore, with qRT-PCR detection, it was shown that METTL3- OE decreased the mRNA expression of COL3A1, while cycloleucine partially recovered METTL3-OE-down-regulated COL3A1 (Figure 5B). As shown in Figure 5C, the results of ELISA showed that the secretion level of Collagen α1(III) in supernatant decreased by METTL3-OE could be recovered by cycloleucine while the METTL3-OE decreased the intracellular level of Collagen α1(III), the cycloleucine could not recover the reduction of the intracellular level of Collagen α1(III). Considering the amount of Collagen α1(III) secreted into the supernatant is greater than the amount in the cell, the total levels of Collagen α1(III) which were decreased by METTL3-OE could be recovered by cycloleucine after normalizing (Figure 5C). These results proved that COL3A1 was down-regulated by METTL3-mediated m6A modification on COL3A1 (Figure 5C).

The role of COL3A1 in TNBC metastasis was further investigated. The influence of COL3A1 on DMFS of TNBC patients analyzed by KM-plotter is shown in Figure 6B, the DMFS of TNBC patients with high-expression COL3A1 was shorter than in those with a low expression thereof. There was no significant difference in the expression level of COL3A1 and DMFS in the overall patients (Figure S1). In addition, when COL3A1 was knocked-down in MDA-MB-231 and MDA-MB-468 cells (Figure 6A), the migration, invasion, and adhesion to ECM were significantly suppressed (Figures 6C–E). These data indicated that COL3A1 played an important role in promoting metastasis in TNBC.

In order to verify the effects of METTL3 and COL3A1 on the prognosis of breast cancer patients in vivo, the expression of METTL3 and COL3A1 was investigated by immunohistochemistry using TMA sections containing 31 TNBC patients and 109 Non-TNBC patients. The results of survival analysis showed that low METTL3 expression was related to short overall survival (OS) (Figure 7A) in TNBC but not in non-TNBC (Figure S2A). The OS of TNBC with COL3A1-high expression was shorter than that with low expression (Figure 7B), which may not be statistically significant because of the small number of TNBC cases. In non-TNBC, COL3A1 has the opposite trend, and patients with high expression have longer OS (Figure S2B). Finally, the relationship between METTL3 and COL3A1 in breast cancer was analyzed with Pearson correlation analysis (Figures 7C,D). The results showed that the expression level of COL3A1 was negatively correlated with METTL3 expression in TNBC patients (R = −0.564, P = 0.001; Figure 7E); however, the expression level of METTL3 and COL3A1 had no significant relationship in the 109 NTNBC patients (R = −0.132, P = 0.170; Figure 7F). This result further confirmed that the metastasis-inhibition function of METTL3 by negatively regulating COL3A1 expression was TNBC specific.