MCF10A, MCF10AT1 and MCF10CA1h cell lines were obtained from Karmanos Cancer center and maintained on 10 cm dishes (CytoOne, USA Scientific) at 37°C, 5% CO₂ in DMEM/F12 (Corning) supplemented with 5% Horse serum, 1% Penn/Strep, 0.2% Sodium Chloride, Insulin and Hydrocortisone, 0.07% Cholera toxin and 0.04% Epidermal Growth Factor.

CRISPR/Cas9/sgRNA all-in-one construct co-expressing mCherry (Genecopoeia) was transfected using Lipofectamine 3000 (Thermo Fisher) according to manufacturer’s directions. Cells received either a control sgRNA construct (pCRISPR-CG01) or one targeting exon 1 of Mettl3 (HCP215070-CG12-3-10-a). Two rounds of transfection and Fluorescence Activated Cell Sorting (FACS) were used to isolate mCherry expressing cells which were then diluted to single cell for clonal selection. Western Blotting was used to confirm knockdown of Mettl3.

Trizol (Life Technologies) was used for all RNA extractions according to the manufacturer’s protocol. RNA for sequencing was further purified and treated with RNase-Free DNase I (Life Technologies) using PureLink RNA Mini Kit (Life Technologies). RNA purity and quantity was determined via NanoDrop 1000 (ThermoFisher Scientific).

Samples were sequenced by the Brody School of Medicine Integrative Genomics Core. RNA quality and concentration were verified by 4200 TapeStation evaluation (Agilent Technologies) and Qubit Fluorometric Quantitation (Thermo Fisher). Stranded cDNA libraries were prepared using TruSeq Stranded Total RNA prep and ligation with Ribo-Zero Plus kit (Illumina) in accordance with the manufacturer’s protocol. Paired-end sequencing (100 bp × 2) was performed on the NextSeq 2000 system (Illumina).

For differential gene expression analysis, sequence reads were pseudo-aligned to the human hg38 reference transcriptome and transcript abundance quantified by using Kallisto (v.0.46.1). Differential gene expression analyses were achieved by using tximport (v1.20.0) and DESeq2 (v1.26.0) packages in R Studio (Build 351 with R v4.1.1) and further analyzed with GSEA and ShinyGo v0.77 (31). The RNA sequencing data presented in this study are openly available in the GEO database with the accession number GSE237066.

For UPLC-MS/MS, purified PolyA+ RNA was digested to individual nucleosides and modified nucleosides were quantified as previously described (20).

EpiQuik m6A RNA Methylation Quantification Kit (Colorimetric) was used to quantify relative m6A levels in 400 ng of total RNA according to the manufactures protocol.

Whole cell lysates prepared in whole cell extract buffer (WCEB: 50 mM Tris pH 7.4, 150 mM NaCl, 5 mM EDTA, 0.1% SDS, and complete protease inhibitor (Promega) were quantified with BCA Assay (Promega). Equal protein amounts were electrophoresed on a mini-PROTEAN any KD acrylamide gel (Bio-Rad Laboratories) and transferred to Hybond ECL nitrocellulose (GE Healthcare). The blot was blocked with 5% nonfat dry milk (LabScientific) in Tris buffered saline with 0.1% Tween 20 (TBST) for one hour at room temperature, followed by primary antibody in blocking buffer overnight at 4° C. After washing with TBST, blots were incubated with appropriate secondary HRP conjugated antibody (GE Healthcare), washed again with TBST, detected using Bio-Rad Clarity Western ECL Substrate, and imaged via Invitrogen Ibright FL1500 imaging system. A list of primary antibodies used can be found in Supplemental Table 1 .

Approximately 2.5x10⁵ cells/well were plated in a 6 well plate. At 24, 48 or 72 hours from plating two wells were lifted with trypsin and counted via trypan blue and hemocytometer in duplicate. Remaining wells were fed with fresh media each day to prevent nutrient/additive deficiency during the time frame of the experiment and to allow for constant proliferation.

Approximately 2.0x10⁶, 1.5x10⁶, and 1.0x10⁶ cells/well were plated for the MCF10A, MCF10AT1 and MCF10CA1h cell line respectively in a 6 well plate. When the cells reach 95-100% confluency they were serum starved for 4 hours, scratched with a p20 pipette tip, washed with Dulbecco’s Phosphate Buffered Saline (DPBS) and fresh serum free media added. Pictures were taken at 0 and 24 hours and wound healing determined by measuring the change in distance between the cells on both sides of the scratch.

Reverse transcription was performed on 1 μg of total RNA with the iScript cDNA synthesis kit (Bio-Rad Laboratories). Quantitative real-time PCR was performed using a Roche Lightcycler 96 with Fast Start Essential DNA Green (Roche Diagnostics Corporation) and primers from Integrated DNA Technologies, Inc. Primer efficiency was verified to be over 95% for all primer sets used. Quantification of mRNA was carried out via ΔΔCT analysis using 18S rRNA and the respective control condition for normalization. All real-time PCR primer sets were designed so the products would span at least one intron (>1kb when possible) to prevent detection of the pre-mRNA and/or DNA, and amplification of a single product was confirmed via melting curve analysis. Primer sequences can be found in Supplemental Table 2 .

First, the m6A levels of each of the cell lines were determined via LC-MS of polyA selected, rRNA-depleted RNA (20). As shown in Figure 1B , the MCF10AT1 cells had the highest percentage m6A-modified adenosines followed by the MCF10CA1h and the MCF10A cells. This initial increase of m6A levels from MCF10A to MCF10AT1 followed by a decrease in m6A levels between the MCF10AT1 and MCF10CA1h suggests that m6A may have a dynamic role in the progression of this breast cancer model.

Next, the cell lines were characterized for the cancer-related phenotypes of proliferation and migration to verify that they modeled breast cancer progression as expected (38). Proliferation was measured by counting the cells every 24 hours over a 72-hour time course while a wound healing assay was used to determine the extent of migration. As expected, there was an increase in proliferation ( Figure 1C ) and migration ( Figure 1D ) as the model progressed from the MCF10A to the MCF10CA1h cells.

Lastly, the state of EMT gene expression was examined to verify that the model also recapitulated known gene expression changes that accompany the epithelial to mesenchymal transition (38). Indeed, there was a decrease in the epithelial marker E-Cadherin and an increase in mesenchymal markers N-Cadherin, Vimentin and MMP2 as the lines progressed from benign to tumorigenic ( Figure 1E ). E-Cadherin was at its highest mRNA expression in the MCF10A line, but its expression was more than halved in the MCF10AT1 cell line and was lowest in the MCF10CA1h cell line which was to be expected as they are the most mesenchymal. There was also a switch from E-Cadherin to N-Cadherin as the model progressed. Interestingly, Vimentin and MMP2 did not show much difference between the MCF10A and MCF10AT1 cell lines but increased dramatically in the MCF10CA1h cell lines.

Overall, the MCF10 breast cancer progression model showed an increase in proliferation and migration between the MCF10A and MCF10AT1, and between the MCF10AT1 and MCF10CA1h cell lines along with a switch from epithelial to mesenchymal markers confirming the model as a valid representation of breast cancer progression.

To investigate the role of m6A in breast cancer progression, Mettl3 was knocked down via transient CRISPR/Cas9 transfection targeting exon 1 coupled with FACS sorting and clonal selection. Knockdown of Mettl3 was confirmed in each cell line via Western blotting and two candidates chosen for each line for use for this study ( Figures 2A–C ). In the MCF10A cell line, knockdown was observed with one candidate having a more apparent knockdown. In the MCF10AT1 cell line there was substantial knockdown in both candidates. TheMCF10CA1h cell line proved to be the most resistant to METTL3 knockdown, as evidenced by only having a moderate level of knockdown in both candidates. Real-time PCR of the lines, while showing lesser of an effect on METTL3 mRNA levels, maintained the relative decrease in METTL3 expression in the knockdown lines ( Supplemental Figure 1 ).

To evaluate the effect of m6A decrease on the phenotype of the cells, proliferation and migration assays were run on two CRISPR knockdown candidates and one negative control CRISPR from each of the three cell lines. With the knockdown of Mettl3 in the MCF10A cell line ( Figure 2A ), there was a significant increase in proliferation of one of the knockdown candidates and a significant decrease (almost 80%) in migration in both Mettl3 knockdown lines compared to the negative CRISPR control. Interestingly, the opposite effect was observed in the MCF10AT1 cell lines ( Figure 2B ). With the knockdown of Mettl3, both MCF10AT1 cell lines showed a significant decrease in proliferation and an increase in migration, with one being significant. The MCF10CA1h cell lines showed no significant difference in proliferation or migration ( Figure 2C ). This suggests that m6A via Mettl3 has a dynamic role in the progression of the model and leads to unique changes in phenotype based on the stage of the cancer.

To determine which genes were most affected by the knockdown of Mettl3, transcriptome wide analysis was performed, and genes identified that differed significantly between the negative CRISPR and one of the knockdown cell lines ( Figure 3A and Supplemental Data 1 ).

In the three cell lines, 1,255 genes in the MFC10A cell line, 2,956 genes in the MCF10AT1 cell line, and 2,308 genes in the MCF10CA1h cell line showed a significant difference in expression with the knockdown of Mettl3. There was a 464 gene overlap (37.0% and 15.7% of the total for MCF10A and MCF10AT1 respectively) between the MCF10A and MCF10AT1. Similarly, between the MCF10A and MCF10CA1h cell lines 438 genes overlapped (35.0% and 19.0% of the total for MCF10A and MCF10CA1h respectively.) Interestingly, the most overlap of significant gene changes was between the MCF10AT1 and MCF10CA1h with 667 genes overlapping (22.6% and 28.9% of the total for MCF10AT1 and MFC10CA1H respectively.) 177 genes were found to be significantly altered in all three cell lines with the knockdown of Mettl3.

Gene set enrichment analysis was then performed using ShinyGo (31). In the MCF10A cell line, the most significant change was seen in the Hallmark EMT genes ( Figure 3B ). In the GO Biological Processes, genes involved in Extracellular Matrix (ECM) and structure (ECS) organization, regulation of migration, as well as cell adhesion and motility were most significantly affected. The MCF10AT1 cell line also showed the same cluster of Hallmark EMT genes having the most significant change in fold enrichment ( Figure 3C ). Similar to the MCF10A lines, the knockdown of Mettl3 caused a significant change in the Biological Processes of cell motility and migration as well cell adhesion and ECM and ECS organization. In the MCF10CA1h cell line, Hallmark EMT genes were the third most significant cluster of genes. The two most affected gene clusters were Hallmark Interferon Alpha and Gamma response ( Figure 3D ). Interestingly, in the 177 significant genes that overlapped between all three cell lines, the largest cluster within this gene set belonged to Hallmark EMT genes ( Supplemental Figure 2 ). This suggests the m6A via Mettl3 plays a significant role in mRNA regulation of EMT messages in breast cancer at all stages.

Next, m6A levels in total RNA of the cell lines with and without the knockdown of Mettl3 were quantified using a colorimetric, antibody-based quantification kit. With Mettl3 knockdown, a non-significant decrease in m6A was seen in the MCF10A cell line, ( Figure 4A ). This may be due to the overall low basal level of m6A in the MCF10A cell lines ( Figure 1B ) being near the detection limits of the assay. In both the MCF10AT1 and MCF10CA1h lines a significant decrease in m6A levels was observed in the knockdown candidates compared to the negative CRISPR lines ( Figures 4C, F ). Interestingly, even with an unremarkable knockdown of Mettl3 in the MCF10CA1h line ( Figure 2C ) significant drop in m6A levels compared to the negative CRISPR line was still seen.

As the EMT pathway was the hallmark most significantly affected by METTL3 knockdown in two of the cell lines (and third in the other), changes in expression of several EMT targets with the knockdown of Mettl3 were examined further by real-time PCR. In the MCF10A cell lines, which typically express high levels of epithelial markers, no change was observed in the mRNA expression of E-Cadherin, but N-Cadherin decreased about six-fold in both knockdowns while Vimentin decreased two to four-fold. Conversely, MMP2 saw a two and half fold increase in both knockdowns of Mettl3 in the MCF10A line ( Figure 4B ). For the MCF10AT1 line, an increase in N-Cadherin, MMP2 and Vimentin was observed in both Mettl3 knockdown lines and a decrease in E-cadherin in one of the lines, consistent with a switch to a mesenchymal phenotype ( Figure 4D ). Vimentin was observed to have thirteen times more expression while MMP2 expression increased about 20-fold with the knockdown of Mettl3. Interestingly, N-Cadherin showed a substantial increase of about 250-fold with the knockdown of Mettl3 compared to the Neg CRISPR in both of the cell lines ( Figure 4D ). Interestingly, similar to the phenotypic results, the Mettl3 knockdown in the MCF10CA1h cell line did not show significant effect on mRNA expression of any EMT targets ( Figure 4G ).

With Vimentin emerging as a regulator of EMT (4) as well as a key player in migration, Vimentin protein levels were determined via Western blot and followed the same trend as the mRNA expression in response to METTL3 knockdown across our three cell lines ( Figure 4E ).