Primary culture establishment and characterization

We performed primary culture of tumor tissue from a patient with MMSCC, identifying triple-negative ductal carcinoma with focal squamous metaplasia but no spindle histomorphology. The tumor cells predominantly exhibited intense E-cadherin (E-cad) staining, with some cells exhibiting decreased E-cad staining and increased Vim staining (Fig. 1A). In the corresponding 37 H primary culture, we observed clustered polygonal E-cad+ Vim− cells, some single epithelioid E-cad− Vim+ cells, and a few E-cad+ Vim+ cells and E-cad− Vim− cells (Fig. 1B). The absence of p63-positive tumor cells indicated that the 37 H cells had a ductal but not squamous origin of carcinoma (data not shown). Sequencing of both the 37 H cells and the primary tumor revealed an identical truncating TP53 mutation (p.V73Rfs*76), confirming that they had a shared cellular origin (Fig. 1C). The presence of E-cad− Vim+ or E-cad+ Vim+ cells adjacent to E-cad+ Vim− cells supported the occurrence of spontaneous EMT (Fig. 1D). The wound healing assay revealed an enrichment of E-cad− Vim+ cells at the migratory front, indicating a correlation between the E-cad− Vim+ phenotype and enhanced migratory ability (Fig. 1E).

Fig. 1

Establishment and phenotypic/functional characterization of 37 H primary culture cell lineages. (A) Histomorphology (hematoxylin–eosin staining) and immunohistochemistry (IHC) of the primary tumor tissue. The tumor was composed of conventional ductal carcinoma with focal squamous metaplasia (inset, upper panel). IHC staining for E-cad (middle panel) and Vim (lower panel) is presented. The carcinoma exhibits focal decreased staining for E-cad (arrowhead, middle panel) and increased staining for Vim (arrow, lower panel). The stroma serves as an internal control for Vim staining (star). (B) Phase-contrast and immunofluorescence staining of 37 H primary culture cells revealing a mix of compact polygonal E-cad+ Vim− cells (arrow) and discohesive epithelioid E-cad− Vim+ cells (arrowhead), with occasional E-cad+ Vim+ [E(+)/V(+)] and E-cad− Vim− [E(−)/V(−)] cells. (C) Sanger sequencing for TP53 in the primary tumor tissue and 37 H cells. Loss of p53 expression in the tumor tissue was demonstrated through IHC (lower panel). Preserved p53 protein staining in the intervening stroma (star) served as the internal control. (D) Immunofluorescence staining for the E-cad/Vim (E/V) phenotypes of 37 H cells. E-cad staining is depicted in the inset image. (E) Wound healing assay results for 37 H cells, with percentages of E-cad+ cells and Vim+ cells before migration (0 h) and during migration (22 h after scratching). F–H. Phenotypes of HE and HM cells demonstrated through immunofluorescence staining (F), flow cytometry (G), and Western blotting (H). I. Migration and invasion ability of HE and HM cells, determined using the Transwell assay with and without Matrigel coating. J. Results of soft agar assay, performed to determine anchorage-independent growth ability (mean ± SD, **p < 0.01, ***p < 0.001, unpaired Student t test). K. Respective percentages of HE and HM cells and their progenies with E-cad+ Vim−, E-cad− Vim−, E-cad− Vim+, or E-cad+ Vim+ phenotype during culture for up to 19 passages. Tumor cells were seeded for immunofluorescence staining every two passages, and at least 300 tumor cells were evaluated from different views (magnification, 400×). Representative immunofluorescence images of HE cells and M-sc1 cells at passage 1 (P1), passage 9 (P9), and passage 19 (P19) are presented in the right panel

37 H cell lineage establishment and characterization

We established 37 H cell lineages enriched with E-cad+ Vim− and E-cad− Vim+ cells, which represented predominant epithelial and mesenchymal traits, respectively. Through dilution cloning, we isolated a clone primarily composed of E-cad+ Vim− cells, hereafter referred to as HE cells (Fig. 1F). Because of the enhanced migratory nature of E-cad− Vim+ cells, we subjected 37 H cells to a Transwell migration assay and collected the cells that moved through the membrane, obtaining a cell population predominantly consisting of E-cad− Vim+ cells, hereafter referred to as HM cells. We characterized the immunophenotypes of these cells through immunofluorescence staining, flow cytometry, and Western blotting (Fig. 1F–H, Supplementary Fig. S1). Functional assays revealed that the HM cells exhibited higher migration, invasion, and anchorage-independent growth than did the HE cells (Fig. 1I, J). These findings confirmed that the HE and HM cells are representative of epithelial- and mesenchymal-predominant 37 H cells, respectively. Subsequently, we performed single-cell cloning of the HE and HM cells. The HE subclones included epithelial-predominant, mesenchymal-predominant, and mixed epithelial and mesenchymal cells, whereas the HM subclones primarily included mesenchymal-predominant cells (Supplementary Fig. S2 and S3). Additionally, mesenchymal-predominant M-sc1 and M-sc5 cells exhibited higher migration, invasion, and anchorage-independent growth than did the epithelial-predominant E-sc60 and E-sc66 cells (Supplementary Fig. S4 and S5). We explored the dynamic changes in the epithelial and mesenchymal phenotypes of the HE- and HM-type cells. All the three HE-type cells demonstrated spontaneous EMT during culture passages, as indicated by the gradual replacement of E-cad+ Vim− cells with E-cad+ Vim+ and E-cad− Vim+ cells (Fig. 1K). By contrast, the three HM-type cells consistently maintained a mesenchymal phenotype. Although the M-sc1 cells briefly underwent mesenchymal–epithelial transition (MET), they gradually reverted to the mesenchymal phenotype.

Identification of transcriptional regulators involved in spontaneous EMT

Because HM cells are the mesenchymal counterparts of HE cells, the differentially enriched TFs in HM cells may serve as EMT regulators. To mitigate potential clonal heterogeneity, we included two subclones from each cell type in our RNA-seq analysis. The HE and HE-derived E-sc60 and E-sc66 cells formed the E group, whereas the HM and HM-derived M-sc1 and M-sc5 cells formed the M group. The clustering heatmap of DEGs revealed a strong genetic relatedness among the cells and subclones within the E and M groups (Fig. 2A). Subsequently, GSEA was conducted to identify differential hallmark molecular pathways enriched in the E and M groups. EMT, MYC, UV response, and angiogenesis were identified as the main mechanisms in the M group (Fig. 2B). MYC is involved in EMT function [22], and angiogenesis has been shown to link EMT-induced cancer stemness to tumor initiation [23]. Additionally, UV DNA damage has been shown to promote EMT-like changes [24]. The upregulation of these pathways in the M group is consistent with its functional correlation with EMT. Furthermore, GSEA performed using external gene sets associated with EMT [25, 26] indicated upregulation of the EMT signature in the M group (Fig. 2C). These results confirmed the suitability of our model for exploring EMT-TFs. To identify potential EMT-TFs in this model, we focused on TFs exhibiting expression levels in the M group that were higher than those in the E group. We combined the results of RNA-seq and the TRRUST database [27] and selected the top two TFs, namely HLX and CREB3L1, and the top two well-characterized EMT regulators, ZEB1 and ZEB2, as candidates for further validation. However, subsequent Western blotting analysis revealed considerably low HLX expression in the HM cells, leading us to exclude HLX from further analysis. To determine the involvement of the selected TFs in spontaneous EMT, we performed shRNA knockdown to silence the corresponding genes in the HM cells. We used pooled cells to reduce the effect of clone heterogeneity. Knockdown of ZEB1 or ZEB2 in the HM cells resulted in a transition from the mesenchymal (E-cad− Vim+) to the epithelial (E-cad+ Vim−) phenotype (Fig. 2D–F, Supplementary Fig. S6), which was accompanied by reduced migration, invasion, and anchorage-independent growth (Fig. 2G). Notably, shZEB1 and shZEB2 downregulated ZEB1, ZEB2 and CREB3L1. By contrast, CREB3L1 knockdown in the HM cells did not result in a reversal to the epithelial phenotype (Fig. 2D–F) and had a moderate effect on invasion; no significant reduction in migration or soft agar colony formation was observed (Fig. 2G). These findings support a combined role for ZEB1 and ZEB2 in spontaneous EMT in this model.

Fig. 2

Identification and validation of spontaneous EMT regulators. (A) Heatmap showing DEG (limma adjusted p < 0.05) clustered by E (HE, E-sc60, and E-sc66) and M (HM, M-sc1, and M-sc5) groups. (B) Bar plot denoting statistical hallmark molecular pathways (FDR < 5%) enriched in the E and M groups, identified using GSEA. The EMT enrichment plot is presented on the right. NES, normalized enrichment score; FDR, false discovery rate. (C) Heatmaps depicting EMT signature genes identified by Tan et al. (PMID 25214461) and Salt et al. (PMID 24302555). D–F. Phenotypic changes in HM cells, assessed through Western blotting (D), flow cytometry (E), and immunofluorescence staining (F), following lentivirus-mediated knockdown of ZEB1 (shZEB1), ZEB2 (shZEB2), or CREB3L1 (shCREB3L1) versus control (shCTRL). HM cells with incomplete knockdown of ZEB1 or ZEB2 (arrow) exhibited a mesenchymal phenotype, as evidenced by immunofluorescence staining results showing decreased E-cad and increased Vim (arrow). G. Transwell assay without (left panel) or with (middle panel) Matrigel coating and soft agar assay (right panel) performed to assess migration, invasion, and anchorage-independent growth, respectively, in HM cells with knockdown of ZEB1, ZEB2, or CREB3L1 compared with that in a control knockdown (mean ± SD, *p < 0.05, **p < 0.01, ***p < 0.001, ns, not significant, unpaired Student t test)

Association between expression of EMT-TFs and spontaneous EMT

We investigated the association between the expression of these TFs and the occurrence of spontaneous EMT. All three TFs were consistently expressed in mesenchymal HM cells. These TFs were also observed in HE cells undergoing spontaneous EMT, as indicated by expression of Vim or loss of E-cad expression (Fig. 3A). Subsequently, we investigated whether the TFs’ high expression levels were correlated with spontaneous EMT during culture passages. In line with the spontaneous EMT observed in the HE-type cells (Fig. 1K), qRT-PCR results indicated the occurrence of spontaneous EMT in the HE cells that was characterized by an increase in VIM expression and a concurrent decrease in CDH1 expression (Fig. 3B). The trend of change in ZEB1, ZEB2, and CREB3L1 expression aligned with spontaneous EMT. By contrast, the mesenchymal cells exhibited a more consistent mesenchymal phenotype. Overall, these findings indicate that ZEB1 and ZEB2 are EMT regulators, whereas CREB3L1 is a marker correlated with spontaneous EMT.

Fig. 3

Expression of ZEB1, ZEB2, and CREB3L1 was correlated with the occurrence of spontaneous EMT. (A) Immunofluorescence staining revealing that ZEB1, ZEB2, and CREB3L1 were consistently expressed in HM cells (upper panel) but only observed in a few HE cells undergoing spontaneous EMT (arrow), as evidenced by a loss of E-cad expression (middle panel) or the presence of Vim (lower panel). (B) qRT-PCR for RNA expression abundance of CDH1 and VIM and the three TFs: ZEB1, ZEB2, and CREB3L1. HE and HM cells were cultivated for 30 passages, and mRNA samples were collected and analyzed at passages 1, 6, 12, 14, 16, 18, 22, 26, and 30 through qRT-PCR. Quantitative values were determined on the basis of the threshold cycle number (Ct), and the fold-change in expression was calculated using the 2−ΔΔCt method. Target gene measurements in all samples were normalized to the internal control gene GAPDH. The expression levels of corresponding genes on day 1 served as the baseline for result adjustments. (C) Spontaneous MET in HM cell xenografts. A schematic outline of the animal study workflow is shown (left panel). For each mouse, 1.5 × 106 HE or HM cells mixed with Matrigel matrix were implanted into the mammary fat pads of 14–17-week-old female NOG mice (n = 5), and all tumors in the mammary fat pads were collected on day 58. Representative hematoxylin–eosin staining (magnification, 40× and 600×) and immunofluorescence staining for CDH1 and Vim with DAPI counterstaining for sample with HE or HM cells are presented (middle panel). Cancer cells exhibiting the CDH1−/Vim+, CDH1+/Vim+, CDH1+/Vim−, and CDH−/Vim− phenotype are marked by an arrowhead, arrow, star, and double arrow, respectively. A lower power view for CDH1 staining is displayed in the inset image. The dot plot (right panel) illustrates the largest tumor size in each mouse implanted with HE or HM cells, measured as length (L) × width (W) (mean ± SD, *p < 0.05, unpaired Student t test)

Essential role of MET in the growth of tumor implants

We conducted a xenograft study to determine whether the tumor growth of HM cells exhibiting EMT phenotypes surpassed that of HE cells (Fig. 3C). Tumor formation was observed in all five HE cell xenografts but only in four out of five HM cell xenografts. Furthermore, lower tumor growth was observed in HM cell xenografts than in HE cell xenografts. In addition, in contrast to the predominant E-cad− Vim+ phenotype of HM parental cells, the majority of the cells in the four tumor implants exhibited an E-cad+ Vim− profile, indicating that EMT did not confer a growth advantage to the implants and that MET is crucial for secondary tumor outgrowth.

Validation of the occurrence of spontaneous EMT and establishment of its correlation with both ZEB1 and ZEB2 through scRNA-seq analysis

Although the HE and HM cells were enriched for the epithelial and mesenchymal phenotypes, respectively, both types of cells may include tumor cells progressing at various stages along the EMT/MET continuum, which could lead to heterogeneity in their gene expression. To investigate the characteristics and underlying regulators of spontaneous EMT and mitigate the potential effect of such heterogeneity, we conducted scRNA-seq on 37 H parental cells representing a spectrum of carcinoma cells advancing at different stages along the EMT course, as indicated by their phenotypes spanning the EMT spectrum (Fig. 4A).

Fig. 4

Overview of 7749 single cells derived from 37 H primary culture parental cells. A. Immunofluorescence staining (left) and flow cytometry (right) revealing expression of E-cad and Vim in 37 H cells. B, C. UMAP plots of the 7749 cells, color-coded by relevant cluster (B) or the expression of the marker gene CDH1 (left) or VIM (right) (C). D. Violin plots illustrating the expression of the epithelial genes VIM and CDH2 and the mesenchymal genes CDH1 and EpCAM across the 11 identified clusters in the UMAP plot. These clusters are categorized as epithelial-predominant (E cluster: clusters 1, 2, 4, and 6–9), mesenchymal-predominant (M cluster: clusters 3), and hybrid (H cluster: clusters 5, 10, and 11) on the basis of the expression levels of these mesenchymal and epithelial markers. E. UMAP plots depicting the three TFs upregulated in HM cells, ZEB1, ZEB2, and CREB3L1, along with the three core EMT-TFs, SNAI1, SNAI2, and TWIST1. F. Violin plots displaying the expression levels of ZEB1, ZEB2, SNAI1, SNAI2, and TWIST1 across the 11 clusters. G UMAP plots generated for cells stratified by VIM expression into six categories: VIM 0–1 (0 < log2Vim expression ≤ 1), VIM 1–2, VIM 2–3, VIM 3–4, VIM 4–5, and VIM 5–6. H. Violin plots illustrating expression of ZEB1, ZEB2, and CREB3L1 and the epithelial marker CDH1, stratified by VIM expression

As indicated by the expression of the mesenchymal markers VIM and CDH2 and the epithelial markers CDH1 and EpCAM, unsupervised UMAP revealed 11 clusters broadly corresponding to epithelial-predominant clusters (E cluster: clusters 1, 2, 4, and 6–9), a mesenchymal-predominant cluster (M cluster: clusters 3), and hybrid clusters (H cluster: clusters 5, 10, and 11; Fig. 4B–D). This finding confirms the presence of EMP, including cells from the CDH1+ VIM− epithelial end to the CDH1− VIM+ mesenchymal end. ZEB1, ZEB2, and CREB3L1, which were upregulated in mesenchymal-predominant HM cells (Supplementary Table S3), were enriched in the M cluster (Fig. 4E).

Whereas elevated expression of ZEB1 and ZEB2 in the M cluster and a lack of evident expression of ZEB1 and ZEB2 in the E cluster were noted, such correlations were not observed in the remaining core EMT-TFs SNAI1, SNAI2, and TWIST1 (Fig. 4E, F) or in the additional well-characterized EMT-associated TFs [4], with the exception of ID1 (Supplementary Fig. S7). We additionally stratified the cells on the basis of their levels of expression of the mesenchymal marker VIM, that is, from the lowest (VIM 0–1) to the highest expression (VIM 5–6; Fig. 4G and Supplementary Fig. S8). ZEB1 and ZEB2 expression were both respectively positively and negatively correlated with VIM and CDH1 expression, whereas CREB3L1 upregulation was observed only in the mesenchymal end (Fig. 4H). Consistently, the expression of the mesenchymal and epithelial markers was respectively positively and negatively correlated with both ZEB1 and ZEB2 expression (Supplementary Fig. S9). Together, these findings indicate that ZEB1 and ZEB2 are spontaneous EMT regulators.

Identification of the TFs upregulated sequentially as cancer cells progress through the EMT course and confirmation of ZEB1’s pivotal role as a regulator of spontaneous EMT

We examined TFs that sequentially and significantly upregulated as cancer cells progressed through the EMT course, which correlated with the increasing expression of VIM. We discovered a highly significant correlation of high VIM expression levels with the NESs of the EMT pathway, as well as the EMT score [21], which provides strong support for the use of high VIM expression as an indicator of EMT status (Fig. 5A and C, Supplementary Table S4, and Supplementary Fig. S10). We identified ZEB1 as the earliest significantly upregulated TF, with its upregulation beginning at the VIM 2–3 stage. Immediately after ZEB1 was upregulated, ZEB2, MXD4, ID1, ID3, SMARCA1, FOXO1, ZBTB16, and HIPK2 were significantly upregulated (VIM 3–4), whereas CREB3L1, SNAI2, CITED2, TWIST1, ZNF300, CREB3, and NFKB1A were significantly upregulated only at the final EMT stage (VIM 5–6; Fig. 5D, E, Supplementary Table S5 and Fig. S11). For the great majority of these TFs, after they were upregulated, they remained upregulated throughout the EMT course. The progressive increase in the expression of these early upregulated TFs (VIM 3–4) during EMT advancement (Fig. 5E), along with their heightened levels in the mesenchymal M cluster and their downregulation following ZEB1-knockdown-mediated EMT reversion (Fig. 5F), substantiates their role involving in acquiring the mesenchymal phenotype during the EMT process.

Fig. 5

TFs that were differentially and significantly upregulated as cancer cells progressed through the EMT course, which correlated with increasing VIM expression. (A) Rank correlation of Hallmark signaling pathways. The X axis presents the expression status of VIM in cells, and the Y axis presents the NESs of molecular signatures. The correlation between VIM expression status and the NESs of molecular signatures was measured using Spearman’s rank correlation. (B) Violin plot illustrating hallmark EMT NESs of cells stratified by VIM expression abundance. (C) Violin plot illustrating EMT score (Byers LA et al.) of cells stratified by VIM expression abundance. (D) Bubble plot depicting differentially significantly upregulated TFs (p < 0.05, log2FC ≥ 1) between each VIM subgroup versus VIM 0–1 subgroup. (E) Violin plots illustrating expression levels of significantly upregulated TFs as cancer cells progress through the EMT course, which correlated with increasing VIM expression. (F) Bar plot illustrating the ratio of RNA level changes for early significantly upregulated TFs in mesenchymal-predominant HM cells following lentivirus-mediated knockdown of ZEB1 (shZEB1) or control (shCTRL), assessed through qRT-PCR. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. ns, not significant. (G) Violin plots illustrating expression of SNAI1, SNAI2, and TWIST1 stratified by VIM expression. (H) Immunofluorescence staining for ZEB1 and Vim in 37 H cells. Carcinoma cells expressing Vim are marked with arrows

Unlike ZEB1 and ZEB2, the core EMT regulators SNAI2 and TWIST1 were significantly upregulated only at the final stage of EMT (VIM 5–6), whereas SNAI1 consistently maintained minimal expression throughout the entire EMT course (Fig. 5G and Supplementary Fig. S12, 13). The precise correlation between ZEB1 and VIM expression during EMT progression (Fig. 5E) was further confirmed through immunofluorescence staining (Fig. 5H). Because the primary tumor exhibited triple negativity, we investigated the effect of ZEB1 on the upregulation of Vim in a TNBC cohort and identified a strong correlation between ZEB1 and Vim expression (Fig. 6A, B). We subsequently examined the involvement of ZEB1 expression in EMT acquisition in a cohort of MpBC with paired ductal carcinomatous and spindle metaplastic components, with spindle metaplastic components being considered the EMT counterpart of paired ductal carcinomatous components [15, 16, 28]. Our observations revealed coexpression of Vim and ZEB1 in all spindle metaplastic components but a significantly lower expression ratio of both markers in the paired carcinomatous components (Fig. 6C–E). The expression of ZEB2 was also enriched in spindle metaplastic components (Supplementary Fig. S14). In addition, we observed consistent expression of Vim and ZEB1 in the carcinomatous components. These findings confirm the pivotal role of ZEB1 as an EMT regulator in breast cancer.

Fig. 6

Correlation of ZEB1 with Vim expression and EMT in the TNBC and MpBC cohorts, and prognostic impact of upregulated gene signatures as cancer cells progress through the EMT process. A, B. Bar plot (A) illustrating the frequency of ZEB1 expression in 56 TNBC cases with or without Vim expression, accompanied by representative IHC (B) depicting Vim and ZEB1 expression. Arrows indicate carcinoma regions. C. Bar plot illustrating the positivity ratio of Vim and ZEB1 staining in the conventional ductal carcinoma (NST) components and spindle carcinomatous (SPS) components across 20 cases of MpBC with paired NST and SPS components. D. Pie plot illustrating the paired status of Vim and ZEB1 staining in the NST components in the 20 cases of MpBC. E. Histomorphology (hematoxylin–eosin staining; upper) and IHC staining for Vim (middle) and ZEB1 (lower) in paired NST and SPS components in a representative case of MpBC. The few ZEB1-stained cells are marked by an arrow (Magnification, 200×). F, G. Kaplan–Meier plots (F) and Cox proportional hazard models (G) used to compare disease-free intervals and overall survival among TCGA TNBC cases, stratified by high and low expression levels of differentially overexpressed EMT-related genes at various stages of VIM (VIM 2–3, VIM 3–4, VIM 4–5, and VIM 5–6) compared with those at VIM 0–2 stages. Samples were assessed through gene set variation analysis based on the expression values of differentially expressed genes and then categorized into high-expression and low-expression groups by using the mean value. The log-rank test was used to calculate p values

Unfavorable prognostic implications associated with gene signatures elevated during the EMT process

We investigated the prognostic implications of gene signatures that are upregulated during the EMT process, particularly those overlapping with genes in the hallmark EMT pathway. We determined that gene signatures overexpressed in cancer cells progressing through the EMT stages (VIM 2–6) compared with those in the VIM 0–2 stages demonstrated a trend toward an unfavorable prognostic effect on survival, DFI, and OS. This trend was particularly significant in the middle (VIM 4–5) and late (VIM 5–6) EMT stages (Fig. 6F, G). The results highlight the unfavorable prognostic significance of gene signatures being elevated during the EMT process, validating the biological relevance of this model.

Elevated EMT gene signatures, along with increased expression of ZEB1, ZEB2, and CREB3L1, in Vim+ compared with Vim− carcinoma cells within the corresponding primary tumor tissue

Finally, we validated our findings in the corresponding primary tumor tissue. Using DSP for RNA-seq [29], we compared the RNA expression in paired Vim+ versus Vim− carcinoma cells collected from nine ROIs in the tumor tissue section (Fig. 7A, B). An unsupervised clustering heatmap revealed a clear distinction in gene expression between Vim+ and Vim− carcinoma cells (Fig. 7C). Subsequently, GSEA revealed that the EMT hallmark played a significant role in distinguishing Vim+ from Vim− carcinoma cells (Fig. 7D). Further analysis of differences in the expression of ZEB1, ZEB2, and CREB3L1, which are included in the GeoMx Cancer Transcriptome Atlas Panel gene list, revealed significant enrichment in Vim+ carcinoma cells relative to that in Vim− carcinoma cells from the nine ROIs (Fig. 7E). For comparison, the expression of SNAI2 and TWIST1 is significantly higher in Vim+ tumor cells compared to Vim− tumor cells, whereas SNAI1 expression shows no significant difference between Vim+ and Vim− tumor cells (Supplementary Fig. S15). Moreover, a heatmap depicting ZEB1, ZEB2, CREB3L1, and CDH1 across the nine ROIs revealed clear segregation between Vim+ and Vim− carcinoma cells (Fig. 7F). These in vivo findings corroborate the data obtained from the primary culture model.

Fig. 7

In situ validation of spontaneous EMT and EMT regulators/markers derived from the primary culture model. (A) Schematic of in situ RNA-seq (digital spatial profiling) of CK+/− Vim+ versus CK+ Vim− tumor cells from the primary tumor sample, determined using GeoMax RNA assays. tcs, target complementary DNA sequence; upl, UV photocleavable linker; do, DNA oligo tag; Ab, antibody. (B) Representative images of ROI and AOI segmentation in the primary tumor section. In total, nine ROIs (ROI1–9) were selected (white line) by using the morphology markers CK (green) to highlight carcinoma nests and Vim (red) to identify carcinoma cells expressing Vim (left panel). A magnified image of ROI8 is presented (middle panel), with Vim+ carcinoma cells within the ROI8 marked by an arrow. The intervening stroma components, which were Vim+ (star), were avoided when ROIs were selected. In ROI8, the AOI segmentation of Vim+ carcinoma cells (right upper panel) was UV illuminated; the DNA oligo tag was collected first, and the AOIs for Vim− carcinoma cells were collected sequentially (right lower panel). (C) Unsupervised hierarchical clustering heatmap presenting clustering of expressed genes from paired Vim+ versus Vim− carcinoma cell AOIs of the nine ROIs. (D) Bar plot depicting statistical hallmark molecular pathways (FDR < 5%) enriched in Vim+ carcinoma cells, determined by using GSEA (upper panel), as well as visualization of the EMT enrichment plot of EMT (left lower panel) and heatmap of leading genes associated with Vim+ group hallmark molecular processes (right lower panel). E Differences in expression of ZEB1, ZEB2, CREB3L1, and CDH1 between paired Vim+ and Vim− tumor cells from the nine ROIs are plotted, with the paired t test used for testing these differences. N.R., normalized reads. F. Heatmap of expression of ZEB1, ZEB2, CREB3L1, and CDH1 in the nine ROIs