Tropomyosin1 isoforms underlie epithelial to mesenchymal plasticity, metastatic dissemination, and resistance to chemotherapy in high-grade serous ovarian cancer

Phenotypic plasticity, defined as the ability of individual cells with stable genotypes to exert different phenotypes upon exposure to specific environmental cues, represent the quintessential hallmark of the cancer cell en route from the primary lesion to distant organ sites where metastatic colonization will occur. Phenotypic plasticity is driven by a broad spectrum of epigenetic mechanisms that allow for the reversibility of epithelial-to-mesenchymal and mesenchymal-to-epithelial transitions (EMT/MET). By taking advantage of the co-existence of epithelial and quasi-mesenchymal cells within immortalized cancer cell lines, we have analyzed the role of EMT-related gene isoforms in the regulation of epithelial mesenchymal plasticity (EMP) in high grade serous ovarian cancer. When compared with colon cancer, a distinct spectrum of downstream targets characterizes quasi-mesenchymal ovarian cancer cells, likely to reflect the different modalities of metastasis formation between these two types of malignancy, i.e. hematogenous in colon and transcoelomic in ovarian cancer. Moreover, upstream RNA-binding proteins differentially expressed between epithelial and quasi-mesenchymal subpopulations of ovarian cancer cells were identified that underlie differential regulation of EMT-related isoforms. In particular, the up- and down-regulation of RBM24 and ESRP1, respectively, represent a main regulator of EMT in ovarian cancer cells. To validate the functional and clinical relevance of our approach, we selected and functionally analyzed the Tropomyosin 1 gene (TPM1), encoding for a protein that specifies the functional characteristics of individual actin filaments in contractile cells, among the ovarian-specific downstream AS targets. The low-molecular weight Tpm1.8/9 isoforms are specifically expressed in patient-derived ascites and promote invasion through activation of EMT and Wnt signaling, together with a broad spectrum of inflammation-related pathways. Moreover, Tpm1.8/9 expression confers resistance to taxane- and platinum-based chemotherapy. Small molecule inhibitors that target the Tpm1 isoforms support targeting Tpm1.8/9 as therapeutic targets for the development of future tailor-made clinical interventions.


INTRODUCTION
Epithelial ovarian cancer (EOC) is the leading cause of death amongst gynecologic malignancies due to its high case-to-fatality ratio [1,2].EOC generally becomes manifest at advanced disease stages, i.e. when metastases have already spread to pelvic organs (stage II), the abdomen (stage III), or beyond the peritoneal cavity (stage IV) [3].Based on the underlying genetic defects, two main EOC subtypes have been recognized [4].Type I tumors are slow growing, mostly restricted to the ovary, and thought to arise from well-differentiated precursor lesions called "borderline" tumors.They are further subdivided into low-grade serous, mucinous, clear cell, and endometrioid subtypes.Mutations in KRAS, BRAF, PTEN, and CTNNB1 (β-catenin) earmarks type I EOCs, often together with a relatively stable karyotype.High-grade serous (HGSOC) and undifferentiated carcinomas are type II EOCs and are frequently characterized by TP53 mutations and by aneuploidy [4].HGSOC represents the most malignant and common ovarian cancer type accounting for up to 70% of all cases with poor prognosis and survival [5].
Of note, EOC is the only cancer type where no physical barrier exists between primary lesion and the main metastatic site, i.e. the intraperitoneal cavity.The dissemination of ovarian cancer cells results in their adhesion to intra-abdominal organs and the peritoneum [6] eventually leading to ascites accumulation due to the obstruction of lymphatic vessels [7].
Epithelial-to-mesenchymal (EMT) and mesenchymal-toepithelial transitions (MET) are thought to underlie local dissemination and distant metastasis in the majority of epithelial malignancies, including ovarian cancer where, as mentioned above, the absence of a physical barrier would suggest alternative shedding mechanisms [8].EMT/MET are regulated by a broad spectrum of epigenetic mechanisms involving chromatin remodeling due to histone methylation/acetylation, non-coding RNAs, promoter DNA methylation, and post-transcriptional mechanisms such as alternative splicing (AS) [9].AS occurs in the majority of human genes and represents a major determinant of protein diversity [10].A broad spectrum of RNA-binding proteins (RBPs) regulate splicing by recognizing specific sequences within (pre) mRNA and are known to play a role in AS through their differential expression among different tissues in homeostasis and disease [11][12][13][14].RBP-driven alternative splicing is thought to play a crucial role in cancer metastasis, as well as in EMT/MET [15].Among the various RBPs, ESRP1 (epithelial splicing regulatory protein 1) was shown to regulate splicing of multiple downstream targets during EMT in different cancer types [16].
Previously, we have shown that ESRP1, together with other RBPcoding genes, are differentially expressed between epithelial and quasi-mesenchymal colon cancer cells as the result of their role in the regulation of alternative splicing downstream of EMT [17].Among several AS targets, the CD44 and NUMB genes were shown to play relevant functional roles to promote EMT and phenotypic plasticity in colon cancer.Here, we have employed a similar approach to study the role of alternative splicing and alternative promoter usage in EOC metastasis.Systematic analysis of the downstream EMT isoform targets only revealed a small overlap with what was previously identified in colon cancer [17].Accordingly, the majority of the targets were ovarian cancerspecific as illustrated by the Tropomyosin 1 (TPM1) gene that encodes for various actin-binding protein isoforms that specify the functional characteristics of individual actin filaments in muscle and non-muscle cells [18].TPM1 was previously shown to suppress tumor development in multiple cancer types through distinct mechanisms [19,20].Here, we established that the low molecular weight isoforms Tpm1.8/9 play a key role in EMT, Wnt signaling, and chemo-resistance in ovarian cancer and represent a candidate target for future therapies as shown by the in vitro results obtained both by isoform-specific siRNA and small molecule inhibitors.

RESULTS
A subpopulation of quasi-mesenchymal cells co-exists with epithelial cells in high-grade serous ovarian cancer Following the experimental strategy previously adopted to identify subpopulations of quasi-mesenchymal cells in immortalized colon cancer cell lines [17,21], we analyzed the HGSOC cell lines OV90, SKOV3, COV504, and CAOV3 by FACS with CD44 and EpCAM antibodies.As shown in Fig. 1A, distinct distributions of CD44 hi EpCAM hi and CD44 hi EpCAM lo cells (from here on referred to as EpCAM hi and EpCAM lo , respectively) earmarked each line albeit in different percentages.Upon sorting and short-term culturing, EpCAM lo cells showed a mesenchymal-like morphology, in contrast with the epithelial appearances of their EpCAM hi counterparts (Fig. 1B).Accordingly, EpCAM lo cells showed increased migration and invasion capacity in trans-well assays (Fig. 1C).
In order to elucidate the global gene expression profiles of the two distinct subpopulations of ovarian cancer cells, RNAseq analysis was carried out on the EpCAM hi/lo cells sorted by FACS from the OV90 and CAOV3 lines.Principal component analysis (PCA) by multidimensional scaling (MDS) revealed a clear separation of the quasi-mesenchymal EpCAM lo cells from their epithelial counterpart in the second dimension (Fig. 1D).Unsupervised clustering of the RNAseq data highlighted differentially expressed genes between the EpCAM lo and EpCAM hi subpopulations from the two cell lines (Fig. 1E).Among these genes, different EMT transcription factors (EMT-TFs; ZEB1/2, RUNX2, SNAI2), EMT target genes (CDH1, VIM), and EMT-related RBPs (ESRP1, RBM24) earmarked EpCAM lo cells in both OV90 and CAOV3.Accordingly, pathway analysis (PA) of the genes whose expression earmarks the quasi-mesenchymal ovarian cancer cells revealed significant associations with EMT, KRAS signaling, and several TNF-and interferon-related inflammatory pathways (Fig. 1F).Pathways with a normalized enrichment score (NES) > 1 and P value < 0.05 were labeled as significant.To confirm the central role played by EMT in the establishment of the EpCAM lo identity as indicated by the PA analysis, ZEB1 expression was downregulated by shRNA in the OV90, SKOV3, and COV504 cell lines (Fig. 1G).As expected, FACS analysis of these cells shows that knockdown of the differentially expressed EMT-TF results in a substantial reduction of the EpCAM lo subpopulation (Fig. 1H).
To demonstrate the clinical relevance of the above data obtained from immortalized cancer cell lines, publicly available single-cell RNAseq data from 42 patient-derived ovarian cancers [22] were employed to search for quasi-mesenchymal cells reminiscent of the EpCAM lo subpopulation.To this aim, we took advantage of the EpCAM lo signature derived from the above RNAseq analysis of the OV90 and CAOV3 cell lines (Supplementary Table 1).After classifying patient-derived cancer cells according to the EpCAM lo signature (Fig. 2A), approx.60% were found to cluster into one of the three subpopulations of HGS ovarian cancer cells defined in the original study by Vazquez-Garcia et al. [22] as EMTlike (labeled as #2, #4, and #6) (Fig. 2B).Of note, a substantial fraction of the EpCAM lo -like cells fell outside the three clusters indicative of large heterogeneity of cellular identities reminiscent of the quasi-mesenchymal state in the cell lines.As shown in Fig. 2C, cells from within the EpCAM lo -like clusters were more represented in metastases and ascites (i.e.non-adnexa including omentum, peritoneum, etc.) when compared with primary tumors.We then evaluated module scores from the Hallmark gene sets across the EpCAM lo -like clusters in primary ovarian cancers (adnexa), ascites, and metastases (non-adnexa).As shown in Fig. 2D, E, EMT-like signature and inflammatory signaling pathways (TNFα, IL6/Jak/Stat3, IFN, and TGFβ) were significantly upregulated in EpCAM lo -like clusters, in particular in cluster #4 in ascites and metastatic lesions.
Collectively, these results indicate that, HGSOC cell lines encompass subpopulations of quasi-mesenchymal cells endowed with increased motility and invasive capacity, and characterized by EMT-TFs the expression of which is central to their cellular identity.Similar subpopulations of ovarian cancer cells are found in patient-derived primary and metastatic lesions, and in malignant ascites.
Differential expression of RNA-binding proteins underlies alternative splicing of a subset of target genes in quasimesenchymal ovarian cancer cells Recently, we showed that several RNA-binding proteins (RBPs) known to be involved in alternative splicing are differentially expressed between EpCAM lo and EpCAM hi cells in colon cancer cell lines and play significant functional roles in controlling E-to-M and M-to-E transitions and phenotypic plasticity during local dissemination and distant metastasis [17].By following an analogous approach, we have identified several differentially expressed RBPs likely to play an active role in the regulation of EMT-related alternative isoforms between epithelial-and mesenchymal-like subpopulations of ovarian cancer cells.As shown in Fig. 3A, ESRP1 and RBM24 were respectively downand upregulated in EpCAM lo cells in both cell lines.Moreover, other RBPs (i.e.ESRP2, RBM47, RBMS3, and QKI) were differentially expressed in one of the cell lines examined.The EpCAM lo -specific up-and downregulation of RBM24 and ESRP1, respectively, was validated by RTqPCR and western analysis of the subpopulations sorted from the OV90, CAOV3 and COV504 cell lines (Fig. 3B, C).
Next, we analyzed differentially spliced genes by MISO (Mixture of Isoforms) [23] and filtered the results by selecting ΔPSI (differential Percentage Spliced In) values > 10% (Supplementary Tables 2 and 3), and by comparing them with the corresponding lists of isoform targets previously found in colon cancer cell lines [17].As shown in Supplementary Fig. 3 Supplement 1, the large majority of ovarian cancer targets was not found in colon cancer.Among the ovarian-specific targets with a known function in EMT (n = 39), diverse cellular components, e.g.extracellular matrix, focal adhesion, and the actin cytoskeleton, and cellular processes, e.g.ECM organization, integrin-and TGFβ-mediated signaling, and cell migration were represented (Supplementary Fig. 3, Supplement 1).In particular, the presence of Tropomyosin 1 (TPM1), a member of a broad family of actin-binding proteins, was noteworthy both because of the high ΔPSI values in OV90 and CAOV3 cells and because of its known cellular function, i.e. in the cytoskeleton of non-muscle cells and in the contractile system of striated and smooth muscles [24,25].Although a direct causative role between TPM1 and EMT has not been reported yet, previous studies have shown that TGF-β signaling increases expression of high-molecular weight tropomyosins together with the formation of actin stress fibers thus affecting cell motility and invasion [26,27].
In order to establish a cause-effect relationship between the differential expression of ESRP1 and RBM24 in EpCAM lo ovarian cancer cells and the observed downstream AS targets using TPM1 as a model, we performed RBP knockdown and overexpression assays in OV90 and COV504 cell lines.First, as predicted by their unique EpCAM hi/lo distribution, the PEA1 and PEA2 cell lines exclusively express RBM24 and ESRP1, respectively (Fig. 3G, right panel).Accordingly, both ESRP1 knockdown and RBM24 ectopic expression in the OV90 and COV504 cell lines resulted in the up-and downregulation of the Tpm1.8/9 and Tpm1.6/7 isoforms, respectively, both at the RNA and protein levels (Fig. 4A, Supplementary Fig. 4-Suppl.1).Vice versa, RBM24 knockdown and ESRP1 ectopic expression resulted in the up-and downregulation of the Tpm1.6/7 and Tpm1.8/9 isoforms, respectively (Fig. 4B, Supplementary Fig. 4-Suppl.1).Of note, the observed changes in RBP expression were also accompanied by a pronounced increase of the EpCAM lo subpopulation as shown by FACS analysis (Fig. 4C, Supplementary Fig. 4-Suppl.2).Concurrent RBM24 overexpression and ESRP1 knockdown had significant effects not only on the TPM1 isoform shift but, more importantly, on the relative percentages of the EpCAM hi/lo subpopulations when compared with the single RBP gainand loss-of-function assays (Fig. 4, Supplementary Fig. 4-Suppl.2).
RBPs have been known to play essential roles not only in alternative splicing but also in the co-transcriptional regulation of gene promoters.As the differentially expressed Tpm1.6/7 and 8/9 isoforms are more likely to result from alternative first exon selection rather than alternative splicing, we performed an in silico analysis of both TPM1a/b promoters using the FANTOM database [28].No ESRP1 or RBM24 binding motifs were detected.However, the same approach indicated potential interaction with the EMT-TFs SNAI1/2, ZEB1, and RUNX2 (Supplementary Fig. 4-Suppl.3A, B).
To further study the alternative usage of the TPM1a (Tpm1.6/7)and TPM1b (Tpm1.8/9)promoters, we cloned both promoters into luciferase constructs to assess their transcriptional activity by transient reporter assays.As shown in Supplementary Fig. 4-Suppl.3C, D, an increase in reporter activity was observed from the TPM1a promoter in the sorted EpCAM hi and in ESRP1overexpressing OV90 cells.Likewise, increased reporter activity was observed from the TPM1b promoter in the sorted EpCAM lo and in RBM24-overexpressing OV90 and COV504 cells (albeit marginally).Conversely, decreased reporter activity was observed from the TPM1a promoter in sorted OV90 and COV504 EpCAM hi cells upon ESRP1-knockdown.Similarly, decreased reporter activity was observed from the TPM1b promoter in sorted OV90 and COV504 EpCAM lo cells upon RBM24 knockdown (Supplementary Fig. 4-Suppl.3E-H).
Overall, these results showed that epithelial and quasimesenchymal subpopulations of HGS ovarian cancer cells are earmarked by the differential expression of specific RBPs known to be involved in alternative splicing and co-transcriptional regulation of alternative promoters.Accordingly, differential isoform patterns of a subset of EMT-related target genes characterize the EpCAM lo cells.While some of the target gene isoforms are shared with those previously identified in colon cancer, the majority appear to be specific for ovarian cancer.Among these, TPM1 is of interest in view of both its function as an actin-binding cytoskeletal protein in various cell types [18] and of its previously reported role as a tumor suppressor in breast cancer [29].

Transcriptional and functional consequences of TPM1 isoforms on quasi-mesenchymal ovarian cancer cells
To assess the functional relevance of the specific TPM1 isoforms, their ectopic expression was induced and validated by RT-qPCR and western blot analysis in multiple ovarian cancer cell lines (OV90, COV504, PEA1 and PEA2) (Fig. 5A, B).Cell viability and proliferation assays indicated that Tpm1.6/7 and Tpm1.8/9 expression was significantly associated with increased and decreased rates of cell division, respectively (Fig. 5C).Of note, ectopic expression of Tpm1.8/9 in OV90 and COV504 cells resulted in a complete growth arrest within 6 days.Moreover, trans-well assays clearly showed that Tpm1.8/9 overexpression resulted in significantly increased migratory and invasive features when compared with Tpm1.6/7 (Fig. 5D).Tpm1.8/9-overexpressing (OE) ovarian cancer cells appeared to invade the collagen layer collectively, as narrow linear strands with "leader" and "follower" cells (Fig. 5E).The latter observation is of interest since in non-muscle cells Tpm1.8/9 are specifically expressed in lamellipodia, i.e. the membrane protrusions found at Fig. 1 A subpopulation of quasi-mesenchymal cells co-exists with epithelial cells in HGS ovarian cancer cell lines.A FACS analysis of the ovarian cancer cell lines OV90, CAOV3, SKOV3, and COV504 with antibodies directed against CD44 and EpCAM.EpCAM/CD44 positive and negative areas were defined as previously described [17,21] using multiple isotype controls and are shown by the quadrants in the plots.Using specific gates, cells were separated in CD44 hi EpCAM hi and CD44 hi EpCAM lo subpopulations.The percentages of CD44 hi EpCAM lo and CD44 hi EpCAM hi cells within each cell line are depicted in each quadrant.Notably, as previously observed for SW480 and HCT116, the ovarian cancer cell lines revealed a continuum of different EpCAM and CD44 expression levels with a large EpCAM hi (or EpCAM lo as in the case of SKOV3) cluster followed by a tail of gradually decreasing (increasing for SKOV3) EpCAM levels.By applying the indicated gates, cells were sorted into EpCAM hi and EpCAM lo subpopulations.Graphs show representative analysis from an individual experiment.B Phase contrast microscopy images of sorted EpCAM hi and EpCAM lo cells from EpCAM hi and EpCAM lo OV90 and CAOV3 sorted cells.Scale bar: 100 μm.While EpCAM hi cells show characteristic epithelial morphology, EpCAM lo cells showed a more spindle-and mesenchymal-like appearance.Scale bar: 100 µm.C Transwell migration assay (upper graph) and invasion assay (lower graph) of EpCAM hi (blue bar) and EpCAM lo (red bar) OV90 and CAOV3 cells.Each bar symbolizes the mean ± SD.D Principal component analysis (PCA) of the RNAseq profiles of EpCAM hi and EpCAM lo cells from the OV90 and CAOV3 lines.E Heatmap of common differentially expressed gene among EpCAM hi/lo and bulk subpopulations from the OV90 and CAOV3 cell lines (abs LFC > 1.5, P value < 0.01).Complete-linkage hierarchical clustering with split by k-means (k = 2) clustering was used.F Hallmark pathways based on the Gene Set Enrichment Analysis (GSEA) of OV90 and CAOV3 EpCAM lo expression profiles compared with EpCAM hi .Plots show only significantly expressed pathways, with a normalized enrichment score (NES) > 1 and P value < 0.05.G RT-qPCR expression analysis of ZEB1 in OV90, SKOV3 and COV504 transduced with an inducible control (shCT) and with a ZEB1-shRNA (shZEB1) construct.shRNA expression was induced with 1 μg/mL of doxycycline.GAPDH expression was used as control.Each bar represents the mean ± SD.P value is indicated.H FACS analysis of the OV90, SKOV3 and COV504 cell lines transfected with the shZEB1 and control constructs using antibodies against CD44 and EpCAM.Cells were induced with 1 μg/mL doxycycline for 72 h prior to the FACS analysis.The percentages of EpCAM lo and EpCAM hi cells within each cell line are depicted in each quadrant.
the leading edge driven by branched as well as unbranched filaments composed of actin and Tpm1.8/9 that promote cell motility [30,31].IF analysis with TPM1 isoform-specific antibodies confirmed the co-localization of Tpm1.8/9 and ARP2, a specific marker for lamellipodia, at the edge of ovarian cancer cells; instead, Tpm1.6/7 were mainly localized in the cytoplasm (Fig. 5F).
In order to carry out a more comprehensive study of the transcriptional and functional consequences of the ectopic expression of the specific TPM1 isoforms, RNAseq analysis was carried out on the OV90 overexpressing cells.Unsupervised hierarchical clustering and principal component analyses confirmed the distinct transcriptional identity of the OV90 parental, Tpm1.6/7-OE, and Tpm1.8/9-OE cells (Fig. 6A, B).Gene set enrichment analysis (GSEA) was then performed to allow the identification of specific signaling pathways and gene ontology functions characteristic of each of the above sample groups.As shown in the heatmap of Fig. 6C, Tpm1.8/9 overexpression resulted in the pronounced activation of Hedgehog, Wnt/βcatenin, TGF-β, and Notch signaling, i.e. pathways known to be involved in the induction and regulation of EMT.Accordingly, EMT was also found to earmark these cells, as also shown by the distinct expression patterns of several EMT-related genes between Tpm1.6/7-andTpm1.8/9-OE samples (Fig. 6D).
To functionally validate the activation of the canonical Wnt/βcatenin signaling pathway, we implemented TopFLASH reporter assays, as previously described [32].As shown in Fig. 6E, a ~10-fold increase in luciferase activity was observed upon ectopic expression of the individual Tpm1.8 and 1.9 isoforms.Likewise, Wnt signaling activity was significantly reduced upon Tpm1.8/9 siRNA-driven knockdown.
Last, to evaluate the in vivo consequences of the ectopic expression of the specific tropomyosin isoforms on their invasive and metastatic capacity, intraperitoneal (IP) injections were performed with Tpm1.6/7-andTpm1.8/9-OE cells to model latestage ovarian cancer with intra-abdominal dissemination and ascites formation.As illustrated in Fig. 6F, G by IVIS imaging and quantification, the results showed that, somewhat surprisingly, IP transplantation of Tpm1.6/7-OE cells resulted in increased ascites volumes when compared with Tpm1.8/9-OE cells.Accordingly, pancreas and omental metastatic lesions were observed in mice injected IP with Tpm1.6/7-OE cells, whereas, no metastases were observed with Tpm1.8/9-OE cells (Fig. 6H).These apparently contradictory results are explained by MET repression in ovarian cancer cells constitutively expressing the EMT-inducing Tpm1.8/9 isoforms.As both E-to-M and M-to-E transitions are necessary for ovarian cancer cell dissemination and metastasis formation, the constitutive induction of EMT in Tpm1.8/9-OE cells prevents them from efficiently colonizing the abdominal cavity.As a confirmation of this, "tumoroids" were derived and cultured from the ascites formed in the animals transplanted with ovarian cancer cells overexpressing the Tpm1.6/7 or Tpm1.8/9 isoforms.As shown in Fig. 6I, the morphology of these floating cell aggregates derived from Tpm1.6/7-OE cells was strikingly structured with multilayered epithelial cells and often with an internal lumen.In sharp contrast, tumoroids from mice transplanted with Tpm1.8/9-OE cells lacked any kind of structure and appeared as random aggregates of floating cells.Accordingly, colony formation analysis of single cells from the ascites-derived tumoroids showed that overexpression of the Tpm1.6/7results in a substantially increased colonization capacity when compared with Tpm1.8/9-OE cells (Fig. 6J).
Overall, these results show that the TPM1 isoforms confer specific functional features on ovarian cancer cells.In particular Tpm1.8/9isoforms are strongly associated with EMT-inducing and inflammatory signaling pathways likely to underlie 'transcoelomic' dissemination of ovarian cancer cells and the formation of ascites.However, their ectopic overexpression is also a potential source of artifacts as EMT is equally essential as is MET in the formation of intra-abdominal metastases.
TPM1.8/9 isoforms confer resistance to taxane-and platinumbased chemotherapy and are expressed in ovarian cancer patients-derived ascites To investigate the role played by TPM1 alternative isoforms in women suffering from ovarian cancer, patient-derived tumor tissues were examined by immunohistochemistry (IHC) and in situ hybridization (ISH) analyses with isoform-specific antibodies and oligonucleotide probes.Among the high-and low-grade serous tumors analyzed (n = 13), Tpm1.6/7 appeared to be consistently expressed in primary and metastatic lesions (Fig. 7A).In contrast, Tpm1.8/9 expression was virtually undetectable above background levels, with few cases showing patchy and enhanced staining.Although based on an admittedly limited number of tumors, these observations seem to suggest that, while the Tpm1.6/7isoforms are mainly expressed in the bulk of epithelial tumor cells, their Tpm1.8/9counterparts are only rarely observed possibly in association with late and chemo-resistant stages of the disease.
The formation of intra-abdominal ascites predominantly occurs in stage III and IV ovarian cancer patients due to the spreading of tumor cells to the peritoneum and the obstruction of lymphatic vessels.As such, ascites-derived tumor cells may encompass quasimesenchymal and chemo-resistant cell types en route to the metastatic colonization of abdominal organs.Hence, we evaluated whether the Tpm1.8/9isoforms are transiently expressed in ovarian cancer patient-derived ascites when compared with Tpm1.6/7.We collected ascites samples from n = 13 patients and sorted the cellular content by FACS according to the [CD45(Lin) − /CD90 + ] and [CD45(Lin) − /CD90 − ] gates, encompassing immune/stromal and cancer cells, respectively (Fig. 7B, C).Total RNA was then extracted from the sorted cells for RTqPCR analysis of TPM1 isoform expression.As shown in Fig. 7D, increased Tpm1.8/9 expression was observed in the ascites-derived cancer cells (CD90 − ) when compared with Tpm1.6/7.Among the immune/stromal cells (CD90 + ), Tpm1.6/7 expression levels were increased when compared with Tpm1.8/9 (Fig. 7D).
Last, we employed expression profiles from the Cancer Genome Atlas (TCGA) project and the TCGA Splicing Variants Database (TSVdb; http://www.tsvdb.com/)and integrated the clinical followup data with the expression of TPM1 (whole gene) and its isoforms.As shown in Supplementary Fig. 7 Suppl.1, Kaplan-Meier analysis was borderline significant for the TPM1 gene and its 1.6/7 isoforms, whereas the p values for Tpm1.8/9 were both >0.05.The latter is not surprising in view of the very low expression level of these low-molecular weight isoforms in primary ovarian cancer from which the TGCA data are derived.
In view of these results pointing at the specific Tpm1.8/9upregulation in ovarian cancer ascites, i.e. at late and recurrent disease stages often in association with chemo-resistance and poor prognosis, we asked whether the same TPM1 isoforms may confer resistance to the platinum-and taxane-based therapies commonly employed in the clinical management of ovarian cancer.We therefore cultured the OV90, COV504, and CAOV3 parental cell line in the presence of two distinct chemotherapeutic agents, namely cisplatin and paclitaxel, and then analyzed TPM1 isoform expression in cells surviving the treatment.RTqPCR Fig. 2 EpCAM lo cells co-exist with epithelial cells in high-grade serous ovarian cancer.A Density plot shows the distribution of patientderived ovarian cancer cells (scRNAseq data from Vazquez-Garcia et al. [22]) earmarked by expression of the EpCAM lo signature.The threshold was set as ≥0.1 for the subsequent analyses.The EpCAM lo signature was defined by genes that were identified as upregulated EpCAM lo cells in both CAOV3 and OV90.EpCAM lo upregulated genes were selected after differential expression analysis with EpCAM hi /bulk cells using the cutoffs log2FC > 1.5 and padj <0.05.B UMAP plot of patient-derived ovarian cancer cells.Cells positive for the EpCAM lo signature are highlighted in red and show enrichment within three clusters (#2, #4, and #6) labeled as EMT-like in the Vazquez-Garcia et al. study [22].Please note that a substantial fraction of the EpCAM lo -like cells appears to fall outside these clusters and is distributed throughout the UMAP.C Violin plots showing the distribution of EpCAM lo -like cells (according to z-score) in different anatomical localization of ovarian cancers (adnexa, ascites, and non-adnexa).D Violin plots showing the expression levels of EMT-like signature of single cell cluster 2, 4 and 6.E Heatmap of hallmark gene sets across the EpCAMlo-like clusters in primary ovarian cancers (adnexa), ascites, and metastases (non-adnexa) based on the GSEA of three EpCAM lo -like clusters and bulk (p val <0.05).Complete-linkage hierarchical clustering was used.
analysis revealed a dramatic downregulation of Tpm1.6/7 expression with both agents when compared with the untreated parental cells.In contrast, Tpm1.8/9 expression appeared to increase in cells surviving both the cisplatin-and paclitaxel-treatment (Fig. 7E).Dose-response curves with OV90, COV504, and CAOV3 cells ectopically expressing the individual isoforms confirmed that Tpm1.8/9-OE cells displayed a 40-fold higher resistance (IC 50 = 3.266/3.414μM) than Tpm1.6/7OE cells (IC 50 = Fig. 3 Differential expression of RBPs ESRP1 and RBM24 regulates TPM1 isoforms in HGS ovarian cancer.A Fold change analysis of RBPs [original list from reference [14]] differentially expressed between EpCAM lo and EpCAM hi cells in OV90 and CAOV3.Red bar indicates RBPs with significant differential expression (log2-fold change >2, and P value < 0.05).The dotted line represents the 1.5 absolute fold change cutoff.B RT-qPCR analysis of ESRP1 and RBM24 expression in OV90, CAOV3, and COV504 EpCAM hi/lo and bulk cells.GAPDH expression was employed as control (means ± SD, n = 3).P values are indicated.C Western blot analysis of ESRP1 and RBM24 expression in OV90, CAOV3, and COV504 EpCAM hi/lo and bulk cells.β-actin was used as loading control for western blots.D. TPM1 exon peak plots relative to the AS analysis of RNAseq data obtained from OV90 EpCAM hi/lo and CAOV3 bulk/EpCAM lo analysis.Each peak indicates the expression of specific exons; the height of each peak is indicative of the expression level of the specific exons.The exon-intron structure of the corresponding TPM1 isoforms is depicted below the exon peak plot.Exon 1a and 2b are specific to Tpm1.6/7 isoforms (brown), while exon 1b is specific to Tpm1.8/9 (green).Exon 6a (red) is only present in the Tpm1.7 and Tpm1.9 isoforms; whereas, exon 6b (yellow) earmarks the Tpm1.6 and Tpm1.8 isoforms.With the exception of exon 6a and 6b, exons 3 to 9d are present in all TPM1 isoforms.E RT-qPCR (histogram panels) and western analysis of TPM1 isoform expression in OV90, CAOV3, and COV504 EpCAM hi/lo and bulk cells.GAPDH expression was employed as control.(Means ± SD, n = 3).P value is indicated.β-actin was used as loading control for western blots.F CD44/EpCAM FACS analysis of the ovarian cancer cell lines PEA1 and PEA2.EpCAM/CD44 positive and negative areas were identified using multiple isotype controls.The percentages of cells within each quadrant are indicated.G RT-qPCR (left histogram panels) and western (right panel) analysis of ESRP1, RBM24, and TPM1 isoform expression in PEA1 and PEA2 ovarian cancer cell line; GAPDH expression was employed as control.(Means ± SD, n = 3).P values are indicated.β-actin was used as loading control for western blots.0.078/0.043μM) upon cisplatin treatment in OV90 (Fig. 7F, Supplementary Fig. 7 Suppl.2).The same was also observed with paclitaxel (IC 50 = 1.28/1.66μM for Tpm1.8/9-OE;IC 50 = 0.017/ 0.015 μM for Tpm1.6/7-OEcells).To validate this observation, we developed siRNA assays to selectively downregulate the Tpm1.6/7 and Tpm1.8/9 isoforms in parental OV90, COV504, and CAOV3 cells and assess their chemo-resistance.As shown in Fig. 7G, H and Supplementary Fig. 7 Suppl.2, the specific downregulation of the Tpm1.8/9isoforms, validated both at the RNA and protein levels, reduced the cisplatin-and paclitaxel-specific IC 50 , while the opposite was true for the siRNA-driven knockdown of Tpm1.6/7.Furthermore, increased resistance to cisplatin and paclitaxel was observed in EpCAM lo -sorted cells upon ZEB1 knockdown (by SH) in the OV90 and COV504 cell lines (Supplementary Fig. 7 Suppl.2).
TPM1.8/9-specific small molecule inhibitors for ovarian cancer therapy Given the newly uncovered role played by the Tpm1.8/9isoforms in ovarian cancer and the broad spectrum of consequences at the cellular and molecular level that their specific inhibition may exert on cell motility and proliferation, EMT/MET, several oncogenic signaling pathways including Wnt, and therapy resistance, the development of small molecule antagonists may provide novel tools in the clinical management of late-stage ovarian cancer.Differences in the N-and Ctermini between tropomyosin isoforms provide an opportunity to develop compounds that preferentially target specific isoforms [33][34][35].Compounds that target the C-terminus of TPM3.1 have been shown to inhibit the function of this isoform both in vitro and in vivo by incorporating into the overlap junction between adjacent dimers in actin/Tpm3.1 co-polymers [36,37].Compounds targeting the N-terminus show similar activity [38].Based on the differences in the N-terminal sequences of exon 1b in the TPM1, TPM3, and TPM4 genes, compounds were developed that target Tpm4.2[39] and Tpm1.8/1.9 (European patent application No. 23187348.0).Virtual screening of compound libraries with the N-terminal model of Tpm1.8/9 was used to identify compounds as potential binders.Targeting of Tpm1.8/9 by virtual screening identified 6 candidate compounds that were tested for biological activity in human fibroblasts and OV90 cells based on their ability to disperse the target Tpm1.8/9away from actin-containing structures in the lamellipodium (Supplementary Fig. 8 Suppl.1).Two compounds, Tpm1.8/9-1 (PubChem CID 6494468) and Tpm1.8/9-3 (PubChem CID 18973468) (hereafter for brevity referred to as compounds #1 and #3) were identified with activity in the low micro molar range.Of note, the effective concentration of an inhibitor is influenced by its affinity for the target, metabolism, and other factors, rather than the concentration of the target in the cell.The compounds do not show interaction with the target in cell-free biochemical assays, compatible with either a slow on-rate or fast off-rate resulting in a low affinity interaction with the tropomyosin dimer followed by incorporation into a high affinity binding state in the overlap junction during polymerization.This proposed model is based on previous studies of the incorporation of Tpm3.1 inhibitors into the overlap junction during polymerization.In order to functionally validate the two compounds, we first determined their most effective concentration in the 0 to 10 μM range both on the OV90 parental cell line and on its sorted EpCAM lo subpopulation.RTqPCR and western analysis showed that compounds #1 and #3 do not affect Tpm1.6/7 or Tpm1.8/9 expression at either the RNA or protein level (Fig. 8A, B).This is not unexpected since displacement of Tpm1.8/9 from actin filaments to the soluble pool in cells does not result in Tpm1.8/9 turnover [40].However, the expression of EMT-related genes was affected by both compounds: ZEB1 and VIM expression was suppressed, whereas, EpCAM was increased, also in agreement with the expected MET-inducing effects of Tpm1.8/9 antagonists (Fig. 8A,  B).Furthermore, the treated OV90 and CAOV3 bulk and EpCAM lo cells showed a 2-5-fold reduction of the cisplatin-and paclitaxelspecific IC 50 values when compared with untreated cells (Fig. 8C-F).Last, compounds #1 and #3 dramatically reduced Wnt/β-catenin signaling as shown by TopFLASH reporter assays (Fig. 8G).As expected, the Wnt-inhibiting effects of the compounds are more clearly illustrated by the EpCAM lo cells because of their Wnt-and EMT-hi transcriptional profiles when compared with the parental OV90 and COV504 cell line where a majority of EpCAM hi (Wnt-lo) and EpCAM lo cells coexist.
Overall, these in vitro results suggest the potential for the future development of novel therapeutic strategies for high-grade serous ovarian cancer centered around the Tpm1.8/9isoforms.

DISCUSSION
It is generally accepted that epithelial-mesenchymal plasticity (EMP), i.e. the transient and reversible identity conferred to cancer cells by EMT/MET processes, underlies tumor progression, local invasion, distant metastases, therapy resistance, and immune evasion.As elegantly shown by Cook and Vanderhyden [41,42], EMT is a highly variable process with a very broad spectrum of upstream signals from the TME, intracellular regulatory mechanisms, and downstream effectors, the identity of which largely depends on the tumor type and its micro-and macroenvironment.The same is true when it comes to the nature of the epigenetic mechanisms that underlie EMP [43].We have recently shown that the differential expression of RNA-binding proteins between the epithelial tumor bulk and subpopulations of quasi-mesenchymal colon cancer cells underlies alterative splicing at a variety of target genes known to play functional roles in EMT, metastasis, and resistance to chemotherapy [17,21].Here, we  applied a similar strategy toward the identification and functional characterization of genes whose isoform patterns are altered during EMT/MET in ovarian cancer.Comparison of alternatively expressed gene isoforms between epithelial and quasimesenchymal colon and ovarian cancer cells revealed few common targets with many ovarian-specific events.This possibly reflect of the distinct modalities of local dissemination and metastatic colonization characteristic of these types of carcinoma.Whereas most colon cancer metastases follow a hematogenous route, unique for ovarian cancer is the 'transcoelomic' dissemination of tumor cells and the formation of ascites fluid in the abdominal and pelvic cavity which provide a favorable tumor microenvironment (TME) for the disseminated cancer cells.Nonetheless, previous studies have indicated that EMT does contribute to ovarian cancer progression and to chemotherapy resistance [44].Hence, even though through distinct cellular and molecular mechanisms, EMT does play a key role in ovarian cancer metastasis and chemo-resistance.Gene ontology analysis of the ovarian cancer specific isoform targets (Supplementary Fig. 3, Supplement 1B-E) revealed an extremely broad spectrum of biological processes, molecular functions, and cellular components likely to collectively contribute to the transition to quasimesenchymal ovarian cancer cells capable of local invasion and distant metastatic colonization.
Previously, AS targets likely to contribute to ovarian cancer progression have been reported such as BCL2L12 [45] and ECM1 [46].Here, among the EMT-related isoforms, we selected TPM1 because of its function in the regulation of cell motility through cytoskeletal modifications [30,31], and its alleged role as a tumor suppressor and even oncogene in multiple cancer types [47,48].TPM1 isoforms are found in multiple tissues [30,49].Our results establish a direct causative relation between the TPM1.8/9isoforms and the activation of EMT in ovarian cancer cells.Whether the RBM24 and ESRP1 RBPs, differentially expressed between epithelial and quasi-mesenchymal cells, also play a direct role as co-transcriptional factors of the TPM1a/b promoters, is at present unclear.Analysis of the FANTOM database, while failing to identify known RBP binding sites, did reveal the presence of known EMT-TFs.The results obtained with the luciferase reporter constructs are admittedly inconclusive due to the confounding role of EMT upon ectopic expression or downregulation of the ESRP1 and RBM24 RBPs.Future ChIP or CUT&RUN analyses will elucidate the mechanisms underlying the regulation of Tpm1.6/7 and Tpm1.8/9 isoforms directly and primarily by EMT-TFs or with the RBPs as co-transcriptional factors.
Because of their localization to the lamellipodia and functional role in cell motility, the TPM1.8/9isoforms are likely to facilitate dissemination from the primary tumor to the intra-abdominal cavity, as also shown by their enrichment in patient-derived ascites.Allegedly as a consequence of their EMT-inducing capacity, ectopic expression of the low-molecular weight TPM1 isoforms confers resistance to taxane-and platinum-based chemotherapy.Accordingly, Tpm1.8/9 are also found to be expressed at very low levels, if any, in primary ovarian cancers and their metastases albeit increased in malignant ascites.
Notwithstanding the above, our attempts to provide in vivo evidence for the metastatic capacity of ovarian cancer cells overexpressing the Tpm1.8/9isoforms failed to show any increase when compared with Tpm1.6/7.This apparently contradictory result can be explained by the transient and reversible nature of EMT along the multistep events that underlie dissemination and metastasis.As previously proposed by Thomas Brabletz and collaborators [50], while the acquisition of quasi-mesenchymal characteristics is required for local invasion and systemic dissemination, METs are equally essential for the colonization of distant organs.In the OV90 cells overexpressing the EMT-inducing Tpm1.8/ 9 isoforms, MET is inhibited thus negatively affecting their metastatic potential.The presence of tumor cells with a strongly compromised colony formation capacity in the malignant ascites from recipient mice transplanted IP with Tpm1.8/9-OE cells supports the hypothesis according to which MET suppression in these cells negatively affects their metastatic potential.Another indication of the potential off-target artifacts caused by the non-physiological expression levels of TPM1 isoforms in these cells became apparent in the analysis of their RNAseq profiles.While several inflammation pathways were upregulated together with Tpm1.8/9 in the EpCAM lo ovarian cancer cells when compared with their epithelial counterpart, ectopic Tpm1.6/7 expression resulted in the activation of similar inflammation-related pathways, e.g.IL6/Jak/Stat3, IFN, and TNF, when compared with Tpm1.8/9OE cells.
Because of the apparent multifunctional role of TPM1 isoforms in ovarian cancer malignancy and resistance to therapy, Tpm1.8/9 forms a potentially relevant therapeutic target.As shown here, the development of small molecule inhibitors will likely prevent EMT and reduce cell motility, and simultaneously inhibit the activation of key signal transduction pathways such as Wnt, known to play a central role in ovarian cancer stemness, EMT, and chemoresistance [51].Last, combined treatment with conventional taxane-and platinum-based chemotherapies may increase their therapeutic efficacy by antagonizing chemoresistance.

RT-qPCR and PCR analyses
Total RNA was isolated using the TRIzol reagent (Thermo Fisher Scientific, 15596018) followed by reverse transcription using high-capacity cDNA reverse transcription kit (Life Technologies, 4368814), according to the manufacturer's instructions.RT-qPCR was performed using the Fast SYBR Green Master Mix (#4368708, Thermo Fisher Scientific) on an Applied Biosystems StepOne Plus Real-Time Thermal Cycling Research device, with three replicates for each analysis.Relative gene expression was determined by normalizing the expression of each target gene to that of GAPDH.Results were analyzed using the 2-(ΔΔCt) method.RT-qPCR primers are listed in Supplementary Table 4.

Flow cytometry analysis and sorting
Single-cell suspensions in PBS supplemented with 1% FBS were incubated with anti-EpCAM-FITC (1:20, #GTX30708, Genetex), and anti-CD44-APC (1:20, #559250, BD Pharmingen) antibodies for 30 min on ice and analyzed on a FACSAria III Cell Sorter (BD Biosciences).CD44 hi EpCAM hi and CD44 hi EpCAM lo OV90 and CAOV3 cells were sorted and incubated in a humidified atmosphere at 37 °C with 5% CO 2 for 3-5 days before RNA or protein were collected as described above.See Fig. 1 for a more detailed protocol and gating specifications.

Cell migration and invasion assays
Migration assays were conducted using 8 μm pore PET Transwell inserts (#353097, BD Falcon™) and TC-treated multi-well cell culture plates (#353047, BD Falcon™).5 ×10 4 cells were seeded in the upper chamber with 100 μL of serum-free culture medium.Culture medium supplemented with 10% FBS was used as a chemoattractant in the lower chamber.After 24 h, cells that migrated to the lower chamber were fixed with 4% PFA (#9713.9010,VWR Chemicals), stained with 0.1% Trypan Blue solution (#15250061, Thermo Fisher Scientific) and counted using a microscope.
The invasion assays were conducted as described here above with the only addition of 200 μL Matrigel mixed with 10-20 μl of a 0.01 M Tris (pH 8.0)/0.7%NaCl solution on top of the Transwell-Clear insert and incubated at 37 °C for 2 h.After having removed the excess liquid, 5 ×10 4 cells were added to the well in 150 μL serum-free medium.Next, 0.5 mL of culture medium supplemented with 10% FCS was added.After 24 h, the cells that invaded the lower chamber were fixed with 4% PFA and stained with 0.1% Trypan Blue solution.Cells were counted using a microscope.
200 μL collagen mix [7.5% 10 ×PBS, 57% collagen, 0.1% 1 M NaOH, 9.5% H 2 O and 25% culture medium] were plated into scaffolds and incubated at 37 °C for 1 h. 5 ×10 4 cells were added on top of the solid collagen and incubated at 37 °C.After 1 week, the entire scaffold with cells and collagen was fixed with 4% PFA and embedded in paraffin.4 μm sections were mounted and counterstained with Hematoxylin.Slides were dehydrated and mounted in Pertex (#00811, Histolab).
Immunohistochemistrycultured cells 0.5-1×10 4 cells were plated into 24-well plates containing glass cover slips coated with 0.2% gelatin.After 6-24 h, culture medium was removed, ice-Fig.8 Small molecule inhibitors directed against Tpm1.8/9isoforms antagonize their effects on EMT, Wnt signaling, and resistance to chemotherapy.A Upper panels: RT-qPCR analysis of TPM1 isoforms and EMT-related gene expression in OV90 parental cells cultured for 24 h in the presence of compound #1 or #3 at 0, 2, 5, and 10 μM.The values were calculated by normalizing with the untreated cells.P values < 0.05 are shown by red bars while gray bars indicate lower values; GAPDH expression was employed as control (Means ± SD, n = 3).Lower panels: western analysis of TPM1 isoform expression in OV90 parental cells cultured for 24 h in the presence of compound #1 or #3 at 0, 2, 5, and 10 μM.Β-actin was used as loading control for western blots.B RT-qPCR analysis of TPM1 isoforms and EMT-related gene expression in OV90 EpCAM lo cells cultured for 24 h in the presence of compound #1 or #3 at 0, 2, 5, and 10 μM.The values were calculated by normalizing with the untreated cells.P values < 0.05 are shown by red bars while gray bars indicate lower values; GAPDH expression was employed as control (Means ± SD, n = 3).Lower panels: western analysis of TPM1 isoform expression in OV90 EpCAM lo cells cultured for 24 h in the presence of compound #1 and #3 at 0, 2, 5, and 10 μM.β-actin was used as loading control for western blots.C Dose-response curves of parental OV90 cells treated with compound #1 or #3 in the presence of different paclitaxel and cisplatin concentrations.IC 50 values were calculated from biological triplicates for each experiment (Means ± SD, n = 3).D Dose-response curves of OV90 EpCAM lo cells treated with compound #1 or #3 in the presence of different paclitaxel and cisplatin concentrations.IC 50 values were calculated from biological triplicates for each experiment (Means ± SD, n = 3).E Dose-response curves of parental CAOV3 cells treated with compound #1 or #3 in the presence of different paclitaxel and cisplatin concentrations.IC 50 values were calculated from biological quadruplicates for each experiment (Means ± SD, n = 4).F Doseresponse curves of CAOV3 EpCAM lo cells treated with compound #1 or #3 in the presence of different paclitaxel and cisplatin concentrations.IC 50 values were calculated from biological quadruplicates for each experiment (Means ± SD, n = 4).G TOP-Flash luciferase reporter analysis of Wnt signaling activity in OV90, COV504 and CAOV3 parental (upper panel) and EpCAM lo (lower panel) cells treated with compound #1 or #3 (Means ± SD, n = 3-5).and DAPI (#D1306, Thermo Fisher Scientific) added.Slides were mounted using VECTASHIELD® (#H100010, VECTOR laboratories) and cells imaged using a LSM-700 (Zeiss) with 20 , 40 , 63 lenses.Images were analyzed using ImageJ.

In vivo study
Mouse experiments were performed according to the Code of Practice -Animal Experiments in Cancer Research, Netherlands Inspectorate for Health Protection, Commodities and Veterinary Public Health, and the Animal Experiment Committee (DEC).6-8-week-old NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ (NSG) female mice were used.Animal randomization was employed.50 μl PBS containing 1 ×10 5 OV90 cells overexpressing Tpm1.6/7/8/9 isoforms was injected IP into each mouse.150 mg/kg D-luciferin (#L2916, Invitrogen) was injected IP for bioluminescence signal.After 10 min of isoflurane-induced anesthesia (#B506, Zoetis), the bioluminescence of the mouse was measured using the IVIS Spectrum imaging system (Caliper Life Science, Hopkinton, MA) and bioluminescence analyzed using LIVINGIMAGE 4.4 software (Caliper Life Science).Ascites was obtained by syringe, and tumoroids collected and washed with RBC lysis buffer.Mice were sacrificed and tissue fixed in 4% PFA for further analysis.specific exon included, termed Percentage Spliced In (PSI or Ψ).The PSI ranges from 0 (i.e.no isoform includes a specific alternative exon) to 1 (i.e.all of the isoforms detected comprise the alternative exon).GRCh37.p13 was utilized while MISO does not support the GRCh38.p14assembly.
We removed alternative events with low expression of related transcript isoforms if less than 3 samples in a dataset had more than 10 informative reads to calculate the PSI.Next, we compared the PSI between EpCAM hi and EpCAM lo groups in the OV90 and CAOV3 ovarian cancer cell lines.AS events were defined as differentially spliced events when the difference of mean PSI between two groups (Δpsi; differential Percentage Spliced In) was >10%.
RNA seq analysis of subpopulations in OV90 and CAOV3 RNA was isolated from sorted populations with Trizol reagent.Libraries were prepared with the TruSeq RNA sample prep kit v2 (Erasmus MC, Biomics).Samples were sequenced with Illumina HiSeq 2000 and adapter sequences were removed with Trimmomatic (v.0.33).Subsequently, the reads were mapped in a two-pass procedure to the human reference genome build hg38 with the RNA-seq aligner STAR (v2.4.2a) [58] using the Homo sapiens GENCODE v23 annotations.Raw counts were imported in DESeq2 (v1.36.0) and normalized with a variance stabilizing transformation (VST) [59].Differential expressed genes were identified by comparing Epcamlow versus Epcamhigh/bulk samples using absLogFC >1.5 and padj <0.05, and visualized with the ComplexHeatmap [60] package (v2.12.1) after a z-score scaling.Pathways activity was evaluated using gene set enrichment analysis (fgsea v1.22.0) on the hallmark gene sets from the Molecular Signature data base [61,62].Complete-linkage hierarchical clustering with split by k-means (k = 2) clustering was used.

scRNAseq analysis of ovarium cancer cells
Publicly available data from Vázquez-García et al. were retrieved from CellxGene portal [22].Downstream analysis was performed in Seurat (v4.3.0)[66].An EpCAM lo signature was evaluated with AddModuleScore based on the previously identified upregulated gene list (N = 38).A threshold (>0.1) was used to annotate cells with the highest association to the EpCAM lo signature ("low-like cells", 4% of cancer cells).Next, low-like cells were visualized on the integrated UMAP embedding from Vázquez-García et al. and pathway activity of clusters encompassing low-like cells were visualized with ComplexHeatmap [60].After approval of a data transfer agreement, FASTQ files from Izar et al. [67] were downloaded from the TerraBio repository and processed with RSEM using the STAR aligner to the hg19 human reference genome with isoform annotation from UCSC [58,63].Files were imported with tximport [64] (v1.24.0) and cells were selected that contained at least 500 different genes (nFeature_RNA > 500).Cells were clustered (kmeans, k = 4) according to the percentage of their respective TPM1 isoform expression and subsequent analysis was performed in Seurat (v4.3.0)[66], where TPM1.7 expressing cells were compared to TPM1.9 with FindMarkers.A gene set enrichment analysis was performed using the Hallmark gene set and pathways were filtered according to similar activity in the OV90 cell line and patient data.Complete-linkage hierarchical clustering was used.

Survival analysis of TPM1 in TCGA OVCA
RSEM processed data from the TCGA cohort was downloaded from the tsvDB [68].Data was log2 transformed and survival analysis was performed with the survival package.Survival curves were generated with the survminer package for the whole TPM1 gene, TPM1.7 (isoform_uc002alk) and TPM1.9 (isoform_uc002alt) based on clinical data on overall survival [69].

Statistical analysis
For statistical comparison, we performed unpaired t test.Statistical analyses were performed using Prism 7 software (GraphPad).Data with statistical significance are as indicated.Information on replicates, independent experiments and statistical test can be found in the Fig.
Legends.Analysis tools were run with all parameters as default unless otherwise stated.

Fig. 4
Fig. 4 The RBPs ESRP1 and RBM24 synergistically regulate TPM1 isoforms and the relative proportion of EpCAM hi/lo cells.A RT-qPCR (histogram panels) and western (lower panel) analysis of ESRP1, RBM24, and TPM1 isoform expression in RBM24-OE (overexpressing) and shESRP1-KD (knockdown) OV90 and COV504 ovarian cancer cell line; GAPDH expression was employed as control (Means ± SD, n = 3).β-actin was used as loading control for western blots.B RT-qPCR (histogram panels) and western (lower panel) analysis of ESRP1, RBM24, and TPM1 isoform expression in ESRP1-OE (overexpressing) and shRBM24-KD (knockdown) OV90 and COV504 ovarian cancer cell line; GAPDH expression was employed as control (Means ± SD, n = 3).β-actin was used as loading control for western blots.C CD44/EpCAM FACS analysis of RBM24-OE/KD and ESRP1-OE/KD OV90 cells.Cells were induced with 1 μg/mL doxycycline for 72 h before analysis.The relative percentages of EpCAM lo and EpCAM hi cells are indicated in each quadrant.D CD44/EpCAM FACS analysis of RBM24-OE/KD and ESRP1-OE/KD COV504 cells.Cells were induced with 1 μg/mL doxycycline for 72 h before analysis.The relative percentages of EpCAM lo and EpCAM hi cells are indicated in each quadrant.

Fig. 5
Fig. 5 Ectopic expression of Tpm1.6/7 and Tpm1.8/9 isoforms results in increased migration and invasion, and decreased cell proliferation.A RT-qPCR analysis of OV90, COV504, PEA1, and PEA2 ovarian cancer cell lines transduced to ectopically express the Tpm1.6/7-OE and Tpm1.8/9-OE isoforms; GAPDH expression was employed as control (Means ± SD, n = 3).P values are relative to the comparison with the parental cell lines.B Western analysis of OV90, COV504, PEA1, and PEA2 ovarian cancer cell lines transduced to ectopically express the Tpm1.6/7-OE and Tpm1.8/9-OE isoforms.β-actin was employed as loading control.C Proliferation assays of OV90, COV504, PEA1, and PEA2 ovarian cancer cell lines transduced to ectopically express the Tpm1.6/7-OE and Tpm1.8/9-OE isoforms.O.D. values are shown from day 1 to 6 (Means ± SD, n = 3).P values are relative to the comparison with the parental cell lines.D Transwell migration assay of OV90, COV504, PEA1, and PEA2 ovarian cancer cell lines transduced to ectopically express the Tpm1.6/7-OE and Tpm1.8/9-OE isoforms.5×10 4 cells were plated on TC-coated membranes and left O/N.The number of cells that migrated to the lower side of the membrane were counted and plotted (Means ± SD, n = 3).P values are relative to the comparison with the parental cell lines.E Confocal images of OV90 and COV504 parental and Tpm1.6/7/8/9-OE cells seeded on collagen layers and incubated for 6 days.As indicated by the arrows, Tpm1.8/9-OE cells appear to invade the collagen layer collectively as narrow linear strands with "leader" and "follower" cells.Scale bar: 250 μm.The number of cells invading the collagen was quantified and plotted (Means ± SD, n = 3).P values are relative to the comparison with the parental cell lines.Plots relative to the PEA1 and PEA2 ovarian cancer cell lines were also calculated (bottom).F Immunofluorescence analysis of OV90, COV504, PEA1 and PEA2 parental cells with antibodies directed against ARP2, Tpm1.6/7 and Tpm1.8/9.Nuclei were visualized by DAPI staining of DNA.Scale bar: 5 μm.