The PF338 line was established from malignant pleural effusion. 5ml of effusion were centrifuged at 1,200 × g at room temperature for 10 min. The pellet was resuspended in RPMI1640 fortified by 10% FBS and 1% penicillin/streptomycin and seeded in a culture flask. More than 15 passages of the adherent cells with a minimum of three freezing-thawing cycles were done before experiments were initiated in order to use a tumor cell culture without non-tumorous cells. The A375 melanoma cell line was purchased from the ATCC and cultured in DMEM supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin in culture flasks. Single Nucleotide Polymorphism (SNP) profiling was performed for PF338 and A375 tumor cell lines by Multiplex Cell Line Authentication (Multiplexion, Heidelberg, Germany) to confirm A375 cell line identity and PF338 unique cell line identity. Selumetinib, galunisertib, olaparib and BEZ235 were purchased from Selleck Chemicals (Houston, TX, United States) and dissolved in DMSO. SAHA and valproate were purchased from Sigma-Aldrich (St. Luis, MO, United States) and dissolved in DMSO and water, respectively. Paclitaxel (Kabi Fresenius, IL, United States) and cisplatin (Accord, Munich, Germany) were dissolved in 0.7% NaCl.

Immunohistochemistry was performed using the Ventana BenchMark Ultra system (Roche Tissue Diagnostics, Grenzach-Vyhlen, Germany). 3 µm sections were prepared from formalin-fixed and paraffin embedded (FFPE) tumors and PF338 cellblock. The following primary antibodies were used: CD10 (Clone 56C6, 1:50, Leica Biosystems, Nussloch Germany), E-cadherin (Clone: NCH-38, Dako-Agilent, Waldbronn, Germany), vimentin (Clone: V9, Dako-Agilent, Waldbronn, Germany), progesterone receptor (Clone: 1E2; RTU, Roche Tissue Diagnostics) and estrogen receptor (Clone SP1, RTU, Roche Tissue Diagnostics). Color development was performed by the OptiView staining kit (Roche Tissue Diagnostics) followed by hematoxylin counterstaining. All stainings were evaluated by a senior pathologist (AB) and representative images were taken.

Total protein amount-based Sulforhodamine B (SRB) assays were performed as follows. 5 × 10³ (PF338) or 2 × 10³ (A375) tumor cells /well were plated on the inner 60 wells of a 96-well plate and first incubated for 48 h. After 72 h of treatment with drugs, 10% TCA was used for fixation, followed by SRB dye (Sigma-Aldrich, St. Louis, MO, United States), and wash out with 1% acetic acid. 10mM Tris puffer dissolved the protein-bound dye and optical density (OD) was read at 570 nm by using a microplate reader (EL800, bioTec Instruments, Winooski, VT, United States). IC₅₀ were calculated by using the CompuSyn software (ComboSyn, Inc., Paramus, NJ). Viability results are illustrated as ratio to control viability. For colony-formation assays, 1,000 tumor cells /well were plated on 6-well plates, incubated for 48 h and subsequently treated every 3–4 days with increasing drug concentrations for 10 days. 10% TCA was used for fixation, followed by SRB dye and wash out with 1% acetic acid. Colonies were counted manually. Experiments were repeated thrice.

In order to test the viability of cells after freezing and thawing at various passages the cell viability was measured on the NucleoCounter NC-3000TM system (Chemometec, Allerod, Denmark) using the cell viability reagents and protocol right after thawing and after 72 h in culture.

For cell cycle analysis, PF338 tumor cells were seeded on 6-well plates in 2 × 10⁵ cells/well concentration and incubated for 48 h, followed by 72 h of treatment. Cells were trypsinized and incubated with lysis buffer containing DAPI for 5 min at 37°C. Stabilization buffer was added, and cellular fluorescence was measured by the NucleoCounter NC-3000TM system (Chemometec, Allerod, Denmark). Cell cycle phases were identified based on the DNA content of the cells.

PF338 tumor cells were seeded into 6-well plates. After a recovery period of 24 h, cells were treated for 72 h with either HDACi (SAHA, valproate), galunisertib or solvent and precipitated with 6% TCA for 1 h, 4°C followed by centrifugation for 10 min at 9000 rpm. The total cellular protein pellets were resuspended in electrophoresis sample buffer (62.5 mM Tris–HCl, pH 6.8, 2% SDS, 10% glycerol, 5 mM EDTA, 125 mg/ml urea, 100 mM dithiothreitol) to be later loaded on 10% acrylamide gels in equal protein amounts. For immunostaining rabbit anti-E-cadherin (Cell Signaling, 24E10, 1:1,000), anti-beta-catenin (Santa Cruz, Sc-7199, 1:500), anti-pSMAD2 (Cell Signaling, 138D4, 1:1,000) and polyclonal anti-beta-tubulin (Abcam, ab6046, 1:1,000) were used. As secondary antibody HRP-conjugated anti-rabbit antibody (Jackson ImmunoResearch, 1:10.000) was used. For development ECL Western Blotting Substrate (Thermo Scientific, Waltham, MS, United States) was applied followed by luminography. Three independent experiments were performed.

DNA from PF338 cells was isolated according to the manual’s instructions by using DNeasy Blood and Tissue Kit (Qiagen, MD, United States). FFPE tissue DNA was isolated according to the manual’s instructions by using QIAamp DNA FFPE Tissue Kit (Qiagen, MD, United States). DNA concentrations were determined by Qubit^® 2.0 Fluorometer dsDNA HS assay kit (LifeTechnologies, CA, United States).

A total amount of 45 ng DNA was used for multiplex-PCR. Multiplex PCR and purification were performed with the GeneRead DNAseq Custom Panel and PCR Kit V2 (Qiagen, MD, United States) and Agencourt^® AMPure^® XP Beads (Beckman, CA, United States). The library preparation was performed with NEBNext Ultra DNA Library Prep Set for Illumina (New England Biolabs, MA, United States), according to the manufacturer’s recommendations by using 24 different indices per run. The pooled library was sequenced on MiSeq (Illumina; 2 × 150 bases paired-end run) and analyzed by Biomedical Genomics Workbench (CLC Bio, Qiagen, MD, United States). For targeted sequencing a customized comprehensive cancer-panel was designed containing regions of interest.

Video microscopy was performed as previously [24] and now described in Supplementary materials.

Two-way ANOVA with Bonferroni posttest was applied to describe significant differences between cell lines and treatment lines. One-way ANOVA with Dunn´s multiple comparison test was applied to identify significant differences between treatment lines. *p < 0.05; **p < 0.01; ***p < 0.001, and ****p < 0.0001 represented significant differences. All calculations were done in GraphPad Prism 8 (GraphPad Software Inc., San Diego, CA).

A 59-year-old female patient was diagnosed with UCS and underwent radical resection yielding a pT3aN0M0 FIGO IIIA tumor containing both a dominant stromal sarcomatous and a focal endometroid carcinomatous component (Figure 1). No adjuvant treatment was applied, however, 2.5 months later the patient developed retroperitoneal recurrence, for which chemotherapy consisting of three cycles paclitaxel/carboplatin was started. Re-staging indicated a tumor response and thus three additional cycles of chemotherapy were applied, followed by resection of the metastatic lesion. Histological analyses at that time revealed positive tumor margins, justifying adjuvant iliac radiation therapy. Due to rapid locoregional spread infiltrating diaphragm, chest wall and pleura accompanied by accumulating pleural effusions, the patient underwent partial resections including laparotomy and video-assisted thoracoscopy. Finally, treatment was switched to supportive chemotherapy, however, the patient continued to deteriorate and succumbed to the disease 12.5 months after initial diagnosis.

To compare the primary tumor lesion at diagnosis with the metastatic tumor we performed immunohistochemical analyses (Figure 2A). At diagnosis, the tumor contained a dominant sarcomatous component positive for vimentin and a focal carcinomatous component positive for E-cadherin. At recurrence, hematoxylin and eosin (H&E) staining revealed mainly spindle-shaped tumor cells diffusely positive for vimentin but negative for E-cadherin. To confirm the origin of the metastatic tumor, we demonstrated that tumor cells were, although only in foci, positive for CD10, a marker for Müllerian system-derived neoplastic mesenchymal cells [25] (Supplementary Figure S1A). In addition, a proliferation rate of up to 60% was detected in hotspot areas by Ki67 staining (Supplementary Figure S1A). Stainings for estrogen- and progesterone-receptor (ER, PR) indicated only heterogeneous nuclear expression patterns (Supplementary Figure S1A).

Taken together, the metastatic resected specimen was highly proliferative, had lost the carcinomatous histological component and was heterogeneously positive for CD10, ER and PR.

At time of progression, we obtained pleural effusion and could successfully establish the PF338 UCS cell line. Congruent to the metastatic tissue, immunohistochemical stainings of the cell block indicated focal positivity for CD10, strong positivity for vimentin and negative staining for E-cadherin (Figure 2A; Supplementary Figure S1B). In vitro, PF338 tumor cells demonstrated a biphasic growth pattern consisting of an epithelial-like component growing in a monolayer and a mesenchymal-like component growing in multiple layers (Figure 2B; Supplementary Video S1). In order to study the effect of multiple freezing cycles on tumor cell viability, we compared viability of PF338 cell passage 16 vs. passage 26 and found no differences (viability right after thawing 81.6 vs. 84.2%, viability after 72 h in culture 98.6% vs. 97.2%).

To illuminate the mutational background of the tumor cell line and the primary/metastatic tumor tissues we performed NGS for a predefined mutational panel that included the most commonly mutated genes in UCS (Supplementary Table S1). We identified a G13C mutation in KRAS, a R130Q mutation in PTEN and a mutation in ARID1A. Interestingly, an R93Q mutation in PIK3CA was only detected in the metastatic tumor and the cell line but not in the primary tumor.

In order to test whether PF338 cells are sensitive to standard-of-care UCS treatment paclitaxel plus platinum-based chemotherapy we performed in vitro sensitivity testing for both drugs. Accordingly, PF338 cells were sensitive to cisplatin and paclitaxel, with IC₅₀ values of 0.96 µM and 3.81 nM, respectively (Figures 3A,B). Interestingly, whereas cisplatin induced morphology changes to a more uniform, flat phenotype, paclitaxel treatment did not interfere with morphology (Figure 3A). Cell cycle analyses for cisplatin revealed a dose-dependent G2/M arrest, whereas paclitaxel induced apoptosis in a dose-dependent manner (Figure 3C). Despite being sensitive to cisplatin, a significant fraction of cells (15%) remained viable when treated with cisplatin at IC₅₀ concentrations for 10 days (Figure 3D; Supplementary Figure S2).

In order to test whether KRAS-mutant PF338 cells are sensitive to MAPK pathway inhibition, we used selumetinib, a MEK inhibitor that has been tested in KRAS-mutant tumors [26]. As sensitive control we used the BRAFV600E mutant A375 melanoma line [27]. We also tested the dual PI3K/mTOR inhibitor BEZ235 due to the mutations in the PI3K pathway and also treated the cells with galunisertib, a TGF-βRI-kinase inhibitor. PF338 cells were resistant to both selumetinib and galunisertib but strongly sensitive to PI3K/mTOR inhibition (IC₅₀: 63.42 nM) (Figures 4A,B). Cell cycle analyses revealed no changes upon selumetinib or galunisertib treatment and only a modest dose-dependent G0/G1 arrest upon BEZ235 treatment (Figure 4C). However, despite the resistance to galunisertib, pSMAD2 expression was abrogated by the treatment (Figure 4D).

Recent work in ovarian cancer demonstrated that mutations in ARID1A confer sensitivity to HDACi [18]. Furthermore, UCS frequently harbor alterations in cell cycle regulators and thus may show susceptibility to certain targeted therapies including PARP inhibitors [7]. Strikingly, PF338 cells were sensitive to both HDACi SAHA and PARP inhibitor olaparib with IC₅₀ of 0.38 and 4.60 µM, respectively (Figures 5A,B). SAHA treatment induced two distinct cell cycle patterns: a dose dependent G0/1 arrest at low drug concentrations and both a G2/M arrest and induction of apoptosis at high concentrations. In contrary, olaparib-treated cells went into G2/M arrest in a dose-dependent manner (Figure 5C). Importantly, PF338 tumor cells changed morphology from the initial biphasic to an epithelial phenotype upon SAHA treatment (Figure 5A; Supplementary Video S2). These phenotypic changes were accompanied by a dose-dependent upregulation of epithelial markers E-cadherin and—to a lesser extent—β-catenin (Figure 5D). Importantly, phenotypic and expression changes were also observed upon treatment with high-dose SAHA (4 µM) or valproate (Supplementary Figures S3A–D).