ISG15 as a prognostic biomarker in solitary fibrous tumour

Background Solitary fibrous tumour (SFT) is a rare mesenchymal malignancy that lacks robust prognostic and predictive biomarkers. Interferon-stimulated gene 15 (ISG15) is a ubiquitin-like modifier, associated with tumour progression, and with poor survival of SFT patients, as previous published by our group. Here, we describe the role of ISG15 in the biology of this rare tumour. Methods ISG15 expression was assessed by immunohistochemistry in tissue microarrays from SFT patients and tested for correlation with progression-free survival and overall survival (OS). The effects of ISG15 knockdown or induction were investigated for cancer stem cell (CSC) characteristics and for drug sensitivity in unique in vitro models of SFT. Results The prognostic value of ISG15 for OS was validated at protein level in malignant SFT patients, prospectively treated with pazopanib and enrolled in GEIS-32 trial. In SFT in vitro models, ISG15 knockdown lead to a decrease in the expression of CSC-related genes, including SOX2, NANOG, ALDH1A1, ABCB1 and ABCC1. Likewise, ISG15 downregulation decreased the clonogenic/ tumoursphere-forming ability of SFT cells, while enhancing apoptotic cell death after doxorubicin, pazopanib or trabectedin treatment in 3D cell cultures. The regulation of CSC-related genes by ISG15 was confirmed after inducing its expression with interferon-β1; ISG15 induction upregulated 1.28- to 451-fold the expression of CSC-associated genes. Conclusions ISG15 is a prognostic factor in malignant SFT, regulating the expression of CSC-related genes and CSCs maintenance. Our results suggest that ISG15 could be a novel therapeutic target in SFT, which could improve the efficacy of the currently available treatments. Supplementary Information The online version contains supplementary material available at 10.1007/s00018-022-04454-4.


Background
Solitary fibrous tumour (SFT) is a rare type of soft-tissue sarcoma (STS), with an estimation of 1 case per million people each year. These tumours were firstly described in the pleura, and they can appear anywhere in the body. In general, SFT consists of a well-encapsulated fibroblastic body that presents a significant collagenous component and prominent branching staghorn vasculature [1]. It is characterized by NAB2-STAT6 gene fusion, which is believed to be key for tumorigenesis [2][3][4]. This mesenchymal neoplasm can manifest in various clinically and histologically different subtypes. On histology, typical and malignant SFT (T-SFT and M-SFT respectively) are distinguished based on the mitotic index and/or presence of necrosis. Of note, dedifferentiated M-SFT (DD-SFT) is an extremely aggressive subtype, which presents an abrupt transition into a high-grade sarcoma [1,5]. However, these histologic features hardly predict SFT clinical course. Therefore, the most recent WHO classification suggests the use of risk-stratification models. The 3-variable risk model takes age at diagnosis, tumour size and mitotic index into account, while the 4-variable model includes also necrosis [1].
Localized T-SFT is commonly indolent, with surgical resection being the most effective treatment. However, very limited therapeutic options are available for advanced disease, when surgical intervention is impracticable. What is more, the monitoring of advanced cases is hindered due to the lack of efficient prognostic and/or predictive biomarkers in SFT. Recently, a phase-II clinical trial carried out by the Spanish Group for Sarcoma Research (GEIS), in collaboration with the French (FGS) and Italian (ISG) sarcoma groups, showed promising activity of antiangiogenics (i.e. pazopanib) in both T-SFT and M-SFT [6,7]. Remarkably, in the M-SFT cohort, ISG15 proved to be a relevant prognostic factor for progression-free survival (PFS) and overall survival (OS), both in the univariate and multivariate analysis.
ISG15 is a 15 kDa ubiquitin-like protein that can be secreted to the extracellular medium, found intracellularly in its free form, or can be covalently bound to other target proteins, in a process known as ISGylation [8][9][10][11].
ISGylation was at first studied for its role in the antiviral immune response. Following viral infection, type-I interferons (IFNα/β) are known to induce ISG15 expression and its conjugation with target proteins [12]. Recent studies have focused on the role of ISG15 in other major cellular processes, such as DNA repair [13][14][15], autophagy [16,17] or protein translation, as well as in pathological contexts like genotoxic stress or tumour development [18][19][20][21][22][23][24][25]. In sarcoma patients, ISG15 expression is up-regulated in tumour tissue compared to normal tissue, and its expression was included in a metastasis-related genetic signature for poor prognosis [26]. In addition, ISG15 and ISGylation have been associated with cancer stem cell (CSC) maintenance and behaviour, which can be translated into poor prognosis in patients. Of note, in pancreatic ductal adenocarcinoma (PDAC), not only does ISG15 extracellular paracrine-signalling play a key role in CSC maintenance [27,28], but also the intracellular ISG15 and ISGylation are required for CSC metabolic plasticity and mitophagy [29]. Besides, ISG15 can be a crucial microenvironmental factor in this malignancy, as tumour-associated macrophages can secrete ISG15, increasing CSC phenotype in tumour cells [30].
On these grounds, we addressed the prognostic value of ISG15 in a well characterized cohort of SFT. In parallel, the role of ISG15 in the maintenance of the CSC-like phenotype and drug resistance was investigated in SFT pre-clinical models.

Immunohistochemistry and tissue microarray (TMA) constructs
Two representative areas (1 mm in diameter) per tumour were selected, based on haematoxylin/eosin staining, for the generation of a tissue microarray (TMA). A TMA instrument (Beecher Instruments; Sun Prairie, WI, USA) was used for TMA assembly. Immunohistochemistry was performed in TMAs 4-µm sections, using an anti-ISG15 monoclonal antibody (sc-166755, Santa Cruz Biotechnology, Dallas, TX, USA). The percentage of ISG15-positive tumour cells was evaluated using the following scoring system: negative (0% positive cells), low (+ , 5-25% positive cells), intermediate (+ +, 25-50% positive cells) and high (+ + + , > 50% positive cells). Staining intensity of ISG15 was graded as negative, weak, or strong. Samples with a negative/low percentage of ISG15-positive cells or negative/weak intensity were included in the low ISG15 group. Samples with an intermediate/high percentage of positive cells or strong intensity were included within high ISG15 group. ISG15 staining was independently evaluated by two sarcoma expert pathologists, blinded for clinical data. Normal liver tissue was used as a positive control for ISG15 staining, according to the manufacturer's instructions. ISG15 expression levels were quantified by HTG EdgeSeq technology as described before [6,7]. Upper quartile Q3 was considered the cut-off value to discriminate between high and low ISG15 expression groups.

Statistical analysis
Overall survival (OS) and progression-free survival (PFS) were measured from the date of initial treatment, within the clinical trial, to the final event (patient death for OS, disease progression according to Choi criteria or death for PFS) and were estimated according to the Kaplan-Meier method. The associations between the variables of interest (i.e., protein/ gene expression and clinical outcomes) were performed by the log-rank test, statistical significance was defined at p = 0.05. Hazard ratios (HR) for the multivariate analysis were calculated following Cox's regression. All the statistical procedures were performed with SPSS 22.0 software (IBM, Armonk, NY, USA).  (ATCC ® CRL-3216™;  ATCC) were used for this study. INT-SFT, 93T449, CP0024,  ICP059 and ICP060 cell lines were cultured in RPMI  medium (Gibco™, Thermo Fischer Scientific, Waltham,  MA, USA), HT-1080 and AA cell lines were maintained in F-10 medium (Gibco™), SW982 cell line was cultured in Leibovitz's L-15 Medium (Gibco™) and HEK293T cell line in DMEM medium (Gibco™). All cell culture mediums were supplemented with 10% FBS (Gibco™), and 100 units/ mL penicillin and 100 µg/mL streptomycin (Sigma-Aldrich, San Luis, MO, USA). Cells cultures were kept at 37 ºC in a 5% CO 2 atmosphere and tested routinely for mycoplasma or fungi contamination. All cell lines were discarded after 2 months, and new lines obtained from frozen stocks.

ISG15 in vitro silencing
HEK293T cells were transfected by the calcium phosphate method, with lentivirus-producing plasmids PMD2.G-VSV-G and pCMV-dR8.91 (Addgene, Watertown, MA, USA); and the plkO.1-puro plasmid containing the sequence for either non-targeting (SHC016-1EA, Sigma-Aldrich) or ISG15 shRNA (TRCN0000007421, Sigma-Aldrich). After 24 h medium was replaced with fresh DMEM. 48 h-post transfection the lentivirus-containing media were filtered through a 0.45 µm syringe filter, supplemented with 4 mg/ ml polybrene (Sigma-Aldrich) and used to transduce INT-SFT or IEC139 cells. Two additional cycles of infection were carried out every 12 h. Transduced cells were selected using 0.5 µg/ml puromycin (Thermo Fisher Scientific) for 5-7 days. For long-lasting silencing, clonogenic cell lines were established through single-cell sorting using BD FAC-SJazz (BD Biosciences, Franklin Lakes, NJ, USA). Silencing was verified by western blot (WB) and real-time quantitative reverse polymerase chain reaction (RT-qPCR) analysis.

Tumoursphere and colony formation assay
For sphere-forming ability assays, INT-SFT or IEC139 cells were seeded in Corning ® Costar ® (NY, USA) Ultra-Low Attachment 96-Well Plate at 1.5 × 10 3 cells/well. Then, images were obtained after 8 days using an inverted microscope Olympus IX-71. In 3D drug resistance assays, INT-SFT and IEC139 clones were plated at different densities, in order to obtain spheroids of comparable size at the time of treatment: 5 × 10 3 cells/well for shNT and 10 4 cells/well for shISG15. 4 days after plating, spheroids were properly formed and were then treated with either 20 µM pazopanib (Novartis, Basel, Switzerland), 0.5 nM trabectedin (Phar-maMar, Madrid, Spain) or 50 nM doxorubicin (Sigma-Aldrich). Drug concentrations correspond approximately to 2X IC50 values for each drug according to cell viability in 2D cultures. After 72 h, drugs were removed. Images were acquired each day using an inverted microscope Olympus IX-71, until appreciable differences in regrowth: 12 days post-treatment for pazopanib and doxorubicin; 20 days posttreatment for trabectedin. Sphere size (area) was determined using the ImageJ tool Analyze Spheroid Cell Invasion In 3D Matrix (RRID:SCR_021204).
For colony formation assay 1 × 10 3 INT-SFT cells were seeded in 100 mm plates. After 8 days, colonies were stained using methyl violet dye (Thermo Fischer Scientific) and counted.

Proliferation, migration, and invasion
A total of 10 3 cells/well were seeded in 96-well plates and were left in the incubator to settle for 24 h. Each day (until day 6), 20 µl of CellTiter 96 ® AQueous One Solution Cell Proliferation Assay (MTS) (Promega, Madison, WI, USA) was added to the media. After 30 min of incubation, absorbance at 490 nm was read using iMark microplate absorbance reader (Bio-Rad, Hercules, CA, USA). For wound healing assays, cells were cultured in 6-well plates until completely confluent. Then, a scratch was gently made across the well using a 200 µl pipette tip, cells and debris were rinsed using PBS. Subsequently, images were obtained with an Olympus IX-71 microscope at various time points. The area of the gap was quantified for each image using ImageJ software and the speed of migration was calculated at µm 2 /h. For invasion assays, cells were cultured in serum-free media for 24 h. Then, 10 4 cells were seeded in the top chamber of 8 µm pore polycarbonate transwells (Corning Costar), previously treated overnight with 0.2X Cultrex Basement Membrane Extract (BME) in coating buffer (Trevigen, McKinley, MN, USA). Serum-containing media was used as chemoattractant in the bottom chamber. After 48 h, the top side of each polycarbonate insert was cleaned using a cotton swab and rinsed with PBS. The membrane was fixed for 5 min with 100% methanol at -20 ºC, then stained using DAPI reagent (Invitrogen) and mounted on a slide to be observed under fluorescent microscopy (Olympus BX-61). Cell nuclei were counted for each condition.

Apoptosis analysis
The number of apoptotic, early apoptotic, and necrotic cells was evaluated in INT-SFT/IEC139 sh non-targeting (shNT) and INT-SFT/IEC139 shISG15 cell lines, after 20 µM pazopanib, 0.5 nM trabectedin or 25 nM doxorubicin 72 h treatment. A FITC Annexin V Apoptosis Detection Kit with PI was used to determine cell death (Immunostep; Salamanca, Spain), following the manufacturer's instructions. Apoptosis levels were determined by flow cytometry, FACSCanto™ II Cell Analyzer (BD Biosciences) and data analysed with both BD FACS Diva and FlowJo software.

RNA extraction and RT-qPCR
RNA was extracted from cell culture with the RNA Pure-Link RNA Mini Kit (Invitrogen, Carlsbad, CA, USA), quantified with the NanoDrop One C spectrophotometer (Thermo Fisher Scientific, Madison, WI, USA) and reverse transcribed to cDNA using the High-Capacity Reverse cDNA Transcription Kit (Applied Biosystems, Thermo Fischer Scientific, Foster City, CA, USA). Expression levels of the selected genes were measured by RT-qPCR, using the following TaqMan RNA probes (Applied Biosystems): ISG15 (Hs00192713_m1), SOX2 (Hs01053049_s1), NANOG (Hs04260366_g1), ABCB1 (Hs01067802_m1), MYC (Hs00153408_m1) and ALDH1A1 (Hs00946916_m1); GAPDH (Hs03929097_g1) was used as housekeeping gene for data normalisation. An ABI Prism 7900HT (Applied BioSystems) real time PCR system was used. The relative expression of genes was expressed using INT-SFT, IEC139 or shNT as a reference, depending on the experiment. ISG15 expression induction was performed in INT-SFT/IEC139 shNT or shISG15 cells after treatment with 250 U/ml human IFN-β1a (Miltenyi Biotec, Bergisch Gladbach, Germany), 20 µM pazopanib, 0.5 nM trabectedin for 48 h or 25 nM doxorubicin.

Immunofluorescence (IF)
A total of 2 × 10 4 cells were plated on round glass coverslips contained in 24-well plates. After 24 h incubation, coverslips were fixed with 3% paraformaldehyde (PFA) and then permeabilised using 0.2% Triton X-100 (Sigma-Aldrich) 2X PBS. For ER staining, cells were exposed in vivo for 30 min with 1X ER Staining Kit-Cytopainter (Abcam), prior to fixation and permeabilization. Blocking was done with 1% BSA 2X PBS for 30 min RT. ISG15 (Santa-Cruz) 1:200 was incubated at 4 ºC overnight. The following secondary fluorescent antibodies were used at 1:1000 for 1 h at RT: Goat anti-Rabbit IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 488 and Goat anti-Mouse IgG1 Cross-Adsorbed Secondary Antibody, Alexa Fluor 546 (Invitrogen). Nucleic acid stain DAPI (Invitrogen) was subsequently added for 15 min. After careful washing, coverslips were mounted onto microscope slides with ProLong™ Gold mounting media (ThermoFisher). Fluorescence microscope Olympus BX-61 was used for image acquiring. For ER/ ISG15 colocalization Z-stack images were obtained using microscope Leica SP5 (Leica, Wetzlar, Germany). For spheroid IF, 2 µM Calcein AM (Cayman Chemical, Ann Arbor, MI, USA) and 33 µM Hoechst 33,342 (Abcam) were incubated for 3 h in cell culture incubator. Subsequently, spheroids were visualized at confocal microscope Leica SP5, using the same acquisition parameters for independent experiments. Dyes were not required to be washed away [32,33]. For signal quantification, intensity density (IntDen)/ spheroid area ratio was determined using ImageJ.
High ISG15 protein expression correlated with worse OS in M-SFT patients (NA; Fig. 1B Table 2 and 3]. Furthermore, high and low ISG15 groups for gene expression and staining intensity showed a positive correlation by χ 2 analysis. ISG15 did not show any prognostic value for the T-SFT cohort ( Supplementary Fig. 1).

ISG15 is overexpressed in M-SFT cell line versus other STS cell lines
ISG15 expression was determined, both at mRNA and protein levels, in a panel of sarcoma cell lines: INT-SFT, IEC139, 93T449, CP0024, AA, HT1080, ICP059, ICP060 and SW982. ISG15 gene expression levels were significantly higher, in INT-SFT cell line followed by IEC139 (both M-SFT) against other sarcoma subtypes ( Fig. 2A). Specifically, INT-SFT presented 3.0-and 7.9-fold higher ISG15 mRNA levels, as compared to SW982 and 93T449 cell lines respectively. When compared to CP0024, AA, HT1080, ICP059 and ICP060, INT-SFT showed 19.6-to 55.6-fold higher ISG15 gene expression. IEC139 expressed between 8.2-and 23.7-fold more ISG15 mRNA levels than these lines and 3.3-fold more than 93T449. Analogous results were obtained at protein level (Fig. 2B). Similarly, INT-SFT showed 4.9-to 33.3-fold more ISG15 protein, when compared to other sarcoma subtypes, but only 1.98fold more when compared to IEC139. ISG15 expression is seemingly not related to SV40 LargeT antigen transformation, as no ISG15 is observed in HEK293T cells (Supplementary Fig. 1). Of note, ISG15 appeared to be differentially overexpressed in two sarcoma datasets (Tumor Sarcoma Mesenchymal-Boshoff-96 and Tumor Sarcoma-Filion-137) versus a normal tissue repository (Normal Various (GNF)-Su-158) ( Supplementary Fig. 1), supporting potential interest of ISG15 expression in sarcoma. In SFT, ISG15 expression is mainly cytoplasmic, with accumulation at vesicles and areas adjacent to nuclei, which does not colocalize with endoplasmic reticulum (Supplementary Fig. 1G). However, nuclear location can also be observed (Fig. 2C).
For IEC139 cells, ISG15 silencing was not substantial enough to prevent induction of the protein by pazopanib, trabectedin or doxorubicin exposure. This translates to sarcoma CSC-related markers (SOX2 being the most reliable) to be upregulated in both shNT and shISG15 when cells were treated with each drug (Supplementary Fig. 2F).
In addition, transcriptomic data on ISG15 silencing was obtained using the microarray Clariom S Assay, human. A gene set enrichment analysis (GSEA) was performed comparing INT-SFT shISG15 and INT-SFT shNT cells. Downregulation of various genes belonging to key CSC pathways was observed, like epithelial-mesenchymal transition,  . E CSC-related gene expression is also inhibited in IEC139 shISG15 cells. F Colony formation ability is impaired by ISG15 silencing. 1000 cells/condition were seeded in 10 mm plates then, after 8 days in the incubator, colonies were stained with methyl violet and counted (n = 4). G Spheroid forming ability is reduced for INT-SFT shISG15#2, which corresponds to greater ISG15 knockdown. IEC139 shISG15 spheroids are also of smaller size. 1500 cells/condition were seeded, after 8 days images were obtained, and sphere size was quantified (n = 3). T-student tests: (*) stands for p-value < 0.05, (**) for p-value < 0.01 and (***) for p-value < 0.001. Error bars are indicative of means ± SD. n.s. not significant TGF-β, p53 and myogenesis. Interestingly, a significant increase was shown in genes related with KRas, MYC or MTORC1 ( Supplementary Fig. 2G).

Drug resistance is related to ISG15 expression
Apoptosis was augmented in shISG15 vs shNT, when M-SFT cells were treated with 20 µM pazopanib (recommended first line antiangiogenic in SFT), 25 nM doxorubicin (first line chemotherapy in STS) or 0.5 nM trabectedin (second line treatment in STS) for 72 h (Fig. 5A, C; Supplementary Fig. 3). To be precise, apoptotic (Annexin V-positive) cell populations showed significant increases of 2.99-fold for pazopanib, 1.20-fold for trabectedin and 2.04-fold for doxorubicin treatments, in INT-SFT shISG15 compared to INT-SFT shNT cells (Fig. 5A, C). Likewise, apoptosis was significantly augmented in IEC139 shISG15 cells by 1.42-fold for pazopanib, 1.10-fold for trabectedin and 2.19fold for doxorubicin treatments (Fig. 5B, C). In parallel, drug resistance was tested in 3D spheroid cultures. At 72 h with pazopanib, INT-SFT shISG15 spheroids exhibited a more drastic shrinkage in area (43.1%) than INT-SFT shNT spheroids (26.3%), due to increased cell death (Fig. 5E, Supplementary Fig. 4). However, no significant size differences were observed, between silenced and control cells, at 72 h for trabectedin or doxorubicin treatments. Besides, haematoxylin-eosin staining revealed a more pronounced fibrotic/necrotic state of shISG15 spheres after pazopanib, trabectedin and doxorubicin treatment (Fig. 5D). In contrast, IEC139 shISG15 spheroids did show a significant reduction in area compared to control, after 72 h trabectedin or doxorubicin, but not for pazopanib treatment (Fig. 5D).
In addition, INT-SFT 3D spheroids were stained using Hoechst 33,342 and Calcein AM which mark nuclei and live cells respectively. Concordantly, the raw intensity signal of Calcein AM/area ratio was significantly lower in ISG15 deprived spheroids versus control, after 72 h drug treatment (Fig. 6). Specifically, Calcein AM signal/area ratio was 68.8% lower in pazopanib, 65.1% in trabectedin and 55.6% in doxorubicin treatments, for shISG15 compared to shNT spheroids. No significant differences were observed for untreated tumour spheres. At this point, to determine cell viability, tumour spheres were released from the drug. Under these conditions, the growth capacity of spheroids was restored and shNT regrew up to reach approximately their original size (95.7%) on day 12 without pazopanib. In contrast, shISG15 spheres not only did not enlarge but also continued shrinking to 34.9% their original size (Fig. 5E, Supplementary Fig. 4). Similar effects were observed when spheroids were treated with trabectedin or doxorubicin, as drug removal resulted in the maintenance of the growth capacity only in the control group ( Fig. 5E; Supplementary Fig. 4). By day 20 without trabectedin, shNT spheres did not show any further shrinking (60.6% their original size) while the size of ISG15-silenced spheres dropped to 33.1% (Fig. 5D, Supplementary Fig. 4). The same pattern was observed for doxorubicin (shNT remained essentially at 57.4% of the original size; shISG15 dropped and treated tumour spheroids for 3 h at 37 ºC. Images correspond to maximum projection of Z-stack, acquired using confocal microscopy. Independent experiments were performed (n = 5). B Quantification using ImageJ of Calcein AM intensity/area for each condition to 30.5% (Fig. 5E, Supplementary Fig. 4). IEC139 tumourspheres were also released from drug exposure after 72 h, and their size was followed for 8 days post-treatment. At this point spheroids were not able to regrow, but IEC139 shISG15 spheroids were smaller relative to their original size than shNT: 29.5% vs 26.6% for trabectedin and 38.3% vs 31.6% for doxorubicin respectively. No significant differences were observed for pazopanib post-treatment (Fig. 5E).

Discussion
The findings presented here validate the prognostic value of ISG15 in M-SFT [6] and the predictive relevance in the context of antiangiogenic treatment. These effects could be related to the important role of ISG15 in drug resistance and maintenance of CSC characteristics, as we observed in our preclinical models of this specific STS type.
ISG15 overexpression significantly correlated with worse OS, in our complete series (T-SFT and M-SFT), a result that was strongly dependent on the expression of this gene in M-SFT. Accordingly, in M-SFT patients, PFS was lower in patients with high ISG15 gene expression. In contrast, in T-SFT ISG15 did not show an impact in survivals. In concordance with gene expression, both high ISG15 protein expression and strong ISG15 protein immunostaining correlated with lower OS, in the M-SFT cohort. However, when taking T-SFT patients into account, ISG15 at protein level loses its significance. The latter, together with ISG15 gene expression being more robust in M-SFT, points at ISG15 as a prognostic marker in more advanced/malignant stages of this disease. Also, a positive statistical correlation is proven between ISG15 expression and number of mitoses, which is a sign of malignancy. Several other studies based on clinical data have associated high ISG15 expression with unfavourable prognosis in cancer patients [19,23,29,34], as well as with higher histological grade, tumour size or invasiveness [24]. Furthermore, in pancreatic adenocarcinoma, peripheral blood ISG15 has also been validated as a potential diagnostic biomarker of cancer patients when compared to healthy controls [35].
As it is well known, the expression of SFT markers STAT6, CD34, CD99 and bcl-2 can be lost in most aggressive tumours, following a process of dedifferentiation [36][37][38][39]. Thus, ISG15 being elevated in more aggressive SFT entities may imply its involvement in dedifferentiation and stemness enhancement. However, no association between SFT markers and ISG15/ISGylation has been described to date.
In our experimental conditions, ISG15 gene silencing inhibited proliferation, but not migration or invasion; in contrast with other published work that observed a decrease of either of these tumoral characteristics in different malignancies [22,23,25,40,41]. However, this might indicate that ISG15 pro-tumoral functions are heterogeneous and depend on the cancer type. On the other hand, ISG15 knockdown decreased the expression of specific markers associated with a CSC-like phenotype in our SFT cell line. Lower levels of sarcoma CSC markers SOX2, NANOG, ALDH1A1, ABCB1 and ABCC1 [42] were observed in silenced cells in 2D and in CSC-enriched cultures (single-cell and 3D/ spheroid).
Additionally, ISG15 induction shows a direct effect in the expression of the mentioned CSC-related genes. This is demonstrated by the increase in the expression of SOX2, NANOG, ALDH1A1, ABCB1 and ABCC1 when ISG15 is induced by IFN-β or by drug treatment, which is not observed in ISG15-deprived silenced INT-SFT cells. Moreover, a greater ISG15 induction, as seen in trabectedin treatment, seems to correspond to a sharper increase in most of said CSC markers. Also, this effect is not replicated in IEC139 cells, which ISG15 silencing is not considerable enough to prevent its induction, and thus CSC markers upregulation. To our knowledge, it is the first time a positive correlation has been described between ISG15 and these CSC-related factors. The upregulation of CSC markers when treated with pazopanib, trabectedin or doxorubicin might warn us about the possible early activation of ISG15-regulated resistance mechanisms in SFT. Besides, some authors advise against the use of doxorubicin in SFT, at least as first-line, due to the addition of genomic instability [43]. Monitoring of ISG15 levels in SFT patients, throughout antiangiogenics or chemotherapy treatment, is needed to prove this hypothesis.
Furthermore, shISG15 cells presented lower tumourinitiating capacity, reflected by impaired spheroid forming ability and clonogenic capacity. In addition, cells expressing higher levels of ISG15 are more resistant to antiangiogenic pazopanib and chemotherapy drugs (doxorubicin and trabectedin) in 2D and 3D cultures; and are more capable of "reforming" spheroids after being damaged by drug administration. All these characteristics have been associated with a CSC phenotype. In general, CSC marker expression, colony/sphere-formation abilities and drug resistance are more reduced in INT-SFT than in IEC139 ISG15 knockdown. This corresponds with a greater ISG15 silencing in INT-SFT shISG15 cells, and might indicate that the CSC characteristics tested here could be proportionally dependent on ISG15 levels. Supporting our findings, other studies claimed that ISG15 promotes CSC behaviour, colony-forming capacity and tumorigenesis in nasopharyngeal and ductal pancreatic cancer cell lines [27][28][29][30]44], which translates into a worse prognosis in these patients. Besides, ISG15 has been described to play a pro-tumoral role in various malignancies like bladder, nasopharyngeal, breast, hepatocellular or pancreatic cancer [18-24, 27-30, 44]. In sarcoma, a recent bioinformatics study into Ewing sarcoma, comparing tumour samples versus non-cancerous samples, identified ISG15 as one of the hub genes in a protein-protein interaction network (PPI) [45]. Conversely, ISG15 showed a protective anti-tumour role in glioblastoma and ovarian models [46,47]. Because of these apparently contradictory functions, ISG15 has been designated as a "double-edged sword" in tumour development [48]; however, our data suggest that ISG15 has a pro-tumoral role in SFT.
ISG15 oncogenic mechanisms are also diverse, as it can act intracellularly in its free [27,28] or its conjugated form [22,25,44]; as well as extracellularly as a microenvironmental modulator [23,28,30]. More specifically, both paracrine secretion from M2 macrophages and autocrine ISG15 secretion from pancreatic tumour cells are able to induce a CSC phenotype in said cells [28,30]. However, the receptor binding to ISG15 or a possible positive correlation between secreted and intracellular/conjugated forms remains unknown. Our data support that autocrine tumoursecreted ISG15 may also play a role in CSC enhancement in an SFT context. Indeed, one of the limitations of this study is precisely related to the multiple physiological forms in which ISG15 can be found: secreted, intracellularly free and conjugated; which cannot be differentiated in patient samples at protein (IHC) or RNA level. Our knockdown experiments showed that all these forms of ISG15 are downregulated in INT-SFT, but ISG15 conjugates are not significantly reduced in IEC139, which could point that free ISG15 plays a more important role. Besides, transcriptomic analysis of ISG15 knockdown in SFT cells showed the downregulation of genes related to key CSC pathways like EMT, TGF-β or p53. However, how each one of these physiological forms can participate in SFT prognosis and the exact mechanism of action needs further research in preclinical models.
Overall, our in vitro studies suggest that ISG15 may be related to poor prognosis in SFT patients due to an enhancement in CSC phenotype. Thus, ISG15 presenting greater prognostic value in M-SFT patients may indicate its involvement in dedifferentiation and stemness processes and might function at latter stages of this tumour development. Moreover, ISG15-targets when ISGylated may vary their function; they can be degraded or protected from degradation at proteasome level, which leaves us with a rather complex situation. The role of ISGylation in SFT and other sarcomas is currently being studied in our laboratory. For future directions, a potential clinical trial could be designed considering ISG15 effect for advanced SFT patients. It would be of great interest to select patients with low ISG15 expression, expecting to improve clinical outcome when treated with efficiency-proven drugs like pazopanib.

Conclusions
ISG15 is validated as a prognostic biomarker in M-SFT patients and could also present predictive value in antiangiogenic-treated patients. Our preclinical results suggest that worse prognosis could be a consequence of ISG15-mediated CSC behaviour and drug resistance mechanisms. These findings provide key information for future SFT clinical trials and a novel therapeutic target in this malignancy.