Novel thrombospondin-1 transcript exhibits distinctive expression and activity in thyroid tumorigenesis

Thrombospondin 1 (TSP1) is known for its cell-specific functions in cancer progression, such as proliferation and migration. It contains 22 exons that may potentially produce several different transcripts. Here, we identified TSP1V as a novel TSP1-splicing variant produced by intron retention (IR) in human thyroid cancer cells and tissues. We observed that TSP1V functionally inhibited tumorigenesis contrary to TSP1 wild-type, as identified in vivo and in vitro. These activities of TSP1V are caused by inhibiting phospho-Smad and phospho-focal adhesion kinase. Reverse transcription polymerase chain reaction and minigene experiments revealed that some phytochemicals/non-steroidal anti-inflammatory drugs enhanced IR. We further found that RNA-binding motif protein 5 (RBM5) suppressed IR induced by sulindac sulfide treatment. Additionally, sulindac sulfide reduced phospho-RBM5 levels in a time-dependent manner. Furthermore, trans-chalcone demethylated TSP1V, thereby preventing methyl-CpG-binding protein 2 binding to TSP1V gene. In addition, TSP1V levels were significantly lower in patients with differentiated thyroid carcinoma than in those with benign thyroid nodule, indicating its potential application as a diagnostic biomarker in tumor progression.


INTRODUCTION
There has been a rise in the number of thyroid cancer cases in the USA (43,800 new cases in 2022) [1] and Korea (59.8 per 100,000 in 2019) [2]. The most common thyroid cancer types are papillary thyroid carcinoma (PTC) and follicular thyroid carcinoma (FTC), which are well differentiated and have indolent properties. There is an urgent need to identify biomarkers to determine whether treatment is necessary and to avoid over-treatment in patients who do not require surgery. PTC can be diagnosed using fineneedle aspiration cytology (FNAC) and BRAF mutation profiling before surgery [3]; however, a prognostic marker is needed because some patients display an aggressive clinical course. Additionally, among the diseases diagnosed as follicular neoplasms with FNAC, the distinction between malignant [follicular thyroid carcinoma (FTC), follicular variant papillary thyroid carcinoma (FVPTC)] and benign [follicular adenoma (FA)] neoplasms is currently only possible in surgical specimens [4,5]. Therefore, a diagnostic marker is required for preoperative diagnosis. Biomarkers can be classified into three categories: DNA, RNA, and protein markers. The identification of new biomarkers is highly desirable for cancer treatment for early detection and correct diagnosis. Several potential thyroid cancer biomarkers have been reported, including the ratio of pro-and mature nonsteroidal anti-inflammatory drug-activated gene-1 expression [6], radio-induced rearrangement in transformation/ papillary thyroid carcinomas (RET/PTC), mutations of BRAFV600E and RAS genes in PTC [7,8], paired box 8-peroxisome proliferator-activated receptor γ fusion gene [9], loss of heterozygosity on chromosome 3p and 7q, and RAS mutations in FTC [10]. The carcinogenic potential of RET/PTC and its expression in different tumors have been extensively studied using different methods in cells and tissues. However, previous research was challenged by the finding that RET/PTC expression is not an absolute PTC marker [11], thereby raising concerns regarding its clinical utility in inconclusive FNAC. RNA biomarkers have not been as extensively investigated as DNA/protein markers and exhibit several unique characteristics, including better specificity and sensitivity than other markers. Most RNA biomarkers that exhibit differential expression between normal and cancerous tissues are derived from alternatively spliced transcripts. In addition, upregulation or downregulation of RNA-binding proteins or splicing factors in cancer is associated with the alternative splicing (AS) of oncogenes or tumor suppressor genes [12]. Thus, AS is an important event associated with tumorigenesis and cancer therapy.
AS can generate multiple gene products from a single primary transcript; thus, it is an important post-transcriptional regulation process. Over 70% of human genes undergo AS, and some splicing abnormalities are also associated with tumorigenesis [13]. AS is regulated by chromatin structure and histone modifications. For example, an AS form of SMAD4 is produced in thyroid cancer cells, which lack some portions of the linker region, thus giving rise to a SMAD4 regulatory mechanism [14]. In association with multiple ligands, thrombospondin 1 (TSP1) is the main player in the tumor microenvironment and regulates diverse processes, such as adhesion, invasion, migration, proliferation, apoptosis, immune response, and treatment response [15][16][17][18][19]. In thyroid cancer, TSP1 plays a critical role in the regulation of cell proliferation, invasion, and migration [20][21][22]. For example, TSP1 participates in the progression of PTC with the BRAFV600E mutation [23]. Therefore, TSP1 might be a potential target for cancer treatment.
In this study, we discovered a new biomarker, TSP1V, in thyroid cancer and provided the molecular mechanism underlying TSP1V transcript generation and its functioning as a tumor suppressor.

Identification of a new alternatively spliced TSP1 in thyroid tissues
To identify a novel biomarker for thyroid cancer, we performed an antibody array using normal human thyroid and tumor tissues.
Among the 507 proteins in the array, TSP1 was highly expressed in the tumor tissue compared to adjacent normal tissue (Fig. 1A). To validate the results obtained from the antibody array, we conducted western blot analysis using proteins extracted from thyroid cancer patients. Western blot analysis revealed higher expression of TSP1 in the tumor tissue (Fig. 1B). This result was consistent with the antibody array data, indicating that TSP1 is highly expressed in tumors compared to adjacent normal tissues. To examine whether the pattern of TSP1 transcript and protein expression was identical in normal and tumor tissues, we isolated total RNA from the three patients and examined TSP1 transcripts. Using the primers for exons 12 and 13, we were able to amplify the wild-type band; however, an extra band was detected in all three samples (Fig. 1C). Although the overall level of RNA transcripts was comparable, the expression pattern of the top band (682 bp) and bottom band (232 bp) varied, indicating that the top band resulted from aberrant expression. Sequencing analysis of these bands revealed that the top band contained F. Fig. 1 Thrombospondin 1 (TSP1) was selected as a strong candidate diagnostic biomarker for thyroid cancer. A Antibody array (RayBio L-Series Human Antibody Array 507, RayBiotech) showed differential expression of TSP1 between normal thyroid and tumor tissues. B Western blotting was performed using the proteins extracted from normal and tumor tissues from patients with papillary thyroid carcinoma (PTC). C Reverse transcription polymerase chain reaction (RT-PCR) was performed using exon 12 forward and exon 13 reverse primers from TSP1 using RNA extracted from the tissues of patients with PTC. D From exon 2 with a start codon to exon 21 with a stop codon of TSP1, primers were designed in five parts. RT-PCR was performed using these primers and HTori-3, immortalized human thyroid cells, as a template (100 bp marker). E, F RT-PCR was performed using HTori-3 as template with exon 7 forward and intron 9 reverse primers set, exon 9 forward and intron 12 reverse primers set from TSP1. G The intron 9 reverse primer from TSP1 was fixed, and HTori-3 was subjected to RT-PCR as a template with exons 6, 3, and 2 forward primers.
exons 12 and 13, along with intron 12. In addition, the bottom band containing TSP1W was highly expressed in the tumor compared to the normal tissue, whereas the spliced form was highly expressed in the normal tissue compared to the tumor tissue, in all three samples. The intron-retained product was named TSP1V. To examine the aberrant TSP1 expression in other regions, we designed primers by dividing the exons of TSP1 into five parts and performed reverse transcription polymerase chain reaction (RT-PCR). As shown in Fig. 1D, the expected bands (no introns) were observed in each part, including two bands with specific primers for exons 12 and exon 13. To further confirm the minor alternatively spliced forms, RT-PCR was performed by fixing the reverse primers in the introns (Fig. 1E-G). The results revealed an additional variant containing intron 8 (Fig. 1F). Since the variant form is minimally expressed in cells (Fig. 1G), we mainly focused on the major variant form containing intron 9 in this study.
TSP1V is secreted into the medium and deposited into the extracellular matrix (ECM) We identified three forms of TSP1 transcripts in thyroid cancer: TSP1W, TSP1V (major), and TSP1Vm (minor). After sequencing and deduced amino acid analysis, TSP1V produced only 63 kDa proteins due to its termination codon in intron 9 ( Fig. 2A). First, three thyroid cell lines, HTori-3, TPC-1, and BCPAP, were used to confirm the TSP1W and TSP1V endogenous levels. TSP1W was highly expressed in TPC-1 cells and barely detected in HTori-3 and BCPAP cells. However, TSP1V expression was extremely weak in all three cell lines (Fig. 2B) and was only observed in the longexposure blot. To determine whether TSP1W and TSP1V were expressed in the cells and media, BCPAP cells (low TSP1W expression) were transfected with an empty vector (EV) and expression vectors. TSP1W was detected in the serum-free medium (SFM) but not in the cell lysates (Fig. 2C, left panel), indicating its rapid secretion into the medium in the absence of serum. TSP1V was neither detected in the media nor cell lysates. However, when the cells were incubated with 2% serum medium, TSP1V was detected in the cell lysate (Fig. 2C, right panel). To determine whether TSP1V is secreted, immunoprecipitation was performed; TSP1V was detected in both the cell lysate and conditioned medium (Fig. 2D). To determine the location of TSP1V, we isolated ECM with different concentrations of serum and found that most of the TSP1V was deposited in the ECM when cells were grown in the absence of serum (Fig. 2E). Overall, our data suggests that TSP1W is secreted into the medium, whereas TSP1V is secreted and deposited into the ECM.

D.
Serum free media  . C BCPAP cells were transfected with empty vector (EV), TSP1W, or TSP1V. These cells were incubated in SFM or 2% serum medium and harvested after 24 h. CM and CL were analyzed using western blotting. D Immunoprecipitation assay was performed using CM (2% serum medium) with rabbit anti-TSP1 and normal rabbit IgG. CM and CL were applied as the input. E Extracellular matrix (ECM) was harvested after 24 h incubation in SFM or 5% or 10% serum medium from transiently transfected BCPAP cells and analyzed using western blotting.

TSP1V-expressing cells exhibit antagonistic activity against TSP1W-expressing cells by inhibiting the focal adhesion kinase (FAK) and Smad pathways
The cell proliferation assay suggested that TSP1V expression suppressed cell growth compared to that of EV and TSP1W (Fig. 3A). TSP1V expression also decreased cell migration and invasion activity when compared to those TSP1W and EV expression ( Fig. 3B, C). TSP1V-expressing cells formed a lower number of spheroids and colonies than cells expressing EV and TSP1W (Fig. 3D, E). Thus, TSP1V overexpression resulted in the opposite effect to that of TSP1W. To determine the signaling pathway affected by TSP1V, we investigated the Smad pathway affects by TSP1W in tumorigenesis [24]. A dual-luciferase assay using the p3TP-luc construct, responsible for the transforming growth factor beta (TGF-β) pathway, was conducted to determine the involvement of Smad in the function of TSP1V. After cotransfection of p3TP-luc and pRL-null cells with EV, TSP1W, or TSP1V, the cells were treated with vehicle, phorbol 12-myristate 13-acetate (PMA), TGF-β, and combination of PMA and TGF-β. In the EV and TSP1W transfected cells, PMA and/or TGF-β treatment increased luciferase activity. However, this increase was significantly suppressed in the TSP1V-transfected samples (Fig. 3F). Stable cell lines were produced with EV, TSP1W, and TSP1V expression. The phosphorylation of Smad2 and FAK increased in a time-dependent manner when TGF-β was added but did not occur in the TSP1V-overexpressing cells (Fig. 3G), indicating that the Smad and FAK pathways may be involved in TSP1V activity.
Finally, to confirm the function of TSP1V in vivo, a stable cell line was xenografted into mice thighs. As shown in Fig. 3H-J, TSP1V inhibited tumor growth without affecting the mouse weight. Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assay was performed using the tumor sections; we found that apoptosis occurred more actively in the TSP1Voverexpressing tumor (Fig. 3H). Analysis of cell lysates isolated from the tumors confirmed that TSP1V was expressed in each tumor tissue (Fig. 3K). Overall, TSP1V suppressed tumorigenesis in vitro and in vivo by partially inhibiting the phosphorylation of Smad2 and FAK.

TSP1V expression is upregulated by anticancer compounds
To confirm the expression of TSP1V in human cell lines, RT-PCR was performed using extracted RNA from thyroid and non-thyroid cell lines (Fig. 4A, B). RT-PCR data revealed that both TSP1W and TSP1V were expressed in all the cell lines tested, including thyroid (HTori-3, TPC-1, and BCPAP), osteosarcoma (U2OS), embryonic kidney (HEK293), colon cancer (HCT116, SW480), and breast cancer (BT-20, MDA-MB-231) cells (Fig. 4B). Notably, most tested cells predominantly expressed TSP1W, whereas HEK293 cells expressed higher TSP1V levels. These results suggest that AS of TSP1 is not only observed in tissues but also in several human cell lines with different ratios of TSP1W and TSP1V expression. Previous research has shown that anticancer compounds affect AS [25], and aberrant AS contributes to tumorigenesis [26,27]. Luciferase activity was measured using the Dual-Luciferase ® system. The Y-axis represents relative luciferase units. Values represent mean ± standard error of the mean. *p < 0.05, and **p < 0.01; n = 3. G-K Stable cell lines of EV, TSP1W, or TSP1V overexpressing BCPAP cells were used in in vitro and in vivo assays. G Cells were treated with TGF-β 5 ng/mL for 0, 0.5, 1, and 2 h, followed by western blotting. H-K Cells were xenografted into mice thighs, and tumors were observed twice a week for 5 weeks. H Tumor weight and IHC analysis results. I Body weight. J Tumor weight and pictures of tumors. K Western blot of protein extracted from tumors. *p < 0.05, and **p < 0.01; n = 5.
TSP1 gene, TPC-1 cells were treated with various phytochemicals and nonsteroidal anti-inflammatory drugs (NSAIDs). All chemicals tested, except meloxicam, increased TSP1V expression compared to dimethyl sulfoxide (DMSO) treatment. A similar pattern was observed in BCPAP cells, albeit with a milder effect (Fig. 4C). We further confirmed that trans-chalcone treatment increased the TSP1V to TSP1W ratio in a dose-dependent manner (Fig. 4D). In terms of TSP1 protein expression, trans-chalcone was the most responsive compound to produce an increased ratio of TSP1V compared to TSP1W, followed by sulindac sulfide (Fig. 4E).
To investigate the mechanism underlying the phytochemicalmediated increase in the expression of the splice variant form, the human TSP1 genomic fragment containing TSP1 exon 12 to exon 13, including intron 12, was subcloned into the pcDNA 3.1 CT-GFP vector (Fig. 4F). The minigene-transfected cells expressed green fluorescent protein (GFP), depending on the splicing event. GFP expression was induced upon removal of the intron by normal splicing, whereas retention of the intron blocked GFP expression owing to the generation of a premature termination codon. First, the minigene was transfected into U2OS cells to confirm GFP expression. A green-fluorescent signal was detected in the transfected cells, indicating a proper splicing event (Fig. 4F). Next, RNA was extracted from minigene-transfected U2OS cells and subjected to RT-PCR using exon 12 forward and GFP reverse primers; the expected bands were observed (Fig. 4G). This observation is consistent with previous data (Fig. 4B), which showed that TSP1W was the predominant form in U2OS cells. Besides, BCPAP cells transfected with the minigene were treated with trans-chalcone to evaluate GFP expression. After performed quantitative analysis using high-content screening and we found that trans-chalcone significantly decreased GFP expression (Fig. 4H). Thus, trans-chalcone indeed increased TSP1V expression, as assessed using the minigene system.

Trans-chalcone induced IR by demethylation of DNA
To elucidate the mechanism of IR induced by anticancer compounds, we transfected BCPAP cells with splicing factors such as RNA-binding motif protein 5 (RBM5) and arginine-rich splicing factor 6. We found that RBM5 inhibited the increase of IR induced by sulindac sulfide (Fig. 5A). To confirm the role of RBM5 in TSP1 splicing, we transfected siRBM5 to knock down RBM5. Transfection of siRBM5 resulted in an increase in TSP1V expression, indirectly calculated by the ratio of TSP1V to TSP1W in both DMSO-and sulindac-sulfide-treated BCPAP cells (Fig. 5B). Since phosphorylation of the RNA-binding motif protein is critical for splicing regulation [28], we examined the change in the phosphorylation status after sulindac sulfide treatment. We found that sulindac sulfide inactivated RBM5 by dephosphorylation, in a dosedependent manner (Fig. 5C). Furthermore, we measured the methylation status of exon 12, intron 12, and exon 13 using combined bisulfite restriction analysis, since DNA methylation is known to directly regulate IR by binding to methyl-CpG-binding protein 2 (MeCP2) [29]. The methylation level in the trans-chalcone treatment group significantly decreased (Fig. 5D). Subsequently, ChIP-PCR was performed with primers designed from intron 12 to confirm the reduction in DNA methylation binding with MeCP2. As a result, intron 12 binding to MeCP2 decreased in T-C treated group compared with its control (Fig. 5E). Additionally, we treated TPC-1 cells with trans-chalcone to identify the association between demethylation and IR. Since DNA (cytosine-5)-methyltransferase 3a (DNMT3a) and DNA (cytosine-5)-methyltransferase 3b (DNMT3b) are involved in DNA demethylation [30], we observed three types of DNMT and MeCP2 expression. We found that after treating the cells with T-C, MeCP2 and TSP1 expression decreased, whereas DNMT3b expression increased in a timedependent manner (Fig. 5F).

TSP1 may serve as a biomarker of follicular thyroid cancer
The critical obstacle in diagnosing follicular neoplasms is the inability to differentiate between nonaggressive follicular adenoma and aggressive FVPTC or FTC. We performed RT-PCR to determine whether alternatively spliced TSP1 transcripts provide cluess to differentiate between benign thyroid nodule (BTN) and differentiated thyroid carcinoma (DTC; Fig. 6A). We calculated the TSP1V to TSP1W ratio and found a significant difference between normal and tumor tissues in the DTC group (12 pairs), which consisted of FTC and FVPTC, whereas no difference was detected in the BTN group (6 pairs; Fig. 6B). RT-qPCR was performed on 32 pairs of tissue samples (BTN: 14, DTC: 18; Table 2). Intron 12 (I12) represented TSP1V expression, and exon 3 (E3) represented sum of TSP1W and TSP1V expression. The ratio of I12/E3 was significantly different between BTN and DTC in tumor tissues, the normal tissues were used as internal control (Fig. 6C).

DISCUSSION
Successful cancer treatment relies profoundly on early detection of promising biomarkers [31]. However, only a few biomarkers are commonly used in clinics [32]. In the case of thyroid cancer, discovery of a successful biomarker would enable us to shift the focus of surgery only on PTC and follicular carcinoma, where worse prognosis is expected. Therefore, these biomarkers can help us in the active surveillance of the remaining patients and their follow-up. The presence of a large number of exons in the TSP1 gene increases the possibility of AS, and this mechanism may control the regulation of expression and function of this gene. Multiple TSP1 forms of different sizes are routinely observed [33]. The origin of these forms has not been thoroughly investigated, and hence, the possibility of AS has not been excluded. Additionally, the potential significance of AS in the regulation of TSP expression has not yet been addressed. Only a few examples have been reported, wherein AS of TSP2 has been shown to skip exon 11, resulting in a decrease in TSP2-metalloproteinase interactions [34,35]. AS of TSP4 is also associated with changes in cancer cell motility [36]. In this study, we identified the role of TSP1V generated by IR and identified that TSP1V expression leads to reduced proliferation, migration, invasion, and colony formation in thyroid cancer cells in vitro. In addition, TSP1V expression resulted in the suppression of tumor growth in vivo. TSP1W and BRAFV600E mutations promote thyroid cancer progression; [23] however, our results indicate that TSP1V may reverse the tumorigenic activity of TSP1W. We show that the antitumorigenic activity of TSP1V occurs through the Smad and FAK pathways. This is the first report suggesting the potential of TSP1V in thyroid cancer therapy, although the detailed mechanisms need to be revealed. Many pharmacological approaches have been proposed over conventional drug therapy; oligonucleotide-mediated and RNAbased therapies may serve as valuable clinical approaches for ASdependent human diseases [13]. In this study, we screened dietary compounds, phytochemicals, and NSAIDs to examine their effects on the AS of TSP1. The TSP1V levels were increased by transchalcone, a phytochemical known to exhibit anticancer activity against several cancers [37,38]. The minigene constructs confirmed the AS event in the presence of trans-chalcone, which indicates that TSP1V expression increased in the presence of phytochemicals (Fig. 4).
Trans-chalcones affect AS through alterations in the spliceosome complex. Some reports have shown that SR proteins undergo phosphorylation to affect RNA-protein interactions and AS [39,40]. Several kinases, including the SR protein kinase and Cdc2-like kinase, are known to phosphorylate SR proteins [41][42][43][44][45]. It is not known whether trans-chalcone alters the activities of these kinases. A number of proteins are involved in the spliceosome complex recognition of the exon-intron boundary. Dozens of serine-arginine factors are known to enhance or repress the AS process [46]. However, the effects of conventional drugs on the spliceosome complex and splicing processes have not yet been studied in detail.
RBM5 was initially cloned from a tumor suppressor gene mapping area on chromosome 3p21.3 [47]. RBM5 is crucial for the activity of the tumor suppressor protein p53 [48]. Overexpression of RBM5 enhances the p53-mediated inhibition of cell growth and colony formation [49]. However, RBM5 level increases in breast and ovarian cancers and is correlated with the HER-2/neu protooncogene [50]. In this study, we demonstrated that RBM5 may enhance IR in the presence of sulindac sulfide (Fig. 5). Sulindac sulfide inactivated RBM5 at the post-translational level, resulting in the induction of TSP1V. It is not clear whether RBM5 directly controls the IR of TSP1. However, our data demonstrate that RBM5 partially affects sulindac-sulfide-mediated IR in the TSP1 splicing process. The identification of cis-acting elements (e.g., exonic and intronic splicing enhancer elements) and trans-acting elements within the pre-mRNA of TSP1 may provide insights into the AS-tumorigenesis relationship. To screen for a potential RBM5binding site on the TSP1 exon 12-13, which is retained by the IR, we analyzed the sequence using the SF map (http:// sfmap.technion.ac.il/index.html) and found no conserved RBM5binding sequence. Thus, the exact biological activity of RBM5 in IR and AS mechanism of individual genes remains unknown.
The role of DNA methylation in AS regulation has been demonstrated, including IR [29]. In this study, we found a strong correlation between MeCP2 and TSP1 downregulation in the presence of trans-chalcones. Accordingly, MeCP2 downregulation by trans-chalcone may enhance IR of TSP1. Decreased MeCP2 binding near splice junctions facilitates IR via reduced recruitment of several splicing factors [29]. Additionally, trans-chalcone treatment increased DNMT3b expression, with reduced levels of  N  T  N  T  N  T  N  T  N  T  N  T   A1862  A2627  A8849  A0465  A3645  A9425   M   NC   N  T  N  T  N  T  N  T  N  T  N  T   V0691  V0937  V0367  V5353  V9693  MeCP2 and TSP1. Since DNMT3b has both DNA methylation and demethylation activities [22] and trans-chalcone decreases MeCP2 expression, these mechanisms might induce DNA demethylation and cause inefficient splicing factor recruitment, thus enhancing the IR in the TSP1 gene (Fig. 6D). We believe that some phytochemicals may increase TSP1V transcript levels and ultimately inhibit TSP1 oncogenic activity in cancer cells. There are two approaches to increase TSP1-mediated inhibition of tumor progression: (1) TSP1-derived peptides, recombinant fragments, and mimics; and (2) antibody blockade and gene therapies. These approaches could also be applied to upregulate TSP1 variant forms for TSP1-mediated inhibition therapy in cells.
Our findings indicate that the use of phytochemicals and NSAIDs can increase the levels of TSP1V at the post-transcriptional level. We also observed that TSP1V was more highly expressed in normal tissues than in adjacent tumor tissues and that these compounds can promote the production of TSP1V. Thus, AS may occur during tumorigenesis, and some anticancer compounds can modulate this phenomenon. Overall, our results demonstrate a new molecular target and a novel molecular mechanism for the post-transcriptional regulation of TSP1 in tumorigenesis. TSP1V splice variants could be used as potential diagnostic and prognostic markers for human thyroid cancer progression because they are highly expressed in normal tissues. The successful determination of direct RNA-binding proteins bound to TSP1 mRNA and elucidation of the function of the wild-type and variant TSP1 in an animal model would allow us to maximize AS in cancer tissues. We believe that our results would have significant preclinical and clinical translational impacts. mRNA processing, which includes mRNA stability and degradation, contributes to tumorigenesis. However, whether splicing itself affects tumorigenesis is questionable. This study is the first to demonstrate that an AS mechanism affects thyroid tumorigenesis, using TSP1 as a model gene. Our findings have significant implications on the future development of cancer drugs that could alter splicing at the post-transcriptional level.
In conclusion, the results highlight the key role of AS in tumorigenesis and reveal a novel TSP1 transcript produced during splicing (Fig. 5). We have demonstrated that TSP1V can be used to differentiate between BTN and DTC (Fig. 6). We believe that the development of a method for distinguishing BTN from DTC will be useful for researchers and clinicians alike. Further, our method may have great social and economic impact if it leads to the discovery of feasible biomarkers.

MATERIALS AND METHODS Antibody array
Whole proteins were extracted from tissues by sonication using radioimmunoprecipitation assay (RIPA) buffer (GenDEPOT, Katy, TX, USA) supplemented with proteinase and phosphatase inhibitors. The antibody array was performed using a RayBio® L-Series Human Antibody Array 507 Membrane Kit (RayBiotech, Peachtree Corners, GA, USA), following the manufacturer's protocol.

Tissue sample
The human thyroid tissues used in this study were provided for research purposes by the National Cancer Center (Gyeonggi-do, Korea) in the form of a pair of tumor tissues and adjacent normal tissues (Tables 1, 2). These tissues were obtained as described previously [6]. This study was approved by the IRB of the National Cancer Center (NCC2014-0003).
Cell proliferation assay BCPAP cells overexpressing TSP1W or TSP1V were seeded in 96-well plates (5000 cells/well) and incubated for 24 h with 100 μL of RPMI-1640 medium. A cell proliferation assay was performed in accordance with the manufacturer's instructions using the CellTiter 96 ® AQueous One Solution

Transwell cell migration and invasion assay
BCPAP cells that overexpressed TSP1W or TSP1V were used for the migration and invasion assay. Cells were seeded in 24-well transwell plates with 8 μm pore size (Corning, NY, USA). For the migration assay, 1 × 10 5 cells in 200 μL of SFM were seeded in the upper chamber, and 600 μL of RPMI-1640 medium was added to the lower chamber. After 24 h, the nonmigrated cells in the upper chamber were detached gently; both the upper and lower sides of the membrane were fixed using 4% paraformaldehyde and stained with 0.2% crystal violet solution (V5265; Sigma-Aldrich). Transwell fields were captured at 100× magnification, and migrated cells were counted using ImageJ software. The results are shown as the percentage of the migrated area compared to the total area. Each experiment was performed in triplicate. For the invasion assay, 50 μL of Matrigel (354234; Corning, NY, USA) diluted in SFM was coated on to a transwell membrane. Cells (1 × 10 5 ) in 200 μL of SFM were seeded in the upper chamber, and 600 μL of RPMI-1640 medium was added to the lower chamber. Further steps were the same as those used in the migration assay.

Colony formation assay
A total of 100 cells from the control, TSP1W-, and TSP1V-overexpressing cells were seeded into 6-well plates and incubated for 14 days until colonies were formed. Cells were fixed and stained with 0.1% methylene blue (A18174; Alfa Aesar, UK) in ethanol. The number of colonies was counted using the ImageJ software (National Institutes of Health, Bethesda, MD, USA). Each experiment was performed in triplicate.
Cell aggregation assay BCPAP cells overexpressing TSP1W or TSP1V were seeded in an ultra-low attachment (ULA) 96-well round-bottom plate (Corning, Kennebunk, ME, USA) (10,000 cells/well in 200 μL of RPMI-1640 medium). To examine the morphology of aggregated cells, spheroid images were captured in each well every 24 h at 40× magnification. Each experiment was performed five times.

Animal studies
We used 10 (5 per group) five-week-old athymic male nude mice for the tumor model using EV-or TSP1V-transfected BCPAP cells. Sample calculator was used to decide the sample size (https://clincalc.com/stats/ samplesize.aspx), which ensures adequate power to detect a pre-specified effect size (Mean, group 1: 1,000; Mean, group 2: 650; Alpha: 0.05; Beta: 0.2; Power: 0.8). The animals were housed as described previously [51]. A total of 3 × 10 6 cells were subcutaneously injected into the thigh. The tumor size was measured twice per week. After 5 weeks, the samples were harvested. Randomization was not used to determine for sample allocation. All procedures used in the animal experiments complied with the standards set out in the Guidelines for the Care and Use of Laboratory Animals of Seoul National University and were approved by the Institutional Animal Care and Use Committee of Seoul National University (SNU-210511-2). All animals were kept and used in accordance with the Animal Research: Reporting of In Vivo Experiments that is, ARRIVE guidelines.

Statistical analysis
Unpaired Student's t-tests and one-way analysis of variance were used for all analyses. Data processing and statistical analyses were performed using Microsoft Office Excel and GraphPad Prism, version 9.5.1. All experimental data are presented as the mean ± standard error of the mean for parametric data. We excluded outliers calculated using Grubbs' test. pvalues < 0.05, < 0.01, and < 0.001 were considered statistically significant. Details of other experimental methods are provided in the Supplementary Methods (available online).