Cancer cell‐derived novel periostin isoform promotes invasion in head and neck squamous cell carcinoma

Abstract It recently has been reported that partial‐epithelial–mesenchymal transition (p‐EMT) program is associated with metastasis in head and neck squamous cell carcinoma (HNSCC). We previously have identified POSTN (which encodes periostin) as an invasion‐promoting molecule in HNSCC. Interestingly, POSTN expression is frequently observed in cancer cells with higher p‐EMT score by using a previous single‐cell transcriptomic data of HNSCC cases. Although it is known that POSTN has 11 splicing variants, the role of them has not been determined in HNSCC. Here, we found that HNSCC cells with EMT features expressed POSTN isoforms, Iso3 (lacking exon 17 and 21) and Iso5 (lacking exon 17), whereas fibroblast expressed Iso3 and Iso4 (lacking exon 17, 18, and 21). The expression of POSTN Iso3 and Iso4 are known to be widely observed in various cell types including stromal cells. Therefore, we focused on the role of novel cancer cell‐derived POSTN isoform, Iso5, in HNSCC. Single overexpression of POSTN Iso5 as well as Iso3 promoted invasion. Surprisingly, Iso5 synergistically promoted invasion together with Iso3. Notably, Iso5 as well as Iso3 upregulated p‐EMT‐related genes. We suggest that a novel cancer‐specific POSTN isoform lacking exon 17 (Iso5) can be a useful marker for detecting cancer cells undergoing p‐EMT. Moreover, a POSTN Iso5 can be a novel target for diagnosis and therapy in HNSCC.


| 8511
accumulated. 3 We previously identified POSTN (which encodes periostin) by comparing the gene expression profiles of parent cells and their highly invasive clone. 4 Periostin encoded by the POSTN gene (GenBank: D13664) is a secreted and a vitamin K-dependent glutamatecontaining matricellular protein. 5,6 To date, POSTN expression has been shown to be induced by growth factors (Transforming growth factor-β1 (TGF-β1), TGF-β2, TGF-β3, BMP-2, BMP-4, Vascular Endothelial Growth Factor (VEGF), connective tissue growth factor 2 (CTGF2)), interleukins (IL-3, −4, −6 and − 13), vitamin K, and valsartan in a cell-specific context. 7 In various cancers, the elevated POSTN expression is observed and is involved in malignant behaviors. 8 We previously found that ectopic periostin promotes invasion, anchorage-independent growth, angiogenesis, lymphangiogenesis, and metastasis. 4,9 Bao et al. demonstrated that periostin overexpression displayed accelerated metastatic growth of colon cancer cells and promoted cancer cell survival via activation of Akt/PKB pathway. 10 Clinical studies in oral, breast, and colon cancers reveal that POSTN is well correlated with angiogenesis and metastasis. [9][10][11][12] In addition, periostin drives the epithelial-mesenchymal transition (EMT) via induction of matrix metalloproteinase 9 (MMP-9), MMP-10, and MMP-13 expression, resulting in local tumor spread via extracellular matrix degradation. [13][14][15] Furthermore, periostin plays a role in remodeling of the tumor microenvironment. 4,[16][17][18][19] On the other hand, metastatic colonization can be achieved by stromal periostin via the interaction between cancer stem cells and their metastatic niche. [20][21][22] EMT is defined by a loss of epithelial cell junctions and polarity and an acquisition of mesenchymal properties. 23 In early stage of metastatic process, cancer cells lose cell-cell adhesion and increase the mobility via EMT. 24 During EMT induction during cancer progression, solid cancer cells exhibit the hybrid epithelial and mesenchymal characteristics, and this process is known as a partial EMT (p-EMT). [25][26][27][28] It is thought that the p-EMT program in cancer cells is involved in enhanced invasive properties, generation of circulating tumor cells, and resistance to anticancer drugs. 29,30 It is believed that p-EMT rather than complete EMT possesses a higher metastatic risk. Single-cell transcriptional profiles from HNSCC patients reveal that a p-EMT program exhibits the expression of extracellular matrix (ECM) proteins and lacks the expression of classical EMT transcription factors. 31 So far, 11 different splice variants of POSTN have been described. 7,32,33 However, functional implications of POSTN splicing variants are not fully understood yet. Potentially, in various tissues and disease states, distinct splicing variants may have different functions. In this study, therefore, we investigated the expression patterns and the role of POSTN splicing variants of HNSCC cell lines. A novel cancer cell-derived POSTN isoform (Iso5) was specifically identified in HNSCC cells and synergistically enhanced invasion with a common isoform (Iso3).

| scRNA-seq data analysis
The processed single-cell RNA-seq data obtained from the GEO database (accession number: GSE103322) was used. 30 The log-transformed expression data from the cancer cells or noncancer cells were re-analyzed using the Seurat (v4.1.1) R package. 31 Clustering analysis was initially performed with default settings using all cells, including cancer or noncancer cells. We used top 15 principal components (PCs) as an input to Uniform Manifold Approximation and Projection (UMAP) analysis. The resolution parameter was set to 0.4 for the FindClusters function. Then, the eight cancer cells in the noncancer cell cluster were removed and the clustering analysis was performed again with cancer cells (PCs:13, resolution:0.6). The second-level clustering for MEEI5/16 cluster was performed with the different parameter (resolution = 0.4) based on the re-calculated top 10 PCs. Differential gene expression analysis between POSTN expressing and nonexpressing cells in MEEI16 tumor was performed with FindMarkers function and top20 differential-expressed genes were visualized.
The p-EMT scores for reported p-EMT signatures 31 were calculated by single sample gene enrichment analysis using the ConsensusTME R package.

| TCGA data analysis
RNA-seq data of 564 (520 tumors, 44 normal) head and neck cancers (TCGA-HNSC) were obtained using RTCGA R package. The fibroblast scores in each TCGA samples were calculated by ssGSEA using the ConsensusTME R package. The three-year overall survival of HNSC patients with higher and lower expression in POSTN were analyzed using survival and survminer R package.

| Western blot analysis
Western blotting was performed as described previously. 4 GFP, periostin Iso3-and Iso5-overexpressing HOC621 cells were treated with 3 μM of monensin (#M5273, Sigma, Burlington, USA) for 24 hours. The intracellular protein expression of periostin is low because of the secretion from intracellular into extracellular. To quantify intracellular protein expression of periostin, we utilized monensin known as a protein transport inhibitor. Monensin blocks intracellular protein transport processes of any secreted proteins and induces the accumulation of these proteins in the Golgi complex. The increased accumulation of secreted proteins enhances the detectability of periostin with western blot analysis utilizing total cell lysate. Monensin treatment is well-characterized method for detecting cytokines in intracellular staining with flowcytometry and immunofluorescence. 35,36 Then, cells were lysed using lysis buffer (50 mM pH 7.6 Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1.5 mM MgCl 2 , 0.5% Nonidet P-40, and 10% glycerol) with protease inhibitor cocktail (#04080-24, Nacalai tesque). After centrifugation at 17,000 g for 20 min, the supernatant was collected. Protein concentration was measured using Bio-Rad protein assay dye reagent (#5000006, Bio-Rad laboratories Inc) by the absorption at 595 nm using a microplate reader (SpectraMax i3, Molecular Devices). Using 5%-20% gradient polyacrylamide gel (#E-T520L, ATTO Corporation), 25 μg of protein were electrophoresed followed by blotting onto a nitrocellulose membrane (#10600093, Cytiva). By using an Immobilon HRP substrate (EMD Millipore Corporation), the signal was detected by a Fusion SOLO 7 S. EDGE (VILBER).

| RNA Interference
Cells were seeded at a density of 1 × 10 5 cells on a 6 cm dish and then transfected with oligonucleotides by using Lipofectamine RNAiMax (#100014472, Invitrogen). After 48 hours of transfection, total RNAs and lysates were prepared and analyzed by qPCR and immunoblotting, respectively. HOC719-NE cells expressed periostin Iso3 and 5. To specifically knockdown POSTN Iso5, siRNAs were designed for targeting POSTN exon 21. For silencing all POSTN splicing variants, a 21-bp duplex oligoribonucleotides corresponding to nucleotides 2397-2417 of the human POSTN mRNA sequence was used (Thermo Fisher Scientific). For silencing POSTN Iso 5, a siRNA targeting 5′-GAAGA UGA AGA AAU UAA AAGA-3′ (siRNA Ex21) was used (GeneDesign Inc.). The negative control siRNA was obtained from Horizon Discovery Ltd.

| In vitro invasion assay
Invasion ability was determined by an in vitro invasion assay using an 8 μm pore cell culture insert (#3097, Becton Dickinson) in a 24-well plate. Twenty micrograms of Matrigel (#354234, Becton Dickinson) was coated on the filter as a reconstituted basement membrane substance. Five hundred microliters of complete media or serum-free medium (for TGF-β1 treatment) were added to the lower compartment. 1.0-1.5 × 10 5 cells were plated on the upper compartment of the cell culture insert with 100 μl of complete media or serum-free medium. After incubation for 24 h at 37°C, the cells were fixed with formalin and stained with hematoxylin (#131-09665, FUJIFILM Wako Pure Chemical Corp). By wiping with a cotton swab, the cells on the upper surface of the filter were removed, and the number of invaded cells (cells on the lower surface of the filter) was counted under a light microscope at 100x magnification. We assayed three times, and counted randomly selected three fields on the filter.

| Statistical analysis
For qRT-PCR analysis, the t-test was used for comparing two groups. For in vitro invasion assay, data were processed in GraphPad Prism 9 (GraphPad) for statistical analyses using one-way analysis of variance. For comparing p-EMT and fibroblast scores and gene expression values between two groups in RNA-seq data, Wilcoxon signed-rank test was used. A p-value of <0.05 was considered as a significant difference.

| Expression of POSTN in HNSCC tissues
In previous studies, POSTN expression was observed in both cancer cells and stromal cells. 37 To know POSTN expression in various cell types within the HNSCC tissue, we used the processed RNA-seq data obtained from the GEO database with single-cell transcriptomes from HNSCC patients. 31 The log-transformed expression data from the cancer cells or noncancer cells was re-analyzed using the Seurat (v4.1.1) R package. 38 Single-cell profiles of malignant and nonmalignant cells highlighted the composition of the tumor microenvironment. By the expression pattern of known marker genes of stromal cells including fibroblasts, endothelial cells, muscle satellite cells (MuSC), T cells, B/plasma cells, macrophages, dendritic cells, and mast cells, the nonmalignant cells were partitioned to annotated clusters ( Figure S1A). POSTN expression was mainly observed in endothelial cells and cancer-associated fibroblasts (CAFs) ( Figure S1B). By analyzing using bulk RNA-seq data (TCGA-HNSC), POSTN expression in tumor tissues was lower than that in normal tissues ( Figure S1C). This may reflect the lower fibroblast abundance in cancer tissues ( Figure S1D). Three-year survival between patients with higher and lower POSTN expression shows no difference, which may be reflected by POSTN expression in noncancer cells.
Then, the malignant cells were clustered according to their origin ( Figure 1A). The cluster composed of two classical subtype tumors (MEEI5 and MEEI16) preferentially expressed POSTN ( Figure 1A, B). As a result of re-clustering, POSTN was preferentially expressed in MEEI16. Notably, previous report has revealed that MEEI16 showed a highest p-EMT score among the HNSCC cases, 28 suggesting that POSTN may be involved in p-EMT program.

| Expression of POSTN isoforms in HNSCC cell lines
In HNSCC cell lines and fibroblasts, the POSTN expression was examined. Our previous reports showed that HOC313, HOC719-NE, and MSCC-Inv1 cells have an EMT phenotype. 39 POSTN expression was detected only in HNSCC cells with an EMT phenotype (HOC313, HOC719-NE, and MSCC-Inv1) and fibroblasts (MRC5) (Figure 2A). As HOC719-NE, MSCC-inv1 and MRC5 cells showed higher level of POSTN expression, we examined the isoforms of POSTN in these cells.
In all cells, primer set F1/R1 recognized one band: Iso3/Iso5/Ios9/Iso10 in HOC719-NE and MSCC-Inv1 cells and two bands: Iso3/Iso5/Ios9/Iso10 and Iso2/Iso4/Iso11 in MRC5 cells ( Figure 3A). Then, primer set F1/R2 recognized Iso3 in MSCC-Inv1 cells, two bands: Iso5/Iso10 and Iso3 in HOC719-NE cells, and two bands: Iso3 and Iso4 in MRC5 cells ( Figure 3B). Iso3 and Iso9/Iso11 could be distinguished by loading PCR products in 3.0% agarose gel. Finally, we performed sequencing the bands for confirming the isoforms. Collectively, MSCC-Inv1, HOC719-NE, and MRC5 cells expressed Iso3, Iso4, or Iso5 ( Figure 3C). All cells showed strongly positive for Iso3. Interestingly, HOC719-NE cells showed strongly positive for Iso5, and MRC5 cells showed strongly positive for Iso4. These findings indicated that Iso3 may be a common in HNSCC cells and fibroblasts, but Iso5 may be a specific isoform in some HNSCC cells. Indeed, previous study showed that Iso3 are common in various types of the cells and Iso4 are frequently observed in periodontal ligament cells 41 (Table 1). However, only one report showed that Iso5 is detected in thyroid tissue and periodontal ligament. 40

TGF-β1
Previous report showed that several growth factors including TGF-β1, BMP, and VEGF induced POSTN expression in a cell-specific context. 7 In HNSCC tissue, TGF-β1 was produced by several types of the cells ( Figure S2A) and its receptor, TGFBR2 as well as POSTN was preferentially expressed in tumors including where p-EMT program was detected (MEEI5, 16, 17, and 28) ( Figure S2B). Then, we treated growth factors including TGF-β1, FGF2, PDGF, HGF, and EGF in HOC621 to examine the induction of POSTN. Among growth factors, TGF-β1 markedly induced POSTN expression ( Figure 4A, B). We examined POSTN induction by TGF-β1 in HNSCC cells with low POSTN expression or without POSTN expression. POSTN was induced by TGF-β1 treatment in HOC621 and HOC313 cells ( Figure 4C). In HOC621 cells, both POSTN Iso3 and Iso5 were induced by TGF-β1 treatment ( Figure 4D). Indeed, the invasion of HOC621 cells was significantly promoted by TGF-β1 treatment ( Figure 4E).

| Synergistic effect of POSTN Iso5 for promoting invasion by Iso3
Screening for POSTN isoforms by combination of several primer sets revealed that HNSCC cells expressed POSTN Iso3 and Iso5 as shown in Figure 3C. Iso3 is known as a common isoform in various types of cancer. Indeed, we previously showed that Iso3 promoted invasion and metastasis in HNSCC. 4,9 However, the role of Iso5 remains unclear. Therefore, here we focused on the role of Iso5 in HNSCC. We generated Iso5-overexpressing cells by using HOC621 cells without POSTN expression ( Figure 5A, B). We also generated empty vector (GFP)-transfected cells and Iso3-overexpressing cells. Ectopic expression of POSTN Iso3 and Iso5 did not change the morphology and cell proliferation, compared with GFP (data not shown). We also detected periostin isoforms after treatment with monensin, which blocks the transport from the medial to the trans cisternae of the Golgi stack ( Figure 5C). Antiperiostin antibody recognized both Iso3 and Iso5, antiperiostin Ex21 antibody recognized Iso5 ( Figure 5C).
By using Iso5-overexpressing cells, we compared the invasion ability with control (GFP-transfected cells) and Iso-3 overexpressing cells. Iso5 as well as Iso3 promoted invasion at the similar level, in comparison with GFPtransfected cells ( Figure 5D). Moreover, we examined the synergistic effect of Iso3 and Iso5 (Iso3/5) in the invasion. Interestingly, Iso3/5 overexpression remarkably promoted invasion comparing with single overexpression, even though the total expression of POSTN was comparable level ( Figure 5A-D). Thus, POSTN Iso5 promoted the invasion and co-expression with Iso3 showed synergistic effect for invasion.

Iso5 depletion
To confirm the role of POSTN Iso5 in the invasion, we designed siRNA (siEx21) targeting exon 21 for the specific knockdown of Iso5 ( Figure 6A). In addition, we designed siRNA (siPOSTN) targeting all isoforms of POSTN ( Figure 6A). We confirmed the knockdown efficiency after transfection of these siRNAs in HOC621 cells overexpressing with POSTN Iso5. POSTN Iso5 expression was significantly downregulated by both siPOSTN and siEx21 siRNAs ( Figure 6B). Next, we transfected these siRNAs to HOC719-NE cells with the expression of both POSTN Iso3 and 5. The total expression of POSTN including Iso3 and Iso5 was suppressed by siPOSTN ( Figure 6C, D). On the other hand, siEx21 specifically suppressed the expression of Iso5 but not Iso3. Then, we performed in vitro invasion assay after depletion of Iso5 by using siEx21 in HOC719-NE cells. Depletion of Iso5 did not affect the morphology and slightly decreased the cell growth (data not shown). In HOC719-NE cells, Iso5 depletion remarkably suppressed the invasion without affecting total POSTN levels ( Figure 6C, E), indicating that POSTN promoted the invasion and that coexpression with Iso3 and 5 showed synergistic effect for invasion.

| Upregulation of p-EMT genes by POSTN Iso3 and Iso5 overexpression
As shown in Figure 2A, POSTN expression was observed only in HNSCC cells with EMT features. Moreover, POSTN was preferentially expressed in MEEI16 with highest p-EMT score among the HNSCC cases ( Figure 1A) suggesting that POSTN may be involved in p-EMT program. A previous single-cell analysis using HNSCC tissues identifies p-EMT-related genes. 31 By using this data, we compared POSTN expression with p-EMT-related genes. Interestingly, cells expressing POSTN transcripts (ON) in MEEI16 tumor expressed multiple common and variable p-EMT genes at higher levels compared to cells not expressing POSTN (OFF) ( Figure 7A). ON cells showed higher p-EMT scores and TGFBI expression, a representative p-EMT gene and homologous to POSTN ( Figure 7B). Moreover, the correlation between POSTN expression and p-EMT score in all type of tumors was examined. POSTN expression was strongly correlated with p-EMT score in basal tumors where the p-EMT program was detected, while both POSTN expression and p-EMT score were lower in classical and atypical tumors 31 ( Figure 7C). There were no differences in TGFBR expression between MEEI16 and other tumors, and between ON and OFF cells, suggesting the existence of a p-EMT induction mechanism specific to POSTN-expressing cells ( Figure S2B, C).

| DISCUSSION
It has been widely accepted that POSTN is closely involved in cancer progression. 35 In this study, POSTN expression is preferentially observed in endothelial cells, CAFs, and HNSCC cells with the highest p-EMT score. Indeed, cells with p-EMT phenotype localize close to CAFs at the leading edge of primary tumor tissues. 31 Moreover, a TGF-βresponsive gene, TGFBI is one of the top-scoring genes in the p-EMT program. 31 Interestingly, POSTN is highly homologous to TGFBI with FAS1 domains. 41 Figure 4C). From these observations, POSTN may be concerned with the p-EMT program. It is well accepted that cancer cells with p-EMT phenotype show a higher metastatic property, compared with complete EMT phenotype. Interestingly, overexpression of POSTN Iso3 and/ or five upregulated p-EMT-related genes. As Iso5 depletion suppressed several p-EMT-related genes, Iso5 may be contributed to p-EMT program. Although POSTN itself could not induce EMT evaluated by morphology and loss of E-cadherin (data not shown), POSTN may promote tumor progression via p-EMT program under TGF-β signaling pathway.
Recently, it has been revealed that POSTN has several different splicing variants. 7,32,33 However, the expression pattern of POSTN isoforms in HNSCC and the functional difference among POSTN isoforms are not fully understood. To date, eight different isoforms of POSTN are registered in NCBI (Table 1). In addition, other three isoforms were also reported in various types of tissues. 40,[42][43][44][45] POSTN isoforms differ in their C-terminal sequences (exons [17][18][19][20][21]. On the other hand, N-terminal sequences are common, and the EMI and Fas1 domains within N-terminus are responsible for integrin binding 37 ( Figure 2B). In previous studies, POSTN isoforms were examined by PCR in the various tissues. 40,[42][43][44][45] However, several types of the cells are included in the tissues. Therefore, here we used the cell lines to know the expression pattern of POSTN isoforms. Indeed, POSTN expression was only observed in HNSCC cells with EMT features. These HNSCC cells and GAPDH was used as a control. (B) POSTN mRNA expression was determined by qRT-PCR in HOC621 cells after TGF-β1 treatment. In each group, data were presented as the mean ± SD of triplicates. *P < 0.05 (C) HNSCC cell lines (HOC621, HOC313, HOC719-NE, HSC2, HSC3, and HSC4) were treated with 10 ng/mL of TGF-β1 for 24 hours. Cells were then collected and analyzed POSTN expression by RT-PCR. GAPDH was used as a control. (D) POSTN isoforms were detected by RT-PCR using primer set F1/R1 or F2/R2 in TGF-β1 treated HOC621 cells. (E) The invasion ability of HOC621 cells with or without TGF-β1 treatment was examined by in vitro invasion assay. Representative images were shown. Graph shows the number of invaded cells as mean ± SEM (n = 3). *p < 0.05. fibroblasts expressed Iso3 and 4. Iso3 was a common isoform in HNSCC cells and fibroblasts and construct of Iso3 has been used for overexpression experiments in previous studies. 40,[42][43][44][45] Iso4 is also frequently observed in various types of the tissue. 40,[42][43][44][45] However, only so far has been reported that Iso5 is detected in fetal kidney. 43 This is the first report that Iso5 is detected in cancer cells. We think that Iso5 may be cancer cell-derived specific isoform in HNSCC cells.
The phenotypes driven by POSTN overexpression may be caused by promoted the downstream signals including migration, invasion, and survival via acting as a ligand for F I G U R E 5 POSTN Iso5 cooperatively promotes invasion with Iso3 which is conventional isoform. (A) HOC621 cells were engineered to overexpressing GFP, POSTN Iso3 and/or Iso5 by transfection with lentiviral gene transfer. The overexpression of GFP was used as a control. Ectopic expression of POSTN Iso3 and Iso5 was quantified by qRT-PCR. In each group, data are presented as the mean ± SD of triplicates. ****p < 0.001, ***p < 0.005, **p < 0.01. (B) Ectopic expression of POSTN Iso3 and Iso5 were detected by RT-PCR using primer set F2/R2. GAPDH was used as a loading control. (C) Ectopic expression of Periostin Iso3 and Iso5 with monensin treatment was examined by immunoblotting with antibodies detecting N-terminus and exon 21 of POSTN. anti-Ex21 periostin antibody can detect POSTN Iso5 but not Iso3 lacking Exon 21. β-Actin was used as a control. (D) The invasion ability of GFP-, POSTN Iso3-, Iso5, and Iso3/5-overexpressing HOC621 cells were examined by in vitro invasion assay. Representative images were shown. Graph shows the number of invaded cells as mean ± SEM (n = 3). ****p < 0.001, ***p < 0.005, **p < 0.01.
integrins. 4,10,11,46,47 In this study, we found co-expression of Iso3 and Iso5 in HNSCC cells with EMT feature and in TGF-β1 treated HNSCC cells. Interestingly, invasion ability driven by co-expression of Iso3 and Iso5 was higher than that by single overexpression. As shown in Figure 2B In each group, data are presented as the mean ± SD of triplicates. ****p < 0.001 (C) POSTN mRNA expression in negative control siRNA (siNC), total POSTN siRNA (siPOSTN), and siRNA targeting POSTN Iso5 (siEx21)transfected HOC719-NE cells were determined by qRT-PCR. In each group, data are presented as the mean ± SD of triplicates. ****p < 0.001 (D) POSTN isoforms were detected by RT-PCR using primer set F2/R2 in siNC, siPOSTN, and siEx21-transfected HOC719-NE cells. GAPDH was used as a control. (E) The invasion ability of siNC and siEx21-transfected HOC719-NE cells were determined by in vitro invasion assay. Representative images were shown. Graph shows the number of invaded cells as mean ± SEM (n = 3). ****p < 0.001. exon 21 may have synergistic effect with Iso3. POSTN activated several signaling pathways such as Akt/PKB pathway, YAP/TAZ pathway via the αvβ3 integrin-FAK/ Src axis. 10,48 POSTN expressing cells in MEEI16 tumors highly expressed the αvβ3 integrin ligands, such as FN1 and CXCL12 and p-EMT genes. 49,50 The αvβ3 integrin signaling has recently been reported to induce p-EMT, and signals from these integrin ligands, including POSTN, may contribute to the promotion of p-EMT program. 51 Hence, it is interesting issue to analyze αvβ3 integrin-FAK/Src signaling by co-treatment of Iso3 and Iso5. In the follow-up studies, we will examine the expression of POSTN Iso5 in a larger number of HNSCC cases by using a specific antibody against exon 21 generated in this study. During EMT, ESRP1 and ESRP2 (Epithelial Splicing Regulatory Proteins 1 and 2) coordinate an epithelial cell-type-specific splicing program via the alternative splicing of FGFR-encoding transcripts. 52,53 Therefore, the change in alternative splicing may be essential for p-EMT program. This is the first report of cancer cell-specific splicing variant of POSTN. However, as a limitation of this study, it is difficult to examine the expression of POSTN isoforms in HNSCC tissues. Previously, gene function analysis in TCGA bulk RNA-seq data using various bioinformatics approaches has been reported. 54,55 In the case of POSTN, such analysis using bulk RNA-seq data is difficult, because multiple cell types express multiple POSTN isoforms. To solve this problem, we should develop in situ hybridization by using a specific probe of each POSTN isoform. In this study, we identified a POSTN isoform lacking exon 17 (Iso5) as a promoting factor of tumor progression via p-EMT program in HNSCC. HNSCC cells having coexpression of POSTN Iso3 and Iso5 may possess more aggressive invasive potential. We also suggest that POSTN Iso5 can be a useful marker for detecting cancer cells undergoing EMT. Moreover, detection of POSTN Iso5 can be a novel diagnostic marker and therapeutic target in HNSCC.