The significant biological characteristics of oral squamous cell carcinoma (OSCC) are strong local invasion and early formation of cervical lymphatic metastasis, which is also a leading cause of death [1, 2]. Accumulating evidence has confirmed that partial epithelial-mesenchymal transition (pEMT) in epithelial-derived tumors is a key factor in initiating cancer invasion, enhancing cell stemness, inducing drug resistance, and promoting metastasis [3, 4]. Moreover, cancer cells that enter the cancer stem cell state by activating the pEMT program usually fail to be eradicated by conventional therapies [5]. Therefore, exploring the key factors that effectively restrain or reverse the pEMT process will develop more effective treatment strategies for OSCC. In this case, we reveal that the transcription factor BarH-like homeobox 2 (BARX2), a member of the Bar-like homeobox gene family, acts as a potent pEMT-restrictor, whose expression is lost during OSCC development.

Current studies demonstrate that the deranged of BARX2 expression in tumors and its effect on EMT of different tumors is contrasting, suggesting its tissue-specific characteristics. Although evidence indicates that low BARX2 expression in related to increased EMT-markers, such as E/N-cadherin switch, vimentin, and S-100, there has been a lack of in-depth research on the underlying mechanisms of BARX2 in the EMT process of OSCC. Here, we presented the innovative finding that restoring BARX2 in OSCC hampers tumor progression by suppressing the pEMT marker - Serpin family E member 2 (SERPINE2).

SERPINE2, also known as protease nexin-1 (PN-1), is a member of the serine protease inhibitor superfamily and has been implicated in functions in hemostasis, thrombosis, and vascular biology [6]. It has been demonstrated that SERPINE2 promotes lung metastatic nodules of mammary tumor, is associated with highly metastatic gastric cancer cells [7], and contributes to poor prognosis of medulloblastomas [8]. Although aberrant expression of SERPINE2 is strongly associated with tumor malignant progression and pEMT program, the regulatory mechanism of SERPINE2 switching during pEMT is not well defined. Our findings provide a microRNA-mediated mechanism activated by BARX2 through which SERPINE2 is strictly controlled.

In this study, we comprehensively demonstrate the relationship between BARX2 and pEMT of OSCC. Notably, we put forward that a novel BARX2-microRNAs-SERPINE2 axis is an essential reverser of the pEMT process in OSCC. Our findings provide a new perspective on rescuing BARX2 signaling as a therapeutic strategy for OSCC.


Loss of BARX2 expression during OSCC development and its tissue-specificity in a pan-cancer dataset

During the transformation from normal epithelium to OSCC, epithelial cells gradually lose original identities and acquire EMT properties, along with the inactivation of pivotal anti-cancer genes. To reveal the potential inhibitors lost during epithelium carcinogenesis progress, OSCC and oral moderate dysplasia epithelia (ODE) tissues were collected for high-throughput RNA sequencing, with the adjacent normal epithelia specimens serving as controls. Compared to adjacent normal epithelia specimens, 185 genes were coordinately down-regulated in four OSCC tissues, and 246 genes were commonly down-regulated in two ODE tissues. Totally, 14 overlapping down-regulated genes were screened in OSCC and ODE, with BARX2 being the sole transcription factor based on the JASPAR Homo sapiens transcription factor database [9, 10] (Fig. 1A, B). Histologically, BARX2 staining was strongly expressed in the nucleus of normal epithelium, while was attenuated in mild-ODE and moderate-ODE, and almost absent in OSCC (Fig. 1C).

Fig. 1: Loss of BARX2 expression during OSCC development and its tissue-specificity in a pan-cancer dataset.
figure 1

A Venn diagram analysis revealed overlapping differentially down-regulated mRNAs among 4 pairs of OSCC tissues, 2 pairs of ODE tissues, and JASPAR Homo Sapiens TF database. B Heatmaps summarized the expression of BARX2 in OSCC, OED, and their corresponding adjacent normal epithelia, respectively. C Immunohistochemical staining of BARX2 in adjacent normal epithelia, mild-ODE, moderate-OED and OSCC revealed nuclear expression in normal epithelium, reduced expression in mild and moderate ODE, and almost absence in OSCC. Magnification: 40× and 400×. D Pan-cancer analysis of the expression profile of BARX2 in TCGA database. BARX2 was significantly downregulated in OSCC, HNSC, COAD, COADREAD, KIRP, READ, BRCA, THCA, GBM, GBMLGG, LGG, KICH and LIHC, and upregulated in CESC, BLCA, ESCA, LUAD, KIPAN, KIRC, LUSC and CHOL. E Kaplan–Meier survival analysis showed that OSCC patients with high BARX2 expression were correlated with favorable 5-year survival (High Exp = 100, Low Exp = 144). F Pan-cancer analysis of correlation of BARX2 expression and its correlation with patients’ 5-year survival in TCGA database and Sangerbox3.0 revealed that low expression was associated with a worse prognosis in KIRC, GBMLGG, KIRP, and THCA, while it was correlated with a favorable prognosis in UVM, PAAD, BLCA, and LUAD. G BARX2 expression decreased with the increased lymph node metastasis stage in OSCC, KIPAN, KIRP, STES, and THCA. H Pan-cancer analysis revealed a negative correlation between BARX2 expression and tumor stemness in HNSC, TGCT, KIPAN, DLBC, CHOL, SKCM, KIRP, LIHC, KIRC, LUSC, LUAD, and P RAD, while a positive correlation was observed in BRCA and STES. Bars represent the means ±standard error. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 (two-tailed Student’s t test). COAD colon adenocarcinoma, COADREAD colon adenocarcinoma/rectum adenocarcinoma esophageal carcinoma, UVM uveal melanoma, PAAD pancreatic adenocarcinoma, KIRP kidney renal papillary cell carcinoma, READ rectum adenocarcinoma, BRAC breast invasive carcinoma, THCA thyroid carcinoma, GBM glioblastoma, GBMLGG lower grade glioma and glioblastoma multiforme, LGG brain lower grade glioma, KICH kidney chromophobe, LIHC liver hepatocellular carcinoma, CESC cervical squamous cell carcinoma and endocervical adenocarcinoma, BLCA bladder urothelial carcinoma, ESCA esophageal carcinoma, KIPAN pan-kidney cohort, KIRC kidney renal clear cell carcinoma, LUAD lung adenocarcinoma, LUSC lung squamous cell carcinoma, CHOL cholangiocarcinoma.

To investigate the expression status of BARX2 in various solid tumors, a pan-cancer analysis was conducted. It focused on BARX2 expression levels between tumor and normal samples, as well as the correlation between BARX2 and tumor prognosis, lymph nodes metastasis (N stages), and tumor stemness. In 27 solid tumors, BARX2 was significantly downregulated in 13 tumors (48.15%) including OSCC and HNSC, whereas upregulated in 8 tumors (30.77%) (Fig. 1D). According to the results obtained from Sangerbox3.0, low BARX2 expression in OSCC predicted a worse 5-year survival (Fig. 1E, P = 1.7e−3). Additionally, low BARX2 expression in KIRC, GBMLGG, KIRP and THCA was associated with an unfavorable 5-year survival, whereas low expression of BARX2 in UVM, PAAD, BLCA and LUAD was linked to a better 5-year survival (Fig. 1F). Furthermore, BARX2 expression was significantly down-regulated with the increasing stage of lymph node metastasis in OSCC, KIPAN, KIRP, STES, and THCA (Fig. 1G). Moreover, BARX2 was predicted to be negatively associated with cancer stemness in 12 tumors, including HNSC, while it exhibited positively associated with stemness in only 2 tumors (Fig. 1H).

Suppression of BARX2 promote pEMT process of OSCC in vitro

Tumor cells commonly express combinations of EMT markers and rarely complete the entire EMT program, suggesting that ‘partial EMT’ represent the norm rather than the exception. To comfirm the role of BARX2-mediated partial-EMT regulation, we employed shRNA to knock down BARX2 expression in CAL27, HSC3, and UM-SCC23 cell lines. The results showed that repression of BARX2 notably enhanced the cell invasion ability (Fig. 2A, P < 0.001). Additionally, BARX2 inhibition significantly increased tumor spheres formation in terms of both quantity and size across all three cell lines (Fig. 2B, P < 0.05). Moreover, attenuation of BARX2 expression was also found to substantially boost the proliferative capacity of these cell lines (Fig. 2C, P < 0.01). Furthermore, the changes in the expression of epithelial and mesenchymal markers were also determined. Notably, the epithelial marker E-Cadherin was partially reduced in BARX2-suppressed cells, while the mesenchymal markers N-cadherin and S-100 exhibited a noticeable increase (Fig. 2D, P < 0.05). Collectively, our findings indicate that a deficiency in BARX2 supports the advancement of pEMT in OSCC cells.

Fig. 2: BARX2 functions as a pEMT reversor of OSCC in vitro and in vivo.
figure 2

A Inhibition BARX2 expression in CAL27, HSC3 and UM-SCC23 promoted cell invasion in vitro. B Sphere formation assay showed that inhibition BARX2 in CAL27, HSC3 and UM-SCC23 increased sphere number and volume. C Clone formation assay showed that inhibition of BARX2 significantly promoted CAL27, HSC3 and UM-SCC23 proliferation. D Protein levels of E-cadherin, N-cadherin and S-100 in shCtrl and BARX2-sh cells. E Restoring BARX2 expression in CAL27 significantly inhibited in vitro cell invasion, (F) cell stemness, and (G) cell proliferation. H Protein levels of E-cadherin, N-cadherin and S-100 in CAL27-con, CAL27-BARX2-ov cells. I Compared to the control group, enforced expression of BARX2 in CAL27 decelerated the weight loss of mice. J Kaplan–Meier survival analysis showed that BALB/c nude mice injected with CAL27-BARX2-ov exhibited favorable survival. K Isolated lungs of BALB/c nude mice injected with CAL27-CON (top) or CAL27-BARX2-ov cells (bottom). Overexpression of BARX2 reduced the volume and weight of lung tumor nodules (right). Bars represent the means ±standard error. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 (two-tailed Student’s t test).

Restoration of BARX2 reverses pEMT process of OSCC in vitro and in vivo

To elucidate the reversal effect of BARX2 overexpression on pEMT, BARX2 was restored in OSCC cell lines. The results demonstrated that recovering BARX2 expression significantly inhibited the invasion of three different OSCC cell lines in vitro (Fig. 2E, P < 0.0001). Additionally, BARX2 induced a significant reduction both in the number (Fig. 2F, P < 0.05) and volume of tumor spheres (Fig. 2F, P < 0.01) in CAL27 cells, indicating reduced stemness. Furthermore, it was observed that increased BARX2 strongly suppressed cell proliferation ability of CAL27 (Fig. 2G, P < 0.01). It also reversed the E/N-Cadherin expression ratio, and decreased the expression of mesenchymal marker S-100 (Fig. 2H). In vivo, the rescue of BARX2 expression significantly inhibited tumor metastasis, as evidenced by the suppression of weight loss (Fig. 2I, P < 0.01), improvement in the survival rate (Fig. 2J, P = 0.0494), and reduction in the volume and weight of lung tumor nodules (Fig. 2K, P < 0.05) in lung metastasis models of BALB/C nude mice.

BARX2 is reversely associated with pEMT signaling and SERPINE2 expression

To explore the essential downstream signaling of BARX2, the mRNA high-throughput sequencing was performed in a BARX2-overexpressed cell line (CAL27-BARX2-ov) and a control cell line (CAL27-con). Totally, 557 genes were differently expressed in CAL27-BARX2-ov, including 371 up-regulated genes and 186 down-regulated genes (Fig. 3A). KEGG enrichment analysis showed that the differential genes were preferentially enriched in the EMT signaling pathway in the Molecular Signatures Database-H: Hallmark Gene Set (MSigDB-H) (Fig. 3B). There were 14 down-regulated genes that log2(fold change)<−1 and 7 upregulated genes log2(fold change)>1 (Supplementary Table 7). The top 10 differential genes were listed in the form of a heat map (Fig. 3B right panel). In order to verify the relationship between expression of BARX2 and genes in EMT-pathway, we supplemented GSEA analysis using RNA-Seq analysis data from the TCGA database. The result showed that genes involved in Hallmark_Epithelial_Mesenchymal_Transition pathway were significantly overrepresented in BARX2-low group (P < 0.001, Fig. 3D). These results suggest that BARX2 participates in EMT signaling by down-regulating EMT-related genes.

Fig. 3: BARX2 is reversely associated with EMT signaling and SERPINE2 expression.
figure 3

A Totally 557 differential genes were detected in CAL27-BARX2-ov, including 371 up-regulated genes and 186 down-regulated genes (|log2(fold change)| > 1.5). B KEGG Enrichment Analysis of differential genes revealed the top10 enriched signaling pathway in MSigDB-H (left panel). Pathway enriched with the most differentially expressed genes was EMT, of which top 10 differential genes were listed in the form of a heat map (right panel). C Quantitative RT-PCR was performed to verify the result of high throughput sequencing. D GSEA analysis on the enrichment of EMT pathway-related genes and BARX2 expression in TCGA database. E The protein levels of SERPINE2 in con/BARX2-ov cells and in F shCtrl/BARX2-sh cells. Bars represent the means ±standard error. ns no significant difference; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 (two-tailed Student’s t test).

To further confirm the downstream pEMT-related genes of BARX2, the top 10 down-regulated genes and top 4 up-regulated genes enriched in EMT signaling pathway were verified in SCC25 and CAL27, respectively (Fig. 3C and Supplementary Fig. S1A). The results identified that seven genes were significantly downregulated in the two OSCC cell lines, with SERPINE2 being the most markedly suppressed (Fig. 3C, P < 0.0001). Furthermore, the levels of SERPINE2 protein were confirmed to be suppressed in three BARX2-overexpressed OSCC cell lines (Fig. 3E, P < 0.001), and increased in three BARX2-sh cell lines (Fig. 3F, P < 0.001), indicating that SERPINE2 is a stable downstream target of BARX2 in OSCC.

SERPINE2 contributes to pEMT process of OSCC

To explore the effect of SERPINE2 on OSCC pEMT process, we established OSCC cell lines that stably overexpressed SERPINE2. It was found that overexpression of SERPINE2 promoted invasion in CAL27 and WHU-TSC-1 cells (Fig. 4A, P < 0.001), facilitated tumor cell stemness (sphere number: P < 0.05; sphere volume: Fig. 4B, P < 0.01), increased cell proliferation (Fig. 4C, P < 0.001), and enhanced Snail and S-100 expression in CAL27 (Fig. 4D, P < 0.01).

Fig. 4: SERPINE2 contributes to pEMT process of OSCC.
figure 4

A Restoring SERPINE2 expression in OSCC cell lines (CAL27 and WHU-TSC-1) significantly promoted cells invasion ability in vitro, as determined by transwell assay. B The sphere formation assay showed that SERPINE2 overexpression promoted CAL27 stemness in terms of sphere number and volume. C The clone formation assay showed that overexpression of SERPINE2 significantly enhanced CAL27 proliferation ability. D The protein levels of SNAIL and S-100 were detected by western blot in CAL27 cells overexpressing SERPINE2. E Immunohistochemical staining of SERPINE2 in cores of adjacent normal epithelium (NE), OSCC center (OC) and invasive front of OSCC (OIF). Magnification: 1.5× and 40×. F Hierarchical clustering analysis and (G) histoscore quantification were used to visualize SERPINE2 expression in NE, OC and OIF. H SERPINE2 were significantly higher in metastatic cases than in nonmetastatic cases, as observed both in OC and OIF. I SERPINE2 were significantly higher in advanced stages, both in OC and OIF. J Kaplan–Meier survival analysis showed that patients with higher SERPINE2 expression exhibits a significantly lower 5-year survival rate (High Exp=74, Low Exp = 37). Bars represent the means ± standard error. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 (two-tailed Student’s t test).

Subsequently, the expression profile of SERPINE2 was determined in tissue microarray that included 120 primary para-cancerous normal epithelium (NE), OSCC centers (OC) and OSCC invasive fronts (OIF). SERPINE2 was strongly expressed in the cytoplasm of tumor cells in OIF, weakly expressed in OC, whereas absent in NE (Fig. 4E–G, P < 0.0001). Moreover, the expression levels of SERPINE2 were significantly higher in metastatic cases than in non-metastatic cases, both in OC (P < 0.0001) and OIF (Fig. 4H, P < 0.01). Furthermore, the expression level of SERPINE2 in stage IVA patients was significantly higher than that in earlier stages, as detected in both OC and OIF (Fig. 4I, P < 0.01). Kaplan–Meier survival analysis showed that patients with relatively high SERPINE2 expression exhibited a significantly lower 5-year survival rate (Fig. 4J, P = 0.0287).

Rescuing SERPINE2 reverts the inhibitory effect of BARX2 on pEMT process and extracellular matrix remodeling

To clarify the role of SERPINE2 in the pEMT process negatively controlled by BARX2, we re-established SERPINE2 expression in the CAL27-BARX2-ov cell line. Restoring SERPINE2 in BARX2-overexpressed tumor cells significantly rescued the defective cell invasion ability (Fig. 5A, P < 0.0001), cell stemness (sphere number: P < 0.05; sphere volume: P < 0.001, Fig. 5B) and cell proliferation (Fig. 5C, P < 0.001). To ascertain whether SERPINE2 could reverse the inhibitory effect of BARX2 on tumor metastasis in vivo, lung metastasis models of nude mice were conducted. In comparison with the control group, nude mice injected with CAL27-BARX2-SERPINE2-ov showed accelerated weight loss (Fig. 5D, P < 0.05), an unfavorable prognosis (Fig. 5E, P = 0.0494), and an augmentation in the volume and weight of lung tumor nodules (Fig. 5F, P < 0.001).

Fig. 5: Rescuing SERPINE2 reverts the inhibitory effect of BARX2 on pEMT process and extracellular matrix remodeling.
figure 5

Restored the expression of SERPINE2 in CAL27-BARX2-ov significantly reverted the inhibitory effect of BARX2 in cell invasion (A), stemness (B) and proliferation (C) in vitro. D The weight of BALB/c nude mice injected with CAL27-BARX2-SERPINE2 overexpression or CAL27-BARX2-VEC tumor cells via tail vein. E Kaplan–Meier survival analysis showed that BALB/c nude mice injected with CAL27-BARX2-SERPINE2-ov were correlated with unfavorable survival. F Isolated lungs of BALB/c nude mice injected with CAL27-BARX2-VEC (top) or CAL27-BARX2-SERPINE2-ov (bottom). Overexpression of SERPINE2 enable CAL27-BARX2 to increase the volume and weight of lung tumor nodules (right). G Protein level of MMP1 in shCtrl/BARX2-sh cells and in H con/BARX2-ov cell lines. I Correlation of the expression of SERPINE2 and MMP1 analyzed in TIMER 2.0. J Protein level of MMP1 in CAL27/WHU-TSC-1-SERPINE2-ov cell line and (K) in CAL27-BARX2-SERPINE2-ov cell line. Bars represent the means ± standard error. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 (two-tailed Student’s t test).

Cells characterized by pEMT are often able to degrade and invade their extracellular matrix by expressing MMPs [11]. High-throughput sequencing showed that MMP1 expression was down-regulated by 5.78-6.74-fold after BARX2 overexpression, a finding further validated at protein level in CAL27 and WHU-TSC-1 (Fig. 5G). In three BARX2-knockdown cell lines, the expression of MMP1 was up-regulated obviously (Fig. 5H). These results indicating that MMP1 is an important target downstream of BARX2.

It has been demonstrated that SERPINE2 can induce the expression of MMP-2/9 and enhance the metastasis of breast cancer and melanoma [12, 13]. According to the results of TIMER2.0 based on TCGA dataset, SERPINE2 and MMP1 were significantly and positively correlated (P < 0.0001, R = 0.345, Fig. 5I). In CAL27 and WHU-TSC-1 cell lines, overexpression of SERPINE2 led to the upregulation of MMP1 (Fig. 5J). Furthermore, restoring the expression of SERPINE2 in BARX2-overexpressed cell line rescued the impaired MMP1 expression (Fig. 5K), indicating that SERPINE2 could reverse BARX2-regulated extracellular matrix reconstruction by re-establishing MMP1 expression.

BARX2 suppresses SERPINE2 via upregulation of miR-186-5p and miR-378a-3p

To analyze the possible signaling pathways through which BARX2 inhibits SERPINE2, we utilized various databases, including JASPAR, ENCORI, PicTar, miRDB, and TANRIC. Based on the database prediction and the analysis of four pairs of OSCC high-throughput sequencing results, miR-150-5p, miR-186-5p, miR-378a-3p and miR-410 were identified as potential intermediaries between BARX2 and SERPINE2. Further experiments confirmed that miR-150-5p, miR-186-5p and miR-378a-3p were significantly elevated in the presence of BARX2 overexpression, while were significantly downregulated in cell lines where BARX2 was suppressed (Fig. 6A). On the other hand, neither RNA-binding proteins nor lncRNAs were predicted to be involved in the BARX2-modulated SERPINE2 suppression, as indicated by database prediction (Supplementary Fig. S1B).

Fig. 6: BARX2 suppresses SERPINE2 through promoting the expression of miR-186-5p and miR-378a-3p.
figure 6

A Screening of the probable microRNAs that mediate the regulation of SERPINE2 by BARX2 (up panel). The expressions of miR-150-5p, miR-186-5p, miR-378a-3p and miR-410 were verified in BARX2-ov and BARX2-sh cell lines by qRT-PCR (down panel). B Diagram of SERPINE2 3’UTR fragments containing miR-150-5p, miR-186-5p and miR-378a-3p binding sites. C MiR-186-5p and miR-378a-3p mimics reduced the luciferase reporter activity of pSiCheck™-2 SERPINE2-3′UTR, respectively, and coordinately. The final normalized luciferase activity was normalized by 3′ UTR/NC control. D Diagram of wildtype and mutation bases of miR-186-5p and miR-378a-3p binding sites in SERPINE2 3’UTR fragments (up panel). Mutation of the target binding sequences led to off-target of miR-186-5p and miR-378a-3p in binding with SERPINE2 3’UTR, resulting in retained luciferase activity. E Diagram of miR-186-5p and miR-378a-3p promoter fragments containing BARX2 binding sites. Red sequences represent mutant bases. F Luciferase assays verified the transcriptional activation of miR-186-5p and miR-378a-3p by BARX2. BARX2 activated the luciferase reporter activity of pGL3-miR-186-5p and pGL3-miR-378a-3p, but failed in mutant sequences. G The protein level of SERPINE2 were detected by western blot in CAL27-BARX2-SERPINE2 under the treatment of miR-186-5p or miR-378a-3p inhibitor. Bars represent the means ± standard error. ns no significant difference; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 (two-tailed Student’s t test).

To identify the inhibition of SERPINE2 by miR-150-5p, miR-186-5p and miR-378a-3p, a 323 bp fragment of SERPINE2 3’UTR containing the binding sites for these specific miRNAs was cloned and inserted into psiCHECKTM-2 luciferase reporter plasmid (Fig. 6B). The result showed that miR-186-5p (P < 0.001) and miR-378a-3p (P < 0.05), but not miR-150-5p (P = 0.0846) inhibited the luciferase activity of SERPINE2 3’UTR (Fig. 6C). Subsequently, miR-186-5p and miR-378a-3p binding sequences on SERPINE2 3’UTR were mutated, leading to off-target of miR-186-5p and miR-378a-3p (Fig. 6D). The results suggested that miR-186-5p and miR-378a-3p degraded SERPINE2 mRNA at the post-transcriptional level.

To verify the transcriptional activation of BARX2 on miR-186-5p and miR-378a-3p, we cloned the promoter sequences of DNA strands encoding pri-miR-186-5p and pri-miR-378a-3p, each containing highly affinity of BARX2 binding sites (Fig. 6E). The results of dual-luciferase assay showed that BARX2 bound to the promoter sequences of miR-186-5p and miR-378a-3p, and stimulated their transcription. Mutation of the binding sequences abrogated the transcriptional activations of miR-186-5p and miR-378a-3p by BARX2 (Fig. 6F). Next, miR-186-5p and miR-378a-3p were silenced to elucidate whether they mediated the protein repression of SERPINE2 by BARX2. The results showed that the inhibitory effect of BARX2 on SERPINE2 was significantly rescued when the CAL27-BARX2-ov cells were transfected with miR-186-5p and miR-378a-3p inhibitors, respectively (Fig. 6G).

To investigate the expression of miR-186-5p and miR-378a-3p in OSCC, we examined their expression in 22 pairs of OSCC tissues and the adjacent normal epithelium. It was observed that miR-186-5p was suppressed in 77.3% (17/22) of OSCC samples, and miR-378a-3p was suppressed in 63.6% (14/22) of OSCC samples, compared with their adjacent normal epithelium (Fig. 7A). Additionally, the relative expression levels of miR-186-5p and miR-378a-3p were positively correlated with BARX2 in tumor tissues (Fig. 7B, C), confirming the transcriptional activation of miR-186-5p and miR-378a-3p by BARX2. Given that the suppressive role of miR-186-5p in pEMT has been established in several solid tumors, we further investigated the impact of miR-378a-3p on inhibiting pEMT in OSCC. Overexpression of miR-378a-3p in CAL27 significantly inhibited cell invasion ability (Fig. 7D, P < 0.001), cell stemness (sphere number: P < 0.05; sphere volume: P < 0.01, Fig. 7E) and cell proliferation (Fig. 7F, P < 0.01).

Fig. 7: MiR-378a-3p and miR-186-5p act as tumor suppressors in OSCC.
figure 7

A Expression of BARX2, miR-186-5p and miR-378a-3p in 22 pairs of OSCC tissues and the adjacent normal epithelium. B, C The correlation between the relative expression of BARX2 and miR-186-5p as well as miR-378a-3p in OSCC tissues. D Overexpression of miR-378a-3p in CAL27 significantly inhibited the cell invasion ability, E cell stemness, and F cell proliferation. G The schematic diagram of BARX2-microRNAs-SERPINE2 axis. In OSCC, loss of BARX2 expression unlashes SERPINE2 via inactivating miR-186-5p and miR-378a-3p, thereby promoting ECM remodeling. Restoration of BARX2 expression activates miR-186-5p and miR-378a-3p, leading to the degradation of the SERPINE2 expression and deceleration of extracellular matrix remodeling. Bars represent the means ± standard error. ns no significant difference; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001 (two-tailed Student’s t test).

In summary, we created a schematic diagram to briefly describe the BARX2-microRNAs-SERPINE2 axis (Fig. 7G).


BARX2, is a widely expressed transcription factor in a variety of tissues, known to influence various embryonic development processes, including the regulation of cell adhesion, actin cytoskeleton remodeling, and cell differentiation [14, 15]. In the field of cancer research, BARX2 has been found to play an opposite role in the development of different tumors. In ovarian, gastric, colorectal and hepatocellular cancers, the loss of BARX2 was identified as a favorable prognostic biomarker [10, 16,17,18]. However, contrasting results have been reported that lung adenocarcinoma and ovarian serous adenocarcinoma, where overexpression of BARX2 expression greatly promoted the tumorigenesis process [16, 19]. These pieces of evidences suggest that the function of BARX2 in tumors is tissue-specific. In this study, BARX2 is revealed to gradually lose expression along with the progression from oral normal epithelia to epithelial dysplasia to OSCC. Moreover, the deficiency of BARX2 is closely associated with different stages of OSCC lymph node metastasis. This indicates that the loss of BARX2 has a great impact on promoting OSCC throughout the stage of its occurrence, development and metastasis.

Recently, studies have also revealed the tissue-specific role of BARX2 expression in regulating pEMT. BARX2 expression is negatively correlated with EMT and invasion in chronic myeloid leukemia, glioblastoma multiforme and uveal melanoma, while it is positively associated with lung adenocarcinoma [19, 20]. Compatible with these findings, we further elucidate the role of BARX2 in cancer pEMT. Notably, pan-cancer analysis reveals that BARX2 has a stronger tendency to inhibit stemness than to promote it (12/37 vs 2/37), highlighting its predominantly negative regulatory effect on EMT in most tumors. It is well known that tumor EMT is rarely a complete state of epithelial-loss and mesenchymal-gain. Actually, most tumor cells harbor both epithelial and mesenchymal characteristics, named as pEMT [21]. Cells in pEMT state contribute to extensive phenotypic heterogeneity and plasticity within the tumor, such as stemness, drug resistance, invasion and metastasis. Given that cancer cells in pEMT status exhibit advanced phenotypes, there is potential value in efforts aimed at reversing pEMT, whether through regaining pEMT impeder or blocking the pEMT driving factors. In our study, we present comprehensive evidences that restoration of BARX2 in OSCC suppresses cell stemness, inverts E/N-Cadherin expression ratio, and inhibits cell invasion and metastasis. These observed phenotypes provide the first compelling evidence that BARX2 acts a potent pEMT-reverser in OSCC.

As our understanding of the mechanisms underlying of pEMT associated transcription factor (EMT-TF) activation advances, it becomes evident that deactivation of EMT-TF program plays a crucial role in impeding pEMT. Researches support the idea that molecules deficiency in tumorigenesis unleashes key EMT-TFs, leading to EMT activation. For instance, in colorectal cancer, the absence of the circadian gene Timeless launches ZEB1 expression [22]. On the other hand, emerging researches show that limiting or reversing pEMT outcomes can arise not only from the deactivation of a previously active EMT-TF program, but also from the presence of specific mechanisms operating through pathways that do not directly involve the well-characterized EMT-TFs [23]. In fact, certain factors whose expression is missing in cancers has been reported to impede the progression of pEMT by restoring its original signaling, without directly affecting EMT-TFs. For instance, in Braf/Pten-mutated mouse models of melanoma, Ambra1 deficiency in melanoma cells influences extracellular matrix remodeling and triggers hyperactivation of the focal adhesion kinase 1 (FAK1) signaling, whose inhibition reduces cell invasion and melanoma growth [24] Also, the loss of 4.1N induces EMT in adherent EOC cells, and its expression inhibits anoikis resistance and EMT by directly binding and accelerating the degradation of 14-3-3 in suspension EOC cells [25]. In our research, the restoration of BARX2 expression in OSCC effectively aborted SERPINE2 expression, suggesting a specific mechanism of pEMT that operates independently of EMT-TFs.

While BARX2 is a transcription factors, limited studies have unveiled its downstream molecular regulatory mechanisms. In non-small-cell lung cancer, BARX2 is reported to inhibit Wnt//β-catenin signaling pathway by suppressing the expression of β-catenin [26]. Chen et al. demonstrated that in esophageal squamous cell carcinoma that BARX2 activates the transcription of pleckstrin homology-like domain family A, member 3, followed by suppression of the PI3K/AKT pathway [27]. In addition, BARX2 has been identified to reduce the phosphorylation levels of MEK and ERK, thereby inactivating the Ras signaling pathway [28]. In this study, we introduce a novel signaling pathway, the BARX2-microRNAs-SERPINE2 axis, which is presented and thoroughly demonstrated in OSCC for the first time.

SERPINE2 is emerging as a factor that actively participates in pEMT [29, 30]. Latest research employing single-cell RNA-seq and multi-omics, has concluded that SERPINE2 exhibits aberrant expression in advanced renal cell carcinoma, where it activates pEMT by suppressing E-cadherin while upregulating N-cadherin, Vimentin and Snail [31]. Similar findings have been discovered in hepatoblastoma and ESCC [30]. Furthermore, SERPINE2 has been identified to promote pEMT by upregulating matrix metalloproteinase, including MMP2 and MMP9 [32]. In this work, we report the upregulation of MMP1 by SERPINE2, adding to the new evidence of SERPINE2’s influences on extracellular matrix remodeling.

In contrast to downstream regulation mechanisms of SERPINE2 have been explored, its upstream regulation remains less studied. It has been reported that SERPINE2 can be activated by EGR1 [33] and LHX2 [34], recognized and degraded by the m6A ‘reader’ [35] and targeted by miRNA-199 [36]. Recently, microRNAs are highlighted to play an important role in pEMT signaling pathways, providing post-transcriptional regulation of splicing and post-translational control of protein stability. According to previous studies, tumor suppressor miRNA, miR-34a, participates in inhibiting pEMT via targeting Snail [37]. The miR-200 family, consist of miR-200a/b/c, miR-141, and miR-429, targets the ZEB1 and ZEB2 3’UTR, while also being suppressed by ZEB1 and ZEB2 themselves, creating a dynamic regulation that driving tumor cell in and out of pEMT [38]. Exist studies have demonstrated the anti-tumor properties of miR-186-5p [39, 40] and miR-378a-3p [41, 42], both of which act as mediators between long non-coding RNAs and pEMT effector [43, 44]. In our research, miR-186-5p and miR-378a-3p, are identified to be a “key node” on the BARX2-mediated suppression of the SERPINE2 signaling pathway. This indicates that BARX2 regulates SERPINE2 at the post-transcriptional level rather than at transcription level. As a result, microRNAs play a crucial role as important intermediaries in the intricate BARX2-mediated pEMT signaling pathway.

Taken together, our current findings strongly indicate that the loss of BARX2 is implicated in the pEMT progression in OSCC. These results suggest a novel avenue for exploring signaling mechanisms in anti-pEMT therapy for OSCC.

Material and method

Clinical sample collection and tissue microarray

Four primary OSCC tissues, two dysplastic epithelium samples, and their corresponding adjacent normal epithelial tissues were collected for high-throughput mRNA sequencing. A total of 120 paraffin-embedded OSCC tissues and adjacent normal tissues were collected and prepared into tissue microarrays (Autor Biotechnology). 22 pairs of fresh OSCC tissues and the adjacent normal epithelial tissues were collected for quantifying miR-186-5p, miR-378a-3p and BARX2. Informed consent was obtained from all patients. All OSCC tissue samples were obtained through surgical resection from the School of Stomatology, Wuhan University, which had been approved by the Ethics Committee of Wuhan University.

Pan-cancer analysis and target prediction of BARX2 and miRNA

Pan-cancer analysis of BARX2 was performed using Sangerbox3.0 tool [45]. All data were downloaded from TCGA, and the gene expression values were transformed by log2(x + 0.001). Data collection and filtering were performed using JVENN [46]. This research focused exclusively on solid tumors. Supplementary Table 1 lists all the web tools used in this article.

The OSCC prognostic data were extracted from the TCGA-HNSC dataset by using R software (version 4.2.1). Cancers occurring in oral cavity, oral tongue, buccal mucosa, lip, alveolar ridge, hard palate and floor of the mouth were selected as OSCC (T = 311, N = 30). For prognostic analysis, patients with a follow-up time of less than 30 days were excluded. Patients were divided into high and low gene expression groups, with the cut-off value for gene expression set at the optimal threshold value.

The ENCORI database was used to predict candidate microRNAs that targeting SERPINE2. The UCSC gene browser home was used to obtain the upstream sequence of pri-miRNA, and the JASPAR database was applied to predict BARX2 binding sites in the pri-miRNA promoter sequence.

Cell culture

CAL27, SCC25, and HEK-293T cells used in this experiment were purchased from American Type Culture Collection (ATCC). WHU-TSC-1 was established by our research group [47]. The UM-SCC23 and HSC3 cell lines were gifts from Dr. Thomas E.Carey (University of Michigan, Ann Arbor, MI, USA) and Professor Chen Qianming (Zhejiang University, Hangzhou, China), respectively. These cells were routinely cultured in DMEM high glucose medium containing 10% serum (Hyclone), 1000 units/mL penicillin and 0.1 mg/ml streptomycin, respectively, in a constant temperature incubator at 37 °C containing 5% CO2.

RNA-seq and gene set enrichment analysis analysis

RNA-seq libraries were prepared with the RNA-seq library Preparation kit (Gnomegen). RNA-seq was performed on Illumina Hiseq 2000 and Illumina GAIIx (BGI). To analyze genes differently expressed in OSCC, ODE and BARX2 overexpression cell line compared to control groups, edgeR [48] in R packages were utilized. For each gene, significance P-value and FDR (Q-value) were obtained based on the model of negative binomial distribution. Fold changes of gene expression were also estimated within the edge R statistical package. Differential gene expression analysis were conducted using DESeq2 [49] in R packages. The criterion for DEG has been set as |log2(fold change)| >1.5 and Q-value < 0.05. Enrichment analysis of differential genes using Msigdb-h was conducted at Differential clustering analysis was performed using HemI 1.0. To verify the relationship between expression of BARX2 and genes in EMT-pathway, gene expression data of 20 patients with the highest BARX2 expression and 20 patients with the lowest BARX2 expression in TCGA database were analyzed by GSEA pathway enrichment.

Immunohistochemical staining, scoring system, hierarchical clustering, and data visualization

The immunohistochemical staining was carried out and scored as described previously [50]. The primary antibodies were used as follows: anti-human BARX2 (1:200, sc53177, Santa Cruz, CA, USA) and polyclonal mouse anti-human SERPINE2 (1:300; OTI2C9, Origene). The stained sections were scanned with an Aperio ScanScope CS scanner and analyzed with Aperio ImageScope, version 11.2.0. Histoscore of SERPINE2 in each field was calculated according to the formula: (total intensity of strong positivity × 3 + total intensity of positivity × 2 + total intensity of weak positivity × 1)/total number of cells. HemI 1.0 software was used for stratified analysis.


Total RNA and cDNA were prepared according to manufacturer’s introduction (YEASEN). Quantitative PCR was performed using Hieff UNICON® Power qPCR SYBR Green Master Mix (YEASEN). For miRNA reverse transcription, 0.5 μg of total RNA was reverse transcribed into cDNA using the miRNA 1st Strand cDNA Synthesis Kit (Vazyme). MiRNA quantitative PCR was performed using miRNA Universal SYBR qPCR Master Mix (Vazyme). Relative gene expression was calculated using equation 2−Δ (ΔCT), where ΔCT = CT (mRNA) -CT (ACTIN) or CT (miRNA) -CT (U6). Primers for RT-qPCR were listed in Supplementary Tables 2 and 3.


The Western Blot was carried out as described previously [50]. The antibodies used were listed: BARX2 (1:500, sc53177, Santa Cruz), SERPINE2 (1:2000, OTI2C9, Origene), E-cadherin(1:500, sc-8426, Santa Cruz), N-cadherin(1:500, sc-393933, Santa Cruz), S-100α/β chain (1:500, sc-58839, Santa Cruz), SNAI1(1:1000, sc-271977, Santa Cruz), MMP1(1:1000, A1191, ABclonal), and β-ACTIN(1:5000, PMK081S, biopm).

Cell invasion assay

For cell invasion assay, the upper transwell chamber of 8.0 μm pore size (Corning) was evenly spread with 40 µL Matrigel (1:4, diluted in DMEM high glucose medium; BD Biosciences). 5 × 105 cells were seeded in serum-free medium in the upper chamber, and 600 µL of DMEM medium containing 20% FBS was added to the lower chamber. After 48 h of culture, cells at the bottom of the upper chamber were fixed in 4% paraformaldehyde and stained with crystal violet. The total number of stained cells attached to the lower part of the membrane was counted under a microscope. Six fields were randomly selected to calculate the mean cell number.

Plasmids construction and transfection

For BARX2 overexpression, the CDs sequence of BARX2 (NM_003658.5) was amplified and inserted into the lentiviral vector pCDH-CMV-MCS-EF1-PURO (Systems Biosciences). For BARX2 knockdown, the lentivirus expression vectors GV-248-BARX2-sh1, 2 and GV-248-shCtrl were purchased from Genechem. For SERPINE2 overexpression, the CDs sequence of SERPINE2 (NM_001136528.2) was amplified and inserted into the lentiviral vector pCDH-CMV-MCS-EF1-NEO (Systems Biosciences). The lentivirus packaging and transfection were performed as previously described [50]. The constructed stable cell lines, together with primers of gene overexpression fractions, and shRNA sequences were listed in Supplementary Table 4. For microRNA overexpression or inhibition, the microRNA mimics/inhibitors (Genepharma) was transfected into CAL27 cells at a final concentration of 20 μM. The transfection of mimics/inhibitors was conducted following previously described procedures [50]. The sequences of miRNA mimics/inhibitors were listed in Supplementary Table 5.

OSCC tumor lung metastasis model in nude mice

Female BALB/c nude mice aged 6-8 weeks were randomly grouped into four groups: CAL27-con, CAL27-BARX2-ov, CAL27-BARX2-ov-vector and CAL27-BARX2-ov-SERPINE2-ov (5 mice each group). A total of 1×106 cells were injected into the tail vein of mice to construct tumor lung metastasis model. Endpoint of the observation was 60 days, or until a losing 25% weight loss occurred. The mice were sacrificed in CO2, and their lungs were isolated, weighed and observed. All procedures were approved by the Animal Ethics Committee of Wuhan University and performed in accordance with the Guide for the Care and Use of Laboratory Animals (approval number A22, Ministry of Science and Technology of China, 2006).

Luciferase reporter assay

The 3’UTR sequence of SERPINE2 variant 2 (NM_001136528.2) was amplified and inserted into pSiCheck™-2 dual luciferase reporter vector (Promega). Additionally, the promoter sequences of pri-has-miR-186-5p and pri-has-miR-378a-3p were amplified and inserted into the pGL3-basic luciferase reporter vector (Promega), respectively. To obtain the corresponding mutant expression vectors, the binding site sequences were mutated using Mut Express II Fast Mutagenesis Kit V2 (Vazyme).

For the SERPINE2-3’UTR luciferase reporter assays, pSiCheck™-2 vectors, pSiCheck™-2-SERPINE2-3’UTR or pSiCheck™-2-SERPINE2-3’UTR-mutant were transiently co-transfected with NC mimics or miRNA mimics into 293T cells. For the miRNA promoter luciferase reporter assays, wild type vectors, miRNA promoter vectors or mutant vectors were transiently co-transfected with either BARX2-ov vectors or control vector into 293T cells. Luciferase activity was measured 48 h post-transfection using Dual-Luciferase® Reporter Assay System (Promega). In SERPINE2-3’UTR luciferase reporter assays, renilla luciferase activity was normalized to firefly luciferase activity. The final normalized luciferase activity was further normalized by 3′ UTR/WT control. In miRNA promoter luciferase reporter assays, firefly luciferase activity was normalized to renilla luciferase activity. The final normalized luciferase activity was further normalized by BARX2-ov/NC control. All primers for luciferase reporter assays were listed in Supplementary Table 6.

Statistical analysis

All quantitative experiments were conducted in triplicate. Data analysis and visualization were performed using GraphPad Prism 9.4.1 software. All data were presented as the mean ± SEM. Other analyses involved two-tailed t-tests to determine statistical differences between groups. The variance is similar between the groups that are being statistically compared. In all cases, p-values lower than 0.05 were considered statistically significant. Survival analysis of patients was obtained by Kaplan–Meier method, and the difference was determined by log-rank test. RNA based Stemness Scores (RNAss) [51] and Pearson’s correlation analysis were used to assess tumor stemness and its correlation with gene expression. In OSCC tissues, the expression levels of BARX2, has-miR-186-5p and has-miR-378a-3p were normalized to the corresponding normal epithelial expression levels. The correlation between the standardized expression levels of BARX2 and the standardized expression levels of has-miR-186-5p and has-miR-378a-3p was analyzed by Pearson’s correlation.