In total, 220 pathologically confirmed ESCC formalin-fixed, paraffin-embedded tissues and 19 matched tumor-adjacent samples were obtained from the Department of Pathology at the Sun Yat-Sen University Cancer Center (SYSUCC). Another 24 pairs of fresh primary ESCC tissues and corresponding normal tissues were collected from the Department of Thoracic Surgery at the First Affiliated Hospital of Sun Yat-Sen University (FAHSYSU). Fifteen normal blood samples and 87 ESCC blood samples were collected from Gaozhou People’s Hospital in Guangdong Province. All patients were untreated and had no history of other tumors. Written informed consent was obtained from all participants, and the research was approved by the Medical Ethical Committee of the SYSUCC.

The eight human ESCC cell lines (EC18, KYSE30/K30, Ec109, KYSE150/K150, KYSE180/K180, KYSE410/K410, KYSE510/K510, KYSE520/K520) and one immortalized normal esophageal cell line (NE1) used in the present study were kindly gifted by Professor Guan (Department of Clinical Oncology, University of Hong Kong). All cell lines were authenticated using the short tandem repeat (STR) technique. Additionally, mycoplasma testing was conducted on the cell lines, and the result is negative. Cells culture was by medium mixed with 10% fetal bovine serum and a 1% penicillin‒streptomycin in DMEM (Gibco, Carlsbad, CA, USA) at 37 ${}^{\circ }$C under 5% CO${}_2$.

Immunohistochemical assays were carried out according to a protocol described by other researchers [30]. A CST1 antibody (diluted 1:200; Proteintech, #16025-1-AP, Chicago, USA) was used to stain for CST1. When staining was completed, two pathologists reviewed and scored the sections independently, and discrepancies in scoring were resolved by consensus. The mean score was used as a cutoff value to divide the samples into high and low CST1 expression groups. The expression level of CST1 in ESCC tissues was scored as the proportion of the area with positive staining (1 for 0–25%; 2 for 25–50%; 3 for 50–75%; 4 for 75–100%;) multiplied by the staining intensity (0 for negative; 1 for weak; 2 for moderate; 3 for intense).

Total RNA was extracted from cells by using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) following the manufacturer’s instructions. The PrimeScript RT Master Mix Kit (Takara Bio, Kusatsu, Japan) was used for reverse transcription. RT‒qPCR was carried out using a SYBR Green Master Mix Kit (YEASEN, Shanghai, China) according to the manufacturer’s instructions, followed by a Roche 96/384-well Real-Time PCR system (Roche Applied Science, Indianapolis, IN, USA). GAPDH was used as an internal control. The primers used are listed in Table 1.

Cells (1 $\times$ 10${}^6$) were plated in 100-mm dishes. After 12 h under normal culture conditions, the cells were starved for 48 h. Next, the conditioned medium was collected and centrifuged at 3000 rpm for 5 min, and the supernatant was concentrated in 250-µL Amicon Ultracentrifuge filters (MWCO = 10 kDa; Merck Millipore, Billerica, MA, USA). CST1 was detected by immunoblotting and other assays.

For this assay, 3 $\times$ 10${}^4$ (migration) or 5 $\times$ 10${}^4$ (invasion) stably transfected cells were seeded in triplicate on the top of a chamber equipped with 8-µm microporous filters with (invasion) or without (migration) a matrix coating. After incubation for 12 h (migration) or 24 h (invasion), the cells were harvested. The number of migrated or invaded cells in five random optical fields from triplicate filters was averaged. For the wound healing assay, 3 $\times$ 10${}^5$ cells were plated in 6-well plates. Twenty-four hours later, wounds were made using a sterile 100-µL pipette tip. Debris was removed, and fresh medium was added. Images were captured at different times (0, 24, and 48 h). Additionally, we conducted another migration assay. Cultured K30-overexpressing and empty vector control cells were treated with isotype (diluted 1:50; Southern Biotech; Cat. No. 0107-01, Birmingham, USA) and CST1-neutralizing antibodies (diluted 1:50; Proteintech, #16025-1-AP, USA) for 12 h.

Western blot experiments were performed according to a protocol described by other researchers [31]. The antibodies used were as follows: CST1 rabbit monoclonal antibody (diluted 1:500; Proteintech, #16025-1-AP, Chicago, USA); GAPDH antibody (diluted 1:1000; Proteintech, #10494-1-AP, USA); MMP1 (Cell Signaling Technology, #54376, Danvers, MA, USA), MMP2 (Cell Signaling Technology, #40994, USA), MMP3 (Cell Signaling Technology, #14351, USA) and MMP10 antibodies (diluted 1:1000; Abcam, #ab261733, Cambridge, MA, USA); and E-cadherin (Proteintech, # 20874-1-AP), Desmoplakin, N-cadherin (Proteintech, # 22018-1-AP), Vimentin (Proteintech, # 10366-1-AP), and Slug (diluted 1:1000; Proteintech, #12129-1-AP, USA), GAPDH was used as an internal reference. Enhanced chemiluminescence reagents were used to visualize the proteins.

An ELISA kit (Cloud-Clone Corporation, product No. SEJ330Hu 96 Tests) for CST1 and the serum sample dilution ratio used were performed according to the manufacturer’s instructions. Specifically, a sandwich enzyme immunoassay for in vitro quantitative measurement of the CST1 protein concentration in human serum, human plasma and cell culture supernatants was used. All the reactions were repeated three times. To make the calculation easier, we plotted the optical density (OD) value of the standard (X axis) against the known concentration of the standard (Y axis), to draw a standard curve. Then bring the detected OD value of the sample into the standard curve to obtain the specific concentration of the sample.

The MTS assay was used to assess cancer cell viability and growth. Cells (1000 in 200 µL of medium) were seeded into a 96-well plate (Jet Biofil; # TCP011096, Guangzhou, China) and cultured under normal conditions for 24 h. Thereafter, we diluted the Cell Titer 96 Aqueous One Solution Cell Proliferation Assay reagent (Promega; # G3581; Madison, WI, USA) at 1:100 with medium, added the diluted reagent to the cells and incubated the cells for 3 h. Cell growth curves were plotted using GraphPad software GraphPad Prism 7.0 (GraphPad Software Inc., San Diego, USA).

The cells were plated in 6-well plates (7.5 $\times$ 10${}^2$/well) and cultured for 12 h to allow the cells to attach and cultured for 10 days until the colonies reached $>$50 cells. Next, the cells were harvested and fixed with methanol for 30 min and then stained with crystal violet for 1 h. Three different experiments were performed.

For the reporter assay, the cells were transiently transfected with the CST1 luciferase reporter plasmid (0.1 µg/well; Genecopoeia; HPRM18606-PG04, Rockville, MD, USA) and SPI1 expression plasmid (0.1 µg/well; Genecopoeia; EX-Z4319-Lv105) or empty vector (EV) in 96-well plates. Secreted Alkaline Phosphatase (SEAP) as an internal control (0.01 µg/well; Genecopoeia; pEZX-PA01). Forty-eight hours later, the luciferase assay was performed using the Secrete-Pair${}^{\text{TM}}$ Dual Luminescence Assay Kit (Genecopoeia, #LF031, USA) according to the manufacturer’s protocol.

Lentiviruses vector plasmids from the GeneCopoeia company (plasmids EX-A3184-Lv105 and EX-NEG-Lv105) were used to create stably expressing CST1 or the vector cell lines, as well as stably knock down shRNA plasmids (HSH003349-21-LVRU6P, HSH003349-22-LVRU6P, HSH003349-23-LVRU6P, HSH003349-24-LVRU6P and CSHCTR001-LVRU6P) were used to construct stable expression cell lines. The targets of CST1 shRNA sequences we used were HSH003349-22-LVRU6P (KD22) 5${}^{\prime }$-GGTACTAAGAGCCAGGCAACA-3${}^{\prime }$ and HSH003349-24-LVRU6P (KD24) 5${}^{\prime }$-CAGAAGAAACAGTTGTGCTCT-3${}^{\prime }$. Overexpressing or knockdown lentiviruses vector plasmids were transfected into 293T cells by using Lipofectamine 3000 (Invitrogen, #L3000150, Carlsbad, CA, USA) as a transfection reagent to produce the lentivirus according to the procedure of Lenti-Pac™ Lentiviral Packaging Systems (GeneCopoeia, #LT001, Rockville, MD, USA). Lentiviruses in 293T cell supernatant were harvested 72 h after transfection and filter through a 0.45-micron filter and freeze in an –80 degree refrigerator. Cells cultured with lentivirus and DMEM Medium (1:1) coupled with 2 µg/mL puromycin (sigma, #58-58-2, St. Louis, Missouri, USA) for stable cell lines selection within two weeks. Quantitative real-time PCR was used to detect the CST1 mRNA level; ELISA and immunoblotting were performed to determine the efficiency of CST1 protein level in overexpression or knockdown stable cells.

After culturing cells with DMEM without serum for 24 h, the conditioned medium (CM) was collected and concentrated tenfold using an ultrafiltration device (from 10 mL to 1 mL) and mixed with CST1 antibody (diluted 1:500; Proteintech, USA) or isotype control (Southern Biotech; Cat: 0103-01) overnight at 4 ${}^{\circ }$C. The cells were then mixed with protein A/G beads for 4 h at 4 ${}^{\circ }$C. Then, the beads were washed 5 times with PBS. Finally, the beads were boiled with 1$\times$ loading buffer for 10 minutes, and immunoblotting was carried out to determine whether CST1 was precipitated with MMP2 from the conditioned medium. The samples immunoprecipitated with CST1 and IgG antibodies were further subjected to mass spectrometry analysis by Shenzhen Weinafei Biotechnology Co., Ltd. (Shenzhen, China).

To obtain cell supernatants cultured in conditioned media, a total of 1 $\times$ 10${}^6$ cells were seeded in 10 $\times$ 10-cm dishes with CM for two days, and total proteins were extracted from CM and separated via PAGE on 10% polyacrylamide gels containing 0.5% gelatin (Macklin, #G9136-10MG, Shanghai, China). The following steps were the same as those in a previous study [32]. The gels were incubated in substrate buffer for 24 h at 37 ${}^{\circ }$C to enable MMP decompose substrate. Gels were stained with Coomassie Blue (BioFROXX, # 1078GR005, Guangzhou, China) for 2 h and washed until clear, and MMP decompose substrate was shown in the bright area.

For the metastasis assay, female BALB/c nude mice aged 5 weeks old were purchased from Guangdong Laboratory Animal Center (Guangdong, China), and spontaneous lymph node (LN) metastasis experiments were conducted as previously reported [33]. Briefly, 2 $\times$ 10${}^5$ cells stably expressing knockdown CST1 sequence with fluorescein tag in 20 µL mix of DMEM: matrigel (6:4) were subcutaneously injected into the footpad of each mouse to generate a primary tumor. Take photos of fluorescent in vivo imaging once a week. After 4 weeks, the experiments were terminated, and the popliteal LNs were isolated and preserved in RNAlater solution (Invitrogen, USA). TRIzol was used for LN total RNA extraction, and real-time qPCR were then performed to assess metastasis using specific primers for hypoxanthine phosphoribosyltransferase 1 (human HPRT), which do not cross-react with the corresponding mouse gene [34] and ACTB (universal for humans and mice). The primer sequences are listed in Table 1.

For the tumorigenicity experiment, 2 $\times$ 10${}^6$ cells suspended in 100 µL of DMEM containing 50% Matrigel (BD Biosciences) were subcutaneously inoculated into the subcutaneous area. The mice were monitored every three days for palpable tumor formation. Five weeks later, the mice were sacrificed and photographed, and the number of tumors was counted. All animal research protocols were approved by the at Sun Yat-sen University Cancer Center (SYSUCC) institutional Animal Care and Use Ethics Committee (IACUC), and the serial number of the protocol approved by IACUC was L102022021003K.

The RNA-sequence data, microarray data and their corresponding clinical information were downloaded from TCGA and GEO databases (https://www.ncbi.nlm.nih.gov/gds), including 260 samples from the TCGA database (80 normal samples and 180 ESCA samples) and 238 samples from GEO datasets GSE53624 and GSE53625 (119 normal samples and 119 ESCA samples).

The ESCA samples were divided into CST1 high expression or CST1 low expression groups based on the cut off value with the median expression level. Kaplan‒Meier survival analysis with the log-rank test was conducted to evaluate the difference in overall survival time between the two groups. The univariate and multivariate Cox regression were used to predict prognostic ability of CST1 expression, age, and grade based on samples from the TCGA and GEO databases. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated accordingly. We constructed a nomogram based on CST1 expression, survival status and clinical characteristics to predict ESCC patient prognosis by R software. The time-dependent receiver operating characteristic (ROC) curve and area under the curve (AUC) were constructed to exam the prognostic predictive value of CST1 by the R software (R Foundation for Statistical Computing, Vienna, Austria), and the specific procedures are similar to we previously published article [35].

Data were analyzed with SPSS (version 21.0) statistical software (IBM Inc., NY, USA). GraphPad Prism 7.0 (GraphPad Software Inc., San Diego, CA, USA) was used to visualize the data. For correlation analysis, the Pearson chi-square test was used to analyze the correlations between CST1 and clinical parameters. A two-tailed Mann‒Whitney U test was used to compare differences between two groups of data. The Kaplan‒Meier method with the log-rank test and the Cox multivariate proportional hazards regression model were applied for survival analysis, and p 0.05 was considered statistically significant. Experimental data are presented as the mean $\pm$ s.d. of three independent experiments. *p 0.05, **p 0.01, ***p 0.001 by two-tailed Student’s t test.

Our research focused on early-stage ESCC to identify new diagnostic or prognostic biomarkers; therefore, we tested the RNA profiles of three pairs of self-paired, early-diagnosis, untreated ESCC tissue and adjacent tumor tissue by RNA sequencing. Further analysis showed differentially expressed genes in a heatmap (Fig. 1A), where CST1 ranked first in overexpression among all the gene profiles. Then, we explored the expression of CST1 in ESCC in many databases. In the UALCAN database (http://ualcan.path.uab.edu/) [36], a differential expression heatmap showed that CST1 was one of the top 25 overexpressed genes in ESCC (Supplementary Fig. 1A), and the violin figure based on TCGA database indicated that CST1 expression was significantly higher in the 180 samples in ESCA group than the 80 samples in the normal group, p 0.001 (Fig. 1B). Similar to TCGA, we downloaded two GEO (Gene Expression Omnibus) datasets (GSE53624 & GSE53625) and found that CST1 expression was significantly higher in the 119 ESCA samples than in the 119 normal group, p 0.001 (Supplementary Fig. 1F).

Fig. 1.

Cystatin SN (CST1) significantly up-regulated in esophageal squamous cell carcinoma (ESCC). (A) A differential expression heatmap showed the top 25 overexpressed genes in three self-paired ESCC and tumor-adjacent tissues based on the RNA-sequencing. (B) The mRNA level of CST1 in 180 esophageal cancer (ESCA) and 80 normal tissues in The Cancer Genome Atlas (TCGA) database, p 0.001. (C~E) Relative mRNA and protein level of CST1 in 12 self-paired ESCC and tumor-adjacent tissues were detected by quantitative PCR (qPCR) and western blotting. (F) CST1 expression level in pan-cancer (in red box) and paired normal tissues (in blue box) and found that CST1 was highly expressed in a variety of tumors, including ESCA based on TCGA database. Data were presented as the mean $\pm$ s.d. of three independent experiments. **p 0.01, ***p 0.001 by student’s t-test.

In the Oncomine database (https://www.oncomine.com/) boxplots, the expression of CST1 in multiple cancers is shown in Supplementary Fig. 1D. Among those cancers, esophageal cancer showed the highest mean expression of CST1. In the ONCOMINE database, data from different investigators also showed that the expression level of CST1 in ESCC tissues was higher than that in normal tissues, and the difference was statistically significant, p 0.001 (Supplementary Fig. 1B,C,E). Therefore, we explored the expression of CST1 in 12 pairs of ESCC self-paired untreated clinical tissue samples (Fig. 1C–E). In the 12 pairs of tumors with adjacent normal tissues, CST1 had significantly higher mRNA levels (Fig. 1C) and protein levels (Fig. 1D,E) in the tumor cells or tissues than in the normal counterparts. Then, we explored the CST1 expression level in pan-cancer and paired normal tissues in the TCGA database and found that CST1 was highly expressed in a variety of tumors, including ESCA (Fig. 1F).

Thus, we conducted IHC to detect the expression levels of CST1 in 220 formalin-fixed, paraffin-embedded ESCC and 19 normal tissue samples. CST1 was mainly expressed in tumor tissues and was absent in normal tissues (Fig. 2A). The median value of the immunohistochemical score was used as the cutoff value, and samples were divided into high and low expression groups. The correlations between CST1 and clinical parameters are shown in Table 2. As shown in Table 2, a total of 118 of 220 (53.64%) ESCC tissue samples exhibited high CST1 expression, while only 2 of 19 (10.53%) normal tissue samples showed high CST1 expression. More importantly, Kaplan‒Meier survival analysis was performed to analyze the correlations between the expression of CST1 and clinical parameters, which showed that ESCC patients with higher CST1 expression had significantly shorter overall survival (OS) and disease-free survival (DFS) than ESCC patients with low expression (Fig. 2B,C). Moreover, the results indicated that the expression of CST1 may be correlated with lymph node stage and TNM stage. Survival curves for the entire ESCC patient cohort stratified by combinations of CST1 expression and clinical stage (Fig. 2D) showed that CST1 expression combined with TNM stage (TNM I/II as early-stage ESCC and TNM III/IV as advanced-stage ESCC) could also predict the prognosis of ESCC patients, with high CST1 expression predicting poor outcomes. Additionally, CST1 combined with tumor differentiation yielded similar results (Fig. 2E). Furthermore, high expression of CST1 was associated with dismal OS when combined with tumor size (Fig. 2F) or lymph node stage (Fig. 2G). In addition, advanced lymph node stage tissues had a higher CST1 expression level than normal and early-stage tissues (Fig. 2G). Moreover, univariate and multivariate Cox regression analyses indicated that CST1 expression and lymph node stage could be independent predictors of ESCC prognosis (Table 3). In addition, survival analysis using clinical parameters revealed that most of the parameters had an impact on the OS of ESCC patients, including TNM stage, lymph node stage, tumor size, age, vascular invasion, family history, and body mass index (BMI). To further investigate the prognostic value of CST1 in different clinical parameter subgroups, we analyzed high and low CST1 expression subgroups within different clinical parameters. In the subgroup analyses, high expression of CST1 was still associated with a poorer prognosis for ESCC (Table 3).

Fig. 2.

Up-regulated CST1 was associated with relatively poor survival based on immunochemistry. (A) CST1 was mainly expressed in tumor tissues and rarely absent in normal tissues by Immunohistochemical staining. Negative CST1 staining in normal esophagus epithelium (negative control), negative, weak, moderate and strong staining of CST1 in ESCC cytoplasm, $\times$100 (4$\times$), $\times$400 (10$\times$). (B,C) Kaplan-Meier survival analysis showed that ESCC patients with higher CST1 expression had significantly shorter overall survival (OS) and disease-free survival (DFS) than ESCC patients with low expression. (D,E) CST1 expression combined with TNM stage (TNM I/II as early-stage ESCC and TNM III/IV as advanced-stage ESCC) could also predict the prognosis of ESCC patients and high CST1 expression predicting poor outcomes combined with tumor differentiation. (F,G) High expression of CST1 was associated with dismal overall survival when combined with tumor size (2F) or lymph node stage (2G). PFS, progression-free survival; pT for tumour stage; pN for node stage; TNM, tumour-node-metastasis stage.

Furthermore, we analyzed the relationship between the risk score and patient follow-up time, and data from the 180 samples in TCGA database suggested that high CST1 expression had an impact on increasing the survival time of ESCA patients (Fig. 3A). In addition, high expression of CST1 was closely associated with poor OS in ESCA patients (Fig. 3B, Supplementary Fig. 1G). The mRNA expression level of CST1 in advanced clinical stage III/IV disease was higher than that in early stage I/II disease (Supplementary Fig. 1H). Data from TCGA indicated that univariate and multivariate Cox regression analyses indicated that CST1 expression, age and clinical stage could be independent predictors of ESCA prognosis (Fig. 3C,D). Hereafter, a nomogram based on these independent prognostic factors was constructed to predict the 1-, 3-, and 5-year survival probability for ESCA patients (Fig. 3E).

Fig. 3.

Up-regulated CST1 was associated with relatively poor survival based on bioinformatics analysis. (A) A model describing the relationship between different risk scores and follow-up time, events, and changes in gene expression in 180 ESCA samples based on TCGA database. (B) Survival curve showed high expression of CST1 was associated with relatively poor survival in ESCA based on TCGA database. (C,D) Data from TCGA indicated that univariate and multivariate Cox regression analyses indicated that CST1 expression, age and clinical stage could be independent predictors of ESCC prognosis. (E) A nomogram was constructed by integrating CST1 expression, age, clinical stage based on 180 ESCA samples in TCGA datasets.

We aimed to explore what causes the upregulation of CST1 in ESCC; thus, we further investigated which transcription factors influence the transcriptional activity of the CST1 gene. There are many online databases for predicting gene transcription factors, including JASPAR (https://jaspar.genereg.net/), hTFtarget [37], and GTRD [38]. We used the three transcription factor databases described above to predict the specific TF that can upregulate the transcription of the CST1 gene. The predicted TFs of the three databases intersected; afterward, the results showed that two common transcription factors were obtained, namely, AR (androgen receptor) and SPI1 (Spi-1 proto-oncogene) (Fig. 4A). Because our previous data revealed no significant difference between CST1 and patient sex, we will investigate the effect of another TF, SPI1, on CST1 transcriptional activity. Our previous study demonstrated that AR protein levels were not significantly upregulated in ESCC [39]. In addition, we analyzed the transcription factor-binding sites (TFBS) of SPI1 in the CST1 promoter region in the JASPAR database (Fig. 4B). Simultaneously, SPI1 can directly bind with the SPI1 binding element (SBE) in the CST1 promoter from the transcription start site (TSS), 388 bp upstream of the TSS, and the sequence of SBE was 5${}^{\prime }$-AGCAAACAGGAAGTCACATA-3${}^{\prime }$ (Fig. 4C). The SPI1 expression plasmid was transiently transfected into ESCC cell lines Ec109 and K30, and the mRNA expression level of CST1 was then detected by qPCR. The results showed that the mRNA level of CST1 was significantly increased in the Ec109 and K30 SPI1 overexpression groups compared with the control (t = 13.37, p = 0.0055; t = 24.15, p 0.001) (Fig. 4D). A dual-luciferase reporter system was used to study the effect of SPI1 on the transcriptional activity of the CST1 promoter. The results showed that the luciferase fluorescein activity of the cotransfection SPI1 expression plasmid and CST1 promoter luciferase plasmid was significantly higher than only empty vector control cotransfection with the CST1 promoter plasmid (t = 3.899, p = 0.0176), which means that SPI1 can positively regulate the transcriptional activity of the CST1 gene (Fig. 4E).

Fig. 4.

SPI1 can positively regulate the expression of CST1. (A) Three online databases for predicting gene transcription factors (TF), including JASPAR, hTFtarget and GTRD, were used to predict the TF which can regulate CST1 expression in ESCC. (B) SPI1 was demonstrated as a TF to regulate CST1 expression and its transcription factors binding site (TFBS) in JASPAR database was showed. (C) SPI1 Binding Element (SBE) in CST1 Promoter was 388 bases upstream away from the transcription start site (TSS) in the picture. (D) The mRNA level of CST1 was significantly upregulated in SPI1 overexpressed ESCC cells than in empty vector control. (E) In a dual-luciferase reporter system experiment, CST1 promoter luciferase activity was increased in SPI1 overexpressed ESCC cells than empty vector control. (F,G) Survival curve showed high expression of SPI1 was associated with relatively poor survival in ESCA based on TCGA and GEO database. (H) The mRNA level of SPI1 in ESCA and normal tissue in TCGA database. (I,J) Further bioinformatics analysis revealed that the mRNA level of CST1 and SPI1 was positively correlated based on 180 ESCA samples in TCGA database (r = 0.355, p 0.001) and 119 ESCA samples GEO database (r = 0.318, p 0.001). Data were presented as the mean $\pm$s.d. of three independent experiments. *p 0.05, **p 0.01, ***p 0.001 by student’s t-test.

In addition, we know that the expression of CST1 is significantly elevated in ESCC, which may be transcriptionally regulated by SPI1. We further explored SPI1 expression in ESCC and its prognostic value based on bioinformatics analysis. Survival curves showed that a high expression of SPI1 was associated with relatively poor survival in ESCA based on TCGA and GEO databases (Fig. 4F,G). The mRNA level of SPI1 in ESCA was significantly higher than that in normal tissue in the TCGA database (Fig. 4H). Further bioinformatics analysis revealed that the mRNA levels of CST1 and SPI1 were positively correlated based on the 180 ESCA samples in TCGA database (r = 0.355, p 0.001) and the 119 samples in GEO database (r = 0.318, p 0.001) (Fig. 4I,J).

qPCR was used to detect the relative mRNA level and Western blotting was conducted to detect the protein levels of CST1 in 8 ESCC cell lines and one normal immortalized esophageal cell line, NE1 (Fig. 5A,B).

Fig. 5.

Over-expression of CST1 facilitated proliferation in vitro and in vivo. (A,B) Relative mRNA and protein level of CST1 in 8 ESCC cell lines and one normal immortalized esophagus cell line NE1 were detected by qPCR and western blotting. (C–F) We constructed stable CST1 knock-down cells in highly CST1 mRNA level Ec109 cell and stable over-expressed cells in low CST1 mRNA level K510 and K30 cells, by stable knocking down with CST1-targeted shRNA (knock down 22 and knock down 24; KD22 and KD24) and scrambled shRNA (SCR) as controls; and stable over-expression by CST1 ORF plasmid and its empty vector as control, respectively. These stable cell lines confirmed by real-time qPCR and ELISA. (C) We detected the concentration of CST1 protein in extracellular conditioned medium (CM) by Enzyme-Linked Immunosorbent Assay (ELISA) to conform the stable expression cells construction system was working. (G–I) Knockdown of CST1 significantly suppressed the proliferation of Ec109 cells compared to the shSCR control groups by MTS assays (G). By contrast, the overexpression of CST1 promoted the proliferation of K510 and K30 cells (H,I). (J–O) Colony formation assays further verified that CST1 can accelerate the proliferation of ESCC cells. (P) In vivo experiment showed CST1 can accelerate the proliferation of ESCC cells, and the stable over-expression of CST1 in K30 cells and its control cells were subcutaneously inoculated into BALB/C nude mice. (Q,R) The over-expression of CST1 could significantly augmented the tumor growth in vivo than its control group, and the tumor weight and volume of CST1 over-expression group were significantly higher than controls in K30 cells. Data were presented as the mean $\pm$ s.d. of three independent experiments. *p 0.05, **p 0.01, ***p 0.001, ****p 0.0001 by student’s t-test. MTS for (3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide).

We constructed stable CST1 knockdown cells in Ec109 cells with high CST1 mRNA levels and stable overexpression cells in K510 and K30 cells with low CST1 mRNA levels by stably knocking down CST1-targeted shRNA (knock down 22 and knock down 24; KD22 and KD24) and scrambled shRNA (SCR) as controls and stable overexpression by the CST1 ORF plasmid and its empty vector as a control, respectively. Since CST1 protein is a secreted protein, we detected the concentration of CST1 protein in extracellular conditioned medium (CM) by ELISA to confirm that the stable expression cell construction system was working (Fig. 5C). These stable cell lines were confirmed by real-time qPCR (Fig. 5D–F). Then, MTS assays were performed to test the proliferation ability of tumour cells and we showed that knockdown of CST1 significantly suppressed the proliferation of Ec109 cells compared with the control groups (Fig. 5G). In contrast, the overexpression of CST1 promoted the proliferation of K510 and K30 cells in vitro (Fig. 5H,I). The results from colony formation assays further confirmed that CST1 can accelerate tumor proliferation in vitro (Fig. 5J–O).

To further investigate whether CST1 affects cell proliferation in vivo, the overexpression of CST1 in K30 cells and its control cells was subcutaneously inoculated into BALB/C nude mice (Fig. 5P). The overexpression of CST1 significantly augmented tumor growth in vivo compared with that in the control group, and the tumor weight and volume of the CST1 overexpression group were significantly higher than those of the controls in K30 cells (Fig. 5Q,R). In addition, we examined the cell cycle and early apoptosis in CST1 knockdown and overexpressing stable cell lines by flow cytometry (FCM), and we found no significant difference in early apoptosis. However, we found cell cycle arrest in G0/G1 phase in CST1 stable knockdown cells (Supplementary Fig. 2A,B,F,H) with its control. The CST1 stable overexpressing cells (Supplementary Fig. 2C,D,G,I) with its control no significant difference in early apoptosis and cell cycle arrest. The cell cycle arrest biomarkers p53 did not seem to be obviously changed and cyclin D1 seems to down-regulated slightly in CST1 knockdown cells by Western blotting (Supplementary Fig. 2E).

To validate the role of CST1 in ESCC motility, a scratch healing assay was used to assess the regulatory effect of CST1 on migration. The results showed that wound closure occurred gradually 72 h after scratching, whereas this effect on wound healing was significantly reduced after CST1 knockdown in Ec109 cells (Fig. 6A,B) and significantly increased after CST1 overexpression in K30 and K510 cells (Fig. 6F,G; Supplementary Fig. 2J,K). The stable knockdown of CST1 in Ec109 cells significantly reduced the migratory (Fig. 6C,D) and invasive (Fig. 6C,E) abilities. Conversely, stable overexpression of CST1 in K510 and K30 cells yielded great enhancement of migration and invasive (Fig. 6H–K; Supplementary Fig. 2L,M) abilities compared with the control cells. These results proved that CST1 was involved in promoting migration and invasion.

Fig. 6.

Overexpressed CST1 can promote invasion and migration of ESCC cells in vitro and in vivo. (A,B) The scratch healing assay was used to assess the regulation ability of CST1 in migration. The results showed that wound closure occurred gradually 72 h after scratching, whereas this effect of wound healing was significantly reduced after CST1 knockdown in Ec109 cells. (C–E) The stable knocking down CST1 cell Ec109 was significantly reduced the migratory (C,D) and invasive (C,E) abilities. (F–G) The wound healing was significantly increased after CST1 over-expression in K30 and K510 cells. (H–K) Conversely, CST1 stable overexpression in K510 and K30 cells showed great enhancement of migratory and invasive abilities compared with the control cells. CST1 stimulated the EMT and accelerated motility of ESCC cells. (L) CST1 knockdown downregulated the expression of mesenchymal markers Vimentin, N-cadherin, and upregulated that of epithelial markers E-cadherin and Desmoplakin in Ec109 cells by immunoblotting assays. In contrast, overexpressing CST1 in K510 and K30 cells upregulated the expression of Vimentin and N-cadherin and downregulated the expression of E-cadherin and Desmoplakin, as well as that of EMT promoting transcription factors Slug. (N–M) Overexpressed CST1 can promote metastasis of ESCC cells in vivo. (O,P) To further explore whether CST1 modulates tumor metastasis in vivo, the knockdown CST1 in Ec109 cells were transplanted into nude mice via footpad injection, and we measured the fluorescence intensity of tumors in the mouse feet once a week. After 4 weeks, the results showed a drastically lower rate of popliteal lymph node metastasis in mice injected with the CST1 knockdown cell lines Ec109-KD24 than in those injected with Ec109-SCR cells. (Q) Examination of popliteal lymph node metastasis by measuring lymph node volume and lymph node RNA extraction then running qPCR with using specific primers for human HPRT, which do not cross-react with the corresponding mouse gene. (Q) The result showed that popliteal lymph node metastasis was markedly suppressed after the downregulation of CST1 in Ec109 cells. Data were presented as the mean $\pm$ s.d. of three independent experiments. *p 0.05, **p 0.01, ***p 0.001, ****p 0.0001 by student’s t-test. Min, minimum; Max, maximum.

Epithelial mesenchymal transition (EMT) plays crucial roles in tumor metastasis, and decreasing epithelial biomarkers and/or increasing mesenchymal biomarkers can enhance the motility of tumor cells [40]. We found that CST1 knockdown downregulated the expression of the mesenchymal markers Vimentin and N-cadherin and upregulated that of the epithelial markers E-cadherin and desmoplakin in Ec109 cells by immunoblotting assays. In contrast, overexpressing CST1 in K510 and K30 cells upregulated the expression of Vimentin and N-cadherin and downregulated the expression of E-cadherin and desmoplakin, as well as that of the EMT-promoting transcription factor Slug. Taken together, these results indicate that CST1 may stimulate EMT to accelerate the motility of ESCC cells (Fig. 6L).

To further explore whether CST1 modulates tumor metastasis in vivo, Ec109 cells with CST1 knockdown were transplanted into nude mice via footpad injection, and the fluorescence intensity of tumors in the mouse feet was measured once a week. After 4 weeks, the results showed a drastically lower rate of popliteal lymph node metastasis in mice injected with the CST1 knockdown cell line Ec109-KD24 than in those injected with Ec109-SCR cells (Fig. 6N,M). Examination of popliteal lymph node metastasis was performed by measuring lymph node volume (Fig. 6O,P) and lymph node RNA extraction and then running qPCR with specific primers for human HPRT, which do not cross-react with the corresponding mouse gene [34]. The results showed that popliteal lymph node metastasis was markedly suppressed after the downregulation of CST1 in Ec109 cells (Fig. 6Q).

CST1 was found to be highly expressed in ESCC cells and tissues. Thus, we aimed to determine whether serum CST1 levels are consistent with those in cells and tissues. As a secreted protein, to further study the role of CST1 in the ECM, we conducted ELISA analysis of serum samples from 15 volunteers who underwent a physical examination (control) and total 87 ESCC patients. Compared with that of the healthy population, the serum CST1 level of the ESCC patients was distinctly increased (Fig. 7A). Given that the number of serum samples was limited, we observed only that the expression level of CST1 in the serum was significantly associated with age and tumor pathological differentiation ($\chi$${}^2$ = 9.079, p = 0.004; and $\chi$${}^2$ = 6.566, p = 0.038, respectively). Elderly ESCC patients aged more than 68 years tended to have higher expression of CST1 in the serum, with a high OR of 9.286 (95% CI 1.985, 43.444). The chi-square test or logistic regression was used to test the cross-table data with the OR and 95% confidence interval (CI), and the cutoff value of CST1 was the mean value in the 38 patients’ serum samples with clinical parameters (Table 4). The relationships between the expression of CST1 detected by ELISA and clinical parameters are shown in Table 4. More importantly, high expression of CST1 may serve as a potential serum diagnostic marker, as we plotted the receiver operating characteristic (ROC) curve using the ELISA experimental data. The area under the curve (AUC) was 0.9748 (p 0.0001), which had high diagnostic value (Fig. 7B). The ROC curve was constructed to investigate the diagnostic ability of CST1 expression in predicting the OS of ESCA patients, which showed that CST1 had a high predictive ability for ESCA patient OS (AUC: 0.805 and AUC: 0.802) based on the GEO dataset and TCGA database, respectively (Fig. 7C,D).

Fig. 7.

The serum secreted CST1 can be a potentially non-invasive early diagnostic method for ESCC patients and the new neutralizing antibody drugs target for ESCC treatment. (A) ELISA analysis of CST1 level in serum samples from 15 volunteers who underwent a physical examination (control) and 87 ESCC patients. (B) The area under the curve (AUC) was 0.9748 (p 0.0001), which had high diagnostic value based on ELISA. (C,D) ROC curve was construct to investigate the diagnostic ability of CST1 expression in predicting overall survival of ESCA patients, which showed that CST1 had a high prediction ability of ESCA patient overall survival (AUC: 0.805 and AUC: 0.802) based on the GEO dataset and TCGA database, respectively. (E,F) The invasion assay by CST1 neutralizing antibody blocking in K30 CST1 over-expression and its empty vector control cells. The invasive ability of ESCC cells was significantly inhibited by antibody neutralization of CST1 secreted in the extracellular medium. (G,H) The RNA profiling of two paired stable knockdown or overexpression CST1 cell lines (Ec109 KD24 and its SCR control; K30 over-expression and its empty vector control) obtained by RNA-sequencing was used for differential analysis to obtain the intersection of genes with significant differences, and further cluster analysis, KEGG and GO analysis were performed. The top 20 GO term showed that the function of CST1 may be related to the extracellular region. Data were presented as the mean $\pm$ s.d. of three independent experiments. **p 0.01, ***p 0.001, ****p 0.0001 by student’s t-test.

We found that CST1 in the ECM may play an important role in the invasive ability of ESCC cells; thus, we conducted an invasion assay by blocking CST1-neutralizing antibody in K30 CST1-overexpressing cells and empty vector control cells. The invasive ability of ESCC cells was significantly inhibited by antibody neutralization of CST1 secreted into the extracellular medium (Fig. 7E,F).

Moreover, two paired stable knockdown or overexpression CST1 cell lines (Ec109 KD24 and its SCR control; K30 overexpression and its empty vector control) were subjected to RNA sequencing. The RNA datasets obtained by RNA sequencing were used for differential analysis to obtain the intersection of genes with significant differences, and further cluster analysis and KEGG and GO analyses were performed. The top 20 GO terms showed that the function of CST1 may be related to the extracellular region (Fig. 7G,H).

Through in-depth analysis of RNA-sequencing data, we found that CST1 may interact with some members of the MMP family in the reactome, and the top 20 reactome enrichment terms are shown (Fig. 8A). Similarly, gene set enrichment analysis (GSEA) also showed differential gene enrichment in the collagen catabolic process pathway, which was closely related to the function of the MMP family (Fig. 8B).

Fig. 8.

CST1 accelerated motility of ESCC cells by interacting with matrix metalloproteinase (MMP) family proteins. (A) The top 20 reactome enrichment terms were showed CST1 may interact with some members of the MMP family in reactome based on RNA-sequencing. (B) Gene Set Enrichment Analysis (GSEA) analysis also showed differential gene enrichment in collagen catabolic process pathway, which was closely related to the function of the MMP family. (C,D) We want to explore how CST1 interact with these MMP family members gene, then we detected the MMP1, MMP2, MMP3, MMP9, MMP10 and MMP13 mRNA level by qPCR in Ec109 KD24 & Ec109 SCR cells and K30 over & K30 vector cells. However, the results showed that CST1 had no apparent effect on regulation the mRNA level of these members of the MMP family. (E) The protein levels of the MMP family (MMP1, 2, 3, 10) were positively associated with CST1 protein level, they decreased in the CST1 knock-down cells and increased in the CST1 over-expressed cells. (F) Further bioinformatics analysis revealed that the mRNA level of CST1 and MMP2 was positively correlated (r = 0.428, p 0.001) based on GEO database. (G) The CST1 and MMP2 protein level of the 87 ESCC patient serum by ELISA, and it turn out that the protein level of CST1 and MMP2 in ESCC patient serum was positively correlation (r = 0.3005, p = 0.0044). Data were presented as the mean $\pm$ s.d. of three independent experiments. n.s. : not statistically significant, *p 0.05, by student’s t-test.

Therefore, we wanted to explore how CST1 interacts with these MMP family members, so we assessed the MMP1, MMP2, MMP3, MMP9, MMP10 and MMP13 mRNA levels by qPCR in Ec109 KD24 and Ec109 SCR cells and K30 overexpression and K30 vector cells. However, the results showed that CST1 had no apparent effect on the regulation of the mRNA level of these members of the MMP family (Fig. 8C,D). In contrast, the protein levels of the MMP family (MMP1, 2, 3, 10) were positively associated with CST1 protein levels, which decreased in the CST1 knockdown cells and increased in the CST1-overexpressing cells (Fig. 8E). Western blot results indicated that CST1 knockdown decreased the protein levels of MMP2 (pro-form, 72 kDa) and cleaved the MMP10 pro-form and turn it into the active form.

Further bioinformatics analysis revealed that the mRNA levels of CST1 and MMP2 were positively correlated (r = 0.428, p 0.001) (Fig. 8F). Therefore, we assessed the protein levels of CST1 and MMP2 in 87 ESCC patient serum samples by ELISA and found that the protein levels of CST1 and MMP2 in ESCC patient serum were positively correlated (r = 0.3005, p = 0.0044) (Fig. 8G).

We aimed to further investigate whether the CST1 proteins immunoprecipitated MMP2 in the ECM. We conducted immunoprecipitation (IP) experiments with a CST1 antibody and IgG control, and the IP samples were then subjected to SDS gel electrophoresis (Fig. 9A). We found that CST1 immunoprecipitated the band between molecular weights 65 and 75 kDa, which we suspect may be MMP2, as CST1 cleaves the MMP2 pro-form (72 kDa) to generate the MMP2 active form (62 kDa). Therefore, we sent the two IP samples (CST1 and IgG samples) for mass spectrometry to identify which proteins could be specifically immunoprecipitated by CST1. The mass spectrometry results are shown in the table, in which MMP2 can be immunoprecipitated specifically by CST1 (Fig. 9B,C). The coimmunoprecipitation (Co-IP) method was further used to verify that both CST1 and MMP2 physically bind to each other in Ec109 cells (Fig. 9D). We were curious about what role CST1 could play when physically combined with MMP2. Could CST1, a cysteine protease inhibitor, regulate the activity of MMP2 gelatinase? We therefore chose to perform gelatin zymography experiments to further validate the effect of CST1 on the activity of MMP2 gelatinase (Fig. 9E) [32]. Through the zymography assay, we found that CST1 can positively regulate the activity of MMP2 gelatinase. Knockdown of CST1 decreased MMP2 activities, while overexpression of CST1 increased MMP2 activity but had no obvious effect on the activity of MMP9 (Fig. 9E). The diagram shows that the transcription factor SPI1 upregulates the expression of the CST1 protein at the transcriptional level by binding to the CST1 gene promoter such that the expression level of the CST1 protein is increased, and the level of secreted CST1 protein is increased in the ECM. The CST1 protein can bind MMP2 and convert precursor MMP2 (pro-MMP2) into active MMP2 protein, leading to the decomposition of gelatin and remodeling of the ECM and further promoting the metastatic ability of ESCC cells (Fig. 9F).

Fig. 9.

CST1 accelerated motility of ESCC cells by up-regulating the quantity and enzymatic activity of MMP2. (A) To further investigate whether the proteins of CST1 and MMP2 directly bind in the ECM. Then We conducted the immunoprecipitation (IP) experiments by CST1 antibody and IgG control, and then the IP samples were subjected to run the SDS gel electrophoresis (Coomassie blue staining). (B,C) We sent the two IP samples (CST1 and IgG samples) for mass spectrometry identification to see which proteins could be specifically immunoprecipitated by CST1. The result of mass spectrometry showed in the table, in which MMP2 can be immunoprecipitated specifically by CST1. (D) The co-immunoprecipitation (CO-IP) method was further used to verify that both CST1 and MMP2 bind to each other in Ec109 cells. (E) The gelatin zymography experiments were used to further validate the effect of CST1 on the activity of MMP2 gelatinase, and found CST1 acts as an activator of MMP2. (F) Diagram showed that the transcription factor SPI1 up-regulates the expression of CST1 protein at the transcriptional level by binding to the CST1 gene promoter, so that the expression level of CST1 protein is increased, and the level of secreted CST1 protein in the is increased in the extracellular matrix (ECM). The CST1 protein can bind MMP2 and convert precursor MMP2 (pro-MMP2) into active MMP2 protein, leading to decompose gelatin and remodel the extracellular matrix, and further promote the metastatic ability of ESCC cells.