RETRACTED ARTICLE: Sorcin Predicts Poor Prognosis and Promotes Metastasis by Facilitating Epithelial-mesenchymal Transition in Hepatocellular Carcinoma

Metastasis-associated recurrence is the main cause for the poor prognosis of hepatocellular carcinoma (HCC). However, the detailed molecular mechanisms underlying HCC metastasis remain elusive. Though some data indicated the oncogenic role of Sorcin in tumors, the prognostic value and biological role of Sorcin in HCC is still unknown. In this study, it demonstrated that Sorcin expression levels were significantly upregulated in HCC tumor tissues compared with matched adjacent nontumorous liver tissues and normal liver tissues, and such expression level correlated with HCC metastasis. High Sorcin expression was significantly correlated with aggressive clinicopathological characteristics such as multiple tumor nodules, high Edmondson-Steiner grade, microvascular invasion, advanced TNM stage and advanced BCLC stage (all P < 0.05). HCC patients with high Sorcin expression had both shorter survival and higher recurrence than those with low Sorcin expression (all P < 0.05). Sorcin expression was an independent and significant risk factor for survival and recurrence of HCC patients. Results of functional experiments showed that Sorcin could promote HCC cell proliferation, migration, and invasion in vitro, and facilitate HCC growth and metastasis in vivo. Mechanistically, Sorcin exerted its role by activating extracellular signal-regulated kinase (ERK) pathway and promoted metastasis by facilitating epithelial-mesenchymal transition (EMT) in HCC.

independent prognosticator for survival and recurrence. Then, results of functional experiments indicated that Sorcin promoted cell migration and invasion in vitro as well as metastasis in vivo. Mechanically, we further demonstrated that Sorcin exerted its function by activating extracellular signal-regulated kinase (ERK) pathway and inducing EMT in HCC.

Sorcin Expression Is Significantly Elevated and Associated with Metastasis in HCC.
Firstly, the expression of Sorcin was determined in sixty pairs of HCC samples including tumors (Ts) and their corresponding adjacent nontumorous liver tissues (ANLTs) in screening cohort (Supplementary Fig. 1 and Supplementary  Table 1) by real-time PCR (RT-PCR) and western blot. Five normal liver tissues (NLTs) from hepatic hemangioma patients were set as loading control. Results showed Sorcin mRNA and protein expression in Ts were markedly higher than in ANLTs or NLTs (P < 0.001; Fig. 1A1,2). The expression levels of Sorcin in HCC tissues were significantly higher in advanced stage patients (TNM stages II/III) than in early stage patients (TNM stage I) Real-time PCR (RT-PCR) and (A2) western blot were used to determine the expression of Sorcin in sixty pairs of Ts and ANLTs in screening cohort. Five NLTs from hepatic hemangioma patients were set as loading control. NLT, ANLT, T represent normal liver tissue, adjacent nontumorous liver tissue, tumor tissue. (B) Sorcin mRNA (B1) and protein (B2) expression in HCC tissues from advanced stage patients were significantly higher (TNM stages II/III) than in those from early stage patients (TNM stage I). (C) Sorcin mRNA (C1) and protein (C2) expression was significantly higher in tumors with microvascular invasion (MVI) than those without MVI. MVI, microvascular invasion; NMVI, without microvascular invasion. (D) Sorcin mRNA (D1) and protein (D2) expression in HCC tissues from patients with early tumor recurrences (within 2 years) was higher than those from patients without early tumor recurrences. ER, NER represent HCC with early recurrence, HCC without early recurrence respectively. **P < 0.01, ***P < 0.001.
(P < 0.01; Fig. 1B1,2). Notably, Sorcin expression was significantly higher in tumors with microvascular invasion (MVI) than tumors without MVI (P < 0.01; Fig. 1C1,2). HCC tissues from patients with early tumor recurrences (within 2 years) also exhibited higher Sorcin expression than that from patients without early tumor recurrences (P < 0.01; Fig. 1D1,2). Taken together, these data proved that the expression of Sorcin was significantly elevated in HCC tissues and may be associated with metastasis in HCC.

High Sorcin Expression Correlates with Aggressive Clinicopathological Characteristics and
Predicts Poor Prognosis in HCC Patients. Next, we assessed the association of Sorcin expression with the prognosis of HCC patients according to ReMARK guidelines for reporting prognostic biomarkers in cancer 14 . Immunohistochemical (IHC) assays of HCC samples from training cohort (Supplementary Table 1 and Supplementary Fig. 1) showed Sorcin expression was mainly located in the cytoplasm (Fig. 2A). The expression level of Sorcin was significantly higher in Ts than that in ANLTs (Fig. 2B), and high Sorcin expression was found in 66.25% (106/160) in Ts, as compared with only 8.13% (13/160) in ANLTs from training cohort (P < 0.001; Fig. 2C1). HCC patients were divided into two groups: the high Sorcin expression group and low Sorcin expression group based on IHC staining score of HCC tissues. Analyzing the relationship of Sorcin expression with clinicopathologic features showed that high Sorcin expression significantly correlated with aggressive clinicopathological variables such as multiple nodules (P = 0.001; Table 1 expression had higher early recurrence (within 2 years) than those with low Sorcin expression (P = 0.003; Fig. 2D3). Next, we further explored the prognostic value of Sorcin in specific subgroups of HCC patients. Patients were also classified into two subgroups based on microvascular invasion (MVI), which is a major prognostic factor for HCC and was recently identified as a better predictor of tumor recurrence 15 Fig. 1 and Supplementary Table 1) further confirmed that the expression level of Sorcin was significantly higher in Ts than that in ANLT (Fig. 2B). High Sorcin expression was found in 63.33% (76/120) in Ts, as compared with only 7.5% (9/120) in ANLTs (P < 0.001; Fig. 2C2) from validation cohort. Sorcin expression was also significantly correlated with aggressive clinicopathological variables including multiple nodules (P < 0.001;

Sorcin Enhances Invasion and Metastasis of HCC cells in Vitro and in Vivo.
The above results indicated that high Sorcin expression was positively associated with metastasis in HCC, so we explored the role of Sorcin in the metastasis of HCC. Firstly, we determined the Sorcin expression in HCC cell lines (HepG2, PLC/ PRF5, Bel7402, Hep3B, MHCC97-H and HCCLM3) and normal liver cells L02 cells. Compared with human normal liver cells L02 cells, Sorcin messenger RNA (mRNA) and protein was highly expressed in 6 human HCC cell lines ( Supplementary Fig. 4A1,2). Notably, Sorcin expression in high metastatic HCC cell lines, such as Hep3B, MHCC97-H and HCCLM3 was higher than in low metastatic HCC cell lines HepG2, PLC/PRF5 and Bel7402 16,17 16,17 . Transwell migration assays showed overexpression of Sorcin significantly increased the ability of migration in HepG2 cells (Fig. 3A1), whereas down-regulation of Sorcin decreased the ability of migration in HCCLM3 cells (Fig. 3A2). The similar results were observed by transwell invasion assays (Fig. 3B1,2). Next, we examined the effect of Sorcin on the proliferation of HCC cells. MTT assays showed that overexpression of Sorcin significantly accelerated HepG2 cell proliferation ( Supplementary Fig. 5A). On the contrary, down-regulation of Sorcin dramatically decelerated HCCLM3 cell proliferation ( Supplementary Fig. 5B). Colony formation assays also shown that the capacity of colony formation of HepG2 Sorcin cell is significantly more potent than that of HepG2 Vector cell ( Supplementary Fig. 5C). However, the capacity of colony formation of HCCLM3 shSocrin cell is significantly weaker than that of HCCLM3 shcontrol cell ( Supplementary Fig. 5D). Above of all, these data indicate that Sorcin displays an oncogenic function and is involved in invasion and metastasis in HCC.
To verify the in vitro results, the role of Sorcin was further evaluated by using subcutaneous xenograft tumor and subsequently by in situ xenograft metastasis mice models. After 6 weeks, HepG2 Sorcin cell-derived tumors at the subcutaneous implantation site grew more rapidly and were larger than HepG2 Vector cell-derived tumors (Fig. 3C1), while HCCLM3 shSocrin cell-derived tumors grew more slowly and were smaller than HCCLM3 shcontrol cell-derived tumors (Fig. 3C2). Consistently, by analyzing the tumor volume of orthotopic xenograft liver tumor, it showed that Sorcin overexpression promoted tumors growth (Fig. 3D1), while Sorcin knockdown inhibited tumors growth in vivo (Fig. 3D2). Further detected the metastatic nodules in livers (Fig. 3E1) and lungs (Fig. 3E2) by H&E, data showed the number of intrahepatic and pulmonary metastatic nodules in mice with tumors derived from HepG2 Sorcin cells was significantly higher than in those with tumors derived from HepG2 Vector cells (Fig. 3F1,2). In contrast, the number of metastatic nodules in mice with tumors generated from HCCLM3 shSocrin Scientific REPORTS | 7: 10049 | DOI:10.1038/s41598-017-10365-3 cells was markedly smaller than in those with tumors generated from HCCLM3 shcontrol cells (Fig. 3F1,2). Taken it together, our studies demonstrated that Sorcin could promote HCC growth and metastasis in vivo.

Sorcin Exerts the Function through Activating Extracellular Signal-regulated Kinase (ERK) Pathway in HCC.
To elucidate the potential signaling by which Sorcin contributes to metastasis in HCC, a multi-pathway reporter array was performed. Sorcin overexpression significantly enhanced the activity of ERK signaling in HepG2 cells (Fig. 4A1), but Sorcin knockdown attenuated ERK signaling activity in HCCLM3 cells (Fig. 4A2). Growing evidence has showed that ERK signaling is crucial for progression of HCC 18,19 , and its activation can be triggered by changing cytosolic Ca 2+ level 9, 20 . As Sorcin also regulates Ca 2+ homeostasis, so we focus on exploring whether Sorcin enhanced metastasis by activating ERK signaling in HCC. Western blot showed the level of phosphorylated ERK1/2 (p-ERK1/2) increased when Sorcin was overexpressed in HepG2 cells (Fig. 4B1). In contrast, knockdown of the level of p-ERK1/2 decreased when Sorcin was knocked down in HCCLM3 cells (Fig. 4B2). However, manipulation of Sorcin expression in HCC cells didn't affect ERK1/2 expression (Fig. 4B1,2  or HCCLM3 cells (Fig. 4B1,2). However, U0126 treatment effectively inhibited Sorcin-induced ERK1/2 activation in HepG2 cells (Fig. 4B1). In HCCLM3 cells, as U0126 treatment decreased the level of p-ERK1/2, Sorcin knockdown also caused the reduced level of p-ERK1/2 (Fig. 4B2). These data proved that Sorcin activated ERK signaling, not its downstream molecules. IHC analysis also showed HCC tissues, with high Sorcin expression, also exhibited high level of p-ERK1/2 (Fig. 4C1). Spearman rank correlation analysis showed the positive correlation of Sorcin expression with the level of p-ERK1/2 (Figs. 4C2,3). These results concluded that Sorcin could activate ERK1/2 signaling in HCC. Further, to study whether Sorcin promotes invasion and metastasis through activating ERK signaling in HCC cells, the effect of Sorcin on HCC cell migration and invasiveness interfered by U0126 was measured. Transwell migration and invasion assays showed that U0126 could significantly inhibited the effect of Sorcin on cell migration and invasion capacity in HepG2 Sorcin cells (Fig. 4D1), while it had little effect on HepG2 Vector cells in which Sorcin expression and ERK activation is low (Fig. 4D1). Similarly, U0126 treatment obviously inhibited the migration and invasion capacity of HCCLM3 shcontrol cells with high Sorcin expression and high level of p-ERK, but it had little effect on HCCLM3 shSorcin cells in which Sorcin expression and p-ERK level is low (Fig. 4D2). In addition, U0126 treatment obviously reduced the effect of Sorcin on the proliferation and colony formation capacity in Sorcin-overexpressed cells HepG2 Sorcin (Supplementary Fig. 6A1 and B1) or high Sorcin expression cells HCCLM3 shcontrol (Supplementary Fig. 6A2 and B2), but it had little effect on HepG2 Vector cells ( Supplementary  Fig. 6A1 and B1) or HCCLM3 shSorcin (Supplementary Fig. 6A2 and B2). Collectively, these results indicate that Sorcin promotes HCC invasion and metastasis by activating ERK signaling.

Sorcin Enhances Metastasis by Activating ERK Signaling via Facilitating EMT. EMT, by which
HCC cells gain cell motility and invasiveness, is defined by loss of epithelial cell polarity and E-cadherin expression, and by the acquisition of fibroblastic mesenchymal morphology and Vimentin expression 21 . It was observed that the HepG2 Sorcin cells exhibited spindle-like mesenchymal morphology, while HepG2 Vector cells mostly look like epithelial cobblestone appearance (Fig. 5A1). On the contrary, HCCLM3 shSorcin cells changed to an epithelial phenotype as compared with mesenchymal like HCCLM3 shControl cells (Fig. 5A1). F-actin immunofluorescence staining was used to analyze the cytoskeleton. Immunofluorescence (IF) analysis showed HepG2 Sorcin cells presented the appearance of F-actin fibers comparing with HepG2 Vector cells (Fig. 5A2). Inversely, HCCLM3 shSorcin cells exhibited shrinkable F-actin fibers relative to HCCLM3 shcontrol cells (Fig. 5A2). Thus, these indicated that Sorcin might be associated with EMT in HCC.
To further study the role of Sorcin in EMT in HCC, immnunofluorescence (IF) analysis revealed that overexpression of Sorcin in HepG2 cells significantly increased the expression level of mesenchymal marker Vimentin, but dramatically reduced the expression level of epithelial marker E-cadherin (Fig. 5B1). Conversely, knockdown of Sorcin in HCCLM3 significantly reduced the expression level of mesenchymal marker Vimentin, but significantly increased the expression level of epithelial marker E-cadherin (Fig. 5B2). Similar with the data of IF, western blot also showed that overexpression of Sorcin increased the expression of Vimentin and decreased the expression of E-cadherin (Fig. 5C1), whereas down-regulation of Sorcin decreased the expression of Vimentin and increased the expression of E-cadherin (Fig. 5C2). IHC analysis of orthotopic liver xenograft tumors of mice exhibited the similar changes of these EMT markers and Sorcin in protein expression ( Supplementary Fig. 7A and B). Furthermore, high Sorcin expression co-located with low E-cadherin and high vimentin expression in HCC tissues was observed by IHC (Fig. 5D). A positive correlation of the expression of Sorcin with the expression of Vimentin and a negative correlation of the expression of Sorcin with the expression of E-cadherin was also observed in the HCC tissues (Supplementary Table 2). These results suggested that Sorcin could promote HCC invasion and metastasis via EMT.
Growing evidence has shown that EMT can be triggered by activation of ERK pathway 18,22 . Further, we asked whether Sorcin promoted EMT through activation of ERK pathway. Interestingly, it was observed by western blot assays that U0126 treatment reversed the effect of Sorcin on decrease in E-cadherin expression, and increase in vimentin expression in HepG2 Sorcin cells (Fig. 5C1) or HCCLM3 shcontrol (Fig. 5C2), but they were not observed in low Sorcin expression cells HepG2 Vector (Fig. 5C1) or HCCLM3 shsorcin (Fig. 5C2) with relative low level of ERK activation, suggesting that activation of ERK is involved in Sorcin-induced EMT. IHC analysis of liver orthotopic transplantation tumors derived from HepG2 Sorcin cells, with mesenchymal tissue phenotype, exhibited high level of p-ERK, whereas liver orthotopic transplantation tumors derived from HepG2 Vector cells, with epithelial tissue phenotype, exhibited low level of p-ERK ( Supplementary Fig. 7B). On the contrary, orthotopic liver exnograft tumors derived from HCCLM3 shSorcin cells, with mesenchymal tissue phenotype, exhibited high level of p-ERK, whereas orthotopic liver exnograft tumors derived from HCCLM3 shcontrol cells, with epithelial tissue phenotype, exhibited high level of p-ERK ( Supplementary Fig. 7B). Moreover, IHC showed that Vimentin, p-ERK expressions level was high in HCC tissues with high level of Sorcin expression, but they were low in tumors with low expression of Sorcin (Fig. 5D). In contrast, the E-cadherin expression was opposite (Fig. 5D). Furthermore, the correlation analysis revealed Sorcin expression was positively correlative with p-ERK level in HCC samples (Fig. 4C2,3). p-ERK level reversely correlated with E-cadherin expression and positively with Vimentin in HCC tissues (Supplementary Table 3). Taken together, these data indicate that Sorcin can promote EMT by activating ERK pathway.

Discussion
In the present study, we demonstrated that both Sorcin mRNA and protein levels were significantly elevated in tumors compared with ANLTs in HCC. Elevated Sorcin expression was found to be positively correlated with the metastatic potential of HCC cells, suggesting Sorcin might play a role in HCC metastasis. Further analysis of the association of Sorcin expression with the clinicopathological characteristics in HCC patients revealed that Sorcin overexpression was significantly correlated with more aggressive clinicopathological characteristics including multiple tumor nodules, poor differentiation, MVI, advanced TNM stage and BCLC stage. Kaplan-Meier analysis showed that the HCC patients with high Sorcin expression in general had worse prognosis than those with low expression. And a multivariable Cox regression analysis indicated that high Sorcin expression is an independent risk factor for recurrence and survival. Notably, we proved that high Sorcin expression was associated with early recurrence. Stratifying the patients into subclinical group by MVI, the prognostic significance of Sorcin was still found to be existed in non-MVI groups. Consequently, our results suggested that the Sorcin expression could be used as a useful indicator for prognostic assessment, especial for early HCC that prognosis is very difficult to predict using conventional clinical indexes, which would substantially improve assessment of tumor prognosis and thus guide therapeutic strategy.
Our results also have shown that overexpression of Sorcin in HCC cells significantly enhanced cell proliferation, migration and invasion in vitro, and facilitated tumor growth and metastasis in vivo, whereas knockdown of Sorcin in HCC cells significantly decreased cell proliferation, migration and invasion in vitro, and inhibited tumor growth and metastasis in vivo. These results consisted with the involvement of Sorcin in progression in other malignancies 11,13 , suggesting that Sorcin possesses oncogenic properties in HCC.
Having identified the role of Sorcin in HCC, we continued to explore the mechanism that Sorcin promote metastasis of HCC. With a multi-pathway reporter array, we screened out ERK signaling. ERK signaling has been widely associated with cell proliferation, differentiation, migration 23 . The core components of ERK signaling are ERK1 and ERK2 24 , and can be activated by stimuli, including growth factors, cytokines, ligands for heterotrimeric guanine nucleotide-binding protein (G protein)-coupled receptors, transforming agents, carcinogens, et al. 23,24 . Our results also provide a link between Sorcin and activation of ERK1/2. Data showed that the level of phosphorylated ERK1/2 was increased when Sorcin was up-regulated in HCC cells. In contrast, an obvious decrease of phosphorylated ERK1/2 was observed in Sorcin knockdown cells. U0126, an effective ERK pathway inhibitor, effectively decreased the levels of phosphorylated ERK induced by Sorcin. By functional experiment, we observed that U0126 treatment obviously reduced the effect of Sorcin on the invasiveness and proliferation in Sorcin-overexpressed cells HepG2 Sorcin or high Sorcin expression cells HCCLM3 shcontrol , but it had little effect on HepG2 Vector or HCCLM3 shSorcin cells. However, previous study found U0126 exerted the inhibition on growth of the parent HepG2 cells 25 . This inconsistent phenomenon may be associated with the following reasons: one reason, as previous studies reported, HCC cell proliferation is association with multiple signal pathways 26,27 . Through selecting stable transfected cells by puromycin for 2 weeks, other signal pathways except for ERK signal pathway in HepG2 Vector cells were activated and actively exert theirs function to make HepG2 Vector resist cell death and enable cells proliferation. Second reason, HepG2 cells exist intra-cell heterogeneities. As microenvironmental discrepancies have been illuminated to affect cell plasticity and heterogeneity in cancer 28 , different culture condition (such as selecting stable transfected cells by puromycin for 2 weeks) may result in heterogeneity of HepG2 Vector cell proliferation activity after cells passaged for several population doublings, comparing with its parent HepG2 cell that other studies used, though short tandem repeat (STR) DNA fingerprinting was used to authenticate all cell lines before the study. Third reason, the introduction of puromycin-resistant gene might make the HepG2 Vector cells insensitive to U0126. Accumulated evidence suggests that activation of ERK1/2 is also linked to an imbalance of intracellular Ca 2+ levels 29,30 . Sorcin, a Ca 2+ -binding protein with multiple E-F hand domains, regulates intracellular Ca 2+ homeostasis by relocates Ca 2+ from the cytoplasm to the sarcoplasmic reticulum 31 . However, additional researches regarding a role of Sorcin-induced disruption of Ca 2+ homeostasis linked to the activation of ERK in HCC are required.
EMT has been generally considered as a hallmark of cancer and found to play a crucial role in facilitating cancer cell invasion and migration 21 . In this study, we found ectopic Sorcin expression changed the epithelial morphology cell to mesenchymal phenotype, decreased the epithelial marker expression and increased the mesenchymal markers expression. However, Sorcin knockdown showed opposite impacts. These findings were confirmed by analysis of the expressions of Sorcin, E-cadherin and vimentin in xenograft tumors in vivo. Data showed that comparing with HepG2 Vector cell-derived tumors tissues, the tissues of HepG2 Sorcin cell-derived tumors showed decrease in E-cadherin staining and increase in vimentin staining. On the contrary, orthotopic liver xenograft tumors derived from HCCLM3 shSorcin cells exhibited high level of E-cadherin expression and low level of vimentin expression, whereas orthotopic liver xenograft tumors derived from HCCLM3 shcontrol exhibited low level of E-cadherin expression and high level of vimentin expression. Obviously, there were more fibroblast-like cells with positive vimentin staining in the tissues of HepG2 Sorcin cell-derived tumors than in those of HepG2 Vector cell-derived tumors, while the number of fibroblast-like cells with positive vimentin staining in tissues of HCCLM3 shSocrin cell-derived tumors was dramatically smaller than in those of HCCLM3 shcontrol cell-derived tumors. It has been proposed that fibroblast-like cells, named tumor-associated fibroblasts (TAF) in tumor microenvironment 32 , undergo an epithelial-mesenchymal transition (EMT) of carcinoma cells, during which cancer cells lose their epithelial properties and acquire a mesenchymal phenotype that consequently favors increased invasive and migratory capabilities. From above all, the data indicated that Sorcin could promote EMT in vivo. Alternatively, TAF may be recruited by cancer cells from resident fibroblasts or from circulating mesenchymal progenitor cells of bone marrow origin 33 . Indeed, vimentin, a major type III intermediate filament protein is also expressed in endothelial and other mesenchymal cells 34 , so part of vimentin staining could exist in cells of tumor blood vessels or TAF recruited by tumor cells. However, distinct recruited populations of stromal fibroblasts usually morphologically identified as fibroblasts surrounding the tumors displayed a thicker capsule with a diffuse border with the tumor parenchyma 35 . So the effect of Sorcin on EMT could be clearly observed as these vimentin-positive fibroblastic cells were not considered in semi-quantitative evaluation under a microscope. By this method, the effect of Sorcin on EMT was further confirmed by analysis of the expressions of Sorcin, E-cadherin and vimentin in HCC tissues. Growing evidence has demonstrated that EMT can be triggered by activation of ERK pathway 36,37 . We also provide the evidence that Sorcin could activate ERK pathway, and thus induce EMT.
In conclusion, our study has identified for the first time the frequently aberrant expression of Sorcin in HCC tissues by quantitative real-time PCR, western blot and IHC. This expression pattern was associated with aggressive clinicopathological characteristics and poor prognosis in HCC. Furthermore, we have demonstrated that Sorcin could promote metastasis via activating ERK signaling in HCC. Collectively, our data indicates that Sorcin plays an important role in metastasis of HCC and provides a novel target for HCC drug discovery.

HCC Samples. The study was approved by the Ethics Committee of the institutional review boards of The
First Affiliated Hospital of Nanchang University, The First Affiliated Hospital of Sun Yat-Sen University, Affiliated Cancer Hospital of Xiangya School of Medicine of Central South University, Xiangya Hospital of Central South University. Prior informed consent was obtained from all participants, and it was conducted in accordance with the Declaration of Helsinki and current ethical guidelines. Three cohorts of HCC samples were enrolled in this study (Supplementary Fig. 1). Firstly, 60 pairs of frozen fresh HCC tumor tissues and corresponding adjacent nontumorous liver tissues (ANLTs) were collected after radical surgical resection at the Department of General Surgery, The First Affiliated Hospital of Nanchang University between January 2012 and December 2014. In addition, normal liver tissues were obtained from 5 patients with hepatic hemangioma for hepatic resection at the same period. These tissues used to screen the expression of Sorcin mRNA and protein were set as screening cohort. Secondly, formalin-fixed, paraffin-embedded paired HCC samples (including tumors and ANLTs) obtained from 160 HCC patients undergoing radical surgical resection at the Department of General Surgery, the First Affiliated Hospital of Nanchang University from January 2006 to December 2010 were set as training cohort. Another HCC sample cohort containing 120 multicenter samples (including HCC tumors and ANLTs) from patients who underwent resection between June 2007 and June 2010 (including 40 patients at the Department of Surgery, The First Affiliated Hospital of Sun Yat-Sen University, 40 patients at Department of Pathology, Affiliated Cancer Hospital of Xiangya School of Medicine, Central South University, and 40 patients at Department of Surgery, Xiangya Hospital, Central South University) was set as validation cohort. These two independent cohorts of subjects were used for prognostic study according to ReMARK guidelines for reporting prognostic biomarkers in cancer 14 . The inclusion criteria used for all samples enrolled in the cohorts were histopathologically diagnosed, had completed clinicopathologic and follow-up data, and without anticancer therapies or distant metastasis before the operation. Clinicopathologic characteristics of the patients in three cohorts were shown in Supplementary Table 1.
Prognostic Study. All HCC patients were regularly followed-up by the experienced and well-trained researchers. The follow-up period was defined as the interval between the date of operation and that of the patient's death or the last follow-up. The median follow-up was 38.5 months (range 3.0-108.0 months) for training cohort and 36.5 months (range 3.0-96.0 months) for validation cohort. Deaths from other causes were treated as censored cases. The recurrence and metastasis was surveillance by clinical examination, serial monitoring of alpha-fetoprotein (AFP) levels and ultrasonography (US) or computed tomography (CT) or magnetic resonance imaging (MRI) scan at a 3 months' interval. Recurrence and metastasis were diagnosed by clinical examination, serial AFP level, and US or CT or MRI scan. Disease-free survival (DFS) was defined as the length of time after liver resection during which a patient survived without sign of HCC recurrence or metastasis. Overall survival (OS) was defined as the interval between operation and death or between operation and the last observation for surviving patients. Data of conventional clinical and pathological variables were also collected for analysis, including gender, age, serum AFP level, HBsAg, liver cirrhosis, Child-Pugh classification, tumor number, tumor size, capsular formation, microvascular invasion (MVI), Edmondson-Steiner grade, TNM stage and BCLC stage. MVI was defined as tumor cells forming a thrombus in peritumoral vessels, can only be assessed after careful histological assessment of the whole surgical specimen 15,38 . The follow-up data were regularly updated in the database for each patient. Patients alive at the end of follow up or dead from causes without sign of recurrence or metastasis were censored. Cell Lines. PLC/PRF/5, Hep3B and HepG2 cells were purchased from the American Type Culture Collection (ATCC, Manassas, VA). L02 and Bel7402 cells were gifted by the Liver Cancer Institute of Xiangya hospital, Central South University (Changsha, China). MHCC97-H and HCCLM3 cells were gifted from the Liver Cancer Institute of Fudan University (Shanghai, China). Short tandem repeat (STR) DNA fingerprinting was used to authenticate all cell lines before the study. All cell lines were routinely cultured with the high glucose DMEM supplemented with 10% fetal bovine serum, and maintained in 5% CO 2 humidified incubator at 37 °C.
Vector Construction and Transfection. The lentiviral vectors (LV) encoding short hairpin RNAs (shRNAs) for Sorcin knockdown, and LV encoding Sorcin gene were purchased from OriGene Technologies Incorporation (Rockville, MD). The sequences of three shRNAs for Sorcin knockdown were as follow: Sorcin-shRNA-Seq. 1: Sense: 5′-CGACTACATCGCCTGCGTCAAACTGA-3′. Sorcin-shRNA-Seq. 2: Sense: 5′-TGACACAGTCTGGCATTGCTGGAGGATAC-3′. Sorcin-shRNA-Seq. 3: Sense: 5′-GTTACTTTGCTGT AGCTGGACAGAT-3′. HCCLM3 cells were transfected with lentiviral vectors encoding the shRNAs, and HepG2 cells were transfected with lentiviral vectors encoding the human Sorcin gene. An empty vector was used as the negative control and was designated as LV-control. The lentiviral vectors were transfected into the HCC cells with a proper multiplicity of infection (MOI). At 48 h after transfection, 2.5 μg/ml puromycin (OriGene) was added, and the cells were incubated for 2 weeks to select stable transfected cells. Overexpression or down-regulated expression of Sorcin was confirmed by qPCR and western blot ( Supplementary Fig. 4B1,2). The inhibitory efficiency of three shRNAs was validated and the Sorcin-shRNA-Sequence2 (named shSorcin in the figures) was adopted for subsequent study because of highly effective inhibition of Sorcin expression in HCCLM3.
Quantitative real-time PCR (qRT-PCR). Total RNA was extracted from HCC cell lines or fresh frozen tumor specimen by using Trizol reagent (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions. qRT-PCR was performed using the SYBR ® Green Realtime PCR Master Mix assay kit (Toyobo, Osaka, Japan) according to the manufacturer's instructions. The primers for Sorcin were as follows: forward, 5′-GGTGATCTTTCCATTGGTG-3′; reverse, 5′-TCCGCTGTATGGTTACTTTG-3′. GAPDH was used as a control using the following primers: forward, 5′-GCACCGTCAAGGCTGAGAAC-3′; reverse, 5′-TGGTGAAGAC GCCAGTGGA -3′. The results were analyzed using the 2 −ΔΔCt method as the following formula: ΔΔCt = ΔCt T − ΔCt ANLT, ΔCt = Ct Sorcin −Ct GAPDH . Western blot. Total proteins were extracted. After protein concentration being determined, the same weight of protein of each sample separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and then transferred onto PVDF membranes (Millipore, Bedford, MA). The blotted membranes were incubated with the primary antibodies and then an appropriate HRP-conjugated secondary antibody (KPL, Gaithersburg, MD) in order. Band was detected with enhanced chemiluminescence regents (Thermo Scientific, Rockford, IL). Beta-actin protein was also determined by using the specific antibody (Sigma, St Louis, MO) as a loading control. Protein expression were quantified by BandScan software (BioRad, Hercules, CA) and defined as the ratio of target protein relative to Beta-actin. Antibodies for Sorcin, ERK, p-ERK, Vimentin, E-cadherin and corresponding secondary antibodies were purchased from Santa Cruz Biotechnology (Santa Cruz Biotechnology, Santa Cruz, CA).
Multi-pathway reporter array. A Cignal Finder 10-Pathway Reporter Array (SABiosciences, Valencia, CA) was employed for finding the potential pathway by which Sorcin contributed to HCC metastasis. Cells were infected with Sorcin expression, Sorcin knockdown or corresponding control lentivirus. Relative firefly luciferase activity was calculated and normalized to the constitutively expressed Renilla luciferase. The assay was performed as protocol provided by the kit.
Immunohistochemistry. The detailed Immunohistochemistry procedures performed as we described before 17 . Tumor tissues except for surrounding fibrosis or capsule and tumor tissues nearby without necrosis were used to made paraffin-embedded tissues. Paraffin-embedded tissues were sectioned into 4 μm slides. Then, the slides were dewaxed, tissues were rehydrated, and antigen retrieval was performed by microwave-pretreated EDTA buffer (1 mM, pH 8.0) for 10 minutes. After blocking, the slides were incubated for Sorcin antibody (Santa Cruz Biotechnology) at 4 °C overnight. The slides were incubated with biotin-labeled secondary and streptavidin-peroxidase (Zhong-shan Goldenbridge Biotechnology, Beijing, China) for 30 minutes. The samples were developed using 3,3′-diaminobenzidine substrate (Zhong-shan Goldenbridge Biotechnology) and counterstained with hematoxylin (Zhong-shan Goldenbridge Biotechnology). Negative controls were without primary antibody incubation during the procedure. The stainings were examined in a blinded fashion by two independent pathologists. The final immunostaining score (IS) defined by the consistency for the grading by two pathologists. The expression levels of Sorcin were scored based on staining intensity (SI) and percentage of positive cells (PP) using the immunostaining score as described before 39 . SI was classified four grades: 0, negative; 1, weak; 2, moderate; 3, strong. PP was defined into five categories: 0, 0% positive cells; 1, 0-25% positive cells; 2, 25-50% positive cells; 3, 50-75% positive cells, and 4, 75-100% positive cells. IS = SI × PP. For categorization of the continuous Sorcin values into low and high, we chose a commonly used cutoff point for the analysis (range 0-12, cut point ≤3 versus >3) as previous study described 39 . For ERK, p-ERK, vimentin, E-cadherin, the procedures as Sorcin and immunostaining score were adopt as described elsewhere 40 . For vimentin expression scoring, vimentin-positive fibroblast-like cells in surrounding fibrosis or capsule and vimentin-positive vascular cells were not considered in semi-quantitative evaluation. Antibodies for ERK, p-ERK, vimentin, E-cadherin were all purchased from Santa Cruz Biotechnology (Santa Cruz, CA).

Cell Proliferation and Colony Formation Assays.
Methyl thiazolyl tetrazolium (MTT) assays were used to determine the level of cell proliferation. For MTT assays, cells were seeded into each well of 96-well plates at a density of 5 × 10 3 cells/well. Three wells of each group were detected every day. 100 μl fresh medium containing MTT (Sigma, St Louis, MO) 0.5 mg/ml was put into each well and incubated at 37 °C for 4 hrs, then the medium was replaced by 100 μl of DMSO and shaken at room temperature for 10 mins. The absorbance was measured at 570 nm. For colony formation assays, 500 cells were seeded into 35 mm dishes (Corning Costar Corp, Corning, NY) and cultured in 5% CO2 for 2 weeks at 37 °C. The number of colonies per dish was counted after staining with crystal violet. Only positive colonies (diameter >40 um) in the dishes were counted and compared 41 . These experiments were performed in triplicate.
Transwell Assays. Transwell migration and invasion assays were determined to HCC cell migration and invasion ability separately. For transwell invasion assays, the upper chamber of the insert was plated with matrigel (BD Biosciences, Franklin Lakes, NJ). But for transwell migration assays, the upper chamber of the insert was without matrigel (BD Biosciences). Briefly, after being preincubated with Mitomycin-C (10 μg/ml) for 1 h at 37 °C to suppress HCC cell proliferation, about 1 × 10 5 cells in serum-free medium were placed into the upper chamber of the insert. After 24 hours of incubation in 5% CO 2 at 37 °C, the cells in upper chamber were removed with cotton swabs, following fixed by 20% methanol, and then stained with a solution containing 0.1% crystal violet (Beyotime Institute of Biotechnology, Beijing, China). The number of cells that adhered to the lower membrane of the inserts was counted. For each experimental group, the assays were performed in triplicates, and five random fields were chosen for analysis. Cellular immunofluorescence. Cellular immunofluorescence was performed as we described previously 17 .
Cells were seeded into the 6-well culture plate (Corning Costar Corp) to prepare for performing cell immunofluorescence (IF). After incubating with primary antibodies, cells then incubated with corresponding fluorescence labeled secondary antibody. After stained with DAPI (Beyotime Institue of Biotecchnology, Jiangsu, China), the slides were photographed using the inverted fluorescence microscope TE-2000S (Nikon, Tokyo, Japan). Primary antibodies for E-cadherin, vimentin were purchased from Santa Cruz Biotechnology (Santa Cruz, CA). All fluorescence labeled secondary antibody were obtained from Beyotime Institute of Biotechnology (Shanghai, China). Rhodamine-conjugated phalloidin (Beyotime Institute of Biotechnology) was used to analyze to cytoskeleton. Cells grown on cover slides were fixed, then incubated with rhodamine-conjugated phalloidin (Beyotime Institue of Biotecchnology) and followed by DAPI (Beyotime Institue of Biotecchnology).
In Vivo Metastatic Assays. For the in vivo metastatic assays, HCC metastatic model in mice was constructed as previously described 42 . Briefly, 5 × 10 6 HCC cells were injected subcutaneously into the left upper flank regions of nude mouse (4 weeks of age, male, BALB/c). After 35 days, the subcutaneous tumors were removed and then divided into commensurate fragments of approximate 1 mm 3 , then implanted into the liver of nude mouse (6 mice in each group). After 7 weeks, mice were killed, and all livers and lungs were harvested. The tumor size was calculated as follows: tumor volume (mm 3 ) = (L × W 2 )/2 43 , where L = long axis and W = short axis. Then all livers and lungs were fixed with 10% phosphate-buffered neutral formalin, sectioned serially and stained with hematoxylin and eosin (H&E) for histological examination. Tumor tissues except for surrounding fibrosis or capsule and tumor tissues nearby were used to made paraffin-embedded tissue for immunohistochemistry. The expression of Sorcin, p-ERK, Vimentin, E-cadherin in orthotropic liver tumor tissues was also determined by immunohistochemistry. Mice were performed and housed in the Animal Institute of Nanchang University according to the protocols approved by the Medical Experimental Animal Care Commission.
Statistical Analysis. All data were analyzed using the statistical software SPSS 18.0 for Windows (SPSS Inc., Chicago, IL). The differences between groups were analyzed by Student's t test between two groups or by one-way analysis of variance (ANOVA) in more than two groups when the variance is homogeneous. If the variance is not homogeneous, the differences between groups were analyzed by Mann-Whitney U test between two groups or by Kruskal-Wallis H test in more than two groups. χ 2 analysis was used to analyze the correlation between Sorcin expression and clinicopathologic features. Spearman's rank analysis was used to analyze the correlations between different protein expressions level. Survival curves were constructed by the Kaplan-Meier method and compared by the log-rank test. The Cox proportional hazards regression model was established to identify independent factors for overall survival (OS) and disease-free survival (DFS) of HCC patients. All the tests were two-tailed and P < 0.05 was considered statistically significant.