Introduction

Gastric cancer (GC) is decreasing but is still the second largest cause of cancer death worldwide during the past decades [1]. While surgical resection and adjuvant therapy have made significant treatment achievements for the well-confined primary GC, the overall prognosis of metastatic GC patients is still dismal because of its resistance of disseminated tumor cells to existing therapeutic agents [2, 3]. Metastases represent the end-stage of a multi-step biological process, which comprises multiple sequential steps: dissemination of cancer cells to anatomically distant organ sites and their subsequent adaptation to foreign tissue microenvironments [4–6]. Recent advances have provided provocative insights regarding these cell-biological and molecular changes and established new paradigms that are likely to guide future research on metastasis [4]. Unraveling the factors driving this process is thus also important for future therapeutic interventions as well as providing novel biomarkers for prognosis and treatment prediction [7].

Protein p21, also known as cyclin-dependent kinase (CDK) inhibitor 1 or CDK-interacting protein 1, is the first identified potent inhibitor of cyclin/CDK complexes [8]. Previous studies showed that the nuclear p21 binds to and inhibits the activity of cyclin dependent kinases CDK1 and CDK2 and blocks the transition from G1 phase into S phase or from G2 phase into mitosis after DNA damage [9]. The nuclear p21 is also found as an important protein for the induction of replication senescence as well as stress-induced premature senescence [10]. Several studies have shown that loss of p21 correlate with carcinogenesis and a poor prognosis in several solid tumors such as squamous cell carcinoma of the lung, colorectal cancer, ovarian cancer, cervical cancer, and head and neck cancers [11–15]. Other studies, in breast, esophageal squamous cell and cervical carcinoma, concluded that increased p21 expression was associated with tumor progression [16–18]. Emerging evidence pointed to a dual role of p21, as a tumor suppressor in the nucleus and as a factor of tumor invasion and metastasis [19]. Subsequent cancer cell studies showed that p21 is phosphorylated in the nucleus by AKT, then translocate into the cytoplasm and inhibit apoptosis [20]. Moreover, cytoplasmic p21 was shown to have anti-apoptotic effect through binding to and inhibiting caspase 3, as well as the apoptotic kinases ASK1 and JNK [21, 22]. In line with these findings, high cytoplasmic p21 was also positively correlated to low survival in breast cancer patients [23] and related to cisplatin resistance in human testicular and ovarian cancer [24, 25]. Although a large body of studies show that low nuclear p21 is associated with poor prognosis in several cancers including GC [21, 26], the prognostic and predictive value of the cytoplasmic p21 expression for GCs is still unclear.

In this context, we used tissue microarray (TMA) to evaluate the subcellular p21 expression by immunohistochemistry (IHC) in three large independent cohorts of patients with GC, which were described in our previous study [27]. Expression of nuclear and cytoplasmic p21 in tumor vs. normal tissues and the correlations between nuclear and cytoplasmic p21 expression and clinicopathological features and their potential prognostic and predictive value for GC patients treated with surgery alone were analyzed. In addition, we further investigated the role of subcellular p21 in GC cell migration and invasion.

Materials and methods

Patients and specimens

There independent retrospective GC sets were studied, and details were provided in our previous study [27]. In brief, the training included 103 patients who underwent radical gastrectomy were recruited from Nantong Cancer Hospital, Nantong City, in the North of Jiangsu Province, China, from 1st May 1990 to 1st June 1995. The testing cohort included 640 surgical patients from Nantong Cancer Hospital from 1st December 2000 to 1st April 2005. The validation cohort consisted of all 1022 surgical cases in Yixing People’s Hospital, Yixing City, in the South of Jiangsu Province, China from 1st January 1999 to 31st December 2006.

The TMA of three GC cohorts were constructed by the GC samples. Due to missing data, samples lost during antigen retrieval or with no tumor cells present in the core or patients who received adjuvant chemotherapy, finally samples from 82 GC patients in the training cohort, 374 cases in testing cohort and 366 cases in the validation cohort were used to evaluate the nuclear and cytoplasmic p21 expression (Tables 1, 2). In addition, 5 pathologically confirmed GC and respective noncancerous fresh-frozen gastric mucosa tissues from recent patients from Yixing Hospital were obtained for Western blot analysis after signed informed consent. Institutional approval was obtained from the Review Board of the respective institutions prior to this study.

Table 1 Relationship between expression levels of nuclear p21 and clinicopathologic features of the individuals in three cohorts of gastric cancers treated with surgery alone
Full size table
Table 2 Relationship between expression levels of cytoplasmic p21 and clinicopathologic features of the individuals in three cohorts of gastric cancers treated with surgery alone
Full size table

Overall survival (OS) was the primary end-point of this analysis. Survival time was calculated from the date of surgery to the date of death or to the last follow-up. Date of death was obtained from patient records or patients’ families through follow-up telephone calls, and was double-verified by local civil affairs department and public security department. Detailed clinicopathological information was obtained from medical records of each case after approved by the ethic committee of the hospital. The histological types of GC were classified according to Lauren [28] and staged according to the Tumor, Node, Metastasis (TNM) guidelines [29].

Construction of TMAs and immunohistochemistry (IHC)

The construction of TMAs and IHC were carried out as described previously [27]. A monoclonal rabbit anti-p21 (1:1000 dilution; Cell Signaling Technology, MA, USA) was used for primary antibody incubation at 4 °C overnight. The slide without primary antibody incubation was used as negative control.

Assessment of immunohistochemistry

Nuclear and cytoplasmic staining of p21 in tumor and non-tumor tissue was scored independently by two pathologists blinded to the clinical data, by applying a semi-quantitative immunoreactivity score (IRS), as reported previously [27]. Nuclear and cytoplasmic staining of p21 in tumor and non-tumor tissue was scored by applying a semi-quantitative IRS. Category A documented the intensity of p21 immunostaining as 0–3 (0, negative; 1, weak; 2, moderate; 3, strong) (supplementary Fig. S1). Category B documented the percentage of immunoreactive cells as 1 (0–25 %), 2 (26–50 %), 3 (51–75 %), and 4 (76–100 %). Multiplication of category A and B resulted in an IRS ranging from 0 to 12 for each tumor or non-tumor. The concordance for IRS staining scores of the nuclear and cytoplasm p21 between the two pathologists was 77 (94 %) in 82 and 75 (91 %) in 82 cases in the training set and the few discrepancies were resolved by consensus using a multihead microscope. The variability in p21 staining was 4 (5 %) in the duplicate cores of 82 tumors.

The optimum cutoff value of IRS was obtained by receiver-operator characteristic (ROC) analysis [30], the area under the curve (AUCs) at different cutoff values of the nuclear and cytoplasmic p21 IRS for different years of OS time were calculated in the training cohort. The optimum value of cutoff points of the nuclear and cytoplasmic p21 IRS was 4 and 2, respectively due to the predictive value of this cutoff point for death was the best in the training cohort (supplementary Fig. S2). Under these conditions, for nuclear p21, samples with IRS 0–4 and IRS 5–12 were classified as low and high expression, and for the cytoplasmic p21, samples with IRS 0–2 and IRS 3–12 were classified as low and high expression, respectively. After establishing the immunohistochemical assessment criteria in the training cohort, the expression of p21 in the other cohorts were scored by the same pathologists with the same cut-off point.

Cell line

Human GC cell line AGS was purchased from the ATCC (Manassas, VA, USA). Cells were cultured in RPMI 1640 medium supplemented with 100 U/ml penicillin, 100 μg/ml streptomycin and 10 % fetal bovine serum. The cells were grown at 37 °C in the presence of 5 % CO2 in a humidified incubator.

Constructs, mutagenesis and transient transfections

The p21 cDNA construct was derived from Flag-p21 plasmid (Addgene, Cambridge, MA, USA), then inserted into pDsRed1-C1 construct (Clontech, CA, USA) to produce pDsRed1-C1-p21 wild type (WT) plasmid. The pDsRed1-C1-p21AAA140–142 mutant construct was generated by site-directed mutagenesis using the following primers: 5′-GAGACTCTCAGGGTGCGGCCGCGCGGCAGACCAGCATGA-3′ (amino acids: RKR140–142 to AAA140–142, mutated nucleotides in italic and bold) according to the previous study [31]. All constructs were confirmed before use by DNA sequencing. The plasmids were transiently transfected into cells using Lipofectamine 2000 transfection reagent (Invitrogen, Shanghai, China) following the manufacturer’s protocol.

Cell migration and invasion assays

The migration and invasion assays were carried out using a modified two-chamber plate with a pore size of 8 μm, which was carried out as described previously [32]. The transwell filter inserts were coated without or with matrigel for the cell migration and invasion assays, respectively. 5 × 104 cells were seeded in serum-free medium in the upper chamber. After 24 h incubation at 37 °C, cells in the upper chamber were carefully removed with a cotton swab and the cells that had traversed the membrane were fixed in methanol, stained with Crystal violet (0.04 % in water), and counted the permeating cells under the inverted microscope and photographed.

Immunofluorescent microscopy

AGS cells was transfected with plasmids of pDsRed1-C1-vector, pDsRed1-C1-p21 WT and pDsRed1-C1-p21 mutant for 48 h, immunofluorescence was carried out as described previously [33]. Then, the cells were incubated with the monoclonal rabbit anti-β-catenin antibody (1:1000 dilution; Epitomics, California, USA) in PBST overnight at 4 °C. After rinsing with PBST, the dishes were further incubated with fluorescein isocyanate (FITC)-conjugated anti-rabbit IgG secondary antibody (KPL Inc., MD, USA) at a 1:200 dilution in PBST for 1 h at 37 °C, and cell nuclei was counterstained with 5 μg/ml 40, 60-diamidino-2-phenylindole (DAPI) (1:1000 dilution; Beyotime, Jiangsu, China). The intracellular distributions of the target proteins were analyzed using confocal fluorescent microscopy. Confocal images were sequentially acquired with Zeiss AIM software on a Zeiss LSM 510 confocal microscope system (Carl Zeiss Inc., Thornwood, NY, USA) with excitation at 488 nm (for FITC), 543 nm (for DsRed) and 340 nm (for DAPI).

Subcellular fractionation and Western blotting

Cytoplasmic and nuclear extracts were obtained using the NE-PER Nuclear and Cytoplasmic Extraction Reagents (Pierce Biotechnology, Rockford, IL, USA). Western blotting was carried out as described previously [27]. The monoclonal rabbit anti-p21 antibody, Histone H3, Tubulin (1:1000 dilution; Cell Signaling Technology, MA, USA), E-cadherin (1:1000 dilution; Epitomics, California, USA), β-catenin (1:10000 dilution; Epitomics, California, USA), and monoclonal mouse anti-β-actin antibody (1:2000 dilution; Boster Biotechnology, Wuhan, China) were used for the primary antibody. Immunoreactive bands were detected with a Phototope-HRP Western blot detection kit (Cell Signaling Technology, MA, USA).

Statistical analysis

The association between nuclear and cytoplasmic p21 expression and clinicopathological parameters were evaluated by Fisher’s exact test. The significance of nuclear and cytoplasmic p21 staining in primary tumors compared with paired non-tumors was assessed by the paired Wilcoxon test (raw scores). Probability of differences in OS as a function of time was ascertained by use of the Kaplan–Meier method, with a log-rank test probe for significance. Univariate and multivariate Cox proportional hazards regression analysis were performed to estimate the crude hazard ratios (HRs), adjusted HRs and 95 %CI of HRs. We then analyzed the predictive value of the parameters using time-dependent ROC curve analysis for censored data and calculated AUC of the ROC curves [34]. We evaluated the performances of different scores by plotting [t, AUC (t)] for different values of follow-up time (t). All the statistical analyses were performed by STATA statistical software (version 10.1; StataCorp, College Station, TX, USA), and R software (version 2.10.1; The R Foundation for Statistical Computing). A p value of <0.05 was deemed statistically significant, and all tests were 2-sided.

Results

Subcellular p21 expression in gastric cancer versus adjacent normal tissues

We investigated p21 expression by IHC using TMAs in 82 GC patients that included GC and paired normal tissues in the training cohort. It was shown that p21 staining was mainly localized in the nuclei in the normal tissues (Fig. 1a), whereas p21 was expressed in both nuclei and cytoplasm in the cancerous tissues (Fig. 1b–d). The distribution of the differences of IRS for nuclear and cytoplasmic p21 expression in non-tumors and matched tumors were shown in Fig. 1e and f. The expression of nuclear and cytoplasmic p21 were significantly decreased in 63 of 82 (77 %) tumors and increased in 51 of 82 (62 %) cases compared with the matched normal gastric tissues, respectively (both p < 0.001) (Fig. 1e, f). The differential expressions of nuclear and cytoplasmic p21 in gastric tumors and paired normal gastric mucosa were validated by Western blotting, which showed that decreased nuclear p21 and increased cytoplasmic p21 expression in gastric tumors compared with the paired normal gastric mucosa (Fig. 1g).

Fig. 1
figure 1

Subcellular p21 expression in gastric cancer vs. adjacent normal tissues. a–d Representative images of immunohistochemical staining of TMA with p21 antibody. Positive nuclear p21 staining in normal gastric tissues (a), positive nuclear (b), cytoplasmic (c) and both nuclear and cytoplasmic p21 (d) staining in tumor tissues, original magnification, ×200. e, f The distribution of the difference of nuclear and cytoplasmic p21 staining (Δ IRS = IRST − IRSN). p21 staining was available from 82 pairs of tissues. p values were calculated with the Wilcoxon test. IRS immunoreactivity score, N normal tissue, T tumor tissue. g The differential expression of nuclear and cytoplasmic p21 expression in gastric tumors and paired normal gastric mucosa were examined by Western blotting. N normal tissue, C gastric cancer tissue

Full size image

In all three independent cohorts of patients treated only with surgery, the expression of nuclear p21 negatively correlated with cytoplasmic p21 expression in the cancerous tissues (p < 0.001 for all correlations, Table 1).

Association of subcellular p21 expressions with clinicopathological features in patients

In the training cohort, we found that nuclear p21 expression was negatively correlated with lymph node metastasis, distant metastasis, TNM stage, histological type (all p < 0.05). This was confirmed in the testing and validation cohorts except for distant metastasis (Table 1). The depth of invasion was significantly related to nuclear p21 expression in the two large sample cohorts. Simultaneously, Fisher’s test was used to reveal the relationship between expression levels of cytoplasmic p21 and clinicopathological features of the individuals in three cohorts of GCs treated with surgery alone (Table 1). In the training cohort, we found that cytoplasmic p21 was positively associated with lymph node metastasis, distant metastasis and TNM stages (all p < 0.01), which were also validated in the two larger cohorts. Moreover, the cytoplasmic p21 expression was significantly correlated with the depth of invasion in the testing and validation cohorts, but not with histological type in three cohorts. In addition, both nuclear and cytoplasmic p21 expression were not correlated with age, gender in all three cohorts (Tables 1, 2).

Correlation of subcellular p21 expression and OS in gastric cancer patients

We constructed Kaplan–Meier survival curves to study whether nuclear or cytoplasmic p21 expression was correlated with OS of GC patients. Our data revealed that the patients with low nuclear p21 expression had poor OS in the training GC cohorts (N = 82) (log-rank test, p < 0.001) (Fig. 2a). These findings were confirmed in two independent and larger (N = 374 and N = 366, respectively) cohorts of GC patients with minimum 5 years follow-up (both p < 0.001, Fig. 2d, g). On the contrary, high cytoplasmic p21 expression was associated with poor OS in three GC cohorts (all p < 0.001, Fig. 2b, e, h). When combining these two markers (nuclear and cytoplasmic p21 expression) as a new variable, we found that the cases with high nuclear/low cytoplasmic p21 expression had the best outcome but those with low nuclear/high cytoplasmic p21 expression had the worst outcome of survival in all cohorts (all p < 0.001 for the three GC cohorts, Fig. 2c, f, i).

Fig. 2
figure 2

Kaplan–Meier curves depicting overall survival according to expression patterns of nuclear p21 (a, d, g), cytoplasmic p21 (b, e, h), and combined with nuclear and cytoplasmic p21 (c, f, i) expression in training cohort (a–c), testing cohort (d–f) and validation cohort (g–i). p values were calculated with the log-rank test. N nuclear p21, C cytoplasmic p21

Full size image

Next, we examined whether nuclear or cytoplasmic p21 expression and the combination of the nuclear with cytoplasmic p21 expression were independent markers for GC prognosis. We performed multivariate Cox regression analysis to study the effects of all factors on patient survival in three GC sets (Table 3). In multivariate Cox regression analysis including six variables (age, gender, TNM stage, histological types, tumor diameter and p21 expression), our results indicated that nuclear and cytoplasmic p21 expression were independent prognostic factors separately or together for GC in all three cohorts (p < 0.001 for all, Table 3). Moreover, combination of low nuclear p21 and high cytoplasmic p21 expression was a more unfavorable prognostic factor than low nuclear p21 or high cytoplasmic p21 alone for patients with surgery alone in two larger cohorts (HR = 22.8 and HR = 2.38 in the testing cohort and validation cohort, respectively, Table 3).

Table 3 Multivariate Cox regression analysis of p21 expression and clinicopathologic variables predicting survival in three cohorts of gastric cancers treated with surgery alone
Full size table

A time-dependent ROC analysis for the censored data was used to further evaluate the prognostic efficacy of subcellular p21 expression. This analysis indicated that the combination of the clinical risk score (TNM stage, histological type and tumor diameter) and nuclear or cytoplasmic or combination of p21 expression contributed much more than either one alone in testing cohorts, which showed that the AUC at year 5 was 0.706 (95 % CI = 0.651–0.760) for clinical risk score, whereas it was significantly increased to 0.930 (95 % CI = 0.903–0.957) when combining the clinical risk score with nuclear plus cytoplasmic p21 risk score (Fig. 3). Additionally, this effect was obvious for 3-year prediction in training cohort, whereas it was not significant in the validation cohort due to the relatively higher AUC (about 0.8) of clinical predictors (supplementary Fig. S3a, b).

Fig. 3
figure 3

Time-dependent ROC analyses. Figures show the time-dependent ROC analysis for the clinical risk score (TNM stage, histological type and tumor diameter), the combined nuclear p21, cytoplasmic p21 or nuclear plus cytoplasmic p21 and clinical risk score in the testing cohort. AUC area under the curve

Full size image

Subcellular p21 expression regulates GC cells migration and invasion in vitro

Based on the finding in our GC cohorts that nuclear and cytoplasmic p21 expression was negatively or positively associated with depth of invasion and lymph node metastasis of GC patients, respectively, the functional roles of subcellular p21 expression were further addressed in the GC cells in vitro. Wild-type p21 (p21 WT) and the RKR140–142 mutant (p21 Mut) were fused to DsRed and transfected to AGS cells, it was shown that p21 WT and Mut were mainly expressed in nuclei and cytoplasm, respectively (Fig. 4a). To test the effect of subcellular p21 expression in GC metastasis, we performed migration and invasion assays in vitro. Our results showed that AGS cells with high nuclear p21 expression decreased the ability of cell migration and invasion by 41 and 43 %, respectively, compared with the vector control cells (Fig. 4b–d). However, AGS cells with high cytoplasmic p21 expression increased the ability of cell migration and invasion by 1.61- and 1.82-fold, respectively, compared with the vector control cells (Fig. 4b–d).

Fig. 4
figure 4

Subcellular p21 expression regulates GC cells migration and invasion in vitro. a Intracellular localization of the p21 protein fused to DsRed. AGS cells were transfected with pDsRed1-C1-p21 wild type (WT), pDsRed1-C1-p21AAA140–142 mutant (Mut) or pDsRed1-C1 vector (Ctrl) for 48 h. Localization of expressed fusion proteins was analyzed by confocal microscopy. Examples of typical localizations of the various p21 proteins are shown. b Cell migration and invasion assay of AGS cells transfected with WT, Mut, or Ctrl plasmids. The representative images of invading AGS cells in the insets of transwell chambers (original magnification, ×200). c, d Quantification of cell migration and invasion. Quantitation of the results was shown in the bar graph with mean ± SD from three independent experiments. * p < 0.05, ** p < 0.01. e AGS cells were transfected with WT, Mut or Ctrl plasmid for 48 h. The abundance of endogenous E-cadherin and β-catenin and the efficiency for p21 WT and Mut transfection were examined by immunoblotting. f AGS cells were transiently transfected with p21 WT, Mut or Ctrl plasmid. After 48 h, the transfected cells were fixed, and immunostained with anti-β-catenin (green), as indicated by the double arrows; the single arrows denote that the p21 WT protein located in nucleus and p21 Mut protein located in cytoplasm. Representative images were photographed and colored using a Zeiss LSM 510 confocal microscope system

Full size image

It has been reported that overexpression of Akt-activated p21 accelerates tumor onset and promotes lung metastasis in MMTV/neu mice through decreasing E-cadherin expression [35]. However, our results showed that neither nuclear nor cytoplasmic p21 could regulate E-cadherin expression (Fig. 4e). However, we found that nuclear p21 decreased but cytoplasmic p21 increased the expression of β-catenin by Western blotting in AGS cells that lost their p21 expression (Fig. 4e). This result was confirmed by immunofluorescence assay, which showed β-catenin was lost in nuclear p21 expressed cells, whereas increased and co-localized with cytoplasmic p21 in p21 mutant transfected cells (Fig. 4f).

Discussion

The combat of improving survival of resectable and metastatic GC has shown some progress by more invasive surgery and combining toxic chemotherapy with targeted treatment [36, 37]. However, with the exception of HER-2 positive tumors, in terms of OS the improvements are small and confined to ill-defined subgroups. Thus, identification of novel prognostic and predictive biomarkers as well as treatment targets is urgent for GC patients. In this study we identified a novel role of subcellular p21 in prognosis of resectable GC and the invasion and migration of GC cell lines.

Metastasis reflects late stage disease and underlying complex biological processes [4]. Within this process, p21 presumably plays an important role. The function of the p21 protein depends on its localization in the cell and plays different roles in nucleus and in cytoplasm. In the nucleus, p21 can function as a tumor suppressor and inhibit cell proliferation and promote cell apoptosis [9], whereas in the cytoplasm, p21 expression inhibits apoptosis, promoting proliferation and tumor metastasis [21, 22, 25, 35]. Previous studies have showed that p21 expression is down-regulated in GC, and associates with poor survival [21, 26]. However, the prognostic and predictive value of the cytoplasmic p21 expression for GCs is still unclear.

In this study, we investigated the roles of nuclear and cytoplasmic p21 expression in GC progression and prognosis. In normal gastric mucosa, we demonstrated that p21 staining was mainly in the nucleus, but in gastric tumors, p21 expression was located both in nuclei and cytoplasm. Consistent with previous reports, the nuclear p21 expression was down-regulated and negatively correlated with clinicopathological features and OS of GC patients [21, 38]. The novel finding here was that high cytoplasmic p21 expression was positively associated with advanced TNM stage, depth of invasion, lymph node metastasis, distant metastasis and shorter OS. When these two markers were combined, we found that the cases with high nuclear p21 and low cytoplasmic p21 expression had the best survival outcome. Furthermore, multivariate Cox proportional hazards regression analysis showed that nuclear and cytoplasmic p21 expression were independent prognostic factors alone or together. The time-dependent ROC analysis showed that the combination of the traditionally clinical parameters and subcellular p21 expression contributed much more prediction efficiency than clinical factors alone. Moreover, ROC analysis also showed a higher predictive value of nuclear p21 plus cytoplasmic p21 expression than nuclear or cytoplasmic p21 expression alone to OS based only on the clinical factors.

In this study, we further explored the roles of subcellular p21 expression in GC cell lines, and found that high nuclear p21 expression significantly inhibited, while cytoplasmic p21 promoted GC cell migration and invasion in vitro, which was consistent with the findings in the GC cohorts. Other researchers showed that amplification or over-expression of HER-2/neu in cancer cells confers resistance to apoptosis and promotes cell growth through requiring the activation of Akt, which associates with p21 and phosphorylates it at threonine 145, resulting in cytoplasmic localization of p21 [39]. Moreover, cytoplasmic phosphorylated p21 at threonine 145 has an oncogenic role in promoting mammary tumorigenesis and metastasis in vivo [35]. In GC, overexpression of HER-2/neu is a clinically relevant target, and its relation to p21 in GC could be of importance. Others identified that cytoplasmic phosphorylated p21 at threonine 145 down-regulates E-cadherin and promote mammary tumorigenesis and metastasis in vivo [35]. It is well known that E-cadherin/β-catenin complex is the notate biomarker of the epithelial-to-mesenchymal transition (EMT) and plays a key role in cancer metastasis [40, 41]. Here we determined whether p21 regulated GC cell migration and invasion via E-cadherin/β-catenin signaling, and interestingly found that nuclear or cytoplasmic p21 expression abrogated or increased β-catenin expression, but neither of them regulated E-cadherin expression. The β-catenin plays a crucial role in cancer development and, can regulate some oncogene expression and increases tumor adhesion and invasion [42]. The accumulation of β-catenin in the nucleus is associated with poor outcome of GC patients [43]. Knockdown of the nuclear p21 expression leads to increased of β-catenin expression to promote tumor invasion and EMT [44]. These data suggest that subcellular p21 expression connected to GC cells migration and invasion may be regulated via the β-catenin pathway, but the exact molecular mechanisms need to be further addressed.

In conclusion, our results demonstrate that subcellular p21 expression significantly correlated with distinct prognosis of GC patients and may regulate GC cell migration and invasion through β-catenin. Thus, subcellular p21 may serve as a promising prognostic marker and a therapeutic target for GC. Nevertheless, despite highly significant results in such three independent GC cohorts, these markers should be validated in different ethnic populations and in prospective studies.