Therefore, it is crucial to identify tumor molecular markers that are predictive of survival and metastasis and can identify ESCC patients who may benefit from surgical resection. Recently, several genes that are associated with ESCC metastasis have been identified, including Gadd45G, glioma-associated oncogene homolog 1, lysine-specific demethylase 1, maspin, PLCE1, and CACNA2D3. These studies have indicated that specific genes may contribute to the metastasis of ESCC. However, the complex mechanisms that are involved in this process are far from being understood. In mammalian cells, the eukaryotic translation initiation factor 2 a subunit is phosphorylated by different eIF2a kinases in response to a variety of stress signals, including anoxia/hypoxia, endoplasmic reticulum stress, amino acid deprivation, and oxidative stress. This phosphorylation event leads to a rapid decrease in global protein biosynthesis that is concurrent with the induced translation of genes that function to alleviate cellular damage from stress, including ATF4. To metastasize, ESCC cells must successfully complete a series of events, including the invasion of the tumor cells into the surrounding tissues, the entry of the tumor cells into systemic circulation, their survival during circulation, the extravasation of the cells to distant organs, and finally, the formation of secondary tumors. During these events, the ESCC cells must avoid stressassociated cell death and thus are prompted to metastasize. Recently, the expression of ATF4 was found to be elevated in hypoxia-induced circulating tumor cells but not in parental cells. In addition, hypoxia was found to stimulate the migration of breast cancer cells via the PERK/ATF4/LAMP3-arm of the unfolded protein response, suggesting a role of ATF4 in cancer metastasis. However, its expression and function in ESCC remains unknown. In the present study, we have determined that ATF4 expression is frequently up-regulated in ESCC tissues compared with adjacent non-cancerous epithelial samples. Using a tissue microarray, we found that ATF4 Trichostatin A overexpression correlated with the TNM stage and lymph node metastasis. In addition, positive ATF4 expression indicated poorer prognoses than negative ATF4 expression in patients with ESCC. Furthermore, we showed that ATF4 promoted the migration and invasion of ESCC cells both in vitro and in vivo. MMP-2 and MMP-7 are both essential for ATF4-induced ESCC cell invasion. Our findings highlight the importance of ATF4 dysfunction in promoting tumor progression and metastasis and implicate it as a potential therapeutic target for ESCC. To determine whether ATF4 can be used as a predictive factor of the clinical outcomes of ESCC patients, immunohistochemistry was performed using 168 paraffin-embedded primary tumor samples and paired adjacent non-cancerous samples. Positive immunoreactivity for ATF4 was observed primarily in the cytoplasm of carcinoma cells and non-cancerous epithelial cells. As summarized in Table 1, among all of the tumor samples that were analyzed, 30 demonstrated strong ATF4 staining, 43 showed moderate staining, 44 had weak staining, and 51 exhibited negative staining. In contrast, the majority of adjacent non-cancerous epithelial samples showed weak or negative staining for ATF4. Next, we explored the association between ATF4 protein expression and the clinicopathological characteristics of ESCC. The segregation of the patients into ATF4-positive and -negative groups did not reveal significant correlations with the clinicopathological parameters of age, sex, drinking habit, tumor site, or tumor differentiation. However, these groups did show significant correlations with TNM stage and lymph node metastasis. Furthermore, we investigated the correlation of ATF4 expression with prognostic data. As a result, we found that the patients with ATF4-positive ESCC had significantly worse prognoses than those that were ATF4-negative. Additionally, the overall survival rates of ATF4-positive patients were significantly lower than those that were ATF4-negative. The staining intensity of ATF4 significantly correlated with shorter overall survival times. As shown in Table 3, univariate analyses showed that overall survival correlated with TNM stage, lymph node metastasis, and ATF4 expression. Furthermore, a multivariate Cox regression analysis indicated that ATF4 expression and lymph node metastasis were independent prognostic factors for overall survival.