The prognostic value of AGR2 expression in solid tumours: a systematic review and meta-analysis

The prognostic value of anterior gradient-2 (AGR2) in tumours remains inconclusive. Here, we systematically reviewed the literature evidence and assessed the association between AGR2 expression and prognosis in solid tumours. The primary outcomes were overall survival (OS), disease-specific survival (DSS), and disease-free survival (DFS)/recurrence-free survival (RFS)/progression-free survival (PFS). All analyses were performed by STATA 12.0, with the hazard ratio (HR) or odds ratios (OR), and 95% confidence interval (CI) as the effect size estimate. A total of 20 studies containing 3285 cases were included. Pooled analyses revealed that AGR2 overexpression had an unfavourable impact on OS (HR 1.93, 95% CI 1.32–2.81) and time to tumour progression (TTP) (DFS/RFS/PFS) (HR 1.60 95% CI 1.06–2.40) in solid tumour patients. Subgroup analyses indicated that AGR2 overexpression in breast cancer patients was significantly associated with poor OS (HR 3.02, 95% CI 1.03–8.81) and TTP (HR 1.93, 95% CI 1.17–3.20). Excluding breast cancer, AGR2 overexpression was also found to have a significant correlation with poor OS in the remaining solid tumour patients (HR 1.51, 95% CI 1.04–2.19). Overall, AGR2 might be a potential biomarker to predict prognosis in solid tumour patients.

Impact of high AGR2 expression on cancer prognosis. Concerning the survival outcomes in patients with solid malignancies, 13 studies evaluated the relationship between AGR2 expression and OS, while 8 studies analysed the association of AGR2 expression with TTP and two studies with DSS. OS was indicated by the percentage of patients who remained alive at a given time point. DSS was indicated using the percentage of subjects who survived a particular disease for a defined period of time 33 . The estimated pooled HRs showed that compared with AGR2 low/ negative expression, AGR2 overexpression/ positivity was highly related to poor OS (HR 1.93, 95% CI 1.32-2.81) and poor TTP (HR 1.60, 95% CI 1.06-2.40) of solid tumour patients (Fig. 2). However, no association was found between AGR2 overexpression and DSS (HR 0.36, 95% CI 0.06-2.14). These pooled analyses were conducted using the random effects model, in which significant heterogeneity of the included studies on OS (I 2 = 83.2%, P = 0.000), TTP (I 2 = 65.2%, P = 0.003) and DSS (I 2 = 77.6%, P = 0.035) was observed, indicating that the choice of this model was appropriate (Table 2).

Subgroup analysis.
To identify the contributing factors underlying heterogeneity, subgroup analyses by cancer types and study location were conducted (Table 2). If only one study provided relevant data on the correlation of AGR2 overexpression with outcome endpoints, the subgroup analysis was not performed. The subgroup analysis by study location indicated that 13 studies reporting OS were stratified into Europe (n = 9), Asia (n = 3), and USA (n = 1); 9 studies focused on TTP were stratified into Europe (n = 6), Asia (n = 2), and Oceania (n = 1). According to the different study locations, subgroup analyses did not reveal any significant correlation between AGR2 overexpression and the OS of solid tumour patients (in Europe: HR1.96, 95% CI 0.99-3.85; in Asia: HR1.91 95% CI 0.95-3.84). Furthermore, the heterogeneity could not be reduced by location stratification either (in Europe: I 2 = 84.8%, P = 0.000; in Asia: I 2 = 86.9%, P = 0.000). However, we observed a significant   13.84). In addition, meta-analyses in prostate cancer and gastric cancer were not conducted, because the number of studies (n = 1) was insufficient. Since nearly half of the included studies focused on breast cancer, we excluded them and performed subgroup analyses for the remaining solid tumours group. AGR2 overexpression still predicted poor OS (HR 1.93, 95% CI 1.32-2.81), but not TTP (HR 1.60 95% CI 1.06-2.40) (Fig. 3). These results demonstrate that AGR2 overexpression can be the prognostic factor for the OS of solid tumour patients. Begg's test and Egger's test as well as funnel plots revealed no obvious publication bias concerning OS and TTP in the subgroup analysis for breast cancer patients ( Fig. 4c and d). However, high heterogeneity among breast cancer studies was identified for OS (I 2 = 81.2%, P = 0.000), but not for TTP (I 2 = 38.7%, P = 0.180). Thus, we set to address the heterogeneity for OS in breast cancer patients by further performing subgroup analyses on study location, follow-up time, estrogen receptor (ER) status, and sample size. The subgroup analysis showed an improved HR in the studies with large sample size ( ≥ 200, HR 4.88, 95% CI 1.60-14.85) and long follow-up time ( ≥ 80 months, HR 6.14, 95% CI 2.53-14.92). Moreover, the heterogeneity across the studies with large sample size (I 2 = 57.3%, P = 0.216) and long follow-up time (I 2 = 50.2%, P = 0.134) were found to be effectively reduced. The subgroup analysis on ER status also confirmed the unfavourable impact of AGR2 overexpression on the OS of ER positive breast cancer patients (HR 2.58, 95% CI 1.06-6.26). In terms of study location, no significant association between AGR2 overexpression and poor OS was identified in breast cancer studies from Europe (HR 2.48, 95% CI 0.79-7.81) ( Table 3).

Correlation of AGR2 expression with clinicopathological features of breast cancer.
To fully elucidate AGR2′s clinical relevance in breast cancer, we assessed the relationship between AGR2 overexpression and clinicopathological features of breast cancer patients. As illustrated in Table 4 Sensitivity and publication bias. The sensitivity analyses were performed by omitting one study at a time to gauge the robustness of our results. We found that the pooled HRs was not significantly altered by excluding any single study, demonstrating that the results of this meta-analysis are statistically robust (Fig. 5). Furthermore, the publication bias in the included studies was assessed by combining the Begg's funnel plot and Egger's test. The results indicated no evidence for publication bias, as all P value for Begg's test and Egger's test were >0.05 (Fig. 4).

Discussion
AGR2 mainly expressed in epithelial cells in human 34 . As a member of PDIs, AGR2 plays a pivotal role in maintaining endoplasmic reticulum homeostasis through regulating the unfolded protein response signalling (UPR) 35 . In addition, AGR2 has been implicated in a range of cell biological processes, in particular cellular transformation, cell migration and adhesion 36 . Moreover, AGR2 reportedly acts as a pro-oncogenic protein overexpressed in various cancers and involved in adenocarcinoma growth, cell metastasis 37,38 and drug resistance 24 . Paradoxically, many studies found that elevated AGR2 levels did not predict the poor outcomes of solid tumour patients 9,19 . Thus, the prognostic value of AGR2 in tumours remains elusive and needs to be clarified.
In this article, we systematically evaluated AGR2 expression and the survival data of 3285 solid tumour patients from 20 different studies. Overall, our results demonstrated that AGR2 overexpression could predict poor OS (HR1.93, 95% CI 1.32-2.81) and poor TTP (HR1.60, 95% CI 1.06-2.40) of all solid tumour patients. These HR estimates were robust with no publication bias. However, high heterogeneity was observed across the studies included for this meta-analysis. The sensitivity analysis in this meta-analysis did not help to elucidate the source of heterogeneity. Therefore, we conducted subgroup analysis by study location and cancer types to address the source of heterogeneity. In subgroup analysis for breast cancer, both HR estimates for OS and TTP (OS: HR 3.02, 95% CI 1.03-8.81; TTP: HR 1.93, 95% CI 1.17-3.20) further indicates that AGR2 overexpression is predictive of poor prognosis in breast cancer patients. Meanwhile, the heterogeneity test showed no obvious heterogeneity for TTP (I 2 = 38.7%, P = 0.180), but it was still significant for OS (I 2 = 83.2%, P = 0.000). Further subgroup analysis was then performed and the results suggest that heterogeneity across the studies with sample ≥ 200 (I 2 = 57.3%, P = 0.216) or with follow-up time ≥ 80 months (I 2 = 50.2%, P = 0.134) can be effectively reduced, and the correlation of AGR2 overexpression with poor OS become significantly evident. Of note, subgroup analyses for the solid tumours group with breast cancer excluded showed that AGR2 overexpression was significantly associated with poor OS, but not with TTP. Thus, AGR2 might serve as a reliable prognostic marker for the OS of pooled patients with solid tumours.
We also analysed the relationship between AGR2 and clinicopathological features of breast cancer. Consistent with the previous study showing that AGR2 expression can be induced by estrogen in ERα expressing breast cancer cell lines 39 , the upregulation of AGR2 was found significantly correlated with positive ER and PgR status as well as low histological grade. As previously reported that AGR2 expression promoted cell lineage differentiation in murine stomach 40 , the co-expression of AGR2 and ER or PgR, and the association with the low histological grade indicated that AGR2 might be a marker of differentiation in breast cancer. However, in breast cancer patients, AGR2's high expression was predictive of poor outcomes. Further functional studies are needed to clarify AGR2's role in breast cancer.
Unlike its prognostic value in breast cancer, clinical effects of AGR2 expression on other tumour types remained inconclusive due to the existence of conflicting results 9 9 . Interestingly, the other two studies focused on colorectal cancer and prostate cancer conducted by the same group (Fritzsche et al.) showed that in colorectal cancer AGR2's up-regulation was strongly associated with improved OS compared with the control, while no prognostic value of AGR2 could be detected in prostate cancer 17,19 . Of note, in contrast to other selected studies, Fritzsche et al. reported a rather opposite observation that AGR2 overexpression predicted longer overall survival time of breast cancer patients 20 . We speculate that these discrepancies might be partially caused by differences in experimental protocols, antibody concentrations, and criteria for high AGR2 expression used in different research groups. Since further analyses could not be performed due to the insufficient number of existing studies, current data led us to a speculation that AGR2 might truly function as a tumour suppressor in some tumour types 19 , which however awaited future experimental verifications, especially in lung cancer, prostate cancer, ovarian cancer, gastric cancer, and colorectal cancer. Alternatively, the different impact of AGR2 expression on solid tumours might be due to an apparently distinct regulation in certain cancer types. AGR2 is the human orthologue of the Xenopus Anterior Gradient-2 (XAG-2) protein 1 . In the Xenopus embryo, XAG-2 plays a key role in patterning the cement gland, a mucus-secreting tissue 41 . Similarly, AGR2 is predominately expressed in normal human colon, rectum, stomach, breast, and prostate 20 , which are the organs capable of secreting a variety of small molecules. Given that AGR2 was originally found in breast cancer specimen 42 , it is not surprising that almost half of the selected studies in this meta-analysis were breast cancer related. In normal mouse mammary glands, AGR2 is required for epithelial cell proliferation and lobuloalveolar development 39 . In breast cancer, AGR2 is co-expressed with estrogen receptor (ER) that directly regulates AGR2 expression 43 , and is significantly correlated with HER2 positive breast cancer 44 . Consistent with the previous studies, our study showed the co-expression of AGR2 and ER or PgR, and AGR2's association with the low histological grade might indicate that AGR2 is a molecular marker of differentiation. This is also supported by the fact that AGR2 expression promoted cell lineage differentiation in murine stomach 40 . Intriguingly, it seems paradoxical to the prognosis of breast cancer patients for whom AGR2' s high expression predicted poor outcomes. A possible explanation would be that AGR2 overexpression might increase breast cancer cells' proliferative and invasive capacity. AGR2 overexpressed tumour cells was showed to have strong propensity for disseminating to lung 39 . Thus, we speculate that the mechanism for AGR2 overexpression predicting breast cancer's poor prognosis could be: AGR2 expression was induced in an ER-or HER-dependent manner at the early stage of tumorigenesis, which led to treatment resistance and metastasis.
Some limitations of our study include: first, the standards for defining AGR2 positivity across the studies vary due to different experimental methods used for assessing AGR2 expression, which might lead to inter-study heterogeneity. Second, the sample size and the number of studies from certain cancer types, such as lung cancer, prostate cancer, ovarian cancer, gastric cancer and colorectal cancer, appear to be quantitatively insufficient. Third, as a secreted protein, the significant elevation of AGR2 in pancreatic juice from pancreatic cancer patients or in urine from prostate cancer patients suggests that AGR2 may also function extracellularly during the development of cancer 45,46 . In addition, both studies by Kyukwang Chung et al. and Valladares-Ayerbes et al. reported that the presence of detectable AGR2 in serum was significantly associated with the poor OS or DFS 25,29 . However, we could not perform a further analysis for the prognostic value of secreted AGR2, because only 2 studies observed serum AGR2 among the included studies. Although the mechanisms and the rate of AGR2 secretion from cytoplasm of tumour cells into the blood serum in different cancers might vary considerably 47 , we speculated that  assessing AGR2 expression in tumour tissue along with a blood test for AGR2 protein may potentially lead to a more accurate and comprehensive assessment of the prognostic role of AGR2 in tumours. In summary, despite the above limitations, our meta-analysis demonstrates that high AGR2 expression can serve as a prognostic predictor of OS for solid tumour patients, especially for breast cancer patients. Furthermore, in breast cancer, high expression of AGR2 trends to correlate with ER positivity, PgR positivity and low

Methods
Search strategy and selection criteria. This present meta-analysis was executed in accordance with the Preferred Reporting Items for Systematics Reviews and Meta-Analyses guidelines 48 .We performed a comprehensive literature search through the electronic databases PubMed, Embase, and Web of Science databases updated to January, 2017. We used the Medical Subject Heading (MeSH) terms and corresponding keywords to make the search strategy. The following combined search terms were used ("Anterior gradient 2"or HAG-2 or "Anterior gradient 2 homolog" or "Anterior gradient protein 2" or AGR2) AND (cancer or carcinoma or tumour or neoplasms). The references from selected articles were also examined by a hand search to find other relevant studies.
To be included in this meta-analysis, studies had to meet the following criteria: (1) The study evaluated the relationship between AGR2 expression and the prognosis of solid tumour patients, such as overall survival (OS) or disease-specific survival (DSS) or relapse-free survival (RFS) or disease-free survival (DFS) or progression-free survival (PFS) and clinicopathological features. (2) The study provided hazard ratios (HRs) with 95% confidence intervals (CIs) directly or these statistics could be calculated based on data presented. (3) A definite cut-off value to classify AGR2 expression as "positive" and "negative" or "high" and "low" was given. (4) Studies published in English or Chinese. (5) AGR2 expression was detected in tumour tissues or serum, rather than in cell lines. The exclusion criteria were as follows: (1) Reviews articles, letters, experimental studies, conference abstracts, case reports, duplicated publications or replicated samples. (2) Lacking sufficient data to estimate the HRs with 95% CIs. (3) Studies without OS or RFS or DFS or DSS or PFS for further quantification.
Two reviewers determined the eligibility of the screened studies independently. Discrepancies were solved by consensus after discussion.  Data extraction and methodological quality assessment. Two authors independently extracted relevant information from each eligible study using a standardized data collection form. The following data were collected: types of cancer, name of first author, publication year, country, the number of patients, age median, detection method, location of AGR2 expression in tumour or serum, follow-up duration, outcome endpoints, and cut-off value to determine AGR2 positivity. When the prognosis was presented only as the Kaplan-Meier curves in some studies, the Engauge Digitizer V4.1 was then utilized to obtain the survival data, and Tierney's method to calculate the HRs and 95%CIs 49 . The methodological quality of included studies was assessed by Newcastle-Ottawa Scale (NOS). The NOS consists of three quality parameters: selection (0-4 points), comparability (0-2 points), and outcome assessment (0-3 points). Studies with an NOS scores ≥ 6 were considered to be high-quality. Two reviewers performed the quality assessments separately.
Statistical analysis. We applied the HRs with 95% CIs to evaluate the impact of AGR2 overexpression on outcomes of solid tumour patients in this meta-analysis. For analysing the association between AGR2 high expression and clinicopathological features, odds ratios (ORs) with 95% CIs were assessed. Begg's funnel plot and Egger's test determined the potential publication bias among selected studies. P values were two-sided and P < 0.05 was considered to be statistically significant. Sensitivity analysis was tested to examine the stability of the pooled results. Inter-study heterogeneity was quantified using Q-tests and I-squared test 50 . In the absence of significant heterogeneity (P > 0.10 or I 2 < 50%), a fixed effects model was appropriately used to calculate the pooled effect, otherwise the random effects model was employed. All statistical analyses were performed with Stata Version 12.0 (Stata Corporation, College Station, TX, USA).