Clinicopathologic significance of legumain overexpression in cancer: a systematic review and meta-analysis

Since reports on the clinical significance of legumain in cancer have shown inconsistent results, we systematically evaluated clinical indicators of legumain in cancer. We searched the Cochrane Library, PubMed, Embase, and EBSCO databases and the Wangfang and CNKI databases in China by using “legumain” and (“neoplasms” OR “cancer”) as search terms. We included case-controlled studies of legumain and cancer. The quality of the studies was evaluated by using Lichtenstein’s guidelines, and valid data was extracted for analysis. In total, 10 articles were included in this study. Meta-analysis showed that legumain was overexpressed in cancer compared with in normal tissue and was higher in stage III–IV disease than in I–II disease. Moreover, legumain overexpression was correlated with poor prognosis and clinical stage. Furthermore, Cancer Genome Atlas data showed that among patients with rectal cancer, those with tumors overexpressing legumain had shorter overall survival than those in the low expression group (P < 0.05). Legumain appears to be involved in tumor development and deterioration; thus, it can potentially be developed into both a marker for monitoring and diagnosing tumors and a therapeutic target.

the invasive potential of human melanoma cells 15 . Yoshio et al. reported that LGMN decreases the invasiveness and aggressiveness of prostate cancer 16 . Additionally, an LGMN DNA vaccine showed a potential protective effect against breast cancer in mice 17 . The results of studies on the clinical significance of LGMN expression in cancer are contradictory. To clarify the clinical and pathological significance of LGMN in cancer, we conducted a comprehensive meta-analysis of all eligible case-controlled studies. We also used The Cancer Genome Atlas (TCGA) (http://cancergenome.nih.gov/) database to confirm our results.

Results
Characteristics of eligible studies. We searched the Cochrane Library, PubMed, Embase, and EBSCO databases and the Wangfang and CNKI databases in China. After reviewing titles and abstracts and analyzing the articles in detail, a total of 10 relevant articles were included in this meta-analysis, 6,13,[18][19][20][21][22][23][24][25] seven of which were in English 13,18,19,[21][22][23] and three of which were in Chinese (Table 1) 20,24,25 . One of the articles compared LGMN expression in ovarian cancer with that in both normal tissues and benign tumors 23 . Although the definitions of LGMN overexpression (LGMN High ) varied among the articles, most characterized LGMN expression in ≥ 25% of neoplastic cells as LGMN overexpression.

Meta-analyses. Cancer tissues vs normal tissues.
Five articles compared LGMN High in cancer tissue and in normal tissue 13,19,20,22,23 , including a total of 670 cancer cases (LGMN High : 51.9%) and 366 normal tissue samples (LGMN High : 13.4%). Since the studies were significantly heterogeneous (P < 0.10, I 2 = 87%), we used a random-effect model for pooled analysis, which revealed significant differences between tumor tissues and normal tissues (RR = 3.32, 95% CI: 1.38-7.96, P < 0.05; Fig. 1). We analyzed each of the five articles individually. All five studies reported that LGMN High rates in cancer tissue were higher than in normal tissue. However, the sample sources were different and the research methods were not completely consistent, likely leading to heterogeneity (P < 0.10, I 2 = 87%). Overall, in our meta-analysis, the LGMN High rate in cancer tissue was higher than that in normal tissue.  6,18,[20][21][22][23] . Since the studies were significantly heterogeneous (P < 0.10, I 2 = 73%), we used the random-effect model for pooled analysis. The results showed no significant difference between the two groups (RR = 1.02, 95% CI: 0.65-1.61; P > 0.05; Fig. 2). All six articles were included in this research; only two reported a significant difference in LGMN High rates between the well-differentiated group and moderate-poorly differentiated groups. One study reported that LGMN High rates in the well-differentiated group was lower than in the moderate-poorly differentiated group 20 . However, the other study showed different results 6 . The remaining four articles reported no significant difference in LGMN High rates between the two groups. Therefore, combined with our meta-analysis, the LGMN High rate in the well-differentiated group was similar to that in the moderate-poorly differentiated group.
Necrosis <10% vs necrosis >10%. Two articles compared the LGMN High rates in cancers with < 10% necrotic tissue (34.5%; 168 cases) and with > 10% necrotic tissue (80.7%; 192 cases) 20,22 . Since the studies were significantly heterogeneous (P < 0.10, I 2 = 97%), we used the random-effect model for pooled analysis, but found no significant difference between the two groups (RR = 0.40, 95% CI: 0.09-1.76; P > 0.05). One study found a significant difference in LGMN High rates between the two groups 20 , but the   other study did not 22 . According to the results of our meta-analysis, the LGMN High percentage in the cancer with < 10% necrotic tissue group is likely no higher than that in the > 10% necrotic tissue group.
Male vs female. Five articles compared LGMN High percentages in cancers in male patients (40.0%; 490 cases) and female patients (32.6%; 313 cases) 6,13,20,21,25 . Since the studies were not significantly heterogeneous (P > 0.10, I 2 = 34%), we used the fixed-effect model for pooled analysis and found no significant difference between the two groups (RR = 1.10; 95% CI: 0.91-1.31; P > 0.05). This meta-analysis showed that among patients with cancer, the LGMN High rate in men was not higher than in women.  13,19,24 . Since the studies were not significantly heterogeneous (P > 0.10, I 2 = 0.0%), we used the fixed-effect model for pooled analysis and found a significant difference between the two groups (RR = 0.66; 95% CI: 0.57-0.76; P < 0.05). We collected 61 cases of patients with rectal cancer from the TCGA database. We analyzed the follow-up data and tumor mRNA data (HiSeq RNASeqV2). Kaplan-Meier survival analysis revealed that the LGMN High group had significantly shorter survival compared to the LGMN Low group (P < 0.05; Fig. 5).
Sensitivity analysis. Even after eliminating one study for a low-quality grade, LGMN expression was significantly different between cancer tissues and normal tissues. Compared with the random-effect model, fixed-effect model analysis showed P < 0.05 in the N0 vs N+ groups and necrosis < 10% vs necrosis  > 10% groups, indicating unstable results. However, we found no significant differences in the P values in other groups, indicating that the results were stable.
Test of publication bias. In funnel plots of cancer tissue vs normal tissue (Fig. 6), Egger's test showed P < 0.05 and Begg's test showed P > 0.05. We considered Begg's test to be more appropriate because the data was heterogeneous. The plots for well vs moderate-poor differentiation (Fig. 7) and male vs female showed an even and symmetrical distribution (Egger's and Begg's tests: P > 0.05 for both), indicating no publication bias.

Discussion
This is the first meta-analysis to examine the association between LGMN and clinicopathological factors of cancer. We included ten studies in this analysis; three studies were in Chinese 20,24,25 and two were doctoral theses 24,25 . One study included patient specimens from a biotechnology company 20 . Specimens used in the other nine studies were from surgical patients. Patient specimens had been pathology confirmed. We found differences in the sample sources and the research methods were not completely consistent. This may have led to heterogeneity (P ≤ 0.1, I 2 ≥ 50%). The scores for two papers were low 6,20 , mainly because of poor case selection and bias. Our analysis showed that the LGMN High percentages in cancer tissues and normal tissues were significantly different, with values of 51.9% and 13.4% respectively, indicating that LGMN was overexpressed in tumors. Sensitivity analysis confirmed that the results were stable and reliable. The rank correlation test  for funnel plot asymmetry showed P > 0.05, indicating no publication bias (Fig. 6). Legumain has been found to be overexpressed in tumor and tumor-associated cells 10 , mouse kidney tumors 26 , gastric cancer 19,20 , ovarian cancer 23 , and colorectal cancer 6,22 . Legumain was also found to be overexpressed in adult zebrafish at lesion sites after injury 27 , suggesting that it was a stress protein. Overexpression of LGMN has also been suggested to affect liver carcinogenesis 12 . Additional studies are required to determine how legumain promotes tumor cell proliferation.
Our meta-analysis demonstrated that LGMN overexpression was not related to tumor differentiation or to lymph node metastasis. Six studies compared LGMN overexpression in well-differentiated and moderate-poorly differentiated cancers. One study 20 reported LGMN High rates in well-differentiated tumors were lower than those in moderate-poorly differentiated tumors. However, one study showed different results 6 . The remaining four studies reported that LGMN overexpression was not related to tumor differentiation. Rong's 25 doctoral thesis also demonstrated that the overexpression of LGMN was not associated with tumor differentiation. Additionally, sensitivity analysis showed that this result was stable. Our analysis also showed that LGMN was not associated with tumor differentiation. Four studies compared LGMN High percentages in patients without (N0) and with (N+ ) lymph node metastases. The results showed no significant difference between the two groups (P > 0.05). However, sensitivity analysis showed significant differences in P values, indicating that this result was not stable.
We found that the LGMN High percentages were not significantly different in tumors with < 10% necrotic tissue compared to in tumors with > 10% necrotic tissue. The LGMN High percentages also did not differ based on patient sex, suggesting that LGMN overexpression was not affected by hormones.
Our study also indicated that LGMN High percentages were associated with clinical stage, further supporting a possible function for LGMN as a stress protein. Sensitivity analysis confirmed that this conclusion was stable. Other studies have demonstrated an association between LGMN and tumor invasion and metastasis. Li and colleagues reported that LGMN overexpression in tumor tissue was associated with large tumor size (P < 0.01) 21 , while Guo et al. showed that LGMN overexpression was associated with hepatic metastasis (P = 0.014) 19 . Cells overexpressing LGMN were found to be more migratory and invasive in vitro 4 , indicating that LGMN expression affects metastasis and invasion. Legumain can promote cell proliferation 12 , which may be related to its effect on tumor proliferation. Knock-down of LGMN decreased cell proliferation, but did not appear to promote cell apoptosis. Knock-down of LGMN in a mouse tumor model significantly reduced tumor growth and metastasis 28 , further supporting LGMN as a potential therapeutic target. Legumain can activate zymogene progelatinase A, which mediates extracellular matrix degradation 29 , this interaction may affect tumor invasion and metastasis 4 .TBX2 is an oncogenic transcription factor that can drive breast cancer proliferation by maintaining LGMN activity 30 . The mechanism by which LGMN affects cell proliferation requires further study. Patients overexpressing LGMN showed lower 5-year survival rates than did patients with low LGMN expression, which agrees with results of reports correlating LGMN overexpression with poor prognosis in cancer 6,13,19,20,[22][23][24] . TCGA data showed that among patients with rectal cancer, the LGMN High group had significantly shorter survival than the LGMN Low group (P < 0.05), further supporting the associations between high legumain expression, tumor progression, and poor prognosis.
Our results showed that LGMN was overexpressed in tumor tissues, affected cancer development and deterioration, and was associated with poor prognosis in cancer patients. We speculated that LGMN was a stress protein. Our meta-analysis showed that LGMN overexpression percentages were no higher in the N+ group than in the N0 group; however, sensitivity analysis indicated that this conclusion was unstable. Considering that LGMN was involved in tumor development and deterioration, we doubted that LGMN was also associated with lymph node metastasis, as it was an indicator of increased malignancy. We also hypothesized that LGMN was not related to tumor differentiation, tumor tissue necrosis, or patient sex. Our analysis data support these hypotheses. These factors share a common characteristic in that they do not effect tumor progression, suggesting that LGMN's function in cancer was limited to tumor progression, unless other mechanisms were involved. Additionally, LGMN was not related to patient age, tumor location, or body mass index, as these factors were found to be unrelated to tumor progression.
Legumain is widely conserved, expressed in both plants and animals, and has unique distribution characteristics 4,6,7,16 . Although low expression of LGMN is observed in normal tissue 13 , it is overexpressed in tumor, matrix, and endothelial cells in the tumor microenvironment 14 . Legumain overexpression in rectal cancer offers both an opportunity and a challenge as a therapeutic target 6 . Knock-down of LGMN was shown to control tumors 28 , further supporting that LGMN can be used as a therapeutic target for cancer diagnosis and therapy 31 . Chen et al. developed an MRI contrast agent and a near-infrared fluorescence probe to monitor LGMN activity in tumors 32 . Numerous studies have been conducted to develop an LGMN vaccine for tumor treatment 17,33,34 . As the target of a multi-peptide vaccine in a breast cancer model, legumain effectively inhibited the tumor load 35 . We previously developed a drug carrier for tumor treatment based on the attachment of endoprotease to LGMN 36 . The TBX2-CST6-LGMN signaling pathway may be a target for the development of novel therapies against breast cancer 30

Conclusion
Our findings indicated that LGMN was overexpressed in cancer and was associated with tumor progression, deterioration, and poor prognosis in cancer. Overexpression of LGMN was not associated with tumor differentiation, tumor tissue necrosis, or patient sex. LGMN may be used as a diagnostic marker or therapeutic target.

Methods
This meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 38 . Study identification and eligibility criteria. We systematically searched the Cochrane Library, PubMed, Embase, and EBSCO databases and the Wangfang and CNKI databases in China by using the following search terms: "legumain" and ("neoplasms" [MeSH Terms] or "cancer"). We included studies published up to July 2015 with no language limitations. A study was eligible in the meta-analysis if it: (1) investigated the association between LGMN and cancer; (2) cancer cells were acquired from humans; (3) LGMN was tested by using immunohistochemical methods; (4) provided sufficient data on the relationship between LGMN expression and clinicopathological variables. We excluded studies that (1) had no control groups; (2) used animal models; (3) were case reports, letters, comments, or review articles; (4) did not allow us to extract needed data; (5) did not conduct immunohistochemical method detection of LGMN; (6) included patients who had received chemotherapy or radiotherapy before the study; or (7) included duplicated data. We also downloaded follow-up data and tumor mRNA data (HiSeq RNASeqV2) for patients with rectal cancer from the TCGA database on April 10th, 2015. After eliminating records with incomplete data, we sorted the remaining 61 cases into three groups based on LGMN expression: 21 studies with the highest expression (LGMN High ), 20 with the lowest expression (LGMN Low ), and 20 with moderate expression (moderate expression group) (Supplemental file 1).

Data extraction. For each study, two authors independently extracted the first author, publication
year, country, numbers of cases and controls, cancer types, LGMN expression, and clinical indicators. Disagreements were resolved by discussion between the two investigators.
Quality score assessment. The quality of the studies was evaluated by the Lichtenstein's guidelines for case-control studies 39 . We evaluated quality based on the following aspects to determine whether: (1) the study design was used the scientific method; (2) inclusion criteria and basic structure characteristics of the studies were clear; (3) processing factors and methods were correct; (4) statistical methods were correct; (5) the existence of bias in the research was discussed. Satisfaction of any one of the above 5 aspects earned 1 point. When a study earned three points or more, the quality was considered reliable. Study quality was evaluated by two independent researchers (Zhao Shuangtao and Ye Zhen). Disagreements between the two researchers were resolved through discussion. Statistical analysis. We used RevMan 5.2 software, provided by the Cochrane Collaboration, to perform the meta-analysis and to merge RR values. Heterogeneity was tested by chi-square analysis. If P ≤ 0.10 for the chi-square test indicated heterogeneity across studies, a fixed-effects model was used for homogeneous data (P > 0.1, I 2 < 50%) and a random effects model for significantly different data (P ≤ 0.1, I 2 ≥ 50%). Publication bias was evaluated by using a funnel plot if a sufficient number of studies (n > 4) was included. We used the Egger's and Begg's tests to evaluate publication bias by using R version 3.1.2 (http://www.r-project.org/). Stability of the data was tested by sensitivity analysis. We eliminated a low-quality document after sensitivity analysis. We also compared the use of the fixed-effect model and random-effect model for sensitivity analysis. We created a Kaplan-Meier survival curve by using the log-rank test for TCGA data to compare the LGMN High group with the LGMN Low group. SPSS 22.0 (SPSS, Inc., Chicago, IL, USA) was used for this statistical analysis. P < 0.05 was considered significant.