Meta-analysis of SIRT1 expression as a prognostic marker for overall survival in gastrointestinal cancer

Sirtuin 1 (SIRT1), a well-characterized NAD+-dependent histone deacetylase, is generally up-regulated in gastrointestinal cancers. However, the prognostic value of SIRT1 in gastrointestinal cancer remains inconclusive. Therefore, we report a meta-analysis of the association of SIRT1 expression with overall survival (OS) in gastrointestinal cancer. PubMed was systematically searched for studies evaluating the expression of SIRT1 and OS in patients with gastrointestinal cancer. Fifteen studies (six evaluating colorectal cancer, three evaluating hepatocellular carcinoma, three evaluating gastric cancer, and one each evaluating pancreatic cancer, esophageal squamous cell carcinoma, and gastroesophageal junction cancer) with 3,024 patients were finally included. The median percentage of gastrointestinal cancers with high SIRT1 expression was 52.5%. Overall analysis showed an association between high SIRT1 expression and worse OS [summary hazard ratio (sHR) 1.54, 95% confidence intervals (CI) 1.21-1.96] in gastrointestinal cancer. However, heterogeneity was observed across studies, which was mainly attributed to cancer type. Subgroup analysis revealed that SIRT1 was significantly associated with worse OS in non-colorectal gastrointestinal cancer (sHR 1.82, 95% CI 1.50-2.21), in particular in gastric cancer (sHR 3.19, 95% CI 1.97-5.16) and hepatocellular carcinoma (sHR 1.53, 95% CI 1.16-2.01), with no evidence of heterogeneity or bias. However, no association was observed in colorectal cancer (sHR 1.15, 95% CI 0.81-1.62). In conclusion, high SIRT1 expression is a potential marker for poor survival in non-colorectal gastrointestinal cancer, but not in colorectal cancer.


INTRODUCTION
Cancers of the digestive system are among the most common types of malignant tumors worldwide [1]. Despite recent advances in the treatment of these malignancies, gastrointestinal cancers, especially liver, colorectal, and gastric cancers, remain responsible for a number of cancer-related deaths [1]. The search for novel molecular prognostic biomarkers in gastrointestinal cancers has been an ongoing task in recent decades. Epigenetics is a promising field for prognostic biomarker research, because aberrant epigenetic modification underlies not only the formation but also the progression of cancers. Numerous studies in this field have linked histone deacetylases, which constitute one class of epigenetic regulators, to clinical outcomes and prognoses in cancer patients [2].
Sirtuins, which are the mammalian orthologs of yeast silent information regulator 2 (SIR2), are NAD + -dependent www.impactjournals.com/oncotarget/ Oncotarget, 2017, Vol. 8, (No. 37), pp: 62589-62599 Meta-Analysis histone deacetylases (class III histone deacetylases). Sirtuin 1 (SIRT1) is the most extensively characterized member of the sirtuin family, and has been reported to participate in various biological processes by the deacetylation of not only histones but also non-histone proteins [3]. Relative to tumorigenesis, the role of SIRT1 is equivocal. Although SIRT1 has been suggested to play a tumor-suppressive role [3], there has been convincing evidence arguing for its oncogenic properties. SIRT1 could deacetylate and suppress the function of several other tumor suppressors, such as p53 [4] and p73 [5], and increase the stability of the oncoprotein N-Myc [6]. In addition, SIRT1 has been shown to promote survival and inhibit apoptosis of cancer cells [7]. Recently, it was reported that transgenic SIRT1 expression promoted carcinogenesis in PTEN-deficient mice [8], whereas enterocyte-specific inactivation of SIRT1 reduced the tumor load in APC +/min mice [9]. These data have further argued for an in vivo tumor-promoting function of SIRT1 during cancer progression.
The deregulation of SIRT1 expression has been found in various cancers [10]. For gastrointestinal cancers, the expression of SIRT1 is generally elevated [11][12][13][14][15][16]. A number of preclinical studies have suggested that blocking SIRT1 activity might be a promising strategy for various cancers of the digestive system [17][18][19][20][21]. However, although considerable attention has been focused on the prognostic significance of SIRT1, there was no conclusive evidence for its prognostic impact in gastrointestinal cancer. Regarding colorectal cancer, a couple of studies have reported significant associations between high SIRT1 expression and poor overall survival (OS) and/or disease-free survival [15,16].
Associations of SIRT1 expression with poor survival have also been found in patients with other gastrointestinal cancers, including liver, pancreatic and gastric cancers [22][23][24]. However, several other reports either showed that there was no association between SIRT1 expression and survival outcome [25,26] or found that the high tumoral expression of SIRT1 predicted better survival [27,28].
Here, we present a meta-analysis that quantitatively summarized the existing evidence to evaluate the prognostic impact of SIRT1 expression on survival in gastrointestinal cancer. The aim of the current study was to estimate the role of SIRT1 in relation to OS in cancers of the digestive system.

Evaluation and expression of SIRT1
All included studies used immunohistochemistry techniques for the assessment of SIRT1 expression.

Overall analysis
The combined analysis of 15 studies showed that high SIRT1 expression was significantly associated with worse OS (Figure 2A), with a summary hazard ratio (sHR) of 1.54 [95% confidence interval (CI) = 1.21-1.96]. Significant heterogeneity was observed among studies [I 2 = 69.7%, P value for heterogeneity (P h ) < 0.001]. Sensitivity analysis which was conducted by omitting one study at a time from the pooled estimate, suggested that none of the individual studies substantially influenced the summary statistic ( Figure 2B).
Potential sources of heterogeneity, including cancer type, geographic area, SIRT1 expression level, primary antibody used for the immunohistochemical evaluation of SIRT1 expression, method of data extraction, confounding adjustment, and quality score, were assessed with metaregression analysis. As shown in Table 2, cancer type had a significant influence on the overall association between SIRT1 expression and OS (P = 0.037), suggesting that cancer type mainly contributed to the heterogeneity in the overall analysis. Several quantitative variables (year of publication, study size, and length of follow-up) were also included in the meta-regression analysis and were not found to be significant sources of heterogeneity.
We also conducted subgroup analyses according to patients' geographic location. As illustrated in Table 3, high SIRT1 expression conferred a significantly worse OS for gastrointestinal cancer patients from Asia (sHR 1.72, 95% CI = 1.31-2.26). However, no association was found in non-Asian patients (sHR 1.09, 95% CI = 0.74-1.58). A further subgroup analysis of Asian patients according to cancer type showed that there was a significant association between SIRT1 and worse

Publication bias
For the overall analysis of OS data, the funnel plot showed an asymmetric distribution ( Figure 4A). Evidence of significant publication bias was detected by Begg's test (P = 0.002), and by Egger's test (P = 0.013). Non-parametric "trim-and-fill" method was utilized to estimating three missing studies ( Figure 4B). After adjustment by "trim-and-fill" method, the estimated sHR was 1.38, with a 95% CI of 1.01-1.89.

DISCUSSION
SIRT1 has been generally over-expressed in gastrointestinal cancers, including liver, pancreatic, and colorectal cancers [11][12][13][14][15][16], suggesting a putative role for SIRT1 consistent with tumor promotion. However, the prognostic value of SIRT1 in these cancers remains inconclusive. The current study meta-analyzed the published data about the expression of SIRT1 in gastrointestinal cancers and its associations with patients' survival. All included studies assessed tumoral SIRT1 expression by immunohistochemistry. Furthermore, our results of metaregression and subgroup analyses suggested that the primary antibody used for the evaluation of SIRT1 expression did not influence the summary estimate. Therefore, there was consistency in the assessment process among the studies.
The results of our overall analysis indicated a significant association of high SIRT1 expression with poor OS in gastrointestinal cancer, although the detected sHR, summary hazard ratio; CI, confidence interval; P h , P value for heterogeneity.
publication bias limited the conclusion that could be drawn. Also, it should be noted that considerable heterogeneity was observed in the overall analysis. The results of the metaregression and subgroup analyses indicated that cancer type might be a major source of this heterogeneity. In noncolorectal gastrointestinal cancers, including hepatocellular carcinoma, pancreatic cancer, and gastric cancers, high expression of SIRT1 was clearly associated with worse OS, and there was no evidence of statistical heterogeneity or bias. Particularly, the result was consistent when we only included patients with hepatocellular carcinoma or gastric cancer. These results argued for a cancer-promoting function of SIRT1 during the oncogenic process in these cancers. There are several mechanisms involved in the oncogenic role of SIRT1. It has been documented that SIRT1 can establish and maintain silent chromatin via the deacetylation of histone proteins, thus protecting cells from apoptosis [34]. Additionally, SIRT1 can repress tumor suppressor genes, such as p53 and FOXO family members, either by directly binding and deacetylating these non-histone proteins or by inducing heritable CpG island methylation at the gene promoter [3,34]. Studies focusing on the non-colorectal gastrointestinal cancers that were included in this meta-analysis have found that SIRT1 could promote chemotherapy resistance [12,20,35] and enhance invasive and metastatic potential by inducing epithelialmesenchymal transition [26]. Inhibition of SIRT1 has been shown either to inhibit the growth of cancer cells or to reduce the tumor burden in animal models [17,18,21]. Additionally, blocking SIRT1 activity with specific inhibitors was suggested to reverse the chemoresistance of both gastric and pancreatic cancers [20,35]. Taken together, the results from our analyses, together with previous findings, supported that SIRT1 is not only an adverse prognostic factor but also a potential therapeutic target, for non-colorectal gastrointestinal cancers. In colorectal cancer, the overall evidence from published studies has been insufficient to establish a correlation between SIRT1 expression and OS. Our results supported the previous studies [25,29], including a recent meta-analysis [36], showing that SIRT1 is not an independent prognostic factor for survival in colorectal cancer. However, the results should be interpreted with caution due to the significant heterogeneity among colorectal studies. This heterogeneity might be partly due to the variation in patient selection among the studies. For instance, one study included more patients with colon cancer and fewer patients with metastasis [27], whereas another study recruited fewer patients with colon cancer and more patients with metastasis [16]. Stratified analyses according to demographic or clinicopathological features (such as anatomic site and disease stage), which were not conducted in this meta-analysis due to the limited number of available studies, can be conducted in the future to further assess the prognostic value of SIRT1 in colorectal cancer when more primary studies are available.
Previous experimental studies using in vitro and in vivo models of human colorectal cancer have shown that SIRT1 exhibited pleiotropic effects, i.e., tumor-suppressive and growth-promoting effects, depending on cellular context [16,37]. The heterogeneity observed in our combined analysis of colorectal studies was also possibly due to the complex role of SIRT1 in this particular cancer. However, the mechanisms involved in such different functions of SIRT1 in colorectal cancer remain unclear. Nevertheless, it is noteworthy that SIRT1 was specifically over-expressed in colorectal serrated lesions with KRAS or BRAF mutations, possibly contributing to their malignant transformation into colorectal cancer [38]. In addition, SIRT1 expression in colorectal cancer has also been associated with microsatellite instability and the CpG island methylator phenotype [25,29], both of which have been linked to prognosis and survival in colorectal cancer [39,40]. Further studies including SIRT1, as well as other molecular features, are required to assess its prognostic role more precisely.
In addition, our subgroup showed that high SIRT1 expression was significantly associated with worse OS in Asian patients, consistent with the result of our overall analysis. However, no association between SIRT1 and OS was observed in patients from non-Asian areas (including Europe and North America). These different associations may be due to the fact that most of the included studies (2/3) conducted in non-Asian areas evaluated colorectal cancer. The further stratified analysis of Asian patients according to cancer type showed a significant association of high SIRT1 expression with worse OS only in non-colorectal gastrointestinal cancer, supporting the results of the subgroup analysis including all patients regardless of geographic location. However, there was a problem of small sample sizes in the analysis of non-Asian patients. Further research is needed to investigate the prognostic role of SIRT1 in colorectal cancer and other gastrointestinal cancer in non-Asian areas.
This study had several limitations, which may affect the interpretation of some of our results. First, there was the problem of heterogeneity not only in the overall analysis but also in the subgroup analysis of colorectal cancer. Second, as we discussed earlier, SIRT1 expression has been associated with other molecular biomarkers for cancer prognosis. Inadequate adjustment for these molecular biomarkers in several included studies might have resulted in spurious associations, whereas the results of metaregression analysis suggested that whether adjusting for confounding factors was undertaken was unlikely to influence the summary statistics substantially. Third, publication bias seemed apparent in the combined analysis of all of the included studies and adjustment for this using the "trim-and-fill" method changed the summary estimate significantly. Nevertheless, no significant publication bias was detected in the subgroup analysis.
In conclusion, our meta-analysis showed that high SIRT1 expression was clearly associated with worse OS in non-colorectal gastrointestinal cancer, in particular in hepatocellular carcinoma and gastric cancer. Moreover, the current study supported the viewpoint that there is no correlation between SIRT1 expression and survival in colorectal cancer. However, this result should be interpreted with caution due to heterogeneity among colorectal studies.
Further studies with large sample sizes and stratified analyses according to clinicopathological characteristics, as well as other colorectal cancer-related molecular biomarkers, are needed to evaluate the prognostic role of SIRT1 more precisely.

Identification and selection of studies
PubMed was searched to identify studies evaluating the expression of SIRT1 and survival outcomes in cancer patients through March 2017. The search terms included "SIRT1" or "sirtuin 1" or "sir2", combined with "survival" or "prognosis" or "outcome", and combined with "cancer". In addition, we used the name of each specific digestive system cancer (for example, colorectal cancer) instead of the search term "cancer" to recognize additional studies. Peer-reviewed studies were eligible and included if they met the following criteria: 1) studies included patients with gastrointestinal cancer; 2) studies explored the relationships between the tumoral expression of SIRT1 and OS; 3) there was sufficient survival data to extract or estimate the individual hazard ratio; and 4) studies were published in English. If the data sets were duplicated, we included only the most recent study.

Data extraction
Two authors independently collected the following information of each eligible study: first author, year of publication, cancer type, number of patients, patient source, primary antibody used for the evaluation of SIRT1, proportion with high SIRT1 expression, follow-up time, survival outcome data, and variables adjusted for in the analyses. When more than one HR was provided, the most adjusted HR was collected.

Quality assessment
Methodological quality was assessed following the REMARK guideline for reporting tumor-marker studies [41]. A scoring system was developed according to previous studies [42]. Briefly, a total of eighteen items that could be grouped into five major classifications, including study design, assay method, confounders, outcome, and analysis, were extracted for assessment. Each item was scored as 0 (no matched description), 1 (partly description), or 2 (complete description). Thus, the final quality score for each of the included studies ranged from 0 to 36, with higher scores reflecting better methodological quality.

Statistical analysis
The association between SIRT1 expression and OS was presented as the HR, with a higher HR reflecting increased hazard of all-cause mortality for OS. The survival data for eligible studies were summarized by using the log HR. For studies that did not provide the numerical data for the estimation of summary statistics, the log HR was estimated based on Kaplan-Meier curves, as previously described [43]. The statistical heterogeneity was assessed by the chi-based Q-test and the I 2 test. The data from individual studies were pooled to generate the summary log HR and variance according to the heterogeneity among studies (I 2 < 50%: the fixed-effects model; I 2 ≥ 50%: the random-effects model). In addition, a sensitivity analysis, in which one study was omitted at a time from the summary estimate, was conducted to assess whether individual studies significantly influenced the summary statistic. The publication bias was evaluated by creating funnel plots, and was estimated by Egger's and Begg's tests.
All statistical analyses were performed using STATA 10.1. P values were two-tailed, and P < 0.05 was considered statistically significant.

Author contributions
YT and YG designed the research and contributed the discussion of results. SW and JJ performed the literature search, carried out the data extraction, and analyzed the data. XW and JL performed the quality assessment. JJ and YG wrote the manuscript. All authors read and approved the final manuscript.