Circulating adiponectin levels in various malignancies: an updated meta-analysis of 107 studies

Early detection of cancers is challenging for lack of specific biomarkers. Adiponectin is an adipokine predominantly derived from adipocytes and hypoadiponectinemia has been reported to associate with risk of many types of cancers. However, available evidence is controversial. Some studies show that increased adiponectin levels correlate with cancer risk. Therefore, we performed a meta-analysis of the association between circulating adiponectin levels and cancer development. A systematic search of PubMed, EMBASE, Wiley Online Library and Cochrane Library was conducted for eligible studies involving circulating adiponectin and malignancies from inception to August 8, 2015. Standard mean differences (SMDs) with 95% confidence intervals (95% CIs) were calculated by use of a random-effect model. Funnel plot and Egger's linear regression test were conducted to examine the risk of publication bias. 107 studies were included with 19,319 cases and 25,675 controls. The pooled analysis indicated that circulating adiponectin levels were lower in patients with various cancers than in controls, with a pooled SMD of −0.334 μg/ml (95% CI, −0.465 to −0.203, P = 0.000). No evidence of publication bias was observed. Circulating high molecular weight adiponectin levels were also lower in cancer patients than in controls, with a pooled SMD of −0.502 μg/ml (95% CI, −0.957 to −0.047, P = 0.000). This meta-analysis provides further evidence that decreased adiponectin levels is associated with risk of various cancers. Hypoadiponectinemia may represent a useful biomarker for early detection of cancers.


INtrODUctION
Cancer, a major cause of human mortality, has been a worldwide public health problem. A variety of factors such as genetic lesions, environmental aspect and increasing adoption of unhealthy lifestyle are considered as crucial causes of cancer [1]. Among them, obesity is an important factor contributing to the occurrence and development of malignancies. According to the literature in 2005, 396 and 937 million people suffer from obese and overweight worldwide, respectively [2]. Epidemiological research reveals that obesity increases the risk of cancer with evidence that obese women have 50% higher incidence rate than normal weight women [3]. In the process of obesity, dysregulated circulating hormones and growth factors may play an important role in carcinogenesis [4]. Among them, aberrant adiponectin concentration is reported to be a vital link between obesity and cancer.
Adiponectin, firstly discovered by Scherer et al. in 1995, is an adipokine predominantly produced by adipocytes with the monomeric subunit containing 244 amino-acids in human and circulates abundantly in plasma [5,6]. Three bioactive forms of adiponectin are produced Review Oncotarget 48672 www.impactjournals.com/oncotarget after post-transcriptional process known as trimeric low molecular weight (90 kD, LMW), hexameric medium molecular weight (180 kD, MMW) and oligomeric high molecular weight (> 400 kD, HMW) adiponectin. Among them, HMW-adiponectin is the dominant form in plasma and has the most biological activity than the other two isoforms [7]. Adiponectin mainly acts on two seven-transmembrane adiponectin receptors, AdipoR1 and AdipoR2. Besides, T-cadherin is also responsible for mediating the role of adiponectin in certain tissues [8,9]. Adiponectin exerts pleiotropic functions in human health such as anti-inflammation, anti-atherosclerosis, and anti-angiogenesis. It also has the properties of insulinsensitizing and balancing glucose and lipid metabolism in various cells [10]. A number of studies reveal that circulating adiponectin levels decrease in metabolic syndrome, whereas overexpression of it can counteract metabolic dysfunctions [10]. Besides, increased weight reduces the plasma adiponectin level and decreased weight upregulates circulating adiponectin level [11].
It was first reported that circulating adiponectin level was lower in patients with breast cancer in 2003 [12]. Since then, the most clinical studies have indicated that hypoadiponectinemia is associated with risk of various cancers including prostate, endometrial, and colorectal cancers [13][14][15][16]. In addition, adiponectin has antiproliferative and pro-apoptotic effects on cultured cancer cell lines [17,18]. These results suggest that adiponectin might be an important regulator in carcinogenesis and progression of cancers. However, unchanged or increased circulating adiponectin levels in pancreatic and hepatocellular carcinoma are also reported [19,20]. Therefore, understanding the exact role of adiponectin in cancer may offer a novel target in tumor diagnosis and therapeutic strategy. In order to gain a more explicit and evidence-based conclusion on the association between circulating adiponectin levels and carcinogenesis, we conducted a comprehensive meta-analysis of current available studies.

Literature selection
The initial comprehensive search yielded 1486 articles, of which 235 articles were excluded for duplication. Then 997 studies were ruled out because of apparent irrelevance after reading titles and/or abstracts.  The remaining 254 studies were included for full-text reading, of which 151 studies were removed for one of the following reasons: (i) reviews, comments or letters (n = 37); (ii) shared population (n = 13); (iii) no report of adiponectin levels and/or SDs for both patients and controls or there was not enough information to calculate them (n = 25); (iv) not case-control study (n = 76). 4 additional studies were included from checking the references list. Finally, 107 studies met the inclusion criteria and were used for further analysis [12,13,15,16,. The flow diagram of this selection process was showed in Figure 1.

circulating adiponectin levels and carcinogenesis
Data from 107 studies were analyzed in a randomeffect model to compare circulating adiponectin levels in people with different cancers and controls. Results showed that circulating adiponectin levels in cancer cases were significantly lower than in the controls with a pooled SMD of -0.334 μg/ml (95% CI, -0.465 to -0.203, P = 0.000). Statistically significant amount of heterogeneity was observed across these studies (I 2 = 97.6%, P < 0.0001), so subgroup analysis was carried out next. These results were presented in Figure 2.
HMW-adiponectin is the dominant form of adiponectin in plasma and correlates with cardiovascular disease, insulin resistance, and obesity [7,113,114]. But few studies have evaluated the relationship between circulating HMW-adiponectin levels and cancer risk. We analyzed data from 8 studies in a random-effect model to compare circulating HMW-adiponectin levels in people with different cancers [33,56,58,72,83,94,107,108]. Results showed that circulating HMW-adiponectin levels in cancer cases were significantly lower than in the controls with a pooled SMD of -0.502 μg/ml (95% CI, -0.957 to -0.047, P = 0.000), which is consistent with the results derived from total adiponectin levels. Statistically significant amount of heterogeneity was observed across these studies (I 2 = 97.0%, P < 0.0001). These results were presented in Figure 3.
Additionally, adiponectin was significantly lower in patients who used serum as test samples (SMD -0.335, 95% CI, -0.483 to -0.186), and in 37 studies who used plasma as testing samples, 26 studies showed the inverse relation of adiponectin to cancer risk. Assay method (radioimmunoassay or enzyme-linked immunosorbent assay) did not affect the results that circulating adiponectin was lower in cancer patients with pooled SMD of -0.316 and -0.266. Study size (more or less than 100 patients) did not change the result of estimated SMD either (SMD -0.135, 95% CI, -0.299 to -0.030; SMD -0.549, 95% CI, -0.825 to -0.273, respectively). Besides, no matter the mean age of cancer patients is older or younger than 60 years, decreased adiponectin levels were still exist in cancer patients (SMD -0.489, 95% CI, -0.689 to -0.288; SMD -0.194, 95% CI, -0.383 to -0.004, respectively).
Next we performed meta-regression to evaluate the effect of the above factors on the estimate of SMD. In meta-regression, none of the examined factors, such as ethnicity, cancer type, study design, blood sample, assay method, study size, study quality and mean age of cancer patients was proved to be significant contributing factors.

sensitivity analysis
Sensitivity analysis was performed by excluding one study at a time and calculating the pooled SMDs for the remaining studies. It was found that the combined SMDs were similar to one another and statistically significant. None of the studies influence the pooled results substantially in this analysis (Table 3).

Publication bias
Publication bias was assessed by funnel plot and Egger's regression test. Funnel plot shapes demonstrated a marginally asymmetrical distribution (Figure 4), accordingly we performed further analysis with Egger's test. The tested result ( Figure 5) showed no evidence of publication bias (P = 0.123).

DIscUssION
By integrating 107 studies, our meta-analysis revealed that lower circulating adiponectin levels were associated with higher risk of cancers. Despite the existence of heterogeneity, the disparity of adiponectin levels between malignant individuals and controls reveals the potential ability of adiponectin to serve as a biomarker for early detection of cancers.
Aberrant adiponectin secretion is associated with tumor progression, metastasis and overall prognosis. Two previous meta-analysis indicated that lower adiponectin levels were associated with higher risk of breast cancer, colorectal cancer and colorectal adenoma [115,116]. By synthesizing 107 studies involving 19,319 cases with different malignancies, the present meta-analysis estimate the inverse association between circulating adiponectin levels and cancer risk. Moreover, through subgroup analysis, we identified that this inverse relation of adiponectin to cancer risk might be more meaningful in Oncotarget 48684 www.impactjournals.com/oncotarget breast, colon, endometrial, prostate, and gastroesophageal cancers. Besides, adiponectin levels tend to decrease as tumor stage increases in gastric cancer [62]. Kang et al. also indicate that breast cancer patients with less than the median adiponectin levels are easy to develop lymph node metastasis [82]. Low adiponectin level is the independent predictor of unfavorable prognosis in colorectal cancer [117]. These findings demonstrate that adiponectin is not only associated with cancer risk, but also correlated with tumor progression. Additionally, in our included 107 studies, 8 studies evaluated the relationship between circulating levels of adiponectin subtypes and cancer risk. The changing trend of total adiponectin was almost same with the three adiponectin subtypes in cancer patients, especially with HMW-adiponectin, that it is inversely associated with cancer risk.
Circulating adiponectin levels are affected by various factors, including inflammatory, dietary, hormonal, genetic, and medicine. One of possible explanations for decreased adiponectin levels in malignancies is the sustained inflammatory status of cancer patients leads to the increased proinflammatory cytokines such as TNF-α and IL-6, which are all reported to suppress adiponectin transcription and translation in adipocyte cell line [118,119]. Besides, in obesity-related cancers, adiponectin may control its own production through a negative feedback loop during the development of obesity [120]. Moreover, dietary with lower intake of fiber and magnesium can also reduce circulating adiponectin levels [121].
However, elevated adiponectin levels are also reported in hepatocellular carcinoma. Since adiponectin is mainly degraded in the liver and adiponectin levels are elevated in advanced disease including cirrhosis and virusrelated cancer [61,122]. One possible explanation for increased adiponectin level in hepatocellular carcinoma might be due to deteriorated hepatic metabolism resulted from repeated necroinflammation and regeneration. Besides, conflicting results also exist in clinical studies of pancreatic cancer that both higher and lower adiponectin levels are reported to be associated with cancer risk [45,50]. After reviewing the pancreatic cancer studies with higher levels of adiponectin, we found that almost half of them were accompanied with jaundice [45]. Since cholestasis would lead to the chronic liver deterioration, it is possible that increased adiponectin levels might be due to the reduced degradation.
The peripheral functions of adiponectin are mainly mediated through AdipoR1 and AdipoR2. The expression levels of AdipoRs vary between malignant tissues and their peritumoral normal counterparts. The upregulation of AdipoR1 and AdipoR2 are reported in gastric carcinoma [123], whereas decreased in prostate cancer tissues compared with the nonmalignant tissues [36]. Increased expression of AdipoRs may be the response of reduced circulating as well as local adiponectin levels and reduced expression suggests that the sensitivity of AdipoRs to adiponectin is decreased in tumor tissues. Yabushita et al. indicate that poor expression of AdipoR1 is associated with tumor invasion and lymph node metastasis, as well as poor prognosis in endometrial cancer patients [124]. A study of non-small cell lung cancer also indicates that patients with higher expression of AdipoR1 have longer overall survival and AdipoR2 expression is inversely correlated with tumor size [125]. Those findings further illustrate the protective role of adiponectin as well as AdipoRs and shed light on exploiting them for cancer therapy. Recently, AdipoRs agonist called 355ADP is identified and might represent a new strategy to replace low adiponectin level in cancer [126].
Despite the inverse correlation between adiponectin and various cancers, the underlying mechanisms of adiponectin in potential cancer suppression are still need to elucidate. Adiponectin decreases low density lipoprotein (LDL) receptor expression in breast cancer cells through promoting autophagic flux and inhibits LDL-cholesterol-induced tumor cell proliferation [127]. Adiponectin induces the phosphorylation of p53, a tumor suppressor, which renders cell cycle arrest and apoptosis in cancer cell lines [128]. Adiponectin also inhibits leptininduced metastasis by downregulating JAK/STAT3 pathway, displaying an inverse correlation with cancer development [129]. In contrast, adiponectin promotes the angiogenesis in human chondrosarcoma by increasing vascular endothelial growth factor-A expression [130]. It is also reported to exert anti-apoptotic effects on pancreatic cancer cells through activation of AMPK/Sirtuin-1 signaling pathway [131]. Taken together, adiponectin might play a complicated role in carcinogenesis and progression of cancers.
Our study has some limitations that need to be addressed when interpreting the results. The significant heterogeneity was observed among the studies thus the conclusion should be more conservative. Although stratified analysis was conducted, none of the factors including ethnicity, cancer type, study design, blood sample, assay method, study size, study quality, and mean age of cancer patients were confirmed to contributing factors. Some possible reasons may partially explain this heterogeneity. Adiponectin levels are changed along with the tumor development. The tumor type, size, histological grade, and lymph node metastasis are the possible contributors caused heterogeneity. It is difficult for us to acquire the detailed information from the included studies. Besides, the subjects were from different regions and the lifestyle combined with diet was varied, which might influence the level of adiponectin. Since adiponectin is mainly secreted from adipose tissue, variables such as age, hormone receptor expression, menopausal status and BMI could contribute to the secretion and those factors were not fully deliberated for the complexity of tumor environment.
Oncotarget 48685 www.impactjournals.com/oncotarget cONcLUsIONs In summary, the present study shows significant difference in circulating adiponectin levels between patients with malignancies and controls. Low circulating adiponectin level is associated with increased cancer risk, which suggests that adiponectin may serve as a potential biomarker for early detection of cancers considering its abundance in blood. Thorough understanding the roles of adiponectin and its receptors in the progression of cancers is helpful to cancer screening and promote individualized treatment.

MAtErIALs AND MEtHODs search strategy
Based on the standard guidelines, a systematic search of English literature from Cochrane library, Wiley online library, PubMed was conducted to retrieve eligible studies until August 8, 2015. Searching terms included Medical Subject Heading (Mesh) and free text words "adiponectin", "ADPN", "Acrp 30", "AdipoQ", "GBP 28" or "apM1" in combination with "neoplasm", "cancer", "carcinoma", "malignancy" or "tumor". Furthermore, we manually searched references of relevant studies to add potential research to this meta-analysis.

Inclusion and exclusion criteria
Studies were included if they met the following criteria: (i) full text case-control studies published in peerreviewed journals evaluating the relationship between circulating adiponectin concentration and carcinogenesis; (ii) all cases were diagnosed as cancer by pathological biopsy or other medical methods with blood sample obtained before any therapies and all the controls were people without any cancers. (iii) circulating adiponectin level and standard deviation (SD) of it were provided or there were enough information to estimate them. Reviews, letters or animal experiments were excluded and articles without key information to carry on further analysis were also beyond consideration. Meanwhile, if replicated patient cohort was published in different studies, only the most recent or complete one was chosen. Since all the studies included were acquired from literature, ethics committee approval was not needed.

Data extraction
Based on the checklist of MOOSE (Meta-analysis Of Observational Studies in Epidemiology) [132], two reviewers (Tai W and Peng Y) extracted the following data independently from eligible studies: the last name of first author, year of publication, geographic region, ethnicity, tumor type, study design, sample type, adiponectin assay method, number of patients and controls, assay source, mean ± SD of adiponectin concentration. Disagreement was resolved by discussion until the two reviewers reached a consensus.

Quality assessment of included studies
Two reviewers (Tai W and Peng Y) independently assessed the quality of each included study according to the Newcastle-Ottawa Quality Assessment Scale (NOS) [133] ranges from 0 to 9 stars. Studies with more than 6 stars were considered as high-quality studies. Any disagreement was resolved by discussion and reevaluation.

statistical analysis
We acquired the mean ± SD of circulating adiponectin levels from cases and controls through three ways. The most accurate method was extracted them from the original research directly. However, a few studies presented the results as median values or standard error. In that case, we regarded median value as mean value considering the large sample size and calculated the SD value by using standard error and population number. If necessary, we contacted the author for detailed information. Standard mean differences (SMDs) and the corresponding 95% confidence intervals (CIs) of circulating adiponectin were calculated for all the eligible studies. Cochran's Q-test was performed to test the heterogeneity of included studies and P < 0.05 was considered statistically significant. Higgins I-squared statistic was applied to offer evidence of heterogeneity with I 2 > 50% suggesting significant heterogeneity. The pooled SMD and 95% CI was calculated using a fixedeffects model if the heterogeneity was not significant, otherwise a random-effect model was employed and subgroup analyses and meta-regression were adopted to detect the potential cause of heterogeneity.
Sensitivity analysis was executed to detect the robustness of the results. Publication bias was evaluated by use of funnel plot and Egger's linear regression test. The Stata 13.0 software (Stata Corporation, College Station, TX, USA) was used to perform all the statistical analysis. All P values were two-sided.