It has been suggested that estrone and 17-βestradiol are endogenous estrogens that are associated with carcinogenesis of various tissue types (Drzewiecka et al., Reference Drzewiecka, Galecki, Jarmolowska-Jurczyszyn, Kluk, Dyszkiewicz and Jagodzinski2014; Santen et al., Reference Santen, Yue and Wang2014; Yager, Reference Yager2014). A study conducted by Yager (Reference Yager2014) indicated that the carcinogenicity of estrogen is related to aberrant activities of both estrogen receptor (ER) and non-receptor signaling pathways, two complementary pathways. The ER regulates cell proliferation, while the non-receptor pathway is involved in the formation of DNA adducts and oxidative DNA damage. Under pathological conditions, excessive cell divisions resulting from both continuous hormonal stimulus and DNA damage-induced genetic instability could promote cancer development. The 17β-hydroxysteroid dehydrogenase type 1 (HSD17B1) gene, located on chromosome 17q12–q21, encodes a 17HSDtype 1 enzyme. This enzyme plays a central role in sex steroid hormone metabolism by catalyzing the conversion of estrone to its most potent form, 17-βestradiol (Puranen et al., Reference Puranen, Poutanen, Ghosh, Vihko and Vihko1997). However, some research has also indicated that individual genetic variation might increase one's susceptibility to cancer (Dai et al., Reference Dai, Xu, Long, Courtney, Xiang, Cai and Shu2007; Zhang et al., Reference Zhang, Yuan, Guan, Li, Liu, Sun and Deng2014).
To date, more than 450 single nucleotide polymorphisms (SNPs) in the HSD17B1 gene have been reported in the dbSNP database (http://www.ncbi.nlm.nih.gov/SNP/). Among these, the Ser312Gly polymorphism (rs605059) has been widely investigated. The non-synonymous A→G polymorphism in codon 312 of exon six changes the amino acid serine to glycine (Mannermaa et al., Reference Mannermaa, Peltoketo, Winqvist, Ponder, Kiviniemi, Easton and Vihko1994; Silva et al., Reference Silva, Cabral, Bezerra de Castro, Pires, Azevedo, Manita and Gaspar2006). Although current evidence indicates that this SNP does not affect the function of the enzyme (Johnson et al., Reference Johnson, Houck and Chen2005; Puranen et al., Reference Puranen, Poutanen, Peltoketo, Vihko and Vihko1994), the level of estradiol as well as the ratio of estradiol to estrone in the blood were dramatically increased in the carriers of AA genotype of the HSD17B1 Ser312Gly polymorphism in comparison to non-carriers (Setiawan et al., Reference Setiawan, Hankinson, Colditz, Hunter and De Vivo2004). Excessive estradiol alters expression of affected genes through the phosphorlation of extracellular signal-regulated kinase (Erk) 1/2, and then further increases cell proliferation, thereby modulating cancer susceptibility (Elloumi-Mseddi et al., Reference Elloumi-Mseddi, Jemel-Oualha, Beji, Hakim and Aifa2014).
Over the past decades, numerous molecular epidemiology studies have been carried out to assess the association of HSD17B1 Ser312Gly polymorphism with susceptibility to a board spectrum of cancers among different ethnic populations, including endometrial cancer (Ashton et al., Reference Ashton, Proietto, Otton, Symonds, McEvoy, Attia and Scott2010; Dai et al., Reference Dai, Xu, Long, Courtney, Xiang, Cai and Shu2007; Setiawan et al., Reference Setiawan, Hankinson, Colditz, Hunter and De Vivo2004), breast cancer (Abbas et al., Reference Abbas, Baisch, Braendle, Brauch, Brors, Bruning and Zapatka2010; Feigelson et al., Reference Feigelson, Cox, Cann, Wacholder, Kaaks, Henderson and Trichopoulos2006; Gaudet et al., Reference Gaudet, Chanock, Dunning, Driver, Brinton, Lissowska and Garcia-Closas2008; Iwasaki et al., Reference Iwasaki, Hamada, Nishimoto, Netto, Motola, Laginha and Tsugane2010; Justenhoven et al., Reference Justenhoven, Hamann, Schubert, Zapatka, Pierl, Rabstein and Brauch2008; Kato et al., Reference Kato, Cichon, Yee, Land and Korczak2009; Mannermaa et al., Reference Mannermaa, Peltoketo, Winqvist, Ponder, Kiviniemi, Easton and Vihko1994; Obazee et al., Reference Obazee, Justenhoven, Winter, Chang-Claude, Rudolph, Seibold and Brauch2013; Plourde et al., Reference Plourde, Samson, Durocher, Sinilnokova and Simard2008; Sakoda et al., Reference Sakoda, Blackston, Doherty, Ray, Lin, Stalsberg and Chen2008; Sangrajrang et al., Reference Sangrajrang, Sato, Sakamoto, Ohnami, Laird, Khuhaprema and Yoshida2009; Setiawan et al., Reference Setiawan, Hankinson, Colditz, Hunter and De Vivo2004; Wang et al., Reference Wang, Li, Li, Yuan, Yang, Lei and Guo2009; Wu et al., Reference Wu, Seow, Arakawa, Van Den Berg, Lee and Yu2003), prostate cancer (Kraft et al., Reference Kraft, Pharoah, Chanock, Albanes, Kolonel, Hayes and Wacholder2005), uterine leiomyoma (Cong et al., Reference Cong, Huang, Cong, Ye, Wang, Zha and Li2012), and lung cancer (Wu et al., Reference Wu, Yin, Guan, Ren, Shen, Su and Zhou2014). However, results were rather inconsistent or even contradictory. The discrepancy among studies is partially due to the relatively small sample size in each of published studies. We performed this comprehensive meta-analysis to provide a more precise assessment of the association between HSD17B1 Ser312Gly polymorphism and the risk of cancer.
Materials and Methods
Literature Search Strategy
We identified eligible publications that examined the association between the HSD17B1 gene polymorphism and cancer risk from PubMed, Embase, China National Knowledge Infrastructure (CNKI) and Wan Fang (WF) databases using the following search items: ‘HSD17B1 or 17β- hydroxysteroid dehydrogenase type 1 or EDH17B1’, ‘polymorphism or variant or variation’ and ‘cancer or carcinoma’ (the search was last updated on July 30, 2015). Extra eligible studies were retrieved by manually searching the reference lists of retrieved studies. No language restrictions were included in the search strategy. In the case of the appearance of the same subjects in multiple studies, we only included the study with the largest sample size or the latest one in our meta-analysis.
Inclusion and Exclusion Criteria
The studies included in our meta-analysis met the following criteria: (1) full-text research article; (2) evaluated HSD17B1 Ser312Gly polymorphisms and cancer risk; (3) case-control study, cohort study, or nested case-control study on human subjects; (4) independent from other investigations; (5) provided adequate data to calculate ORs and the corresponding 95% CIs. Case-only studies were excluded. In addition, reviews, conference abstract, case reports, studies without detailed data, and meta-analyses were also excluded.
Data Extraction
Information was extracted from eligible studies by two authors working independently. Conflicts were resolved by discussion between these two investigators until consensuses were reached, or a third investigator would step in to adjudicate potential disagreement when needed. The following information was collected: the surname of first author, publication date, country of origin, ethnicity, source of control groups (HB or PB), genotyping method, p values of the chi-square test for Hardy–Weinberg equilibrium (HWE) for the control subjects, and genotype counts of cases and controls (Table 1).
TABLE 1 Characteristics of Investigations Included in Meta-Analysis
PB = population based; HB = hospital based.
Quality Assessment
The two authors evaluated the quality of each study using the quality assessment criteria derived from previously published meta-analyses conducted by Thakkinstian et al. (Reference Thakkinstian, McKay, McEvoy, Chakravarthy, Chakrabarti, Silvestri and Attia2011) and Xue et al. (Reference Xue, He, Zhu, Ma, He and Jia2014); see Table 2. A score ranging from 0 to15 was used to weigh up the quality of studies. Zero was deemed to the lowest quality, while 15 indicated the highest quality. Investigations with scores ≤9 were considered as low quality, while those with scores >9 were categorized as high quality. Disagreements were judged impartially by a third investigator.
TABLE 2 Quality Assessment Criteria For Article
Statistical Analysis
The crude ORs and 95% CIs were used to evaluate the strength of associations between the HSD17B1 Ser312Gly polymorphism and cancer risk. The pooled risk estimations were performed under different genetic models, including homozygous model (Gly/Gly vs. Ser/Ser), heterozygous model (Gly/Ser vs. Ser/Ser), recessive model (Gly/Gly vs. Gly/Ser+ Ser/Ser), dominant model (Gly/Gly + Gly/Ser vs. Ser/Ser), and allele comparison model (Gly vs. Ser). Deviation from the HWE in controls were examined by chi-square goodness-of-fit test, and p < .05 was considered a significant deviation from the HWE (Rohlfs & Weir, Reference Rohlfs and Weir2008). Heterogeneity across studies was tested by a chi-square-based Q test (heterogeneity was considered to exist between studies when p < .10 and I2 > 25% for the Q test, with 25% < I2 < 50% regarded as low heterogeneity, 50% ≤ I2 ≤ 75% regarded as moderate heterogeneity, and I2 > 75% as high heterogeneity). We calculated pooled ORs with a fixed-effects model if there was no heterogeneity; where the p value was <.10, a random-effect model was adopted (Tan et al., Reference Tan, Xian, Chen, Shi, Wang, Guo and Chen2014). If heterogeneity was observed, meta-regression analyses would be performed to explore the potential source of the heterogeneity (Higgins & Thompson Reference Higgins and Thompson2004; Thompson & Higgins, Reference Thompson and Higgins2002). Stratification analyses were conducted to explore the effects of covariates on the association of interest, such as ethnicity (Caucasians, Asians, and Mixed, which contained more than one ethnic group), source of control, quality score of investigations, and type of cancer (if a cancer type encompassed less than three individual investigations, it was classified into the ‘other cancers’ group).
The Begg's funnel plot and Egger's regression asymmetry test were conducted to detect the underlying publication bias: a p value less than .05 was considered as significant publication bias (Peters et al., Reference Peters, Sutton, Jones, Abrams and Rushton2006). We performed sensitivity analysis to estimate the stability of the results by consecutively excluding eligible studies individually and recalculating the pooled ORs and 95% CIs.
The significant results further underwent a false-positive report probability (FPRP) analysis (Hua et al., Reference Hua, Li, Liang, Zhu, Zhang, Ye and Sun2014; Tan & Chen, Reference Tan and Chen2014). To determine the FPRP value for an association between the studied SNP and cancer risk, we adopted a prespecified prior probability of 0.1 and statistical power to determine the OR of 0.67 (for protective effects) or 1.50 (for risk effects). The threshold of a FPRP value was 0.2. Only the significant associations with an FPRP value of less than 0.2 were deemed as noteworthy.
The FPRP was performed using SAS software (version 9.1; SAS Institute, Cary, NC). The rest of statistical tests were conducted by STATA version 11.1 (Stata Corporation. USA). All p values were two-sided, and a p value less than .05 was considered statistically significant.
Results
Characteristics of Relevant Studies
As shown in Figure 1, using search terms described in the methods section, a total of 68 relevant articles were initially retrieved from EMBASE and PubMed databases, and an additional three articles were identified from the reference lists of the retrieved studies; a further seven articles written in Chinese were obtained from the NCKI and WF databases. Of the total number of articles, 51 were excluded after title and abstract screening. Among the remaining 27 articles, two reviews, one meta-analysis, and two articles with a case-only design were further excluded. Another two articles were excluded due to failure to provide detailed genotype data. As a result, 19 articles that met the inclusion criteria remained for final analysis. Of these, one publication was considered to consist of two individual studies, since it reported data on breast cancer and endometrial cancer separately. Ultimately, 19 publications, consisting of 20 individual studies with 29,460 cases and 36,687 controls, were included in the present meta-analysis (Table 1). Among these studies, there were 14 studies conducted on breast cancer, three on endometrial cancer, and three on ‘other cancers’ (prostate cancer, uterine leiomyoma and lung cancer; Table 1). In terms of ethnicity, nine studies were performed among Caucasians, seven among Asians, and four among mixed ethnic groups (only one study was conducted on African Americans, which was sorted into the ‘mixed’ category). When sorted by the source of control, 13 studies were PB and seven were HB (Table 1). Furthermore, 17 studies were classified as high quality and three as low quality. All cases were confirmed pathologically.
FIGURE 1 Flow diagram of selection of studies included in the current meta-analysis for the association between HSD17B1 gene polymorphisms and cancer susceptibility.
Meta-Analysis Results
As shown in Table 3 and Figure 2, a pooled analysis did not yield any significant association between HSD17B1 Ser312Gly polymorphism and cancer risk (homozygous: OR = 0.966, 95% CI = 0.925–1.010; heterozygous: OR = 0.978, 95% CI = 0.943–1.015; dominant: OR = 0.975, 95% CI = 0.942–1.009; recessive: OR = 0.984, 95% CI = 0.948–1.021; allele comparing: OR = 0.984, 95% CI = 0.963–1.006).
TABLE 3 Meta-Analysis of the HSD17B1 Gene Polymorphism and Cancer Risk
aThe pooled p value; b p value for heterogeneity test. PB = population based; HB = hospital based.
FIGURE 2 Forest plots of effect estimates for HSD17B1 Ser312Gly polymorphism and cancer susceptibility (GG vs. AA). For each study, the estimation of OR and its 95% CI are plotted with a box and a horizontal line. e, pooled ORs and its 95% CIs.
However, subgroup analysis by ethnicity indicated that this variant significantly decreased the risk of developing cancer among Caucasians (dominant: OR = 0.958, 95% CI = 0.919–0.998; allele comparing: OR = 0.973, 95% CI = 0.947–0.999). Moreover, stratification analysis by source of control revealed that this variant is associated with a significantly decreased risk of cancer for PB studies (homozygous: OR = 0.951, 95% CI = 0.908–0.997; allele comparing: OR = 0.976, 95% CI = 0.954–0.999). No significant association was found in the subgroup analysis by quality of studies and cancer type. In addition, in light of this study strongly comprised of breast cancer samples (14 out of 20 investigations), we further conducted subgroup analyses for breast cancer studies by source of control, ethnicity and quality score, but observed no evidence of the association in any of subgroups (Table 4).
TABLE 4 Meta-Analysis of the HSD17B1 Gene Polymorphism and Breast Cancer Risk
aThe pooled p value; b p value for heterogeneity test. PB = population based; HB = hospital based.
The significant results then underwent FPRP analysis. Since the HSD17B1 polymorphism was associated with protection against cancer, FPRP values were computed using a preset prior probability of 0.1, statistical power to detect an OR of 0.67, and our significant results (ORs and p values). The FPRP values suggested that the significant association between decreased cancer risk and the HSD17B1 protective genotypes in PB subgroup was not noteworthy (homozygous model: FPRP = 0.24; allele comparison model: FPRP = 0.279). Likewise, for Caucasians, the FPRP value for the significantly decreased cancer risk associated with this variant was: dominant model: FPRP = 0.274; allele comparison model: FPRP = 0.228. The result of the FPRP indicated some bias due to the small sample size in subgroups. Thus, further validation of these findings by studies with larger sample sizes are needed.
Heterogeneity and Sensitivity Analyses
A chi-square Q test and I2 statistic were performed to evaluate the heterogeneity among the studies. There was no heterogeneity under the heterozygous model, dominant and recessive models (p = 0.353, I2 = 8.3%; p = .198, I2 = 20.7%; p = .132, I2 = 26.7%) in the overall analysis. In contrast, significant heterogeneity was observed under the homozygous and allele comparison models (p = .090, I2 = 31.4%; p = .071, I2 = 33.7%). Then, a meta-regression was performed to determine the source of heterogeneity by types of cancer, source of controls, ethnicity, and quality score. We found that none of these factors contributed to the heterogeneity in the current meta-analysis (data shown in Supplementary Table 1). Moreover, sensitivity analysis, by consecutively excluding studies one at a time, indicated that the removal of any of studies did not substantially alter the pooled ORs, suggesting the stability of the meta-analysis results.
Publication Bias
For the significant results, the shape of the funnel plot for Caucasians in the dominant and allele comparing model, as well as PB controls in the homozygous and allele comparing model, did not show obvious asymmetry (Figure 3). Moreover, Begg's and Egger's tests suggested no evidence of publication bias in these models, respectively (Caucasian subgroup: homozygous model, Begg's test, p = .835, Egger's test, p = .308; heterozygous model, Begg's test, p = .211, Egger's test, p = .707; dominant model, Begg's test, p = .211, Egger's test, p = .674; recessive models, Begg's test, p = .404, Egger's test, p = .201; allele comparison model, Begg's test, p = .835, Egger's test, p = .289 and PB controls subgroup: homozygous model, Begg's test, p = .222, Egger's test, p = .784; heterozygous model, Begg's test, p = .088, Egger's test, p = .344; dominant model, Begg's test, p = .067, Egger's test, p = .292; recessive models, Begg's test, p = .714, Egger's test, p = .899; allele comparison model, Begg's test, p = .393, Egger's test, p = .876). Similarly, in the pooled analysis and other subgroup analysis, the shape of the funnel plot did not show any obvious asymmetry for the Ser312Gly polymorphism and cancer risk.
FIGURE 3 Funnel plot analysis to detect publication bias for HSD17B1 Ser312Gly polymorphism by allele comparison model for Caucasian subgroup analysis. Each point represents a separate study for the indicated association.
Discussion
In this current meta-analysis for the association between HSD17B1 Ser312Gly polymorphism and cancer risk, 20 studies with a total of 29,460 cases and 36,687 controls were included. This meta-analysis observed no significant association between HSD17B1 Ser312Gly polymorphism and overall cancer risk in all the genetic models. However, further stratified analyses by ethnicity indicated that the Ser312Gly polymorphism decreased cancer risk among Caucasians, but not other ethnic groups. When the analyses were stratified by the source of control, such association was observed for a PB, but not a HB, subgroup. There was no significant association between HSD17B1 Ser312Gly polymorphism and the risk of cancer when subgroup analyses were performed by cancer type and quality score. To date, this is the first meta-analysis to assess the relationship between HSD17B1 Ser312Gly polymorphism and cancer susceptibility.
The HSD17B1 gene is located at chromosome17q21.2. The protein product of the gene is a metabolizing enzyme that converts estrone to estradiol, the more bioactive form of estrogen (Vihko et al., Reference Vihko, Isomaa and Ghosh2001). So far, several SNPs in the HSD17B1 gene have been extensively investigated. Among these SNPs, the Ser312Gly polymorphism in the exon 6 that may modulate genetic cancer predisposition (Gaudet et al., Reference Gaudet, Chanock, Dunning, Driver, Brinton, Lissowska and Garcia-Closas2008; Iwasaki et al., Reference Iwasaki, Hamada, Nishimoto, Netto, Motola, Laginha and Tsugane2010) is the most common non-synonymous polymorphism. The non-conservative amino acid substitution of serine to glycine in the HSD17B1 protein caused by this variant may influence transcriptional activation of the HSD17B1 enzyme. Potentially functional polymorphisms in the estrogen-metabolizing genes may influence their biological function, and thereby influence the susceptibility to hormone-related cancer (Abbas et al., Reference Abbas, Baisch, Braendle, Brauch, Brors, Bruning and Zapatka2010; Tsuchiya et al., Reference Tsuchiya, Nakajima and Yokoi2005). A meta-analysis including 13,987 cases and 17,066 controls was carried out in 2010 to estimate the relationship between HSD17B1 Ser312Gly and risk of breast cancer, and no significant association was observed (Yao et al., Reference Yao, Cao, Qiu and Yu2010). With the inclusion of more investigations, our meta-analysis replicated this previously reported null association. However, we failed to confirm some of the significant results presented in the previous meta-analysis. For example, our study could not duplicate the finding that the HSD17B1 Ser312Gly polymorphism was associated with decreased breast cancer risk among Caucasian. The association was no longer significant upon the inclusion of five more investigations in our meta-analysis. Alternatively, significant association was obtained in stratified analyses by ethnicity, probably owing to biological or genetic differences between various races. Furthermore, genetic and environmental differences by race may play an important role in the disparities of cancer susceptibility (Pinheiro et al., Reference Pinheiro, Holmes, Pollak, Barbieri and Hankinson2005; Reding et al., Reference Reding, Chen, Lowe, Doody, Carlson, Chen and Malone2012; Setiawan et al., Reference Setiawan, Haiman, Stanczyk, Le Marchand and Henderson2006). The source of control may also contribute to the results. For example, controls who were recruited from hospital could not fully represent the population from which the cases originated, which may weaken the value of risk. Moreover, although a subgroup analysis by ethnicity and control source found an association between Caucasian and PB controls, the result should be considered with caution as the p value was only moderate.
The FPRP analyses were performed for the significant findings to rule out false positive findings resulting from multiple comparisons. While FPRP values at the preset prior probability of 0.1 suggested that Ser312Gly polymorphism might not be able to influence the cancer risk, there are several possible interpretations of these outcomes. First, multiple comparisons and the limited sample size in the subgroups would make a certain effect in some findings. Second, the significant association for PB studies and Caucasians was based on a moderation p value. Third, between-study heterogeneity was observed in the pool analysis. However, the meta-regression did not determine the source of the heterogeneity, which is probably because only slight heterogeneity existed across studies. Moreover, the sample sizes were reduced when stratification analysis was performed to identify the source of heterogeneity. Several factors may contribute to the heterogeneity among the pooled analysis. Discrepancies in lifetime environmental factors and genetic background could conceivably contribute to surveyed racial differences in the incidence of cancer (Laiyemo et al., Reference Laiyemo, Doubeni, Pinsky, Doria-Rose, Bresalier, Lamerato and Berg2010; Reding et al., Reference Reding, Chen, Lowe, Doody, Carlson, Chen and Malone2012; Sucheston et al., Reference Sucheston, Bensen, Xu, Singh, Preus, Mohler and Taylor2012). Furthermore, the same SNP may play a different role in tumor susceptibility because multi-genetic and multi-factorial factors are involved in carcinogenesis. The disparity of pathogenesis in different tumors may also interpret the heterogeneity. Heterogeneity should be explored in more investigations in the future.
There were also some limitations to be addressed in this meta-analysis. First, there were some studies with a relatively small sample size (less than 500) included in our study. Second, in some subgroup analysis, only a moderate number of individual studies were included. For instance, subgroups with endometrial cancer and other cancers only comprised three studies. As a result, the statistical power to detect the potential association was reduced. Third, conference abstracts and unpublished studies were omitted. Finally, this current meta-analysis was solely dependent on unadjusted estimates. If possible, ORs adjusted for gender, age, smoking status, drinking, occupational exposures, as well as hormonal replacement therapy, would be more accurate.
In summary, no significant association was found between HSD17B1 Ser312Gly polymorphism and overall cancer risk. Nevertheless, stratified analysis by ethnicity suggested that this variant might confer decreased cancer risk among Caucasians, indicating that HSD17B1 Ser312Gly might be an ethnicity-specific SNP. Finally, large-scale studies should be performed to confirm these findings.
Supplementary Material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/thg.2016.6.S1832427416000062.
