Polymorphisms in the mTOR Gene and Risk of Sporadic Prostate Cancer in an Eastern Chinese Population

Background The mTOR gene regulates cell growth by controlling mRNA translation, ribosome biogenesis, autophagy, and metabolism. Abnormally increased expression of mTOR was associated with carcinogenesis, and its functional single nucleotide polymorphisms (SNPs) may regulate the expression of mTOR and thus contribute to cancer risk. Methodology/Principal Findings In a hospital-based case-control study of 1004 prostate cancer (PCa) cases and 1051 cancer-free controls, we genotyped six potentially functional SNPs of mTOR (rs2536 T>C, rs1883965 G>A, rs1034528 G>C, rs17036508 T>C, rs3806317 A>G, and rs2295080 T>G) and assessed their associations with risk of PCa by using logistic regression analysis. Conclusions/Significances In the single-locus analysis, we found a significantly increased risk of PCa associated with mTOR rs2536 CT/CC and rs1034528 CG/CC genotypes [adjusted OR = 1.42 (1.13–1.78), P = 0.003 and 1.29 (1.07–1.55), P = 0.007), respectively], compared with their common homozygous genotypes, whereas mTOR rs2295080 GT/GG genotypes were associated with a decreased risk of PCa [adjusted OR = 0.76 (0.64–0.92), P = 0.003], compared with wild-type TT genotypes. In the combined analysis of the six SNPs, we found that individuals carrying two or more adverse genotypes had an increased risk of PCa [adjusted OR = 1.24 (1.04–1.47), P = 0.016], compared with individuals carrying less than two adverse genotypes. In the multiple dimension reduction analysis, body mass index (BMI) was the best one-factor model with the highest CVC (100%) and the lowest prediction error (42.7%) among all seven factors. The model including an interaction among BMI, rs17036508, and rs2536 was the best three-factor model with the highest CVC (100%) and the lowest prediction error of 41.9%. These findings suggested that mTOR SNPs may contribute to the risk of PCa in Eastern Chinese men, but the effect was weak and needs further validation by larger population-based studies.


Introduction
Prostate cancer (PCa) is the second most frequently diagnosed cancer and the sixth leading cause of cancer death in males according to the latest report released by the International Agency for Research on Cancer (IARC) in 2008 [1]. It has been well established that PCa is one of the pronounced geographically and ethnically related human malignancies, with a much higher incidence observed in the Western world than in Asian countries [2]. Recently, accumulated evidence from genome-wide association studies (GWASs) suggests that more than 40 single nucleotide polymorphisms (SNPs) are associated with human PCa risk, some of which were also confirmed in Chinese male populations. However, almost all the candidate SNPs are reported to be in weak associations with PCa risk to date [3][4][5][6]. Therefore, it is still not fully understood to what extent genetic factors and their interactions with environmental attributes may play a role in the etiology of PCa.
The phosphoinositide-3 kinase-AKT-mammalian target of rapamycin pathway (PI3K/AKT/mTOR) is a major pathway controlling cell growth and tumogenesis [7,8]. As a key downstream effector of PI3K/AKT/mTOR pathway, the mTOR has been confirmed to be a central regulator of vital cellular processes, such as cell growth, proliferation, metabolism, migration, and apoptosis, based on the in vivo and in vitro investigations [9][10][11][12]. Structurally, mTOR contains several important domains across the whole protein, of these, the rapamycin-binding domain and the kinase domain was considered closely relevant to carcinogenesis [13]. Additionally, several studies have demonstrated that mTOR targeted therapies can be designed to block the induction of the proliferative, prosurvival, and oncogenic functions of mTOR [14]. Therefore, it was speculated that mTOR is a possible driver gene in carcinogenesis, and a promising target point and prognosis marker in cancer treatment as well.
Somatic aberrations of PI3K/AKT/mTOR pathway genes have been commonly observed in a variety of malignancies, including PCa [8]. And the mutations in the mTOR gene have been identified in a few human cancers [15]; however, the mechanism has not been well established to date. PCa harboring almost the same well-known mutations often presents with heterogeneous clinicpathologic characteristics. By the same token, genetic factors, such as naturally occurring polymorphic genetic variants or SNPs in mTOR, may be contributing to the variation in individual susceptibility to PCa and the progression of this disease.
Given that mTOR is one of the most important downstream components of the mTOR pathway, which can also receive signals from other pivotal pathways. Several studies have demonstrated that mTOR can serve as a promising therapeutic target in the future cancer treatment. And there have been few studies to date addressing the role of common, functional variants in the mTOR gene as PCa susceptibility factors, together with some variations of other pivotal genes in this pathway have been investigated as weak or null associations with cancer risk. we performed a case-control study by genotyping six potential functional SNPs in mTOR using genomic DNA from 1004 patients with prostate adenocarcinoma and 1051 cancer-free controls in an Eastern Chinese Han population. We tested the hypothesis that risk of PCa may be associated with SNPs in the mTOR gene and their interactions with environmental factors.

Patients and controls
We recruited PCa patients and the matched cancer-free controls from genetically unrelated Chinese Han participants between January 2008 and January 2012. This analysis included 1004 patients who were inhabitants of the administrative regions of eastern China (including Shanghai city, Zhejiang province, Jiangsu province and the surrounding areas) and have been histopathologically confirmed primary prostate adenocarcinoma at Fudan University Shanghai Cancer Center (FUSCC). All cases had received no prior chemotherapy or radiotherapy upon recruitment. The clinical stages were determined and categorized into stage I (T1a-bN0M0), stage II (T1c-2N0M0), stage III (T3-4N0M0), and stage IV (T1c-4N1M0-1 or T1-4N0-1M1) according to the Tumor-Node-Metastasis system, and pathological grades of the PCa were determined according to the WHO criteria [16]. The entire document, including the Gleason score, serum PSA level at diagnosis, and clinical staging (TNM) was abstracted from the archival medical records. The male control group was comprised of 1051 cancer-free individuals, frequency-matched with the cases by age (65 years) and geographical regions, recruited from the Taizhou longitudinal study (TZL) [17] during the same period. Individuals with a known test of serum PSA .4 ng/mL present with or without abnormal digital rectal examination were excluded from the control group.
All of the participants were interviewed with a self-administered questionnaire after a written informed consent was obtained. Blood samples were collected and processed, with a written informed consent from participants, as a routine practice by the Institutional Tissue Bank at Shanghai Cancer Institute (for cases) and the TZL study (for controls). Response rate was 92% and 91% for cases and controls, respectively. The research was approved by the Institutional Review Board of FUSCC.

Single nucleotide polymorphisms selection
Among all of the reported mTOR SNPs, potentially functional SNPs of interest were selected from the NCBI dbSNP database (http://www.ncbi.nlm.nih.gov/projects/SNP) and SNPinfo (http://snpinfo.niehs.nih.gov/snpfunc.htm) according to the following criteria: 1) the minor allele frequency (MAF) reported in HapMap was$5% for Chinese populations; 2) affecting the functional regions of the gene, including transcription factor binding site (TFBS), potential miRNA binding site, splicing regulation locus, and stop codon; 3) the linkage disequilibrium (LD) coefficient r 2 ,0.8 between SNPs; and) not included in the published GWASs studies. Ultimately, six variants were selected for the present study, including rs2536 T.C, rs1883965 G.A, rs1034528 G.C, rs17036508 T.C, rs3806317 A.G, and rs2295080 T.G, of which four (rs1034528 G.C, rs1883965 G.A, rs2295080 T.G, and rs3806317 A.G) located in the first intron region may affect the transcription factor binding site (TFBS) activity, two (rs2536 T.C and rs17036508 T.C) located in the 39-untranslated region (39 UTR) region may affect the miRNA binding site activity, and SNP rs17036508 T.C locus also predicted locate at the potential splicing site. Bioinformatics analysis was performed with HaploView software 4.2 to estimate the haplotype block for Chinese population (CHB) data of HapMap (HapMap Data Rel 27 Phase II+III), and no LD was found between any of these SNPs described above. All these six selected SNPs were genotyped by the TaqMan real-time PCR method as described previously [17], and the results with .98% call rates and 100% concordance for duplicated specimens were acceptable for further genotyping data analysis.

Multifactor Dimensionality Reduction (MDR) Analysis
Evidence indicated that gene-gene and gene-environment interactions are difficult to be fully characterized by using logistic regression model. And statistic power would decrease and type II errors would increase when detecting interactions by LR in casecontrol studies with relatively small sample sizes [18]. By contrast, The MDR analysis can overcomes some of the limitations of logistic regression model for the interactions by collapsing highdimensional data into a single dimensional variable with two levels. In the present study, we performed the MDR analysis, as described previously [19]. We used a model of 100-fold crossvalidation and repeated the complete analysis for 10 times under different random seeds, and then the test was repeated 1000 times under the null hypothesis of no association. As a result, the model employing the minimized prediction error together with the maximized cross-validation consistency (CVC) was recommended. This analysis was performed by using the MDR V2.0 beta 8.2 software (http://www.multifactordimensionalityreduction.org/).

Statistical analysis
Hardy-Weinberg equilibrium (HWE) for evaluation of genotype distributions of the controls was performed by a goodness-of fit x 2 test. Differences in the frequency distributions of the alleles, genotypes and the selected categorical variables between cases and controls were evaluated by Pearson's x 2 test under various genetic models (including dominant model, recessive model, and additive model). Crude and adjusted odds ratios (ORs) and 95% confidence intervals (CIs) were calculated according to the significant genetic models by univariate and multivariate uncon-ditional logistic regression models, respectively, to evaluate associations between the genotypes and risk of Pca with and without adjustment for by confounding factors. Given the present study was single ethnicity, and all of the SNPs loci were agree with HEW, the confounding factors which should be adjusted for was age, smoking status, and body mass index (BMI). Further stratification analyses were conducted to calculate the associations of SNP genotypes with PCa risk by demographic and clinicpathologic variables, followed by the homogeneity Q-tests to detect any difference in the risk estimates between the strata. Based on the observed genotypes, haplotype frequencies and individual haplotypes were generated using Statistical Analysis Software PROC HAPLOTYPE, with a reference group of common haplotype, to calculate ORs for haplotypes associated with PCa risk in logistic regression analysis. For all the significant findings observed in our study, we calculated the false-positive report probability (FPRP) with prior probabilities of 0.0001, 0.001, 0.01, 0.1 and 0.25 to detect the possible false-positive associations [20]. Statistical power was estimated to detect an OR of 1.50/0.67 (for a risk/protective effect), with an a level equal to the observed P value. Only significant results with FPRP value less than 0.2 were considered a noteworthy association. All statistical analyses were performed with SAS 9.1 statistical software (SAS, Cary, NC, USA). All P values were two-sided with a significance level of P ,0.05.

Characteristics of the subjects
The distributions of demographic characteristics of the subjects are presented in Table 1. Briefly, there were no statistical differences in the distributions of age and smoking status between 1004 cases and 1051 controls. The body mass index (BMI) for overweight (. 24.0 kg/m 2 ) was more evident in controls than in cases (P , 0.0001), which was further adjusted for in subsequent multivariate logistic regression analyses. Among the case subjects, 178 (17.7%) cases were PSA#10 ng/ml, 312 (31.1%) cases were Gleason score#7 (3+4), and 601 (59.9%) cases were Gleason score$7 (4+3). For tumor staging, five (0.5%) cases had stage I disease, 431 (42.9%) had stage II disease, 140 (13.9%) had stage III disease, and 351 (35.0%) had stage IV disease. However, some cases had missing data because of the insufficient documented records, including 87 (8.7%) lacking serum PSA values, 91 (9.1%) lacking Gleason scores, and 77 (7.7%) lacking clinical staging status.

The mTOR allele and genotype distributions and associations with PCa risk
The genotype and allele distributions of the six selected SNPs among cases and controls are summarized in Table 2. The observed genotype frequencies of the six SNPs in controls agreed with the Hardy-Weinberg equilibrium. Furthermore, significant differences in genotype distributions were observed between cases and controls for rs2536 T.C (P = 0.007), rs1034528 G.C (P = 0.022), and rs2295080 T.G (P = 0.012). Interestingly, the heterozygote genotypes of the above three SNPs were more likely to be significantly associated with PCa risk with adjusted OR (95% CI) and P value of 1.45 (1.15-1.84) and 0.002 for rs2536 TC, 1.31 (1.08-1.59) and 0.005 for rs1034528 GC, and 0.77 (0.64-0.93) and 0.006 for rs2295080 TG, respectively, compared with their respective wild-type genotypes, respectively. Additionally, we also found significant associations with PCa risk for SNPs in special genetic models, including rs2536 T.  Table 3).

Stratification analysis of PCa risk associated with mTOR SNPs
In stratification analyses, as shown in Tables 4 and 5, the multivariate logistic regression analyses indicated, by assuming a dominant genetic model, that both mTOR rs2536 CT/CC and rs1034528 CG/CC genotypes were associated with an increased risk of PCa, particularly in subgroups of age#69, BMI#24 kg/ m 2 , ever smokers, Gleason score#7 (3+4), Gleason score$7 (3+4), and stage III/IV disease, compared with their homozygous wildtype genotypes, respectively. The rs17036508 CT/CC genotypes were also associated with an increased risk of PCa among subgroups of BMI#24 kg/m 2 , Gleason score#7 (3+4), and stage

Haplotype analysis of the mTOR SNPs
Based on the genotyping results to infer possible haplotypes, we used the four SNPs (rs2536 T.C, rs1034528 G.C, rs17036508 T.C, and rs2295080 T.G) that were statistically significantly associated with PCa risk in the single locus analysis ( Table 6). When the common ''TGTT'' haplotype was used as the reference, the ''CCCG'' haplotype was associated with an evidently increased PCa risk [adjusted OR = 1.31 (1.03-1.66), P = 0.026] However, the ''TGTG'' and ''TGCG'' haplotypes were associated with an evidently decreased but not increased PCa risk, with the adjusted OR of 0.39 (0.27-0.56), P ,0.0001 and 0.63 (0.43-0.91), P = 0.014, respectively. The findings of haplotypes ''CCTG'' and ''CCCT'' may not be reliable due to their relatively small numbers of observations.

Association of high-order interactions with PCa
To further explore high-order interactions, we performed the MDR analyses by including the genotypes of four significant mTOR SNPs (i.e., rs2536 CT/CC, rs1034528 CG/CC, rs17036508 CT/CC, and rs2295080 GT/GG vs. their wild-type homozygotes, respectively) and three risk factors (i.e., age at diagnosis, smoking status, and BMI). The results showed that BMI was the best one-factor model with the highest CVC (100%) and the lowest prediction error (42.7%) among all seven factors. Likewise, the interaction between BMI, rs17036508 T.C, and rs2536 T.C was the best three-factor model involving both environmental and genetic factors with the highest CVC (100%) and the lowest prediction error of 41.9% (Table 7).
Finally, the FPRP values at different prior probability levels for all significant findings are summarized in Table 8. When the assumption of prior probability was 0.01, the association with rs2536 (CT/CC vs. TT) was noteworthy in subgroups of#24 kg/ m 2 BMI and stage III+IV (FPRP = 0.112 and 0.055, respectively), and the similar results can be observed in the association with rs1034528 (CG/CC vs. GG) in subgroups of#24 kg/m 2 BMI and stage III+IV (FPRP = 0.132 and 0.043, respectively) as well as the association with subgroup of stage III+IV (FPRP = 0.165) by rs17036508 (CT/CC vs. TT). In contrast, some greater FPRP values for the other significant associations between mTOR variants and prostate cancer risk suggested some possible bias in the findings, which need further validation in larger studies.

Discussion
In this large, ethnic specific single institutional case-control study, we investigated the associations between six potentially functional SNPs of the mTOR gene and PCa risk, and we found that the rs2536 C, rs1034528 C, and rs2295080 G variant genotypes were associated with PCa risk, and the effects were more evident in subgroups of age#69, BMI#24 kg/m 2 , and eversmokers. Additionally, the variant genotypes were more common in patients with high-grade diseases (stage III+IV), indicating their likely involvement in the development and progression of PCa. To the best of our knowledge, this is the first post-GWAS study that focused on the associations of these six potentially functional mTOR SNPs with PCa risk.
The mTOR gene, located on chromosome 1p36.2, encodes a protein kinase product of 289 kDa and has emerged as a critical cell growth effector by controlling mRNA translation, ribosome biogenesis, autophagy, and metabolism [21][22][23]. Studies have shown that there are some important domains ranging from the Nto the C-terminus of mTOR. For example, the N-terminus of mTOR contains two tandem repeated HEAT motifs that can mediate interactions between proteins, the FAT domain that can facilitate focal adhesion to the targeting domain, and the FRB domain that is regarded as one high-affinity binding site for the inhibitory complex FKBP12-rapamycin, whereas there are two regulatory domains that are located at the C-terminus of the protein including PtdIns 3-kinase related catalytic domain and FATC domain [13]. To date, there are 2651 SNPs that have been observed scattered across the whole mTOR gene. Given the critical role of mTOR in the PNET/AKT/mTOR signaling pathway, it    is biologically plausible that functional SNPs affecting the pivotal domains described above may contribute to cancer susceptibility. However, in addition to the published GWAS studies, only a few reported post-GWAS studies have investigated the associations between functional SNPs of the mTOR gene and risk of PCa. In a Chinese study with 666 PCa and 708 cancer-free controls, Chen et al. [24] indicated that mTOR rs2295080 GT/GG genotypes had a protective effect on PCa risk, compared with the TT genotype, which was recently shown by Xu et al. in gastric cancer [25] and by Cao et al. in renal cell cancer [26]. These findings are consistent with those of the present study with a larger sample size. Additionally, Hildebrandt et al. found that individuals carrying the rs2295080 GG genotype had reversed clinical outcomes in Caucasian esophageal cancer patients treated with chemotherapy, compared with the TT homozygous wild-type genotype [27]. All studies that focused on Chinese populations indicated an association between rs2295080 GT/GG and cancer risk, suggesting an possibly ethnic-specific association. Nevertheless, the associations between mTOR rs2536 C variant or genotypes and cancer risk in Chinese populations were various in the literatures; for example, the mTOR rs2536 CT heterozygous genotype was found to be associated with decreased risk of Chinese childhood acute lymphoblastic leukemia [28]; however, this association was not observed in other tumor types, such as gastric cancer [29], prostate cancer [24], and esophageal squamous cell carcinoma [30]. On the contrary, in the present study, we found that the mTOR rs2536 CT/CC genotypes were associated with an increased PCa risk under a dominant genetic model, different from the findings of another previously published PCa study (666 cases and 708 controls), in which a null association was reported [24]. We speculated that the disagreement might be due to the different sample size or different inclusion criteria for the participaion, which needs large and better designed studies to confirm. Studies have shown that the rs2295080 T allele could enhance the transcription activity of mTOR in HEK293, 786-O, HeLa, and GES-1 cell line in vitro [25,26]. Likewise, individuals carrying the TT genotype had higher levels of mTOR expression as well [25,26]. These suggest that the rs2295080 T allele could increase the affinity of special transcription factors to this region of the mTOR promoter and subsequently contribute to the increased mTOR activity in humans. Theoretically, miRNAs can bind to the 39 UTR of target genes and inhibit gene expression translationally and/or by destabilizing the target mRNA. Based on a bioinformatics web server (http://snpinfo.niehs.nih.gov/cgi-bin/snpinfo), the SNP rs2536 T.C was predicted to bind to miRNA-576 at the T variant allele or bind to miRNA-767 at the C allele. Therefore, we speculated that the expression of mTOR depended on the proportions of these two miRNAs or the affinity between miRNA and SNP rs2536 T.C, which has a growth advantage of immortalized cells and induces neoplastic transformation. It was indicated that disease-associated functional intronic variants may alter mRNA levels of the genes by affecting the transcriptional efficiency, RNA elongation, or splicing [31][32][33]. On the other hand, the SNP rs1034528 G.C located in the first intron region of the mTOR gene was also found to be associated with risk of PCa; however, both rs1034528 G and C alleles were predicted to bind to different transcription factors, respectively, in this region. Therefore, the exact mechanisms of the rs1034528 G.C underlying the observed PCa risk need additional functional studies.
There is evidence in the literature that each SNP may have a weak effect but the combination of multi-SNPs may present much stronger effects than any of the SNPs. This is particularly true in the present study, in which the haplotype and combined analyses confirmed the multi-SNPs effects in PCa. In the logistic regression model, a locus dose-response was found for the increased PCa risk with the increasing number of adverse genotypes of all studied SNPs. Additionally, we noticed the combined effects was more pronounced among subgroups of age#69 and BMI#24 kg/m 2 . These findings agreed with the hypothesis that genetic susceptibility contributes to the risk of developing cancer in those who had an early age onset and minor exposures. Although the interaction between smoking and mTOR SNPs was not observed in the present study, we did find an obvious effect of the combined unfavorable genotypes on PCa risk, particularly among subgroups of ever smoker, suggesting that the effect of the tobacco smoke-related carcinogens may also depend on genetic factors.
In the present study, the number of positive findings from the stratified analyses was obviously decreased in the FPRP assessment. There are several possible explanations for the false positive findings. Firstly, some findings in the stratified analyses may be a chance finding due to the limited sample size in the subgroups. Secondly, some missing information and potential confounding factors might result in the false positive associations. Therefore, all positive results should be explained with caution. Extensive evidence from previous epidemiology studies has indicated that several genetic variant and environmental factors are involved in the initiation and development of cancer [34][35][36][37]. We also found the similar interactions by using logistic regression and MDR approaches (Table S1 in File S1). In the MDR analysis, BMI was found to be the most noteworthy factor in one-factor model; however, the exact mechanisms for the association between BMI and PCa risk have not been established. Possible hypotheses include the effect of hormones, PSA, and adipose-related proteins [38]. In the present study, we found some evidence of the interactions between environmental factor (BMI) and genetic factors (rs17036508 T.C and rs2536 T.C), as shown in the best three-factor model, we speculated that those variations might alter the expression of mTOR and the subsequent synthesis of adiposerelated proteins, but this finding needs to be validated in larger studies.
In summary, the present study investigated the associations between six selected potentially functional mTOR SNPs and PCa risk with a relative large sample size. However, several methodological issues and limitations of the present study should be discussed. Firstly, some participants might be misclassified due to the lack of PSA serum information; for example, some silent tumors (stage A1, usually asymptomatic) may have been included as normal controls, which could subsequently bias the results to the null. Secondly, although hormonal, occupational, dietary, inflammation and other factors have been suggested as etiological factors of PCa, we did not adequately documented these covariables for adjustment. Thirdly, only six potentially functional SNPs of mTOR were investigated in the present study, which did not cover all variants in the mTOR gene. Therefore, additional larger and welldesigned studies are warranted to confirm our findings.

Supporting Information
File S1 File includes: Supplementary Table S1 for Stratification analysis of significant SNPs by age, smoking status, and BMI; S1-1 Stratification analysis of significant SNPs by age; S1-2 Stratification analysis of significant SNPs by smoking status; and S1-3 Stratification analysis of significant SNPs by BMI. (DOCX)

Author Contributions
Conceived and designed the experiments: DY QW. Performed the experiments: QL CG MW. Analyzed the data: JH MZ TS XZ. Contributed reagents/materials/analysis tools: YZ LJ YY JW. Wrote the paper: QL QW DY.