TERT-CLPTM1L Polymorphism rs401681 Contributes to Cancers Risk: Evidence from a Meta-Analysis Based on 29 Publications

Background Some common genetic variants of TERT-CLPTM1L gene, which encode key protein subunits of telomerase, have been suggested to play a crucial role in tumorigenesis. The TERT-CLPTM1L polymorphism rs401681 was of special interest for cancers risk but with inconclusive results. Methodology/Principal Findings We performed a comprehensive meta-analysis of 29 publications with a total of 91263 cases and 735952 controls. We assessed the strength of the association between rs401681 and overall cancers risk and performed subgroup analyses by cancer type, ethnicity, source of control, sample size and expected power. Rs401681 C allele was found to be associated with marginally increased cancers risk, with per allele OR of 1.04 (95%CI = 1.00–1.08, P heterogeneity<0.001) and an expected power of 1.000. Following further stratified analyses, the increased cancers risk were discovered in subgroups of lung, bladder, prostate, basal cell carcinomas and Asians, while a declined risk of pancreatic cancer and melanoma were detected. Conclusions/Significance These findings suggested that rs401681 C allele was a low-penetrance risk allele for the development of cancers of lung, bladder, prostate and basal cell carcinoma, but a potential protective allele for melanoma and pancreatic cancer.


Introduction
Telomeres are repetitive (TTAGGG)n sequences into arrays of up to 25 kb and cap the end of linear chromosomes in human cells. They play a key role in counteracting the end-replication losses that occur as a consequence of semiconservative replication of linear DNA molecules [1,2]. They also protect against coding sequence erosion and consequent DNA damage repair, which results in genome instability, chromosomal fusions, and rearrangements [3]. Telomerase and the control of telomere length are intimately linked to the process of tumourigenesis in humans. Telomerase have been showed to play a role in tumor progression and metasis by activation of the glycolytic pathway and suppression of cancer cell differentiation. Abnormal telomere length has been demonstrated in many cancers [4,5].
5p15. 33, which was commonly suggested to mediate the function of telomerase, contains two key genes: Telomerase reverse transcriptase (TERT) gene, and cleft lip and palate transmembrane 1 like gene (CLPTM1L; alias CRR9; MIM 612585). TERT is one of the main functional subunits of the telomerase enzyme and a key regulator of telomerase. Making use of the telomeric RNA subunit of telomerase as a template for the synthesis of single stranded DNA within the telomere, TERT thereby produce (TTAGGG)n tandem nucleotide repeats [6,7]. TERT is reactivated in cancer cells. Mutations in the coding regions of TERT can affect telomerase activity and telomere length, and generate severe clinical phenotypes, including bone marrow failure syndromes and a substantive increase in cancer frequency [8]. Although the function of the CLPTM1L is largely unknown, studies have demonstrated that it may be involved in the apoptotic response to genotoxic stress induced by cisplatin, as it encodes a transcript whose over-expression has been shown to induce apoptosis in cisplatin-resistant-sensitive cells [9,10,11,12]. Moreover, the CLPTM1L variants are hypothesized to enhance the metabolic activation of the reactive metabolites and/or formation and persistence of DNA adducts [13]. According to  [14,15,16,17]. A TERT-CLPTM1L SNP, rs401681 (C.T, located in the intron 13 of CLPTM1L and 27 kb from the TERT gene), is one of the most extensively studied SNPs. It has been reported to be associated with an increased risk of lung cancer through genomewide association studies (GWAS) [18,19,20,21,22]. However, the reported genetic effects varied across the published studies. For example, an early study reported that the rs401681 T allele(the minor allele) was not associated with risk of lung cancer in 341 cases and 431 controls in Caucasians (P trend = 0.259) [10], but another study with 2396 lung cancer cases and 3001 controls showed that the T allele was associated with a remarkably decreased risk of lung cancer in the same ethnicity (per allele OR = 0.87; 95% CI = 0.84-0.92) [23]. Additionally, a recent GWAS composed with 20726 cancer patients and 134650 controls suggested that the rs401681 C allele was associated with increased risk of lung, bladder, prostate and basal cell carcinomas [13]. More recently, the rs401681 C allele was inversely showed protective effect on melanoma risk in a study of 3843 cutaneous melanoma patients and 41963 controls [24], while another study did not find any significant association between rs401681 and risk of melanoma [25]. As above, the results remain controversial and ambiguous. Meanwhile, a single study might have been underpowered to detect the overall effects. A quantitative synthesis of the accumulated data from different studies is important to provide evidence on the association of rs401681 polymorphism with cancers risk. Thus, in this study, a comprehensive meta-analysis including the latest and relevant articles was conducted to explore whether rs401681 contribute to cancers risk.

Materials and Methods
We conducted a systematic review and meta-analysis in accordance with the guidelines provided by the Human Genome Epidemiology Network [26].

Inclusion and Exclusion Criteria
Articles which met the following criteria were included: (1) published in English; (2) the outcome was cancers; (3) tested for rs401681 polymorphism of TERT-CLPTM1L locus; (4) reported race and numbers of affected and unaffected subjects; (5) sufficient data for calculating an odds ratio (OR) with 95 percent confidence interval (95% CI) in additive model (two studies showed allelic ORs were also included because of large sample size and powerful influences [13,24]).
Exclusion criteria were:(1) investigations in subjects with family cancer risks or cancer-prone disposition; (2) unpublished studies; (3) abstract, case report, comment, review and editorial; (4) Whenever reports pertained to overlapping patients, we retained only the largest study to avoid duplication of information.

Data Extraction
The following information from each study was extracted by two investigators independently: (1) publication data, first author, year of publication; (2) cancer types; (3) ethnicity, source of control group and genotyping method; (4) minor allele frequency(MAF), genotype information (first priority) and/or additive OR and 95% CI, as additive model could recruit the largest number of subjects compared with any other genetic models in this meta-analysis; (5) for studies including subjects of different cancer types or ethnicities, data were extracted separately; (6) several included articles reported consortium or multistage results with multiple independent populations, if the summary OR did not show, these populations were listed as separated data sets. Discrepancies were resolved through discussion.

Statistical Analysis
Deviation from the Hardy-Weinberg equilibrium (HWE) among controls subjects was tested by a x 2 -test and a P,0.05 was considered as significant disequilibrium. The strength of the association between rs401681 polymorphism and overall cancers risk was measured by OR and corresponding 95%CI. Heterogeneity across studies was checked using the Cochran's Q-test and considered significant at P,0.05 [27]. When homogeneity existed, the fixed model (Mantel-Haenszel method) was used to calculate the summary ORs and 95% CIs; otherwise, the random-effects model (the DerSimonian and Laird method) was utilized [27]. The quantity I 2 that presents the percentage of total variation across studies as a result of heterogeneity was also calculated [28].The potential source of heterogeneity among studies was explored by stratification and meta-regression analyses. Studies were categorized into different subgroups by type of cancer, ethnicity, source of control, sample size and expected power. If one cancer type contained less than three individual studies, it was combined into the ''other cancers'' group. When it comes to ethnicity, data sets were categorized as Caucasian, Asian and others. Meta-regression was performed to explore the source of heterogeneity among covariables,such as ethnicities, genotyping methods, source of controls, cancer types, sample size (,1000 and$1000 subjects) and expected power (,0.5, 0.5 to 0.8 and.0.8) [29,30]. Inverted funnel plots and the Egger's test were used to examine publication bias [31]. Additionally, sensitivity analyses were performed by including and excluding studies not in HWE, and by removing sequential of individual studies. What's more, we estimated the expected power of each individual study and subgroup analyses as determined by the probability of detecting a true association between rs401681 and cancers risk at the 0.05 level of significance, assuming OR of 1.2 (for a risk effect) or 0.83 (for a protective effect), with an alpha level equal to the observed P value [32]. All the P values were two sided, and all analyses were done in STATA statistical software (version10.0; Stata Corporation, college Station, Texas).

Characteristics of All Included Studies
As shown in Figure 1, 34 eligible original studies which through a comprehensive literature search up to the end of July, 2012, seemed to meet the inclusion criteria. However, after further examination, five studies were excluded because: three studies used subjects with family cancer history or cancer-prone disposition [33,34,35]; one study was overlapped with the study of Rafnar et al and had a smaller sample size [36], one study did not provide sufficient data [37]. Final data pool was consisted of 29 publications with 52 data sets. The research strategy was illustrated in Figure 1 and Table S1.
Among 52 data sets included in this meta-analysis, 41 were conducted in Caucasians, eight in Asians, one was in Africans and Figure 2. ORs of overall cancer risks associated with rs401681 under the additive model by random effects. For each data set, the OR and 95% CI was plotted with a box and a horizontal line. The symbol filled diamond indicates pooled OR and its 95% CI. Stacey1-3 represented studies for cancers of basal cell, squamous cell carcinomas and melanoma, respectively; Rafnar1-17 represented studies for basal cell, lung, bladder, prostate, cervical, breast, colorectal, melanoma, endometrial, kidney, lymphoma, multiple myeloma, ovarian, pancreatic, squamous cell, stomach and thyroid, respectively; Gago-Dominguez1-2 represented studies for bladder cancer in Caucasians and Asians, respectively; Pooley1-3 represented studies for breast, colorectal cancers and melanoma, respectively; Nan1-3 represented studies for melanoma, squamous cell and basal cell carcinomas, respectively. doi:10.1371/journal.pone.0050650.g002

Association between the rs401681 and Overall Cancers Risk
Rather than using ORs adjusted by covariables, our estimations were based on the raw data when possible. To assess the effect of adjustment, summary effect of ORs with and without adjustment were compared. Although there were subtle differences between these two sources of ORs in each study, the pooled ORs (95%CI) were nearly identical ( Figure S1). These differences were of low impact to the synthesis, which were also suggested by other researchers [59,60,61,62]. Meanwhile, we tried to explore the gap between allelic and additive ORs, and found them were amazing close to each other ( Figure S2). As a result, two researches, which only reported allelic ORs for rs401681, were also included to final meta-analysis [13,24].
As shown in Figure 2, the overall meta-analysis showed that rs401681 allele C marginally increased overall cancers risk in additive model (OR = 1.04; 95%CI: 1.00-1.08; P heterogeneity ,0.001 and I 2 = 87.1%), with an expected power of 1.000.
In terms of subgroup analyses by ethnicity, the associations were significant in Asian populations (per allele OR = 1.14; 95%CI: 1.09-1.19; P heterogeneity = 0.382 and I 2 = 6.30%), while it was bordline significant in Caucasians (per allele OR = 1.03; 95%CI: 0.99-1.07) with high heterogeneity (Q = 322.82, P,0.001; I 2 = 87.6%). Stratified analysis by cancer types was performed in Caucasians, the results for lung, melanoma, squamous cell, pancreatic; bladder and basal cell carcinomas were the same in Caucasians as that of the overall population because these studies were mostly conducted in Caucasians ( Table 2).
Further analyses also showed marginally significant results in hospital-based, population or community-based studies and studies of large sample size or high expected power ( Table 2).

Evaluation of Heterogeneity
The source of heterogeneity across studies was explored among covariables, such as ethnicities, genotyping methods,source of controls, cancer types, sample size and expected power. Interesting, cancer types were found to contribute to the heterogeneity across the studies in the overall (Table S3) and subgroups metaanalyses of Caucasians (data not shown).

Sensitivity Analyses
A one-way sensitivity analysis was conducted to assess the influence of each individual study on the combined OR, with each particular data set dropped at a time. A random-effect model was employed when heterogeneity was indicated. Stability of odds ratio estimates was confirmed for association between rs401681 and cancers risk ( Figure S3). Meanwhile, after the omission of the study departure from HWE, the results did not alter notably (data not shown).

Publication Bias
Finally, funnel plots and the Egger's test were used to assess publication bias. In the funnel plot analysis, the shape of the funnel plot seemed symmetrical ( Figure 3). Furthermore, an Egger's test did not detect any publication bias for rs401681 in the overall or subgroup analyses ( Table 2). Therefore, there was no significant publication bias in the studies included in current analyses.

Discussion
In the present meta-analysis, our results suggested that the carriers of rs401681 allele C had increased cancers risk, especially for lung, bladder, prostate and basal cell carcinomas, such effect was still found in subgroup of Asians; whereas decreased risk for melanoma and pancreatic cancer. Further exploration of the functional explanation of this locus is warranted to understand the mechanisms for these associations.
These findings have some degree of biological plausibility. The TERT gene is mapped to chromosome 5p15.33 and consists of 16 exons and 15 introns spanning 35 kb of chromosome 1. It encodes the catalytic subunit of telomerase, functions as telomere maintenance and may play a role in the determination of cancer risk [63]. TERT protein shows a high-level of expression in many tumors and it possibly contributes to unlimited cell division and carcinogenesis [64,65]. The CLPTM1L gene has been documented to be upregulated in cisplatin-resistant cell lines and linked with cisplatin-induced apoptosis [9], and over-expression of CLPTM1L mRNA have been observed in many cancers [12,13,66]. Variants in this locus are hypothesized to mediate telomere length and be associated with multiple malignancies, including cancers of lung, prostate, urinary bladder, cervix and pancreas [13,20,21,22,67].
The heterogeneity among studies in this meta-analysis was dramatically reduced in stratified analyses by cancer types. It suggested a potential modified effect of TERT-CLPTM1L polymorphisms by tumor origins and the rational of stratified analyses. Therefore, we can infer that rs401681 had cancer-specific contributions and may play different roles in the etiology of different tumor sites [10,19,22,23,25,26,41,42,44,48,50,68]. For example, strikingly increased risk was found in smoking related cancers, such as lung and bladder cancers. It may be explained that CLPTM1L protein may be involved in the apoptosis response of genotoxic stress [9,10,11,12]. Further more, Nan and collaborators observed a suggestive positive relationship between rs401681 C allele and shorter relative telomere length [50]. Rafnar et al. suggested rs401681 C allele might be associated with an acceleration of the gradual shortening of telomeres with age [13]. Rs401681 was also demonstrated to be associated with risk of pancreatic cancer for chromosome ends lacking telomeric repeat sequences was observed in this cancer [69,70]. Possible links between shorter telomeres and decreased risk of melanoma were reported [50,67]. This might be because shorter telomere length conferring a shorter replicative lifespan in melanocyes, thus providing a more stringent barrier to unlimited cell division [67]. Declined melanoma risk might also due to the reduction of nevi size and count in individuals with shorter telomeres [67,71]. Compared to the basal and squamous keratinocytes, melanocytes have a higher tendency to senescence in response to oncogenic stress, rather than undergoing cell apoptosis. In addition, shorter telomere length was reported to be associated with an increased risk of basal cell carcinoma [72]. This is probably suggested the different roles of replicative senescence in basal keratinocyets and melanocytes [72]. However, the exact biological function of rs401681 has not been clarified now. It may be in strong linkage disequilibrium (LD) with other potential functional or causal SNPs. For example,rs402710 located in the intron 15 of CLPTM1L,it is in strong LD with rs401681 (r 2 = 0.70 in CEU, r 2 = 0.89 in CHB and r 2 = 0.759 in YRI), was predicted to have potential regulate function by SNPinfo and reported to be associated with higher levels of bulky aromatic and hydrophobic DNA adducts [10,73]. Meanwhile, some SNPs in high LD with rs401681 are also predicted to have potential function by SNPinfo, including rs31490 (r 2 = 0.910 in CEU, located in the transcription factor binding site of CLPTM1L) and rs414965 with high regulatory potential scores (located in the intron 11 of CLPTM1L, r 2 = 0.759 in YRI) [73]. Further investigations were required to explore the role of rs401681 or SNPs in high LD with it in carcinogenesis, especially in various cancer types.
Furthermore, the different LD pattern of these potential functional SNPs with rs401681 in different ethnic populations may explain the different associations between rs401681 and cancer risk. Therefore, genetic backgrounds might explain, to some extent, the somewhat conflicting associations in different populations.
In terms of the control's sources, the subtotal ORs and 95%CIs for rs401681 varied. The pooled ORs in population or community-based studies were 1.07(95%CI: 1.00-1.14) in additive model with an expected power of 1.000. Thus, these findings emphasized that the advantages of population based studies, including greater efficiency in sample recruitment, external validity than other study designs [74,75].
There were still some limitations which need to be addressed. First, although no any publication bias was showed in the funnel plot and Egger's tests, selection bias might still exist as non-English literatures were excluded. Second, the number of published studies was still insufficiently for the subgroup analyses of some particular cancer sites, such as endometrial, colorectal and testicular germ cell carcinomas. It might mask or exaggerate possible true associations. Third, due to insufficient genotype frequencies, we were unable to calculate the pooled ORs in other genetic models except additive model. Four, ORs with and without adjustment were pooled together, although there was no substantial changes between these two kinds of ORs in this synthesis, it might be a consideration source of heterogeneity. However, after restricting to crude estimations, only 24 data sets were available for synthesis. It might be non-representative and bias, which could lead to vastly different conclusions [76]. Additionally, matching was often assumed to account for confounders in case-control studies (the majority of studies were designed as case-control studies and had matched cases and controls in this meta-analysis). Thus, unadjusted findings represent adjustment that was accounted for by study design and not by the statistical methods [61]. However, we placed more emphasis on assessing biases across studies and tried to reduce potential sources of heterogeneity via stratification and sensitivity analyses. High expected powers of significant findings in this meta-analysis revealed great noteworthy and robustness. In view of this, we were confident that the findings in this metaanalysis were reliable and reasonable.
In conclusion, cumulated evidence suggested that rs401681 C allele was a tumor susceptibility allele in the development of lung, bladder, prostate and basal cell carcinomas, but a potential protective allele for pancreatic cancer and melanoma. These results suggested that the TERT-CLTMP1L polymorphism rs401681 may be potential biomarkers of cancer susceptibility. However, the effect on cancers risk may be modified by ethnicity, cancer type, source of controls and sample size. Future studies are required to validate the current findings. Figure S1 Additive ORs and corresponding 95%CI with and without adjustment were nearly identical for rs401681. Gago-Dominguez1-2 represented studies for bladder cancer in Caucasians and Asians, respectively; Nan1-3 represented studies for melanoma, squamous cell and basal cell carcinomas, respectively. (TIF) Figure S2 Additive ORs (95%CI) and corresponding allelic ORs (95%CI) of each data set for rs401681. Gago-Dominguez1-2 represented studies for bladder cancer in Caucasians and Asians, respectively; Nan1-3 represented studies for melanoma, squamous cell and basal cell carcinomas, respectively. (TIF) Figure S3 One-way sensitivity analyses. The pooled odds ratios were calculated by omitting each data set at a time.