Abstract
The association between telomere length (TL) in peripheral blood cells and cancer risk remains inconclusive. We carried out a meta-analysis on prospective studies. The study-specific RR estimates were first transformed to a common comparable scale and then were pooled by a random-effects model. The dataset was composed of 13,894 cases and 71,672 controls from 28 studies in 25 articles. In the comparison of the longest versus shortest third of TL, we observed a marginally positive association between longer TL and higher risk of total cancers [OR = 1.086; 95% confidence interval (CI), 0.952–1.238]. Subgroup analyses showed that the association was stronger in lung cancer (n = 3; OR = 1.690; 95% CI, 1.253–2.280), in men (n = 6; OR = 1.302; 95% CI, 1.120–1.514) and in studies with more precise methods for DNA extraction (phenol–chloroform, salting-out or magnetic bead, n = 6, OR = 1.618; 95% CI, 1.320–1.985) and TL measurement (multiplex Q-PCR, n = 8; OR = 1.439; 95% CI, 1.118–1.852). Our meta-analysis suggested longer TL in peripheral blood cells is a likely risk factor for lung cancer or cancers in men. Accurate DNA extraction and TL measurement methods make it more liable to find significant associations between TL and cancer risk and thus should be taken into consideration in future epidemiologic studies. Cancer Epidemiol Biomarkers Prev; 26(9); 1381–90. ©2017 AACR.
Introduction
Telomeres, repetitive DNA sequences coated by capping proteins at the ends of linear chromosomes, play vital roles in the maintenance of chromosome integrity and stability during cell division. In human somatic cells without telomerase activity, telomere genetic material would lose approximately 30 to 200 bp after each mitotic cycle because of the asymmetric replication of DNA (1). With continuous shortening, when their length reaches a critical point, telomeres signal the arrest of cell proliferation, as well as senescence and apoptosis. Thus, telomere length (TL) has been suggested to be a “cellular mitotic clock” that records the number of cellular turnovers within an individual (2) and serves as a biomarker for biological age in epidemiologic studies. In addition to chronologic aging, TL is influenced by genetic factors (3) and varies considerably between individuals, even those with the same chronologic age. TL is also affected by numerous environmental factors (4); nutrients, foods, and dietary patterns (5); cigarette smoking (6); air pollution (7); physical activity (6, 8); and chronic inflammation (9).
The relationship between TL and cancer risk is debatable because of the dual role of telomeres in carcinogenesis (10, 11). In general, telomere attrition theoretically leads to genome instability, which is a fundamental event in the origin of tumors (12). However, long telomeres may allow for more cell divisions, promote immortality of the cells, and lead to aberrant cell proliferation and tumor formation (13, 14). The empirical evidence from epidemiologic studies also remains controversial. For example, some studies showed that short telomeres increased the risk of breast cancer (15, 16), whereas others revealed opposite or no associations (17, 18).
The true relationship between TL and cancer risk may be obscured by several population and study-specific characteristics. The first and foremost is age at TL measurement. For example, the prospective study of 47,102 Danish general population participants, with follow-up times of up to 20 years, showed a strong association between shorter TL and higher cancer risk in a crude model (HR, 1.74; 95% CI, 1.58–1.93, shortest vs. longest quartiles), but the association was attenuated to nonsignificance after adjusting for age at blood sampling (HR, 0.98; 95% CI, 0.88–1.08; ref. 19). The second is study design. Cancer per se and cancer treatment can shorten TL; therefore, the negative associations between TL and cancer risk reported in retrospective studies and prospective studies with short follow-up times may reflect reverse causation. The third is the accuracy of TL assessment; for example, TL measured by Q-PCR from QIAamp (column)-extracted DNA was less than from either PureGene (Salting-out) or phenol–chloroform (P < 0.001), and the association was more evident in DNA samples extracted by the latter two methods (20). The fourth is that in statistical models TL has always been treated as a categorical variable (e.g., quantiles) according to its distribution in the control subjects, but the uncommon scales of categorization make the results not easily comparable. In addition, the relationship between TL and cancer risk are highly variable across cancer types (21). A recent meta-analysis pooling unadjusted risk estimates from both prospective and retrospective studies found no association between TL and overall cancer risk and the validity was further attenuated because the study-specific RR was not transformed to a common scale before combination (22).
Given the inconsistent results and the potential sources of heterogeneity from study-specific characteristics, we conducted a systematic review and meta-analysis on prospective studies to provide a precise assessment of the relationship between TL and cancer risk based on fully variable-adjusted models. We also aimed to compare these associations across a wide range of study-level characteristics.
Materials and Methods
Search strategy
To identify studies exploring the relationship between TL and cancer risk, we conducted a systematic literature search on PubMed and Web of Science. The search strategy used the following terms: “telomere” or “telomeres,” “telomeric,” “T/S ratio,” or “T/C ratio”; “neoplasms,” “neoplasm,” “tumor,” “tumors,” “neoplasias,” “cancer,” “cancers,” “benign neoplasms,” “neoplasms, benign,” “benign neoplasm” or “neoplasm, benign”; and “risk,” “risks,” “relative risk,” “relative risks,” “risk, relative” or “risks, relative.” The search results were updated through April 2, 2017. Additional studies were sought by a manual search according to the list of references within eligible studies or reviews in the field.
Study selection
Studies eligible for inclusion in our meta-analysis had to meet the following criteria: (i) full reports in English, (ii) having evaluated the relationship between TL and cancer risk, (iii) prospective studies (e.g., nested case-control and prospective cohort studies in which DNA samples used for TL measurement were collected before diagnosis of cancer), (iv) presenting multivariate-adjusted risk estimates, that is, RR, OR, or HR across TL category groups or per unit change for cancer risk, (v) DNA extracted from peripheral blood, (vi) independent from other studies (if two or more reports used data from the same cohort, only the one with the largest sample size and/or longest follow-up time was included), and (vii) the reference group in the categorical analysis being the longest or shortest TL group. In addition, studies were excluded if participants had cancer susceptibility or had previously suffered from cancer. There were no restrictions on cancer types. Eligible studies were identified by two researchers (X. Zhang and W. Zhu) independently, and a third researcher (B. Zhang) dealt with discrepancies. The details of the process of study selection are shown in Fig. 1.
Data extraction and quality assessment
Two researchers (X. Zhang and W. Zhu) extracted data independently and reached a consensus on all projects. If two or more independent studies were reported in one article, they were treated separately in our meta-analysis unless the combined results were provided in the original study. A standardized information sheet was used to extract characteristics from eligible studies, including the first author, year of publication, geographic location, race, cancer type, population source, age at baseline or blood draw, age at diagnosis, time interval of follow-up, cohort time, number of participants (the number of cases and controls), gender, DNA extraction method, TL assay methods, number of experimental replicates, TL (mean or median), telomere units, RRs (HRs, risk ratios, and ORs) and corresponding 95% CI, and main findings. If the TL was divided into more than one category in one study, such as dichotomies, tertiles or quartiles, we used the results in tertiles. Study quality was evaluated by the Newcastle–Ottawa scale, and a study scoring equal to or greater than eight was regarded as high quality.
Statistical analysis
We made a plan for data analysis and performed analyses in strict accordance with the plan. Meta-analyses were performed with Stata version 12.0 (Stata Corp.). HRs and RR were assumed to have the same meaning as ORs. We calculated the summary OR for the association between TL and cancer risk by a random-effects model. We adopted methods previously described (23) to transform effect sizes, such as ORs, RRs, and risk ratios, into the same comparable values before meta-analysis, namely the longest versus shortest tertile of TL. Under the assumption that TL is normally distributed and has a log-linear relationship with cancer risk, the log OR for the longest versus shortest third of the distribution of TL is expected to be approximately 1.27 times the log OR for the top versus bottom half in TL, 0.86 times the longest versus shortest fourth, 0.78 times the longest versus shortest fifth and 0.62 times the longest versus shortest tenth (Supplementary Table S1). Data transformation was performed with Matlab version 7.0. The 95% CI of the transformed log OR was calculated by using the same multiplication coefficient as the originally reported 95% CI. When studies reported risk estimates with different degrees of statistical adjustment for covariates, we used the fully adjusted result.
A funnel plot and Egger's tests were used to investigate publication bias. The between-study heterogeneity was assessed by the Cochran Q test and I2 statistic and considered significant if P < 0.05 or I2 > 50%. To evaluate the effect of a single study on the result, sensitivity analyses were performed to recalculate the OR and 95% CI after excluding each study individually. Sources of heterogeneity were assessed by subgroup analyses and meta-regression. The study-level characteristics tested as sources of heterogeneity included cancer types, geographical location of the study (Europe, North America, and Asia), study quality (Newcastle–Ottawa scale score), age at blood draw, time interval of follow-up, age at cancer diagnosis, gender, DNA extraction method (QIAamp vs. Non-QIAamp method, that is, phenol-chloroform, salting-out and magnetic bead method), TL detection method (singleplex or multiplex Q-PCR), number of experimental replicates, and sample size (<500, 500–1,000, or s,000).
Results
Characteristics of included studies
As shown in Fig. 1, 2,598 publications were originally identified. After duplicate checking and title or abstract reviewing, 63 remained. Thirty-eight publications were then removed after reading full texts, resulting in 25 publications that were included in the meta-analysis (17, 18, 24–46). Liang and colleagues' study (24) included three independent nested case-control studies, skin squamous cell carcinoma (SCC) cases within the Health Professionals Follow-up Study (HPFS) and two groups of skin basal cell carcinoma (BCC) cases within the Nurses' Health Study (NHS), and Pooley and colleagues' study (18) reported results of breast cancer and colorectal cancer separately, so they were treated as separate entries, then there were totally 28 independent studies included in our meta-analysis. Han and colleagues' study (35) contained three types of skin cancer: melanoma, squamous cell carcinoma (SCC), and basal cell carcinoma (BCC), but melanoma and BCC cases were later reported in Liang and colleagues' study (24); therefore, only SCC cases were included.
The essential information of the included studies is listed in Table 1 and Supplementary Table S1. In total, there were five studies for skin cancer, three for pancreatic cancer, three for breast cancer, three for lung cancer, two for colorectal cancer, two for blood cancer, two for prostate cancer, two for all cancer risk, and six studies for other cancers (including renal cell carcinoma, glioma, endometrial, bladder, liver, and ovarian cancer). Ten studies were conducted in Europe, 15 in North America, and three in Asia. The population of the 28 studies were from 14 cohorts and eight cohorts were used at least twice and at most nine times to analyze different cancer types. We have seriously checked the data source to avoid repeated reports. We found that Seow and colleagues′ study (41) combined lung cancer cases from three cohorts: the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO), the Alpha-Tocopherol, Beta-Carotene Cancer Prevention trial (ATBC), and the Shanghai Women's Health Study (SWHS), but the subjects in the latter two studies were reported separately in Shen and colleagues' (27) and Lan and colleagues' (28) studies. Thus, only PLCO subjects in Seow and colleagues' study were included. Regarding the gender of subjects, six studies were only composed of men (ATBC and HPFS were male cohorts), 10 only of women [NHS, WHS, SWHS, WHI-OS, Women's Health Initiative Observational Study (WHI-OS) were female cohorts], and 12 of both men and women. All studies provided results adjusting for either age at blood draw or age at cancer diagnosis or both. DNA extraction method was clearly illustrated in 22 studies, among which QIAamp method was used in 16 studies and non-QIAamp methods were used in six studies: four studies using phenol–chloroform (25, 27, 28, 30), one using PureGene (Salting-out; ref. 41) and one using magnetic bead (36). All studies measured TL using quantitative PCR (Q-PCR) method. Except Zeng and colleagues′ study (45), 27 studies used Cawthon method that was originally developed in 2002 (singleplex Q-PCR; ref. 47) and updated in 2009 (multiplex Q-PCR; ref. 48), expressing TL as the ratio (T/S) of the telomere repeat (T) copy number to the single copy gene (S) copy number. Among them 19 studies used singleplex Q-PCR method (47) and eight used multiplex Q-PCR.
Characteristics of studies included in the meta-analysis
. | . | . | . | . | . | . | . | Age at baseline or blood draw (mean± SD) . | . | . | . | . | . | . | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Author . | Year . | Location . | Ethnic . | Cancer location . | Population source . | Sex . | Case/control . | Case . | Control . | Age at diagnosis(mean or median) . | Time intervals . | DNA extraction method . | TL detection method . | Main findf . | NOS quality score . |
McGrath et al. (33) | 2007 | USA | Caucasian | Bladder | HPFS, NHS | Mixed | 184/192 | 64.1 | 64.1 | NR | NR | QIAamp | Singleplex | Linear(-) | 8 |
Lan et al. (30) | 2009 | Finland | European | Blood | ATBC | Male | 107/107 | 58.3 | 57.5 | NR | NR | P-C | Multiplex | Linear(+) | 9 |
Mirabello et al. (34) | 2009 | USA | Caucasian | Prostate | PLCO | Male | 612/1049 | 64 | 64 | NR | NR | QIAamp | Singleplex | No | 8 |
De Vivo et al. (17) | 2009 | USA | Caucasian | Breast | NHS | Female | 1,122/1,147 | 58.4 | 58.4 | 65.7 | NR | QIAamp | Singleplex | No | 8 |
Han et al. (35)a | 2009 | USA | European | SCC | NHS | Female | 285/273 | 59.3 | 58.8 | 64.7 | NR | QIAamp | Singleplex | No | 8 |
Prescott et al. (32) | 2010 | USA | Caucasian | Endometrial | NHS, WHS | Female | 279/791 | NR | NR | NR | 5.9 years | QIAamp | Singleplex | No | 8 |
Willeit et al. (31) | 2010 | Italy | European | Cancer risk | BS | Mixed | 92/695 | 62.6 | 62.6 | 67 | NR | NR | Singleplex | Linear(-) | 9 |
Pooley et al. (18)b | 2010 | UK | European | Breast | EPIC | Female | 199/420 | 64 | 64 | NR | 3.3 years | NR | Singleplex | No | 8 |
Pooley et al. (18)b | 2010 | UK | European | Colorectal | EPIC | Mixed | 185/406 | 64 | 64 | NR | 3.3 years | NR | Singleplex | No | 8 |
Lee et al. (37) | 2010 | USA | White | Colorectal | WHS | Female | 134/357 | 60.10 | 60.66 | NR | 5.80 years | QIAamp | Singleplex | No | 8 |
Nan et al. (39) | 2011 | Europe | European | Melanoma | WHI-OS, HPFS, NHS | Mixed | 557/579 | 60.7 | 60.7 | 65.9 | NR | QIAamp | Singleplex | Linear(-) | 9 |
Shen et al. (27) | 2011 | Finland | Finnish | Lung | ATBC | Male | 229/229 | 59 | 58 | NR | 5.23 years | P-C | Multiplex | No | 8 |
Liang et al. (24)b | 2011 | USA | Caucasian | SCC | HPFS | Male | 241/241 | 63.1 | 63.1 | 69.0 | NR | QIAamp | Singleplex | No | 8 |
Liang et al. (24)b | 2011 | USA | Caucasian | BCC1 | NHS | Female | 260/260 | 56.3 | 56.3 | 66.1 | NR | QIAamp | Singleplex | No | 8 |
Liang et al. (24)b | 2011 | USA | Caucasian | BCC2 | NHS | Female | 363/1,683 | 61.0 | 59.5 | 69.3 | NR | QIAamp | Singleplex | No | 6 |
Walcott et al. (40) | 2013 | Europe | European | Brain | PLCO | Mixed | 101/198 | NR | NR | NR | NR | NR | Singleplex | No | 8 |
Lynch et al. (25) | 2013 | Finland | Finnish | Pancreatic | ATBC | Male | 193/660 | 58 | 58 | NR | 6.3 years | P-C | Multiplex | Linear(+) | 8 |
Hofmann et al. (26) | 2013 | USA | Mixed | Renal | PLCO | Mixed | 209/410 | NR | NR | NR | NR | QIAamp | Multiplex | No | 7 |
Qu et al. (38) | 2013 | China | Chinese | Breast | SWHS | Female | 601/695 | 52.7 | 53.4 | NR | NR | QIAamp | Multiplex | Linear(-) | 9 |
Lan et al. (28) | 2013 | China | Chinese | Lung | SWHS | Female | 215/215 | NR | NR | NR | NR | P-C | Multiplex | Linear(+) | 9 |
Hosnijeh et al. (29) | 2014 | USA | Caucasian | Blood | EPIC | Mixed | 464/464 | 56.58 | 56.59 | 61.5 | 4.6 years | NR | Singleplex | Linear(+) | 8 |
Campa et al. (36) | 2014 | Europe | Mixed | Pancreatic | EPIC | Mixed | 331/331 | 57 | 57 | 63 | 5.3 years | Salting-out | Singleplex | No | 8 |
Seow et al. (41)c | 2014 | USA | Caucasian | Lung | PLCO | Mixed | 403/403 | 64.1 | 63.6 | NR | 7.41 years | Magnetic Bead | Multiplex | Linear(+) | 8 |
Julin et al. (43) | 2015 | USA | Caucasian | Prostate | HPFS | Male | 922/935 | 63.6 | 63.5 | 69.5 | 5.5 years | QIAamp | Singleplex | Linear(+) | 8 |
Rode et al. (42) | 2016 | Denmark | Danish | Cancer risk | CCHS, CGPS | Mixed | 4,510/56,773 | NR | NR | NR | 5 years | QIAamp | multiplex | No | 8 |
Bao et al. (44) | 2016 | USA | European | Pancreatic | Five cohortd | Mixed | 386/896 | NR | NR | NR | 6.7 years | QIAamp | Singleplex | Linear(-) | 8 |
Zeng et al. (45) | 2017 | China | Chinese | Liver | CSP | Mixed | 268/536 | NR | NR | 53.48 | NR | NR | NR | No | 8 |
Yang et al. (46) | 2017 | USA | Caucasian | Ovarian | NHS, NSHDS | Female | 442/727 | 54.03 | 54.59 | 63.66 | 9.7 years | QIAamp | Singleplex | Linear(-) | 8 |
. | . | . | . | . | . | . | . | Age at baseline or blood draw (mean± SD) . | . | . | . | . | . | . | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Author . | Year . | Location . | Ethnic . | Cancer location . | Population source . | Sex . | Case/control . | Case . | Control . | Age at diagnosis(mean or median) . | Time intervals . | DNA extraction method . | TL detection method . | Main findf . | NOS quality score . |
McGrath et al. (33) | 2007 | USA | Caucasian | Bladder | HPFS, NHS | Mixed | 184/192 | 64.1 | 64.1 | NR | NR | QIAamp | Singleplex | Linear(-) | 8 |
Lan et al. (30) | 2009 | Finland | European | Blood | ATBC | Male | 107/107 | 58.3 | 57.5 | NR | NR | P-C | Multiplex | Linear(+) | 9 |
Mirabello et al. (34) | 2009 | USA | Caucasian | Prostate | PLCO | Male | 612/1049 | 64 | 64 | NR | NR | QIAamp | Singleplex | No | 8 |
De Vivo et al. (17) | 2009 | USA | Caucasian | Breast | NHS | Female | 1,122/1,147 | 58.4 | 58.4 | 65.7 | NR | QIAamp | Singleplex | No | 8 |
Han et al. (35)a | 2009 | USA | European | SCC | NHS | Female | 285/273 | 59.3 | 58.8 | 64.7 | NR | QIAamp | Singleplex | No | 8 |
Prescott et al. (32) | 2010 | USA | Caucasian | Endometrial | NHS, WHS | Female | 279/791 | NR | NR | NR | 5.9 years | QIAamp | Singleplex | No | 8 |
Willeit et al. (31) | 2010 | Italy | European | Cancer risk | BS | Mixed | 92/695 | 62.6 | 62.6 | 67 | NR | NR | Singleplex | Linear(-) | 9 |
Pooley et al. (18)b | 2010 | UK | European | Breast | EPIC | Female | 199/420 | 64 | 64 | NR | 3.3 years | NR | Singleplex | No | 8 |
Pooley et al. (18)b | 2010 | UK | European | Colorectal | EPIC | Mixed | 185/406 | 64 | 64 | NR | 3.3 years | NR | Singleplex | No | 8 |
Lee et al. (37) | 2010 | USA | White | Colorectal | WHS | Female | 134/357 | 60.10 | 60.66 | NR | 5.80 years | QIAamp | Singleplex | No | 8 |
Nan et al. (39) | 2011 | Europe | European | Melanoma | WHI-OS, HPFS, NHS | Mixed | 557/579 | 60.7 | 60.7 | 65.9 | NR | QIAamp | Singleplex | Linear(-) | 9 |
Shen et al. (27) | 2011 | Finland | Finnish | Lung | ATBC | Male | 229/229 | 59 | 58 | NR | 5.23 years | P-C | Multiplex | No | 8 |
Liang et al. (24)b | 2011 | USA | Caucasian | SCC | HPFS | Male | 241/241 | 63.1 | 63.1 | 69.0 | NR | QIAamp | Singleplex | No | 8 |
Liang et al. (24)b | 2011 | USA | Caucasian | BCC1 | NHS | Female | 260/260 | 56.3 | 56.3 | 66.1 | NR | QIAamp | Singleplex | No | 8 |
Liang et al. (24)b | 2011 | USA | Caucasian | BCC2 | NHS | Female | 363/1,683 | 61.0 | 59.5 | 69.3 | NR | QIAamp | Singleplex | No | 6 |
Walcott et al. (40) | 2013 | Europe | European | Brain | PLCO | Mixed | 101/198 | NR | NR | NR | NR | NR | Singleplex | No | 8 |
Lynch et al. (25) | 2013 | Finland | Finnish | Pancreatic | ATBC | Male | 193/660 | 58 | 58 | NR | 6.3 years | P-C | Multiplex | Linear(+) | 8 |
Hofmann et al. (26) | 2013 | USA | Mixed | Renal | PLCO | Mixed | 209/410 | NR | NR | NR | NR | QIAamp | Multiplex | No | 7 |
Qu et al. (38) | 2013 | China | Chinese | Breast | SWHS | Female | 601/695 | 52.7 | 53.4 | NR | NR | QIAamp | Multiplex | Linear(-) | 9 |
Lan et al. (28) | 2013 | China | Chinese | Lung | SWHS | Female | 215/215 | NR | NR | NR | NR | P-C | Multiplex | Linear(+) | 9 |
Hosnijeh et al. (29) | 2014 | USA | Caucasian | Blood | EPIC | Mixed | 464/464 | 56.58 | 56.59 | 61.5 | 4.6 years | NR | Singleplex | Linear(+) | 8 |
Campa et al. (36) | 2014 | Europe | Mixed | Pancreatic | EPIC | Mixed | 331/331 | 57 | 57 | 63 | 5.3 years | Salting-out | Singleplex | No | 8 |
Seow et al. (41)c | 2014 | USA | Caucasian | Lung | PLCO | Mixed | 403/403 | 64.1 | 63.6 | NR | 7.41 years | Magnetic Bead | Multiplex | Linear(+) | 8 |
Julin et al. (43) | 2015 | USA | Caucasian | Prostate | HPFS | Male | 922/935 | 63.6 | 63.5 | 69.5 | 5.5 years | QIAamp | Singleplex | Linear(+) | 8 |
Rode et al. (42) | 2016 | Denmark | Danish | Cancer risk | CCHS, CGPS | Mixed | 4,510/56,773 | NR | NR | NR | 5 years | QIAamp | multiplex | No | 8 |
Bao et al. (44) | 2016 | USA | European | Pancreatic | Five cohortd | Mixed | 386/896 | NR | NR | NR | 6.7 years | QIAamp | Singleplex | Linear(-) | 8 |
Zeng et al. (45) | 2017 | China | Chinese | Liver | CSP | Mixed | 268/536 | NR | NR | 53.48 | NR | NR | NR | No | 8 |
Yang et al. (46) | 2017 | USA | Caucasian | Ovarian | NHS, NSHDS | Female | 442/727 | 54.03 | 54.59 | 63.66 | 9.7 years | QIAamp | Singleplex | Linear(-) | 8 |
Abbreviations: NR, not reported; P-C, phenol–chloroform method.
Study acronyms: HPFS, Health Professionals Follow-up Study; NHS, Nurses' Health Study; ATBC, Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study; PLCO Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial; PHS, Physicians' Health Study; WHS, Women's Health Study; BS, Bruneck Study; EPIC, European Prospective Investigation into Cancer and Nutrition; CCHS, Copenhagen City Heart Study; CGPS, Copenhagen General Population Study; CSP, Taiwan Cancer Screening Program cohort; NSHDS, the Northern Sweden Health and Disease Study.
aHan and colleagues' study included SCC, BCC, and melanoma. We speculated that BCC and melanoma cases were reanalyzed in Nan and colleagues' study and Liang and colleagues' study, respectively, so only SCC was included in the current meta-analysis.
bThese data were reported separately in original article.
cOnly samples from PLOC were included to avoid repeated reports.
dIncluding the NHS, HPFS, the PHS I, WHI-OS, the WHS.
fLinear(+) represent longer TL is a risk factor for cancer, Linear(−) represent shorter TL is a risk factor for cancer.
Meta-analysis results
When all eligible studies were combined, the pooled OR for cancer risk comparing the longest versus shortest third of TL was 1.086 (95% CI, 0.952–1.238) with moderate heterogeneity between studies (P for heterogeneity test <0.001, I2 = 70.4%), as shown in Fig. 2. Table 2 shows the results of subgroup analyses. Pooled ORs differed significantly by cancer types, gender, DNA extraction method, and TL measurement method. The associations between longer TL and higher cancer risk were statistically significant in lung cancer (n = 3; OR = 1.690; 95% CI, 1.253–2.280), in men (n = 6; OR = 1.302; 95% CI, 1.120–1.514), and in studies with more precise method for DNA extraction (non-QIAamp methods, i.e., phenol–chloroform, salting-out or magnetic bead, n = 6; OR = 1.618; 95% CI, 1.320–1.985) and TL measurement method (multiplex Q-PCR, n = 8; OR = 1.439; 95% CI, 1.118–1.852).
Forest plot of pooled studies for the association between TL and cancer risk. NOTE: Weights are from random effects analysis.
Forest plot of pooled studies for the association between TL and cancer risk. NOTE: Weights are from random effects analysis.
Associations between TL and risk of cancer stratified by chosen factors
. | . | . | . | Pheterogeneity . | . | |
---|---|---|---|---|---|---|
Variables . | No. of studies . | Sample case/control . | Longest vs. shortest OR (95% CI) . | P . | I2 . | P for meta-regression . |
All | 28 | 13,894/71,672 | 1.086 (0.952–1.238) | 0.000 | 70.4% | |
Cancer type | 0.5540 | |||||
Pancreatic cancer | 3 | 910/1,887 | 1.086 (0.650–1.815) | 0.009 | 78.7% | |
Breast cancer | 3 | 1,922/2,262 | 0.983 (0.607–1.592) | 0.035 | 70.1% | |
Lung cancer | 3 | 847/847 | 1.690 (1.253–2.280) | 0.547 | 0.0% | |
Colorectal cancer | 2 | 319/763 | 1.005 (0.675–1.496) | 0.661 | 0.0% | |
Skin cancer | 5 | 1,706/3,036 | 1.208 (0.860–1.697) | 0.024 | 64.3% | |
Others | 12 | 8,190/62,877 | 0.979 (0.799–1.200) | 0.000 | 79.0% | |
Location | 0.4688 | |||||
Europe | 10 | 6,504/60,398 | 1.119 (0.843–1.484) | 0.000 | 77.7% | |
North America | 15 | 6,306/9,828 | 1.013 (0.856–1.198) | 0.000 | 66.4% | |
Asia | 3 | 1,084/1,446 | 1.445 (0.903–2.314) | 0.085 | 59.3% | |
Study quality | 0.4870 | |||||
<8 | 2 | 572/2,093 | 0.901 (0.706–1.150) | 0.740 | 0.0% | |
≥8 | 26 | 13,322/69,579 | 1.105 (0.960–1.271) | 0.000 | 71.9% | |
Study design | 0.0970 | |||||
Nested case–control study | 27 | 9,292/14,204 | 1.128 (0.978–1.302) | 0.000 | 66.4% | |
Cohort | 2 | 4,602/57,468 | 0.585 (0.172–1.991) | 0.000 | 93.2% | |
Age at blood draw | 0.5136 | |||||
<60 | 12 | 4,676/7,367 | 1.208 (0.963–1.514) | 0.000 | 69.9% | |
≥60 | 10 | 3,529/5,277 | 1.013 (0.777–1.322) | 0.000 | 76.9% | |
NR | 6 | 5,689/59,028 | 0.965 (0.742–1.256) | 0.025 | 61.1% | |
Time interval of follow-up | 0.9467 | |||||
<5 | 3 | 848/1,290 | 1.134 (0.533–2.415) | 0.004 | 81.9% | |
≥5 | 10 | 7,829/62,102 | 1.081 (0.918–1.274) | 0.004 | 62.9% | |
NR | 15 | 5,217/8,280 | 1.063 (0.847–1.334) | 0.000 | 73.9% | |
Age at cancer diagnosis | 0.7206 | |||||
<65 | 8 | 2,491/4,011 | 1.192 (0.892–1.592) | 0.001 | 70.3% | |
≥65 | 11 | 4,478/7,126 | 1.008 (0.790–1.287) | 0.000 | 72.9% | |
NR | 9 | 6,925/60,535 | 1.082 (0.864–1.355) | 0.000 | 73.4% | |
Gender | 0.2849 | |||||
Men | 6 | 2,304/3,221 | 1.302 (1.120–1.514) | 0.287 | 19.4% | |
Women | 10 | 3,900/6,568 | 1.002 (0.817–1.229) | 0.024 | 52.9% | |
Mixed | 12 | 7,690/61,883 | 0.997 (0.772–1.289) | 0.000 | 80.0% | |
DNA extraction method | 0.0210 | |||||
QIAamp | 16 | 11,107/67,008 | 1.017 (0.891–1.161) | 0.000 | 62.8% | |
Non-QIAamp | 6 | 1,478/1,945 | 1.618 (1.320–1.985) | 0.464 | 0.0% | |
NR | 6 | 1,309/2,719 | 0.829 (0.474–1.449) | 0.000 | 82.9% | |
Sample size | 0.8123 | |||||
≤500 | 7 | 1,211/1,539 | 1.213 (0.828–1.776) | 0.003 | 69.6% | |
500–1,000 | 11 | 2,889/4,858 | 1.054 (0.778–1.429) | 0.000 | 72.6% | |
≥1,000 | 10 | 9,794/65,275 | 1.046 (0.892–1.226) | 0.000 | 73.3% | |
Method of TL measurement | 0.0825 | |||||
Singleplex qPCR | 19 | 7,159/11,644 | 0.971 (0.813–1.160) | 0.000 | 73.0% | |
Multiplex qPCR | 8 | 6,467/59,492 | 1.439 (1.118–1.852) | 0.002 | 68.3% | |
Others | 1 | 268/536 | 0.900 (0.523–1.549) | – | – |
. | . | . | . | Pheterogeneity . | . | |
---|---|---|---|---|---|---|
Variables . | No. of studies . | Sample case/control . | Longest vs. shortest OR (95% CI) . | P . | I2 . | P for meta-regression . |
All | 28 | 13,894/71,672 | 1.086 (0.952–1.238) | 0.000 | 70.4% | |
Cancer type | 0.5540 | |||||
Pancreatic cancer | 3 | 910/1,887 | 1.086 (0.650–1.815) | 0.009 | 78.7% | |
Breast cancer | 3 | 1,922/2,262 | 0.983 (0.607–1.592) | 0.035 | 70.1% | |
Lung cancer | 3 | 847/847 | 1.690 (1.253–2.280) | 0.547 | 0.0% | |
Colorectal cancer | 2 | 319/763 | 1.005 (0.675–1.496) | 0.661 | 0.0% | |
Skin cancer | 5 | 1,706/3,036 | 1.208 (0.860–1.697) | 0.024 | 64.3% | |
Others | 12 | 8,190/62,877 | 0.979 (0.799–1.200) | 0.000 | 79.0% | |
Location | 0.4688 | |||||
Europe | 10 | 6,504/60,398 | 1.119 (0.843–1.484) | 0.000 | 77.7% | |
North America | 15 | 6,306/9,828 | 1.013 (0.856–1.198) | 0.000 | 66.4% | |
Asia | 3 | 1,084/1,446 | 1.445 (0.903–2.314) | 0.085 | 59.3% | |
Study quality | 0.4870 | |||||
<8 | 2 | 572/2,093 | 0.901 (0.706–1.150) | 0.740 | 0.0% | |
≥8 | 26 | 13,322/69,579 | 1.105 (0.960–1.271) | 0.000 | 71.9% | |
Study design | 0.0970 | |||||
Nested case–control study | 27 | 9,292/14,204 | 1.128 (0.978–1.302) | 0.000 | 66.4% | |
Cohort | 2 | 4,602/57,468 | 0.585 (0.172–1.991) | 0.000 | 93.2% | |
Age at blood draw | 0.5136 | |||||
<60 | 12 | 4,676/7,367 | 1.208 (0.963–1.514) | 0.000 | 69.9% | |
≥60 | 10 | 3,529/5,277 | 1.013 (0.777–1.322) | 0.000 | 76.9% | |
NR | 6 | 5,689/59,028 | 0.965 (0.742–1.256) | 0.025 | 61.1% | |
Time interval of follow-up | 0.9467 | |||||
<5 | 3 | 848/1,290 | 1.134 (0.533–2.415) | 0.004 | 81.9% | |
≥5 | 10 | 7,829/62,102 | 1.081 (0.918–1.274) | 0.004 | 62.9% | |
NR | 15 | 5,217/8,280 | 1.063 (0.847–1.334) | 0.000 | 73.9% | |
Age at cancer diagnosis | 0.7206 | |||||
<65 | 8 | 2,491/4,011 | 1.192 (0.892–1.592) | 0.001 | 70.3% | |
≥65 | 11 | 4,478/7,126 | 1.008 (0.790–1.287) | 0.000 | 72.9% | |
NR | 9 | 6,925/60,535 | 1.082 (0.864–1.355) | 0.000 | 73.4% | |
Gender | 0.2849 | |||||
Men | 6 | 2,304/3,221 | 1.302 (1.120–1.514) | 0.287 | 19.4% | |
Women | 10 | 3,900/6,568 | 1.002 (0.817–1.229) | 0.024 | 52.9% | |
Mixed | 12 | 7,690/61,883 | 0.997 (0.772–1.289) | 0.000 | 80.0% | |
DNA extraction method | 0.0210 | |||||
QIAamp | 16 | 11,107/67,008 | 1.017 (0.891–1.161) | 0.000 | 62.8% | |
Non-QIAamp | 6 | 1,478/1,945 | 1.618 (1.320–1.985) | 0.464 | 0.0% | |
NR | 6 | 1,309/2,719 | 0.829 (0.474–1.449) | 0.000 | 82.9% | |
Sample size | 0.8123 | |||||
≤500 | 7 | 1,211/1,539 | 1.213 (0.828–1.776) | 0.003 | 69.6% | |
500–1,000 | 11 | 2,889/4,858 | 1.054 (0.778–1.429) | 0.000 | 72.6% | |
≥1,000 | 10 | 9,794/65,275 | 1.046 (0.892–1.226) | 0.000 | 73.3% | |
Method of TL measurement | 0.0825 | |||||
Singleplex qPCR | 19 | 7,159/11,644 | 0.971 (0.813–1.160) | 0.000 | 73.0% | |
Multiplex qPCR | 8 | 6,467/59,492 | 1.439 (1.118–1.852) | 0.002 | 68.3% | |
Others | 1 | 268/536 | 0.900 (0.523–1.549) | – | – |
To explore the sources of heterogeneity, we conducted meta-regression by various clinically relevant characteristics and methodologic factors. DNA extraction method (P = 0.021), TL measurement method (P = 0.083), and study design (P = 0.097) might explain some source of heterogeneity for overall cancer risk. The sensitivity analysis indicated that no single study could change the pooled OR qualitatively.
Publication bias
A funnel plot was drawn for 28 studies, and results showed no obvious asymmetry (Fig. 3), which was consistent with Egger's test and Begg's test (P = 0.897; Supplementary Figs. S1 and S2), suggesting that there was no evidence of publication bias.
Discussion
We found a marginally positive association between longer TL and higher cancer risk in the present meta-analysis of 28 prospective studies in 25 articles. The associations were more significant in lung cancer, cancers in men, and in studies with more precise methods used in DNA extraction and TL measurement.
Comparison with previous meta-analysis and strengths
Our findings are not consistent with Zhu and colleagues' meta-analysis published recently (22), which found no statistically significant association between peripheral TL and overall cancer risk and short telomeres might be risk factors for the tumors of digestive system. Several factors may contribute to these disparities. First, we included 28 prospective studies in contrast to 16 prospective and 46 retrospective studies in Zhu and colleagues' study. Thus, more information is available and reverse causation is expelled at most in our study. Second, our study primarily used adjusted risk estimates, which made the results more robust than Zhu and colleagues' study, in which crude data were combined. Third, we transformed risk estimates to a common scale of comparison before meta-analysis, which improved the power to detect the relationship between TL and cancer risk by reducing the probability of misclassification.
Causal inference
Although our meta-analysis was based on the results from prospective studies which are less prone to reverse causation bias than retrospective studies, the associations we found could be subject to residual confounding factors or other bias induced by meta-analysis itself. Thus, it is cautious about causal inference. Recently, Mendelian randomization studies, using germline genetic variants as instrumental variables for their random distribution in the general population, are performed to appraise the causal relevance of TL for cancer risk. Seven SNP robustly associated with white blood cell (WBC) TL were identified in a recent genome-wide association study (GWAS; ref. 49). Although these variants were believed to explain only a small proportion of the total variation in TL (0.08%–0.36%), a genetic risk score (GRS) based on them was used as an instrumental variable in Mendelian randomization studies to explore the causal relationship between TL and risk of cancer. Machiela and colleagues (50) showed that genetic variants related to a longer TL are associated with an increased lung cancer risk among never-smoking females in Asia (OR, 1.51; 95% CI, 1.34–1.69, comparing upper quartile to lower quartile of GRS predicting longer TL). Luu and colleagues (51) found short TL determined by six of the seven lead SNPs was associated with a decreased risk of breast cancer in Asian women (OR, 0.83; 95% CI, 0.72–0.95, for tertile 3 vs. tertile 1 of GRS predicting shorter TL). The telomeres Mendelian randomization collaboration recently reported that increased TL due to germline genetic variation was significantly associated with increased risk for nine site-specific cancers, that is, glioma, serous low-malignant-potential ovarian cancer, lung adenocarcinoma, neuroblastoma, bladder cancer, melanoma, testicular cancer, kidney cancer, and endometrial cancer (21). These estimates from genetic analyses, which in general avoid reverse causation bias and confounding factors in observational studies that directly build the relationship between TL and cancer risk, can be viewed as causal (52) and are consistent with our results, supporting that longer TL is a risk factor for cancer, at least for some types of cancers.
Implications of subgroup analyses
There are four study-specific characteristics, that is, cancer type, gender, DNA extraction method, and TL measurement method, that might modify the relationships between TL and cancer risk. Our findings of the tissue (lung)-specific association between longer TL and higher cancer risk are compatible with the results of the recent Mendelian randomization studies described previously: not every cancer risk increased with genetic determinants of long TL (21, 42). Such tissue-specific relationship need to be confirmed by future studies and the relevant mechanisms are also worthy of research.
In the stratification analysis of gender, we observed that longer TL was significantly related to higher cancer risk only in men. The underlying mechanism for such effect modification by gender is not fully known. Taking into account that women have longer TL than men (53) and that men have greater exposure to toxic factors that can shorten TL (e.g., smoking, alcohol, and stress) (54), the male-specific relationship needs further confirmation.
As for DNA extraction and TL measurement methods, Cunningham and colleagues (20) and Hofmann and colleagues (55) demonstrated that DNA extraction method is crucial to TL measurement and DNA extraction using phenol–chloroform, salting-out or magnetic bead methods produced more authentic and longer TL than QIAamp method which has more vigorous vortexing and multiple washes, and is prone to destroy DNA integrity, so the studies using QIAamp-extracted DNA are the least likely to find an association between TL and cancer risk (20). Similarly, multiplex Q-PCR method measured both T and S in each reaction well, providing improved consistency compared to singleplex Q-PCR which measured T and S in separate reaction wells, so the amount of DNA pipetted into the T and S reaction wells results in variation in T/S. Our meta-analysis indicates that accurate assessment of TL, from DNA extraction to Q-PCR, is critical to find true associations between TL and cancer risk. Thus, it is recommended that a standard protocol for DNA extraction (phenol–chloroform) and TL measurement (multiplex Q-PCR) should be made in further epidemiologic studies.
It is worth noting that cancer types, gender, DNA extraction method, and TL measure method are not randomly distributed across the included 28 studies. They are interrelated and coexist in some studies, so it is scarcely possible to disentangle these factors and identify the vital source of heterogeneity. For example, five of eight studies adopting multiplex Q-PCR used non-QIAamp method to extract DNA; all three lung cancer studies, as well as three male studies (half of the total six male studies) from ATBC cohort which is composed of male smokers, used both non-QIAamp and multiplex Q-PCR methods.
Biological mechanism of association
Our meta-analysis supports the viewpoint that longer TL is related to higher cancer risk, which is compatible with known biology. First, telomere elongation must rely on telomerase activity, but telomerase may have telomere-independent functions when reactivated in carcinogenesis. For example, Ghosh and colleagues (56) found that telomerase directly regulates NF-κB-dependent gene expression, such as that of IL6 and TNFα, cytokines that are critical for inflammation and cancer progression. Telomerase could also regulate MYC-driven oncogenesis by stabilizing MYC levels on chromatin through regulating MYC ubiquitination and proteasomal degradation independent of its reverse transcriptase activity and role in telomere elongation (57), these results indicate TL is likely a surrogate of telomerase activity and telomerase may play a vital role in cell transformation. Second, although most studies have observed shorter TL in relation to environmental and occupational chemical exposures (4), some studies have reported carcinogen exposures [e.g., arsenic (58), persistent organic pollutants (59)], occupational exposure to benzene (60), and short-term exposure to particulate matter (61, 62) to be related to longer TL. These paradoxical phenomena reflect the complicated roles of TL and its regulatory signals during chemical carcinogenesis.
Limitations
Although our study has some advantages, as described previously, several limitations need to be mentioned. First, the method we used to harmonize the reported associations across studies to a common scale has three prerequisites: approximate normality of the distribution of TL, equal standard deviation of TL in cases and controls, and a log-linear association between TL and cancer risk (23). We have no original data to judge whether every study met these preconditions. Second, we were unable to test the nonlinear associations between TL and cancer risk. For example, Cui and colleagues' study (63) revealing a U-shaped association between TL in peripheral blood cells and colorectal cancer risk was excluded from the current meta-analysis. U-shaped or J-shaped associations indicate that either excessively shorter or longer TL may contribute to cancer development. The inflection point is unknown and the delicate TL balance in cells is worthy of detailed study. Third, average TL across all chromosomes was used in the included studies; however, it is not the average TL but the shortest telomeres that contribute to telomere dysfunction (64), and it is further hypothesized that chromosome-specific telomeres may be involved in specific cancer etiology (65). A technique for measuring chromosome-specific TL is available (66) but is less feasible for large scale epidemiologic studies. Fourth, meta-analyses of observational studies are subject to biases inherent in the original studies, and we could not access the individual data to adjust for the potential confounders across studies, for example, time interval from blood draw to cancer diagnosis. Thus, the heterogeneity among studies was not fully explored. Finally, 22 of the 28 included studies clearly described the average age at blood draw. All of them were older than 50 years. Given the negative association between TL and age and the individual difference of TL erosion dynamics, TL at different ages, especially at younger age, even at birth may serve as a predictive biomarker for cancer risk.
In conclusion, our meta-analysis including prospective studies showed a marginally positive association between TL and over all cancer risk. The association was stronger in lung cancer, cancers in men and in studies using precise method extracting DNA and measuring TL. More large-scale and well-designed prospective cohort studies are required to further explore the predictive values of TL on cancer risk and uncover the underlying biological mechanisms.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Grant Support
This work was supported by the National Natural Science Foundation of China (to B. Zhang, 81273099).