Cigarette smoking and telomere length: A systematic review of 84 studies and meta-analysis

1Telomeres are ribonucleoprotein structures at the end of linear chromosomes essential for maintaining genome stability. Consisting of tandem arrays of TTAGGG sequence, telomeres serve as dispensable DNA sequences that shield genomic DNA from inevitable shortening during replication. In addition, the special cap structure at the end of telomeric repeats, formed by 3’ G-strand overhang and telomere associated binding proteins, prevent recognition of the linear chromosome ends as DNA double strand break by the DNA repair machinery that may result in chromosome fusions.

2Human telomeres shorten with each cell division and as telomeres become critically short, cells will cease proliferating and become senescent. As such, telomere length has long been considered as a marker of cellular aging. In addition to genetic factor, environmental influences play an important role in determining telomere length. Tobacco smoking is a well-known health risk factor and exposure to harmful chemicals in cigarettes may induce oxidative stress and irreparable damage to the telomeric DNA. Despite this biological link, there have been inconsistencies in the literature regarding association between telomere length and smoking, with some studies showing shorter telomeres with smoking whereas a lack of association was reported in other studies. We therefore performed this systematic review and meta-analysis to determine whether combined evidence supports association between telomere length and smoking.

3The meta-analysis was conducted according to the MOOSE And PRISMA guidelines. MEDLINE (Ovid) and EMBASE (Ovid) databases were searched from their inception to 29 April 2016, with the final search performed on 02/05/2016. We applied a search strategy as follows: (smoking OR cigarette*) AND (telomere OR telomeres) as free text. Searches were limited to studies conducted in humans. No language restriction was applied. References from eligible studies were hand-searched for additional studies. Two investigators independently identified eligible studies, and any discrepancies were resolved by consensus with a third investigator. There was no prior review protocol published for this study.

4We included studies that investigated an association between cigarette smoking status (including smoking status e.g. smokers, former smokers and never smokers and smoking intensity) and telomere length in humans, in which smoking status and telomere length were measured in the same subjects. Studies were either cross-sectional, cohort, or case-control studies in humans. We included studies in which smoking or telomere length was used as an adjustment variable if individual estimates of association between smoking and telomere length were available.

5Duplicated publications or additional studies of already included studies were excluded. We also excluded studies which did not fulfil any inclusion criteria, for instance, those which did not provide estimates for association between smoking and telomere length.

6Data from eligible studies were independently extracted using a standard form. The following information was collected: first author, year of publication, type of study, description of study population (age, sex, race, country of study), method of telomere length measurement, source of sample used, description of smoking exposure assessment, sample size, comparison method, main results including maximally adjusted effect size and standard error or confidence intervals, any adjustment variables, and any other relevant information.

7When information was available in the included studies, estimates for the following comparisons were collected: 1) current smokers and non-smokers, the latter of which consisted of former and never smokers, 2) ever smokers, which included both current and former smokers, and never smokers, 3) current smokers and former smokers, 3) current smokers and never smokers, 4) former smokers and never smokers, 5) smoking intensity, expressed as pack-years of cigarette, defined a product of packs of cigarettes smoked per day and smoking duration in years, 6) levels of cotinine, a metabolite of nicotine. When multiple measurements were available, we collected smoking status and telomere length measured at the same time, or closest to each other.

8For studies which only reported estimates for categories e.g. quartiles of pack-year of cigarettes, we assigned interval scores of categories from the original studies based on medians or means when available. Category midranges were applied for the remaining closed-ended categories. For upper open-ended categories with bi as the lower bound of the ith interval and the intervals indexed by i = 1,…,n, interval scores were assigned as bn + 0.5 (bn - bn-1). Correspondingly, interval scores for the lower open-ended categories were assigned as b2 − 0.5 (b2 - b1).

9Although quantitative scores have been used for meta-analysis of observational studies, interpretation could be challenging. We adapted assessment criteria from items in Critical Appraisal Skills Programme (CASP) questionnaires to assess cohort and case-control studies and use these criteria to assess included studies: (i) Did the study address a clearly focused issue? (ii) Did the authors use an appropriate method to answer their question? (iii) Was the exposure accurately measured to minimise bias? (iv) Was the outcome accurately measured to minimise bias? (v) Have they taken account of important confounding factors in the design and/or analysis? (vi) Do the results fit with other available evidence? Each item was answered with ‘Yes’, ‘No’ or ‘Don’t know’, according to information presented in the publications.

10Assessment for publication bias was carried out by assessing funnel plot asymmetry for comparisons including at least 10 studies. Data points were derived from estimates and standard errors from individual studies in relation to the pooled estimate effect. Asymmetrical distribution of data points for smaller studies (those with wider standard errors) indicates small study effects, which may be caused by publication bias. In addition to visual inspection of the funnel plot, we also conducted Egger's test, which applies weighted linear regression analysis to test for funnel plot asymmetry. A p-value of <0.1 was considered to represent significant asymmetry. Where asymmetry was indicated, sensitivity analysis was performed to seek potential sources of asymmetry.

11The studies were evaluated clinically and methodologically to assess if it was reasonable to consider combining data. Statistical heterogeneity was measured by the visual inspection of the forest plots and statistically through an assessment of homogeneity based on the Chi2 test. The I2 measurement was calculated as an indicator of the amount of statistical variation not attributable to sampling error. A value of more than 75% was considered to represent high heterogeneity.

12A random effects meta-analysis was performed to obtain pooled results for each aforementioned comparison. Because different methods were used to assess telomere length, a standardised mean difference (SMD) approach was applied in the analysis. Summary results were obtained from final values and their variance in individual studies. A sensitivity analysis was performed by excluding studies one at a time. Where difference in means were presented for categories of exposure, e.g. for pack-year of smoking, we performed a two-stage meta-analysis approach. First, a dose-response model was estimated within each study taking into account the covariance of SMD corresponding to each assigned interval score. In the second stage, study-specific estimates were combined using a multivariate random effects model. Linear and quadratic dose-response associations were assessed, and Aikake information criterion (AIC) values were used to compare model fit. Where necessary, we performed a sensitivity analysis excluding one study at a time to determine potential sources of any observed heterogeneity. Analyses were performed using the package metafor and dosresmeta in R version 3.3.2 (R Foundation for Statistical Computing, Vienna, Austria Foundation for R Foundation for Statistical Computing, Vienna, Austriatatistical Computing, Vienna, Austria) and OpenMetaAnalyst software.

13As shown in Fig. 1, we identified a total of 616 studies from initial search, of which 373 study duplicates were removed. After title and abstract screening, full-text articles for 190 studies were retrieved. Among these articles, we further excluded 110 studies for reasons reported in Fig. 1 and full-text articles for the remaining 80 studies were retrieved for detailed reviewing. Finally, 4 additional studies identified among references in retrieved studies were included, resulting in a total of 84 articles selected for systematic review. Of these, 30 reported sufficient data for inclusion in meta-analysis.

14Main characteristics of the included studies are summarised in Table 1, and references containing included studies are included in the Supplementary Information. Forty three studies were conducted in Europe, 25 in North America, 12 in Asia, 2 in Africa, and 2 in South America. In total, these studies involved 144,504 adults (age 18 and older) and most studies include both men and women (Table S1, Supplementary Information). Although some studies were longitudinal, all collected smoking status and telomere length were assessed cross-sectionally. Most studies used a single assessment of telomere length. Smoking status was self-reported in these studies. Measurements of telomere length were performed in the majority (75%) of the included studies and peripheral blood or leukocytes were mostly used as tissue source of DNA. Overall, among the 84 studies included, thirty-three reported shorter telomere length with smoking, one study reported longer telomere length with smoking, and fifty found a lack of association between smoking and telomere length. Only one included study and a secondary analysis of study already included in the review assessed the association between cotinine levels, in which shorter telomeres were indicated with higher cotinine levels. A meta-analysis was not possible because mean difference in telomere length was only available in one study. Time since quitting smoking was assessed in two studies, and weak positive trends with telomere length were reported albeit different comparison methods hampered meta-analysis to be conducted.

15In addition to cross-sectional assessment of smoking status and telomere length, three studies had two repeated measures of telomere length. When adjusting for baseline telomere, telomere length measured 6 years later was inversely linked with number of cigarettes smoked at baseline, but not smoking status. Müezzinler and colleagues showed less telomere attrition over 8 years among current compared to never smokers. Meta-analysis was not conducted due to the different types of measurements and interval time assessed.

16To assess the risk of bias, we used criteria adapted from the CASP questionnaires for observational studies. As shown in Table S2 (Supplementary Information),all studies demonstrated clear aims, and adjustment for important confounding variables was applied in 42 out of 84 studies. Accurate measurements of exposure and outcome was indicated in the majority of studies, with seventy-six studies clearly described their methods in assessing LTL, and seventy one studies described how smoking was measured.

17Assessment of publication bias was conducted for the studies included in the meta-analysis which compared telomere length between smokers and non-smokers (Fig. 2A) as well as ever and never smokers (Fig. 2B), which each contained more than 10 studies. Both visual inspection and formal test for funnel plot asymmetry indicated symmetrical distribution (Egger's test p-values = 0.12 in analysis comparing smokers and non-smokers and 0.41 when comparing ever and never smokers), indicating that small study effects, which may reflect publication bias, were unlikely.

18Meta-analyses were performed for 30 studies which provided comparable estimates for association between self-reported smoking exposures and telomere length. As several studies reported individual analyses either between smokers (current smokers) and non-smokers (former and never smokers) or ever (current and former smokers) smokers and never smokers, meta-analyses were carried out separately for each comparison. We found that telomere length were lower among smokers compared to non-smokers, with a pooled SMD of −0.17 (95% CI −0.24 to −0.09) (Fig. 3A). Moderate heterogeneity was found (I2 = 60.47%). Similarly, telomere length was lower among ever smokers than never smokers (Fig. 3B), with a summary SMD of −0.11 (95% CI −0.16 to −0.07), with I2 indicating a lack of substantial heterogeneity (I2 = 46.51%). A sensitivity analysis excluding one study at a time one at a time was performed in analysis comparing smokers and non-smokers. This analysis showed reduced heterogeneity (I2 = 26.87%) when a study by Wang and colleagues was removed, with summary SMD of −0.13 (95% CI −0.17 to −0.07) (Fig. S1).

19We further evaluated difference in telomere length across current, former and never smokers. Telomere length was suggested to be lower among current compared to never smokers (summary SMD: −0.15, 95% CI: −0.28 to −0.02) and or compared to former smokers (summary SMD: −0.06, 95% CI: −0.13 to 0). No difference in mean telomere length was observed between former and never smokers (Fig. 4). There was moderate heterogeneity when comparing current to never smokers and high heterogeneity (I2 > 75%) when comparing former to never smokers. In the sensitivity analyses excluding one study at a time, we found that exclusion of a study by Needham et al. lowered the heterogeneity substantially with I2 less than 20% in comparison of telomere length between current and never smokers (Fig. S2, summary SMD: −0.20, 95% CI −0.30 to −0.10). Similar reduction in heterogeneity was shown by excluding a study by Muezzinler et al. when comparing former and never smokers (Fig. S3), with summary results indicating no difference in telomere length between the two groups (summary SMD: −0.01, 95% CI −0.09 to 0.07).

20Five studies provided comparable estimates for mean difference of telomere length by categories of pack-years of smoking (Table S3, Supplementary Information), and they were included in a dose-response meta-analysis assessing the association between pack-year of smoking and telomere length. As shown in Fig. 5, assuming a linear relationship, there was an inverse trend between pack-years of smoking and telomere length. Each increase in pack-years of smoking corresponded to −0.01 decrease in mean telomere length (95% CI: −0.03 to −0.002) and high heterogeneity was indicated (I2 = 98.2%). Use of a quadratic model did not show any improvement of the model fit. In the sensitivity analysis, excluding any of the studies in the sensitivity analysis did not remove the observed heterogeneity.

21To our knowledge, this is the first systematic review and meta-analysis on cigarette smoking and telomere length. We found that telomere length was shorter among ever smokers compared to those never smoked. Among those who ever smoked, telomere length was indicated to be shorter among current smokers compared to those who quitted smoking. Using a dose-response meta-analysis, we found a weak inverse association between pack-years of smoking and telomere length. When assessing risk of bias, we found that half of included studies have adjusted for confounding variables in their analysis. Substantial heterogeneity was shown in some of the findings.

22The observed associations may suggest that biological ageing, indicated by telomere length, is enhanced with active smoking. A plausible mechanism is that free radicals generated following cigarette smoking may induce oxidative stress and inflammation, resulting in significantly shortened telomeres and lead to cellular senescence and apoptosis. Our results also indicated the benefit of quitting smoking compared to being current smokers with regards to telomere length. There was a weak indication that former smokers who quitted smoking for a longer period of time may have longer telomeres compared to those who quit more recently or current smokers, although a meta-analysis was not feasible. Additionally, higher pack-years of smoking, which reflect lifelong cigarette smoking exposure, corresponded to shorter telomeres although heterogeneity was shown. Altogether, these findings indicate the need to investigate whether there are cumulative effects of smoking on telomere length, and correspondingly, any reversal of such effects with longer period of smoking cessation. Studying the dynamic of telomere shortening may add understanding into such impact of smoking and cessation since it better reflects the biological impact of environmental influences compared to a single measurement of telomere length. In the study by Müezzinler and colleagues in which repeated measures of telomere length 8-year apart were assessed, greater telomere loss was shown with longer period of quitting smoking among ever smokers, albeit non-significant. The same study also showed that those who were current smokers at baseline had subsequent less telomere attrition than never smokers. Such conflicting finding may be explained by the “thrifty telomere” hypothesis, which suggests that telomere length may act as a marker of one's limited resources over the life course or “disposable soma”. This theory states that having shorter telomeres to start with results in a more thrifty investment in maintenance efforts and reduced cell proliferation, which may explain greater telomere length but more pronounced subsequent telomere loss with longer time of smoking cessation. Further investigations regarding the role of smoking and its cessation on telomere length and attrition is therefore necessary to confirm such complex association.

23The strength of this review lies in the use of different measures of smoking exposures ranging from self-report to objective biomarker measurements, and the inclusion of studies which included smoking as covariates when appropriate in addition to those evaluating smoking as the main exposure. We were able to estimate the dose-response relationship between smoking intensity, calculated as pack-years of smoking, with telomere length. Nevertheless, there was high heterogeneity when comparing telomere length between current and former smokers. Differences in population demographics may have explained such heterogeneity since exclusion of one Asian study removed the heterogeneity observed when comparing ever and never smokers. However, less heterogeneity for other analyses were observed when excluding population-based studies which mostly included Caucasians, which may indicate that there might have been factors other than ethnicities which explained the observed heterogeneity. It is unlikely that study design or telomere tissue explained this heterogeneity as all included studies in the analysis measured smoking status and telomere length cross-sectionally and used peripheral blood cells. Another possible source of heterogeneity might be the inter-laboratory technical variation of quantitative PCR (qPCR) to measure relative telomere length, which tend to yield higher variability than Southern blotting, although a study by Martin-Ruiz et al. suggested that both techniques show comparable intra- and inter-assay inconsistencies. Previous meta-analyses on cross-sectional associations between other lifestyle-related factors such as hypertension or obesity and leukocyte telomere length also reported high heterogeneity, which may indicate the need to address methodological problem in telomere association study that should involve standardisation of qPCR and Southern blotting protocols.

24Another limitation was that most included studies only used a single assessment of telomere length, which indicates the need to extend such analysis with repeated measurements. Furthermore, all studies included in the meta-analysis assessed circulating telomere length in blood. We were unable to assess any potential effect modification by sample source of telomere length such as saliva, which has also been linked to smoking. Leukocyte telomere length has been widely used as a representative parameter of individual telomere status and that shorter leukocyte telomere length has been linked with increased mortality in several studies. Nevertheless, the evidence regarding correlation of telomere length in leukocytes and other tissues is conflicting. There are discrepancies in both telomere length and attrition rate of granulocytes and lymphocytes subpopulation. Additionally, a study by Svenson et al. demonstrated fluctuation in leukocyte telomere length during the course of 6 months. This fluctuation may be due to natural cycle or environment factors, such as acute infection, that changes the composition of leukocyte subpopulations and their replicative state. Leukocyte is therefore not necessarily an ideal surrogate marker for other tissues in cross-sectional or short longitudinal studies. Most of our studies were of cross-sectional nature, therefore, our results only suggest association instead of causation. Having certain diseases, such as chronic obstructive lung disease, may result in smoking cessation and thus confound association between smoking status and telomere length. Estimates collected from included studies were adjusted for age, when information was available, in order to take into account the effect of age-related disease. However, residual confounding may still have occurred. We were able to capture lifelong exposure of smoking by pack-years of smoking, but this was based on self-report and therefore prone to recall bias. However, this was likely to result in non-differential misclassification because our outcome was a biomarker rather than any existing disease. Although we were unable to do a meta-analysis for cotinine levels, high agreement (~90%) has been shown between self-report active smoking status and cotinine levels. To further tease out this association, a life course approach which allows assessment of smoking exposure prior to telomere length assessment, and use of repeated telomere measures may enhance understanding into the mechanisms of which smoking may influence telomere length.

25In summary, this review showed that telomere length was shorter among ever smokers compared to those who never smoked, and among current smokers compared to former smokers. As shorter telomere may be a marker of early changes in ageing-related disease, the benefit suggested among former smokers compared to current smokers may further indicate the need to optimise policies targeting smoking cessation, particularly at earlier age. However, heterogeneity found in some of our findings indicates the need for further studies to quantify the implication of specific smoking exposures on telomere length and biological ageing.

26None declared

27WW is employed under grant from the UK Medical Research Council [MC_UU_12019/2; MC_UU_12019/4]. YA is supported by the Indonesian Endowment Fund for Education.

28YA, AW and WW conceptualised the review. YA, AW and WW performed literature search, title and abstract screening, full text screening and data extraction. YA, JW and WW performed statistical analysis. YA and WW wrote the first draft of the manuscript. All authors provided feedback, reviewed and approved the final version of the manuscript.