Association between air pollution and telomere length: A study of 471,808 UK Biobank participants

Previous research suggested an association between air pollution and shortened telomere length (TL), a biomarker of oxidative stress and inflammation. However, supporting results are challenged by the small sample size and heterogeneity in participant characteristics. To comprehensively evaluate the association of long-term exposure to air pollution with telomere length, we studied 471,808 participants from UK Biobank with measurements on leukocyte telomere length (LTL). Air pollution data on PM 2.5 , PM 10 , NO 2 , NO x , SO 2 , and CO before baseline at 1 km spatial resolution were collected and linked to each participant’s residential address. We applied mixed-effects linear regression models to examine the association between long-term air pollution exposure and LTL. Compared to the lowest quartile (Q1) of air pollutants, the estimated percentage changes of age-corrected LTL were -2.71% [95% confidence interval (CI): -3.78, -1.63] for SO 2 , -0.82% (95% CI: -1.87, 0.23) for NO 2 , -1.17% (95% CI: -2.23, -0.11) for NO x , and -0.47% (95% CI: -1.45, 0.53) for CO in the highest quartile groups (Q4). Decreasing trends in age-corrected LTL following the increase in PM 2.5 and PM 10 leveled off during high levels of air pollutants. Among participants with lower household income, lower educational attainment, and higher BMI, a stronger association was found between air pollution and LTL. Our findings suggest a negative association between air pollution and LTL and provide insights into the potential pathways linking air pollution to age-related diseases.


INTRODUCTION
Ambient air pollution exposure is the world' s largest environmental risk factor for disease and premature death. 1 In 2019, it was estimated that 99% of the world' s population is exposed to polluted air that exceeds WHO air quality guidelines levels. 2 Globally, approximately 4.5 million premature deaths could be attributable to ambient air pollution in 2019, corresponding to 7.8% of total mortality. 3 The health effects of air pollution have led to much interest in understanding the potential pathways of air pollution-induced preclinical cellular responses. Several studies suggest air pollution may accelerate biological aging, during which process systemic inflammation and oxidative stress could be involved. 4,5 Telomere is a ribonucleoprotein complex of repetitive nucleotide sequences found at the ends of chromosomes. 6 It protects chromosomes from damage and end-to-end fusion and plays an important role in the maintenance of chromosome integrity and prevention of the loss of genetic information. Telomere length (TL) is known to be affected by environmental exposures throughout the life span. 7 TL is considered an indicator of the overall cellular turnover independent of chronological age, and it has been linked to a range of human biological traits and degenerative diseases (e.g., dementia, diabetes, and stroke). 8 At the cellular level, TL has been proposed as a potential indicator of oxidative stress and cellular inflammation, 9,10 which are recognized as underlying mechanisms through which air pollutants can induce organ damage and contribute to a variety of health issues. Recent studies show that shortened TL could be an important early health signal following air pollution exposure. 11 It has been argued that air pollution exposure might be a determinant of telomere shortening. Most prior research has focused on the effects of maternal exposure to air pollution on newborn TL, 12,13 or particular populations such as women, 4,14,15 people with specific occupations, 16,17 nonsmokers, 18 or diabetes patients, 19 and thus may not be generalizable to the wider population. Many of these studies have shown conflicting results, perhaps due to diversity in study design, exposure assessment, analytic strategy, and most importantly, small or moderate sample sizes. To the best of our knowledge, few prior studies of long-term air pollution exposure have ever been performed among adults. Despite an attempt made in a recent study to investigate the correlation between air pollution and LTL, 20 potential biases were present due to the misclassification of exposure. One notable example is the utilization of air pollution data from 2010 as the exposure metric, while the LTL data primarily consisted of measurements taken between 2006 and 2009, leading to a potential mismatch and inaccuracies in the analysis. Further research is necessary to validate the association between air pollution and LTL. UK Biobank is a prospective cohort with measurement of leukocyte telomere length (LTL) in over 0.47 million adult residents across the UK and comprehensive information on lifestyle, behavior, and socio-economic status. In this population-based study with a large sample size, we aimed to explore the association between long-term exposure to air pollution and LTL among adults and identify factors that can potentially modify this relationship.

MATERIALS AND METHODS Study design and participants
In this population-based study, we used cross-sectional baseline data from the UK Biobank cohort which recruited about 0.5 million residents aged between 37 and 73 years from 2006 to 2010 across the UK. The scientific rationale and study protocol have been described elsewhere. 21 Among the 502,414 adult participants with available data in the current study, we excluded participants without LTL data (n = 29,882) and air pollution data (n = 724). A total of 471,808 participants were finally included in the final analysis ( Figure S1). UK Biobank has ethical approval from the North West Multicenter Research Ethics Committee (reference 16/NW/0274). Informed consent has been obtained from all participants. The generation and use of the data presented in this paper were approved by the UK Biobank access committee under UK Biobank application number 55257.

Outcome -leucocyte telomere length
Detailed information on LTL measurements in UK Biobank has been provided elsewhere. 22 Briefly, using an established multiplex qPCR methodology from available DNA samples, LTL was reported as the ratio of telomere repeat copy number (T) relative to that of a single copy gene (S, Hgb). The resulting T/S ratios were further calibrated and validated to minimize technical variation. Finally, the LTL measurements were log e -transformed to follow a normal distribution and Z-standardized to allow comparison of the results across studies.

Exposure -air pollution
UK Biobank provides air pollution estimates for each participant. However, the availability of UK Biobank pollution data is limited to specific years. For instance, the data for PM 2.5 and NO x exposure were only accessible for the ARTICLE year 2010. Considering that 83% of UK Biobank participants were enrolled before 2010, using the UK Biobank air pollution data in 2010 would result in potential exposure misclassification for most participants. Therefore, we used an air pollutants database across the UK covering the period of 2001-2019 provided by the UK' s Department for Environment, Food and Rural Affairs (DEFRA) (https://uk-air.defra.gov.uk/data/pcmdata). Air pollutants include particulate matters of up to 10 μm diameter (PM 10 and PM 2.5 ), nitrogen dioxide (NO 2 ), total nitrogen oxides (NO x ), sulfur dioxide (SO 2 ), and carbon monoxide (CO). The annual average concentrations of air pollutants before the recruitment year were derived from the UK-wide air pollution maps modeled on a 1 km × 1 km grid square. These maps were developed by a pollution climate mapping (PCM) model based on concentrations from different layers, e.g., point sources modeled using the air dispersion model, diffuse emissions from unspecified locations derived by a dispersion kernel using emissions estimates from the National Atmospheric Emissions Inventory, and secondary inorganic aerosol interpolated from sulfate (SO 4 ), nitrate (NO 3 ), and ammonium (NH 4 ). 23 These concentrations were calibrated and validated by comparison with observed measurements from the national monitoring network sites. Three-year average air pollution estimates before baseline were assigned to each participant based on the 1km grid square in which their address (grid co-ordinates rounded to a 1km distance) was located.

Meteorological data
We extracted hourly temperature data from the European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA-5) reanalysis data set with a spatial resolution of 0·1°×0·1° (ECMWF). We mapped meteorological data to the participant' s geocoded residential address at baseline. Daily temperature data were calculated by averaging hourly data within each day. Daily temperature data were then aggregated into yearly averages.

Covariates
Baseline data collected by the UK Biobank include demographics, lifestyle factors, socioeconomic status, and anthropometric measurements. Based on the previous evidence, 24-26 we considered age, sex, white blood cell count, body mass index (BMI), assessment centers, and self-report covariates including ethnicity, cigarette smoking, alcohol consumption, healthy diet score, educational attainment, average total annual household income before tax as potential confounders (the field ID in the UK Biobank are shown in Table S1). Ethnicity was classified into five categories (white, mixed, Asian or British Asian, black or black British, and others). Educational attainment was categorized as degree, General Certificate of Education Advanced levels (A levels)/National Vocational Qualification (NVQ)/other, General Certificate of Education Ordinary levels (O levels)/Certificate of Secondary Education (CSE), and none of the above. Smoking and drinking status were both categorized categorically (never/ever). Annual household income was classified into five groups (< £18,000, £18,000-£30,999, £31,000-£51,999, £52,000-£100,000, and > £100,000). A healthy diet score was calculated based on the following dietary factors: vegetable intake ≥ four tablespoons/day; fruit intake ≥ three pieces/day; fish intake ≥ twice/week; unprocessed red meat intake ≤ twice/week; and processed meat intake ≤ twice/week. One point was given for each favorable dietary factor, and the healthy diet score ranged from 0 to 5. 24 BMI was calculated from objectively measured weight and height as weight over height squared. Weight status was defined based on BMI as underweight (below 18.5 kg/m 2 ), normal weight (18.5-24.9 kg/m 2 ), overweight (25.0-29.9 kg/m 2 ), and obese (above 30 kg/m 2 ). 27 We merged underweight and normal weight as less than 1% of participants were classified as underweight. We also collected information on the presence of chronic disease (hypertension, diabetes) and self-report history of medication use (anti-hypertensive medication, insulin, and cholesterol-lowering medication) for sensitivity analyses. Disease status was based on self-reported information and medical records.

Statistical analysis
We fitted a mixed-effects linear regression model with a random effect of participants nested within assessment centers for each air pollutant. We used the age-corrected LTL as the outcome variable to reflect a cumulative measure of the history of oxidative damage after removing the correlation between LTL and age as a continuous variable. 28,29 The model is described by Z-standardized leukocyte telomere length, mean (SD) 0.0 (1.0) White blood cell count, mean (SD) 6.9 (2.0) Abbreviations: SD, standard deviation; BMI, body mass index.

ARTICLE
The Innovation Medicine 1(2): 100017, September 21, 2023 3 www.the-innovation.org/medicine the following: is the age-corrected LTL for participant from assessment . Telomere length 4 10 15 20  income, blood white cell count, alcohol consumption status, cigarette smoking status, weight status, and healthy diet score. Models were estimated unadjusted (without , model 1); adjusted for age group, sex, ethnicity, education attainment, and annual household income (model 2); and additionally adjusted for potential mediating biomarkers and lifestyles, including blood white cell count, alcohol consumption status, cigarette smoking status, weight status, and healthy diet score (model 3).
To identify potential modifiers of the association between air pollution and LTL, we further identified subgroups vulnerable to air pollution through stratification analyses by age group (<65 versus ≥ 65), sex (male versus female), ethnicity (White, Mixed, Asian, Black, and others), weight status (≤ 24.9, 25~29.9, ≥ 30), education attainment (degree, A levels, O levels, and none of the above), annual household income (< £31,000 versus ≥ £31,000), alcohol consumption status (never versus ever), and drinking status (never versus ever). Finally, we performed sensitivity analyses to test the robustness of our results. We considered alternative exposure windows of 1-year (using the 1year mean exposure preceding the baseline) and 2-year period (using the 2year average exposure preceding the baseline). We examined the effect of further adjustment for the presence of chronic medical conditions (hypertension, diabetes) and the history of medication use (anti-hypertensive medication, insulin, and cholesterol-lowering medication). As there are missing values in some covariates, model 2, model 3, and subgroup analyses were performed only among participants with complete data with the sample size decided by covariates included in these models. Therefore, we performed sensitivity analyses by using complete data after multivariate imputation by chained equations. To handle missing data, we created five imputations, each consisting of five iterations. Different methods (Bayesian linear regression, binary logistic regression, proportional odds model, and polytomous logistic regression) were employed for imputation depending on the type of data. In the primary analyses, we combined the underweight and normal weight categories due to the low prevalence of underweight participants (less than 1%). To assess the potential impact of this combination on our estimates, we conducted a sensitivity analysis by excluding the underweight individuals. Finally, we further adjusted for mean temperature in the model as temperature could be a confounder of pollution-health effect associations. 30,31 R (version 3.6.2) was used for all analyses. Two-sided P-values < 0.05 were considered statistically significant.

Role of the funding source
The funding source had no role in the design, data analyses, or results interpretation of this study. The corresponding author is the one individual who has access to all the data in the study and the final responsibility for the decision to submit for publication.

RESULTS
The baseline characteristics of the participants are shown in Table 1. Among 471,808 participants, 45.8% were males. The mean age [standard deviation (SD)] was 56.5 (8.1). A total of 94.3% of participants were white and of European ancestry. Figure 1 shows the participants' residential locations and the distribution of air pollutants. The participants were widely distributed across the UK. Participants living in England tended to experience higher annual average concentrations of air pollutants. The 3-year average estimates (SD) of PM 2.5 , PM 10 (Table 2). Notable variations were observed in both NO 2 and NO x levels across different assessment centers. Conversely, the concentrations of other pollutants remained relatively stable across these assessment centers (Table S2). From 2001/2002 to 2010, NO 2 , NO x , and SO 2 showed a decreasing trend. In contrast, the levels of other pollutants displayed significant fluctuations during the same time period ( Figure S2). Of all covariates, income had the most missing data (missing rate: 14.6%),

ARTICLE
followed by healthy diet score (2.4%), education attainment (1.2%), and BMI (0.4%). At baseline, mean LTL values were different across groups (Table 3). Higher mean LTL values were found in females (0.09) than males (-0.10); in those aged below 65 years (0.07) than those aged above 65 years (-0.28). We also observed higher mean LTL values among those who identified as black, those with lower BMI levels, higher healthy diet scores, higher household income, higher educational attainment levels, non-drinkers, and non-smokers (all p-value < 0.05). Participants with missing values in BMI and educational attainment were very similar to complete cases in terms of LTL, while those with missing values in healthy diet score and annual household income had relatively lower LTL values (Table S3). Figure 2 presents the exposure-response relationships for age-corrected LTL. Generally, we observed a downward trend in age-corrected LTL with the increase in each air pollutant. A stronger linear relationship for SO 2 was observed compared to other air pollutants. The curves of PM 2.5 , PM 10 , NO 2 , and NO x showed an evident decrease trend in age-corrected LTL at concentrations below the median, but a relatively flatter shape after the median with increasingly larger confidence bands.

Subgroup analyses
The subgroup analyses on associations between air pollutants and agecorrected LTL are presented in Figure 3 and Figure S3-8. In the fully adjusted model, those aged below 65 years had a stronger negative association between exposure to air pollution (especially for PM 2.5 and PM 10 ) and agecorrected LTL ( Figure 3A). A significant decreasing trend in exposure to SO 2 was observed among those aged below 65 years, with the percentage change in age-corrected LTL of 0.59% (95% CI: -0.47, 1.66) for Q2 of SO 2 , -0.76% (95% CI: -1.91, 0.41) for Q3, and -2.78% (95% CI: -3.94, -1.59) for Q4.  Figure 2. Exposure-response curves of the relationship between air pollutants and age-corrected leukocyte telomere length based on fully adjusted regression models using restricted natural cubic splines to model air pollutants with 3 degrees of freedom. Models were adjusted for age group, sex, ethnicity, education attainment, annual household income, blood white cell count, alcohol consumption status, cigarette smoking status, weight status, and healthy diet score. Zero on the y-axis represents the mean effect, and the fraction of the curve below zero indicates a smaller estimate than the mean effect. Vertical dashed grey lines are the 25 th percentiles, median, and 75 th percentiles. Abbreviations: PM, particulate matter; SO 2 , sulfur dioxide; NO 2 , nitrogen dioxide; NO x , nitrogen oxides; CO, carbon monoxide. PM 2.5 and PM 10 effect estimates were larger in magnitude among females compared to males ( Figure 3B). Among other effect modifiers, we observed that those with lower household income, lower educational attainment, and higher BMI had a stronger association between age-corrected LTL and most air pollutants ( Figure S3-8). The exact values of subgroup analyses are shown in Table S4.

Sensitivity analyses
Using alternative exposure windows did not lead to substantively different conclusions but did widen the confidence interval (Table S5). Additional adjustments for the presence of chronic medical conditions and the history of medication use produced almost identical results ( Figure S9). Similar results for PM 2.5 , NO 2 , NO x, and SO 2 were obtained based on the complete data after imputation while more significant associations were found for PM 10 and CO ( Figure S10). After excluding those underweight and participants who have resided at their current address for less than five years, the results changed slightly (Figures S11-12). Results changed slightly after further adjustment for mean temperature ( Figure S13).

DISCUSSION
In this uniquely large and geographically diverse sample of adult UK residents, our findings revealed that long-term exposure to air pollutants was associated with shortened LTL, although the relationship is not always strictly linear. We also identified women, younger adults, those with lower household income, lower educational attainment, and higher BMI as having a stronger association of LTL with most air pollutants. To our knowledge, this is one of the largest studies to date assessing the association between air pollution and LTL.
Growing epidemiological evidence reveals that air pollution was associated with the LTL, while results are inconsistent between different exposure windows of air pollutants, particularly PM. A study among 166 non-smoking older adults showed a 16.8% (95% CI: 7.4, 26.0) decrease in LTL associated with every 5 ug/m 3 increment in annual PM 2.5 concentration, 18 as evidenced by the present study but with relatively lower effect estimates. Another study of occupational PM 2.5 exposure reported a non-significant decrease in relative LTL associated with an increase in cumulative PM 2.5 in the year before blood draw. 16 By contrast, short-term exposure to PM 2.5 is more likely to be associated with an increase in TL. 32 The opposite direction in TL change following short-term exposure to PM 2.5 versus long-term exposure could be partly explained by transient telomere lengthening during the acute inflammatory response to maintain cellular differentiation and preserve the replicative lifespan in germinal center (GC) B cells. 33,34 After long-term exposure to PM, oxidative stress and cellular inflammation could also arise. In the long run, air pollution induces oxidative stress by the generation of reactive oxygen species (ROS) which further causes DNA damage and single-strand breaks at the G-rich telomeres. 11 On the other hand, telomere shortening is associated with the suppression of Sirtuin 1 (SIRT1) and the acetylation of p53, the most widely known substrate of SIRT1. The accumulation of acetylated p53 can accelerate cellular senescence and apoptosis. 35,36 The shapes of the concentration-response relationship between LTL and SO 2 , NO 2 , CO have rarely been assessed in the literature. To our knowledge, only one study among diabetes patients in China observed a positive but nonsignificant association of LTL with short-term exposure to SO 2 and NO 2 . 19 This result was inconsistent with our results potentially because we used  www.the-innovation.org/medicine long-term exposure windows to air pollutants. In this study, we observed a significant linear decrease in TL as related to increasing levels of SO 2 . Conversely, the decreasing patterns of other pollutants (PM, NO 2 , and NO X ) leveled off from the third quartile of pollutant concentrations. Two potential reasons could explain these findings. Firstly, telomeres naturally shorten with each cell division. When telomeres reach a minimum length (also called a critical limit), they lose their ability to protect chromosomes from fraying, leading to cellular senescence. 37 Individuals exposed to high levels of air pollutants tend to have lower LTL, which may not decrease dramatically after prolonged exposure to air pollution. According to the data presented in Table 2, the concentration of SO 2 is significantly lower in comparison to other air pollutants. The other reason might relate to the healthy worker effect. As suggested by the present study, those with higher age, pre-existing medical conditions, and unhealthy lifestyle behaviors tended to have lower LTL than the general population. The former might be excluded from recruitment due to severe illness and disability, with participants recruited in the UK Biobank study being healthier or less vulnerable to air pollution. Nevertheless, future research is required to support these hypotheses. Our findings could be  . Associations between air pollution and age-corrected leukocyte telomere length stratified by age and sex. Results are presented as percentage changes per 10 ug/m 3 increase in PM 2.5 , PM 10 , SO 2 , NO 2 , NO x , and per 0.1 mg/m 3 increase in CO. Analyses stratified by age were adjusted for sex, ethnicity, education attainment, annual household income, blood white cell count, alcohol consumption status, cigarette smoking status, weight status, and healthy diet score; analyses stratified by sex were adjusted for age group, ethnicity, education attainment, income, blood white cell count, alcohol consumption status, cigarette smoking status, weight status, and healthy diet score. Abbreviations: PM, particulate matter; SO 2 , sulfur dioxide; NO 2 , nitrogen dioxide; NO x , nitrogen oxides; CO, carbon monoxide.
complemented by future research when more repeated measurements of LTL are available.
Although emerging evidence suggests a negative association between long-term air pollution and LTL, few have performed subgroup analyses to identify populations with stronger associations between air pollution and LTL. In the present study, we observed some attenuation in effect estimates after sequential adjustment for covariates. This finding suggests that certain covariates may act as mediators in the relationship between air pollution and LTL. By adjusting for these potential mediators in our statistical model, the estimated effects tended to reflect the direct effect of air pollution on LTL. As suggested by our results, the association of higher air pollution exposure with shortened LTL is notable in younger adults and participants identified as white ethnicity. The non-significant estimates for older adults aged above 65 years and participants identified as non-white ethnicity could be partly due to the moderate sample size, that is, less than 20% for those aged ≥ 65 years and less than 10% for non-white participants. Interestingly, we observed a relatively stronger effect of air pollution on LTL in women, those with lower household income, lower educational attainment, and higher BMI. Compared to those with higher socio-economic status (SES), populations with lower SES are often disproportionately exposed to higher concentrations of air pollution. 38 Such findings, coupled with the fact that telomeric DNA is highly susceptible to accumulation of oxidative stress following air pollution exposure, 39 have made people living in lower SES areas more vulnerable to air pollution-induced telomere shortening. In addition, higher BMI and alcohol drinking are shown to be associated with shortened LTL. 40 Among the explanations of associations between LTL and both BMI and alcohol drinking, increased oxidative stress and systemic inflammation are involved, which could act synergistically with air pollution.
LTL is one of the widely used hallmarks of aging, which can provide additional information on biological aging that differs from chronological age. Exploring the association between air pollution and LTL helps us to understand why some subjects prematurely die from different causes of diseases after exposure to air pollution. As evidenced by our study, air pollution might accelerate biological aging first, then leading to a wide range of diseases. On the other hand, as oxidative stress is presumed to be a major cause of telomere shortening, the significant association of air pollution with LTL could also support the assumption that oxidative stress and inflammation are involved in the adverse health impact of air pollution. In addition, we identified vulnerable populations who had a stronger association between air pollution and LTL, indicating that the health impact of air pollution, to some extent, could be mitigated; and factors with a negative effect on oxidative stress and inflammation could be a potential modifier to reduce the adverse effect of air pollution.
Our study had several strengths and limitations. The strengths of our study include its sufficiently large number of participants to enable a comprehensive and robust analysis of the association between air pollution and telomere length. Due to the detailed lifestyle and behavior data available in the UK Biobank, we were able to provide comprehensive subgroup analyses to reliably detect vulnerable populations who are at greater risk of shortened LTL associated with air pollution. Some limitations still need to be acknowledged. As we used air pollution data with a resolution of 1 km × 1km which may not fully apprehend individual-level exposure, our results could be underestimated toward the null due to non-differential misclassification. 41 In addition, our utilization of the modelled pollution maps provided by Defra introduces the possibility of bias in pollution concentration due to the use of very local emission sources not adequately represented in the emission inventory. In this volunteer-based cohort, the participants are more likely to be healthier residents who were identified as white, which limits the ability to extrapolate our findings to the general population across the globe. Future studies are necessary to evaluate this association in other populations.
In the study on almost half a million adults, we observed that long-term exposure to various air pollutants was associated with shortened LTL, although the relationship is not always strictly linear. Some vulnerable populations were identified as those with low SES status and higher BMI. Our findings suggest air pollution may shorten the LTL and provide insights into the potential pathways linking air pollution to age-related diseases.