Study population
The Betula cohort is a longitudinal, population-based study on dementia, memory and ageing which was initiated in 1988 to investigate health and cognitive trajectories in a representative fraction of the adult and elderly population residing in Umeå county, located in the Västerbotten region of Sweden. The comprehensive recruitment process for the Betula study has been extensively detailed in previous publications16. In summary, participants in the Betula cohort have undergone examinations up to seven different time waves (T1 to T7), spaced at five-year intervals from 1988 to 2017. Each assessment includes the administration of health-related questionnaires, examinations, and cognitive evaluations. For the purposes of the present study, participants enrolled during the time periods of 1988 to 1990 (test wave 1; T1) and 1993 to 1995 (test wave 2; T2) were selected. In the Betula study, dementia status was assessed at baseline and every five years using DSM-IV criteria, identifying when cognitive symptoms impaired daily functioning. Diagnoses were based on study visit evaluations, supplemented by comprehensive medical records, including neuroimaging and CSF biomarkers when such information was available in medical records. The cohort (n = 4,445) was evaluated after each test wave (T1–T5), with diagnoses coordinated by the same senior geropsychiatrist throughout. Predetermined criteria—such as MMSE ≤ 23, cognitive decline, functional impairment, or reported memory loss—triggered extended evaluations17,18. We included participants from T1 and T2 of the Betula cohort as the parent sample (labeled as ‘Parent sample’ in Fig. 1) for this study. From this sample, we identified subsets with available data on telomere and air pollution.
Flow chart illustrating the participant selection process for the study. *Overlap of both datasets. €Based on residualization of telomere against age and gender variables.
Relative leukocyte telomere length
The relative leukocyte telomere length (rLTL) was measured by quantitative polymerase chain reaction (qPCR) in DNA from whole blood extracted with the Kleargene XL blood DNA extraction kit (LGC Genomics Ltd., UK), drawn during the second test wave (T2) between 1993 and 1995 for a total of 509 included participants. rLTL measurements were conducted in 2014 by the qPCR method for rLTL measurement originally described by Cawthon in 200219. In this study, this method was used with some minor modifications20. Each sample was evaluated by Telomere (TEL) and single copy gene hemoglobin subunit beta (HBB, Gene ID:3043) PCR reactions. A TEL/HBB value were calculated using the 2−ΔCt method, in which ΔCt = average CtTEL-average CtHBB. The rLTL value were obtained by dividing the TEL/HBB value of each sample with the TEL/HBB value of a reference cell line (CCRF-CEM) DNA included in all runs. The rLTL values were further normalized for potential plate effects (between-run variability) by subtracting mean-centered plate effects, estimated through a mixed-effects model with age, gender, age × gender interaction and plate as fixed effects, and individuals as a random effect.
Air pollution exposure assessment
The annual mean concentrations of fine particulate matter (PM2.5) and Black Carbon (BC) were computed by the Swedish Meteorological and Hydrological Institute (SMHI)21. Local and regional emission inventories served as inputs for the national dispersion modelling system, and SMHI provided the necessary data for the Gaussian dispersion model simulations to calculate annual mean concentrations of PM2.5 and BC. The model exhibited high spatial resolution, with concentrations modelled near major roads and close to smokestacks down to 35 m by 35 m. Additionally, the model incorporated emissions from industrial sources and shipping as point sources in its simulations. Emission factors for various types of vehicles were determined based on the Handbook on Emission Factors for Road Traffic, version 3.122. Non-exhaust emissions, which encompass road wear and some contributions from brake wear and tire emissions, were also quantified. To estimate emissions from residential wood combustion, data from inventories of individual stoves and boilers, along with information gathered from chimney sweepers and interviews regarding wood burning habits, were utilized23. The assessment of long-range transport air pollution relied on data from rural background monitoring stations. This involved calculating the difference between the total concentrations measured at these monitoring stations and the modelled local particle concentrations at the same locations. The total PM2.5 concentrations used in this study were based on dispersion modeling previously validated against measured concentrations in multiple Swedish cities, showing reasonably good agreement with an R² of 0.65 (21), which supports the reliability of the exposure estimates used.
Finally, the resulting concentrations of air pollution were linked to each participant’s residential address at sampling, and the modelled value of year 1990 was used as marker for long-term exposure for the participants. Although the exposure assessment was conducted in the early 1990s, the levels in the study area were, and continue to be, low by both national and international standards. Importantly, we examined within-city contrasts, which remain relevant today despite reductions in overall pollution levels. As we were able to model source-specific concentrations of locally emitted PM2.5 and BC from vehicle exhaust and wood smoke, sources that are still dominant in the area, our findings are likely more generalizable to current urban environments than studies relying on total PM2.5 alone. The inclusion of BC, a key pollutant linked to combustion and climate impacts, further strengthens the relevance of our results for contemporary public health policy.
The six exposure variables included total concentrations of PM2.5 and BC (PM2.5_total, BC_total) as well as source-specific concentrations of PM2.5 (PM2.5 _exhaust, PM2.5 _woodburning) and BC (BC_exhaust, BC_woodburning). The study area, Umeå municipality, is located in Northern Sweden, and has moderate levels of PM2.5, with a total concentration of 9.81 µg/m3 in the present study (Table 1). This is below the former WHO recommended air quality guideline from 2005 of 10 µg/m3, but over the revised WHO (in 2021) air quality guideline of 5 µg/m3.
Statistical analysis
The calculation of blood lymphocyte proportion involved dividing the lymphocyte count by the total count of white blood cells, which encompasses neutrophils, eosinophils, basophils, lymphocytes, and monocytes. To mitigate the influence of age and gender-related variations, relative leukocyte telomere length (rLTL) was residualized against the participant’s age at the time of rLTL measurement and their gender, using a linear regression model. This yielded a residualized telomere length, denoted as rLTL, which was then utilized to investigate the relationship between air pollution and telomere length. We considered age, gender (female, male), lymphocyte proportion, education level (compulsory, high school, university), and smoking status (smoker or former smoker, non-smoker) as a potential confounding factors to be included in this study. Air pollution exposure, rLTL as well as all potential confounders were assessed at test wave T2, whereas dementia was assessed with follow-up to February 2022.
Linear regression analysis was employed to assess the impact of both total and source-specific particle concentrations on rLTL. Model 1 was an unadjusted model. Model 2 incorporated adjustments for lymphocyte proportion, age and gender. Model 3 (the main model) additionally accounted for individual-level baseline potential confounders, including education level and smoking status. We created Q-Q plots to evaluate the assumptions of linearity and normal distribution in linear regression (Figure. S1, S2, and S3) in the Supplementary material.
Furthermore, subgroup analysis based on future dementia diagnosis (Dementia, No dementia) was carried out to evaluate whether the association between air pollution and rLTL differed depending on dementia status. Here, a linear regression model was used and a multiplicative interaction term was included in the main model between each of pollutants and dementia status to explore potential effect modification. All statistical analyses were performed using R version 3.4.8.
The study was approved by the Ethical Review Board in Umeå with Dnr: 2022-04608-01, and written informed consent was obtained from all Betula participants. The data for the present study were accessed for research purposes on March 15, 2023. The researchers analyzing the data did not have access to information that could identify individual participants. All methods were conducted in accordance with relevant guidelines and regulations.