Introduction

Cognitive decline refers to reduction in cognitive abilities, including memory, attention, use of language, decision making and awareness1. Although slow, gradual cognitive decline is part of the normal ageing process in some cases it can predict mild cognitive impairment or dementia2. Given the potentially concerning nature of cognitive decline in itself and the fact that it may be associated with later dementia, there is an interest in understanding potentially modifiable risks factors for cognitive decline3. One such potential risk factor is increased severity4 and accumulation5,6 of affective symptoms such as anxiety and depression. Given the high overlap between anxiety and depression and the fact that they are often not diagnosed, measures of affective symptoms enable the researcher to tap into anxiety and depression without a formal diagnosis7 and are a useful predictor for studies investigating longitudinal associations between affective disorders and cognitive function8.

A gap in the literature concerning the link between affective symptoms and cognitive decline is that there is not clarity on mechanisms (which could be biological, social or psychological in nature) underpinning this8. Understanding these mechanisms is important to appropriately target interventions to ameliorate cognitive decline. Previous work testing biological mechanisms has suggested that cardiometabolic risk may partially mediate the association9 but other biological mechanisms are poorly understood. One potential biological mechanism is telomere attrition, recognised as one of the key hallmarks of aging10. Telomere attrition has been implicated in various age-related pathologies, including cognitive decline, suggesting that investigating its role could provide valuable insights into the biological mechanisms of neurodegenerative processes.

Telomeres are tandem repeat DNA sequences located at the end of chromosomes that play a vital role in protecting DNA from damage during recombination11. Telomeres are associated independently with both cognitive decline12, dementia13,14 and depression15,16. Han et al. 17 showed that both early subjective depressive symptoms and cognitive complaints in relatively healthy elderly individuals were associated with a relatively shorter TL in the randomised controlled prospective SUPERBRAIN study. However, to our knowledge, no longitudinal study has investigated the potential role of telomere shortening in the observed association between affective symptoms and cognitive decline.

This is the purpose of the current research. We test the hypotheses that TL or RTS may:

  1. 1.

    Represent an indirect pathway between the association of affective symptoms and later-life cognitive decline (suggestive of TL and/or RTS mediating the relationship between depression and cognitive decline).

  2. 2.

    Contribute to both affective symptoms and cognitive decline (suggestive of a common cause mechanism whereby depression and cognitive decline might both be influenced by TL and/or RTS).

  3. 3.

    Act via both mediator and common cause pathways.

The aim of this study was to test these possibilities using data from a nationally representative birth cohort study with follow up available over 69 years (NSHD).

Results

Descriptive statistics and missing data

Participant characteristics are presented in Supplementary Table 1 and a flowchart for the study sample is provided in Fig. 1. At age 53, the cohort size dropped to 3035 due to participant death untraceability or refusal18. At age 69, the cohort dropped to 2149 people. Of the remaining participants, 32% had taken part in all 23 previous data collections, 55% had participated in 22 or more and 69% in at least 2119. By age 69, 928 members of the cohort had complete data for all waves of affective symptoms and cognitive outcomes. Cognitive measures at age 69 varied across cognitive domains, whereby 1482 participants completed the verbal recall task, 1496 completed the processing speed task and 1255 completed the comprehensive ACE-III examination. The telomere data-collection at ages 53 and 60–64 included 2479 and 1004 participants respectively.

Fig. 1: Flow diagram of participant follow-up and analytical sample selection.
figure 1

Flow diagram illustrating participant follow-up, attrition, and availability of telomere and cognitive data from the original 1946 Birth Cohort through to the final analytical samples used in unadjusted and fully adjusted models.

Full size image

The sample with complete data differed from the sample with missing data on key variables. Participants that had experienced multiple affective symptoms were more likely to have missing data or to drop out of the study. Participants with missing cognitive scores were more likely to be male with lower childhood cognitive scores and lower educational attainment at age 26. Some individuals with telomere data at age 53 were missing telomere data at age 60–64. The characteristics of the groups with and without telomere data at both ages were similar.

The analytical sample available for this study based on FIML was 1770 (for unadjusted models), and 1019 (for adjusted models). Supplementary Table 1 shows the descriptive statistics for cohort members included in the fully adjusted models.

Path analysis

The adjusted model using TL fit the data well (n = 1019 χ2 = 0.174, P = 0.00; CFI = 0.998; TLI = 0.964; RMSEA = 0.027). The unadjusted model was an adequate fit to the data (n = 1770; χ2 = 0.011, P = 0.00; CFI = 0.991; TLI = 0.861; RMSEA = 0.055). The adjusted FIML models using RTS fit the data well, (n = 1019; χ2 = 0.189, P = 0.00; CFI = 0.999; TLI = 0.968; RMSEA = 0.026). The unadjusted model was an adequate fit to the data (n = 1770; χ2 = 0.011, P = 0.00; CFI = 0.991; TLI = 0.859; RMSEA = 0.056).

Mediation hypothesis

The unadjusted model including TL showed significant direct effects of the accumulation of affective symptoms on all three cognitive measures in the model, ACE-III (β = −0.070, s.e = 0.030 p = 0.020), verbal recall (β = −0.063, s.e = 0.029, p = 0.027) and search speed (β = −0.076, s,e = 0.028, p = 0.007). However, after adjustment for all covariates, the only significant direct effect of the accumulation of affective symptoms remaining, was search speed (β = −0.070, s.e = 0.032, p = 0.031). Results were similar for the RTS model, as shown in Supplementary Table 2.

As shown in Figs. 2 and 3 and Supplementary Table 3, there was no evidence for the mediation hypothesis as there were no significant indirect effects through either of the telomere measures (TL or RTS).

Fig. 2: Path models showing direct, indirect and total associations betweenaccumulation of affective symptoms and cognitive measures viatelomere length.
figure 2

a Unadjusted model (n = 1770). b Adjusted model (n = 1019). Covariances were included between cognitive measures andbetween each cognitive measure and affective symptoms age 69. Significant pathways are represented with solid arrows.Nonsignificant pathways are represented by dotted arrows.

Full size image
Fig. 3: Path models showing direct, indirect and total associations betweenaccumulation of affective symptoms and cognitive measures via rateof telomere shortening.
figure 3

a Unadjusted model (n = 1770). b Adjusted model (n = 1019). Covariances were included between cognitive measures andbetween each cognitive measure and affective symptoms age 69. Significant pathways are represented with solid arrows.Nonsignificant pathways are represented by dotted arrows.

Full size image

In the unadjusted models, there were significant total effects of affective symptoms on all three cognitive measures ACE-III (β = −0.071, s.e = 0.030, p = 0.018), verbal recall (β = −0.063, s.e = 0.029, p = 0.027) and search speed (β = 0.070, s.e = 0.028, p = 0.014).

Common cause hypothesis

As shown in Figs. 2 and 3 and Supplementary Table 3, no significant association was found between TL or RTS and affective symptoms at age 69, before or after adjusting for the covariates. In both the adjusted and unadjusted models, no significant association was found for either of the telomere measures (TL or RTS) and ACE-III and verbal recall at age 69. There was a significant association between TL and search speed at age 69 in the unadjusted model (β = 0.114, s.e = 0.041, p = 0.005), and this association remained significant in the adjusted model ((β = 0.117, s.e = 0.043, p = 0.006). There was a significant association between RTS and search speed at age 69 in the unadjusted model (β = 0.081, s.e = 0.041, p = 0.049), this association was no longer significant after adjusting for the covariates.

Discussion

To our knowledge this was the first study to examine whether TL or RTS acts as mediators or a common cause variable in the association between accumulation of affective symptoms and cognitive function. We found no evidence to support this. This finding aligns with previous studies reporting null associations between telomere shortening in adulthood and both affective symptoms and cognitive decline20,21. However, some longitudinal studies have found an association between TL and cognitive measures22. Where associations have been found, the effect sizes are often small, suggesting that the predictive ability of TL for cognitive decline may be limited23.

Given that many papers have shown a significant association between short TL and risk of dementia13,14, it is possible that the role of TL is dependent on the extent of cognitive impairment in the individual and the presence of degenerative disorders. Since this cohort at age 69 is considered cognitively healthy, this hypothesis that TL is only important in neurodegenerative processes and not normal cognitive ageing, could explain the null results in this study. Another possibility is that this cohort are still relatively young, it is possible that associations may only become apparent at older ages. At age 69, cognitive outcomes largely reflect a preclinical stage of cognitive ageing, and the low prevalence of clinical cognitive impairment may have limited the ability to detect associations involving telomere dynamics. One important consideration in interpreting the findings of this study is the potential limitation related to statistical power. Although our analyses included ~1000 participants in the fully adjusted models, this sample size may not have been sufficient to detect subtle associations between TL/RTS and cognitive function or affective symptoms. More longitudinal research with larger sample sizes is needed to robustly investigate how the role of telomere attrition differs across various populations across the lifespan.

Ascertaining biological factors that influence cognitive decline is key to enhancing healthy cognitive ageing. Beyond telomeres, other potential biological pathways are being investigated to explain the association between affective symptoms and cognitive decline. Lopez-Otin et al.10 suggests that mechanisms such as epigenetic alterations and mitochondrial dysfunction could also play a role. For example, there is emerging evidence that affective symptoms may impact epigenetic modifications, which could, in turn, contribute to cognitive decline24. Affective symptoms are linked to increased oxidative stress and impaired mitochondrial bioenergetics25, with studies showing that such dysfunction can reduce energy production and elevate oxidative damage in the brain, contributing to cognitive impairment and accelerated aging.

Research in the 1958 birth cohort has found support for various biomedical pathways including cardiometabolic risk9 and inflammation26. No support was found for the role of cortisol8. Identifying the mechanisms behind this association could lead to implementation of treatment strategies for depression that alter the risk of cognitive decline. It will be important to understand whether treatment for affective symptoms can improve cognition, and whether these treatments can be included as part of multi-domain interventions designed for dementia prevention. In addition to mechanisms discussed, lifestyle, socio-behavioural and social factors may play a role in the observed association. It is generally thought that there is not one pathway driving this association, and that the association is driven by synergistic processes. Telomere dynamics need to be further elucidated in terms of protection against attrition and potential downstream effects of healthy ageing.

To the best of our knowledge, this is the first longitudinal study investigating the potential role of TL/RTS as a mediator in the relationship between affective symptoms and cognitive function. A key strength of this study is the use of a nationally representative cohort, with data available over a period of almost 7 decades from birth through to later-life. Attrition is unavoidable in long-running studies, but previous analyses of this cohort has shown that the samples at 53 and 60–64 years remained broadly representative of the British-born population of that age27.

Since the yearly rate of change of TL is extremely low compared to the ability of telomere assays to detect small differences in TL28, a strength is the 10-year follow up period, allowing the detection of inter-individual variations in TL with confidence. Since the pathology underlying neurodegenerative diseases has been found to begin decades before the symptoms can be detected, measuring TL at 53 and 60–64 provides a clinically relevant time point to predict decreased cognitive function. Given that limited studies have examined the association of TL and cognitive function in cognitively healthy cohorts, this study provides key insights to the literature. By avoiding using AD as an outcome measure and measuring cognitive function instead, this study allows the investigation of effects of TL on cognitive function prior to the emergence of symptoms of neurodegenerative diseases. This is important because if associations were indeed observed, this may indicate the presence of a critical window prior to cognitive impairment in which early intervention may be particularly effective. However, since cognitive ability was the outcome measure, any conclusions regarding AD and other neurodegenerative diseases are made cautiously.

Another strength is the use of the ACE-III examination, which is one of the most comprehensive measures of global cognitive state available29. Furthermore, this study controlled for many covariates, including childhood cognition, limiting bias from reverse causation30. The use of FIML has been shown to be superior to complete case analysis31, is easily implemented and performs well in structural equation modelling (SEM).

This study has a high percentage of missing data with selection bias found within the dataset and many groups were underrepresented. Attrition was higher in people with lower SEP and for people with high affective symptoms. This could have resulted in underestimating the association between the accumulation of affective symptoms and cognitive function or have impeded detection of significant associations with TL or RTS. The study lacked ethnic diversity, with a 100% White population.

This is a psychologically and cognitively healthy sample, the cohort did not have high rates of case-level affective symptoms with few participants showing high levels of cognitive impairment. It is possible that significant affective symptoms and significant cognitive impairment need to be present before the role of telomeres can be observed. The measures of affective symptoms implemented across this study were inconsistent, with long-time frames between the various measures. In terms of measuring cognitive change, previous work in this cohort shows that quadratic models fit the rate of cognitive change better than linear models32. However, since telomere data is only available from age 53, modelling a quadratic curve would have resulted in modelling backwards temporal associations. The solution used in this analysis was to measure if RTS from age 53 to 60–64 predicted change in cognitive function from age 60–64 to 69. These measures were within a relatively narrow time frame, possibly explaining the null association.

To our knowledge, this is the first known longitudinal study investigating the potential role of telomere shortening in the relationship between affective symptoms and cognitive function. This study presents evidence which contributes towards the continued mission to improve our understanding of the biological mechanisms involved in ageing and the factors that explain healthy ageing. These findings accentuate the value of monitoring affective symptoms for the progression of cognitive decline. Further work is needed to identify environmental and genetic mechanisms that contribute to TL and RTS. More longitudinal studies are required across different samples and ages to demonstrate how TL and RTS play a role in the association between the accumulation of affective symptoms and cognitive function. Further work is needed to elucidate the mechanisms behind this association to enable effective interventions and prevention strategies for dementia. Given that a decade has passed since the last telomere data collection, future data collection in this cohort of all variables important to this study, may reveal the intricacies in the role of telomere dynamics and help determine if telomeres are important in the association between affective symptoms and later-life cognition.

Methods

Participants

Data were used from the National Survey of Health and Development (NSHD) birth cohort. This maternity survey recorded 13,687 births across England, Scotland and Wales during one week in March, 1946. From this sample, a socially stratified sample of 5362 singleton babies (male and female), who were born to married parents, were selected to form the NSHD. There have been 24 follow-ups with the most recent data-collection at age 69 (n = 2149)33. This study is considered approximately representative of the British-born population at the same age. Data has been collected via interview, examination, and questionnaires to determine sociodemographic factors and cognitive, physical, medical, and psychological health. Ethical approval for NSHD came from the North Thames Multicentre Research Ethics Committee and research was conducted in accordance with the Declaration of Helsinki. Participants provided written informed consent at each data collection. For further details see ref. 19. The present study involved secondary analysis of anonymised NSHD data provided by the Medical Research Council (MRC) Lifelong Health and Ageing Unit at UCL.

Measures

Measures of cognitive function were assessed at ages 53, 60–64 and 69. The short-term verbal memory test comprised of a 15-item word recall test including immediate and delayed components34. Processing speed and accuracy was tested using a visual search task known as the Letter Cancellation Task35 adapted from the MRC Cognitive Function and Ageing Study36, these have been detailed extensively elsewhere5. At age 69, in addition to these measures, the Addenbrooke’s Cognitive Examination-III (ACE-III)37 was used to measure cognition. The ACE-III consists of five domains which include attention and orientation, memory, language, verbal fluency, and visuospatial function. The scores for each domain are added together to provide a total score, whereby a high score indicates good cognitive function. The measures used in this cohort have been described in detail elsewhere32. A schematic overview of cognitive, affective, telomere and covariate measures across assessment waves in the NSHD is provided in Fig. 4.

Fig. 4
figure 4

Flowchart of the measures included in the NSHD.

Full size image

Assessment of affective symptoms assessment varied over time. The earliest measurements were teacher ratings of emotional problems using a forerunner of Rutter A Scale38, at age 13 and 15. A short version of the Present State Examination was used at age 3639 to determine both the frequency and severity of affective symptoms. At age 43, the Psychiatric Symptom Frequency scale was used40. The final three affective symptom assessments at ages 53, 60–64, and 69, were determined using the 28-item General Health Questionnaire (GHQ28)41.

To capture long-term exposure, we used a binarised measure of accumulated affective symptoms from age 13–53, coded as follows; 0 = no case level episodes, 1 = 1 case level episode, 2 = 2 case level episodes and 3 = 3 or more case level episodes. See John et al.6 for further details. This approach reflects evidence that the cumulative burden of affective symptoms across the life course is more predictive of later-life cognitive outcomes than symptoms in specific periods5,6.

Using Puregene DNA isolation kits, DNA was extracted from frozen ethylenediaminetetraacetic acid (EDTA) blood samples to determine leukocyte telomere length (LTL), which was measured as a ratio to standard reference DNA. The quantitative Polymerase Chain Reaction (qPCR) technique was used to measure TL due its low DNA input threshold and high throughput capabilities42. Further details regarding LTL measurement are comprehensively documented elsewhere43. LTL was measured at age 53 in 2479 individuals and at 60–64 years in 1004 individuals. LTL was measured in the same laboratory at both time points to reduce variability. Based on a validated procedure by Masi et al.44, for the individuals that participated in both data collections, RTS was calculated as show in Eq. (1):

$$((\Delta \mathrm{LTL\; from}\,53\,\mathrm{to}\,60-64)/{(\mathrm{LTL\; at}\,53))}^{* }100$$
(1)

LTL was log10 transformed to improve the normality of the distribution and the outliers excluded from the analysis were the 0.9% of participants who had a baseline LTL <2000bp or >11,000bp44.

Models were adjusted for sex45, father’s occupation46, childhood cognition47, educational attainment by age 2648, adult socioeconomic position (SEP)18 and antidepressant medication use49. Fathers’ occupation at age 4, 11 and 15 was used as a measure of SEP using the Registrar General’s classification to separate participants by trade: unskilled, partly skilled, skilled (manual), skilled (non-manual), intermediate and professional. Childhood cognition was a composite score of four tests of verbal and non-verbal ability collected at ages 8, 11 and 15. Educational attainment was measured as the highest level of academic achievement attained by 26: none; vocational or GCSE; and A-level or higher. Adult SEP was defined as the participants current or most recent occupational social class at 43 using the same Registrar General classification used for child SEP. Self-report antidepressant medication use was collected at age 31, 36, 43, 53, 60–64 and 69. This was split into those that had never reported use of antidepressant medication and those that had at least once from ages 31–69. Information of medication use in adolescence was not available5.

Data analysis

A path model (Fig. 5) was run to investigate the role of telomere in the association between accumulation of affective symptoms and later-life cognition. Specifically, the model included direct associations between accumulation of affective symptoms (age 13–53) and the three cognitive measures at age 69. To test the mediation hypothesis, indirect pathways running through TL at age 60–64 and RTS over 10 years were included in the model. To test the common cause model, direct associations were investigated between both telomere measurements (TL and RTS) and affective symptoms at age 69 and both telomere measurements and the three cognitive outcomes at age 69. A direct association was tested between accumulation of affective symptoms from 13 to 53 and affective symptoms at age 69 to test the hypothesis that accumulation of affective symptoms from 13 to 53 are associated with later-life affective symptoms at age 69. Covariances were included between the three cognitive measures and between cognitive measures and affective symptoms at age 69.

Fig. 5
figure 5

Path model examining the role of telomere in the association between affective symptoms and cognitive function.

Full size image

Standard fit statistics were used to test model fit. These included the Tucker–Lewis index (TLI), comparative fit index (CFI), root mean square error of approximation (RMSEA) and chisquared (χ2) goodness of fit test. To improve fit, nonsignificant covariances were removed from the model, these included covariances between affective symptoms age 69 and all cognitive measures. The models were first run unadjusted, and consequent models were adjusted for all covariates. Given that significant differences were found between the sample with complete data and the sample with missing data (Section 3.1), full information maximum likelihood (FIML) was used to maximise the sample size and account for missing data31.

All statistical analyses were performed using Stata 17.0 and RStudio 4.1.2. The data used in this study is accessible to researchers on request, review and approval through the NSHD Data Sharing Committee. Further information is available here: https://www.nshd.mrc.ac.uk/data/data-sharing/.