The longitudinal relationship between cortisol responses to mental stress and leukocyte telomere attrition 2

Context: Chronic psychological stress has been associated with shorter telomeres in some studies, but 26 the underlying mechanisms are poorly understood. One possibility is that the neuroendocrine responses 27 associated with stress exposure are involved. 28 Objective: To testing the hypothesis that greater cortisol responsivity to acute stressors predicts more 29 rapid telomere attrition. 30 Design: We measured salivary cortisol responses to two challenging behavioral tasks. Leukocyte 31 telomere length was measured at the time of mental stress testing and 3 years later. 32 Participants: We studied 411 initially healthy men and women aged 54-76 years. 33 Main outcome measure: Leukocyte telomere length. 34 Results: Cortisol responses to this protocol were small, we divided participants into cortisol 35 responders (n = 156) and non-responders (n = 255) using a criterion (≥20%) previously shown to 36 predict increases in cardiovascular disease risk. There was no significant association between cortisol 37 responsivity and baseline telomere length, although cortisol responders tended to have somewhat 38 shorter telomeres (β = -0.061, standard error 0.049). But cortisol responders had shorter telomeres and 39 more rapid telomere attrition than non-responders on follow-up, after controlling statistically for age, 40 gender, socioeconomic status, smoking, time of day of stress testing and baseline telomere length (β = - 41 0.10, standard error 0.046, p = 0.029). The association was maintained after additional control for 42 cardiovascular risk factors (β = -0.11, p = 0.031). The difference between cortisol responders and non- 43 responders was equivalent to approximately 2 years in aging. 44 Conclusions: These findings suggest that cortisol responsivity may mediate in part the relationship 45 between psychological stress and cellular aging.


Introduction 49
Telomeres are complexes of DNA and proteins situated at the ends of chromosomes that protect the 50 genomic DNA of eukaryotic cells (1). Telomeres shorten with each cell division, and telomere length is 51 a marker of cellular aging. Telomere function is impaired when shortening becomes critical, leading to 52 cell senescence, genome instability and apoptosis. Leukocyte telomere length is associated with 53 increased risk of cardiovascular disease, cancers, diabetes, dementia and all-cause mortality (2-4). 54 These relationships have been confirmed by studies of inherited telomere syndromes (5), and by 55 Mendelian randomization studies (6). 56 Several environmental and lifestyle factors are associated with telomere shortening, including 57 smoking, obesity and physical inactivity (7). There is growing interest in the relationship of leukocyte 58 telomere length with psychiatric conditions and psychological stress as well. Large scale investigations 59 indicate that individuals with major depressive disorder have shorter telomeres independently of 60 demographic factors and health behaviors , although findings across studies have been variable (8). 61 Anxiety disorders may also be associated with reduced telomere length (8), while a meta-analysis of 22 62 studies documented a small statistically significant relationship between greater perceived stress and 63 shorter telomeres (9). Exposure to early life adversity has been linked with reduced telomere length in 64 some studies (10), but not in all (11). Associations with low social support (12) and hostility (13) have 65 also been described. 66 Evaluation of the importance of links between stress exposure, mental health, and telomere 67 dynamics would be strengthened by better understanding of potential underlying mechanisms. 68 Unhealthy habits such as smoking, excessive alcohol consumption and inactivity might play a role, but 69 many studies have observed associations with leukocyte telomere length after these factors have been 70 taken into account (8,9,14). The physiological responses associated with mental stressors may also be 71 involved. Cortisol plays a central role in the stress response because of its multiple effects on immune, 72 metabolic, and vascular processes . Animal studies indicate that embryonic exposure to corticosteroids 73 elicits increased oxidative stress and shorter telomeres in later life (15). There are large individual 74 differences in the magnitude of cortisol responses to standardized mental stress tests, and these reflect 75 variations in the capacity of neuroendocrine regulatory processes to adapt to challenge. A small number 76 of studies have shown that larger cortisol responses to mental stress are associated with shorter 77 telomeres in adults and children (16)(17)(18) However, these studies of telomeres and stress physiology have been cross-sectional. It is possible that 81 heightened cortisol responsivity drives telomere attrition, or conversely that greater cortisol responses 82 are characteristic of people with shorter telomeres. Null associations have also been described (20). 83 In the present study, we evaluated the relationship between cortisol responses to mental stress 84 and differences in telomere length measured at the time of mental stress testing and three years later. 85 We tested the hypothesis that cortisol stress responders would show greater telomere attrition over time 86 than non-responders. This hypothesis was examined in a sample of healthy men and women aged 54-87 76, since biological aging processes are particularly relevant to disease risk as people progress into 88 older age. We used a measure of cortisol responses to mental stress tests that has previously been 89 shown to predict the progression of subclinical coronary atherosclerosis as indexed by coronary 90 calcification (21), and the development of hypertension (22). Our analyses also took into account 91 sociodemographic and physiological factors that might also contribute to telomere shortening over 92 time. 93 94

Materials and Methods 95
Participants 96 We analyzed data from the Heart Scan Study, a sample of 543 men and women of white European 97 origin of the Whitehall II epidemiological cohort recruited between 2006 and 2008 to investigate 98 physiological responsivity to mental stress testing and subclinical coronary artery disease. Participants 99 were selected as having no history of coronary heart disease, and no previous diagnoses or treatment 100 for hypertension, diabetes, inflammatory diseases, or allergies. We used civil service employment 101 grade as an indicator of socioeconomic status (SES), and recruitment was stratified to include men and 102 women from higher, intermediate and lower employment grades. The women in the study were 103 postmenopausal. Participants were invited for reassessment 3 years after mental stress testing (mean 104 1087 days interval). Ethical approval was obtained from the University College London Hospital 105 Committee on the Ethics of Human Research, and all participants gave signed informed consent. All 106 procedures were carried out in accordance with approved guidelines. 107 Figure 1 shows a flow chart summarizing participant progression through the study. Telomere 108 length was measured in 501 (92.3%) respondents an average 36.2 months after stress testing. Of these, 109 411 also had telomere length measures at the time of stress testing, since assessments were not 110 introduced at the start of data collection. They constitute the sample for this study. There were no 111 differences on any measures between individuals included and not included in the telomere length 112 analyses. 113 114

Laboratory mental stress testing 115
We tested participants individually in a light and temperature-controlled laboratory, with sessions 116 beginning either in the morning at 8:30-9:30, or in the early afternoon at 13:30-14:30. Participants were 117 instructed not to drink caffeinated beverages or smoke for at least 2h before testing and to avoid 118 vigorous exercise and alcohol from the previous evening, and not to have taken any anti-inflammatory 119 or anti-histamine medication for the 7 days before testing. They were rescheduled if they reported colds or other infections on the day of testing. At the start of the session, we measured height, weight, waist 121 and hip circumference using standardized techniques, and body mass index (BMI) was computed. After 122 a 30 min rest period, baseline blood pressure (BP) was measured with an automated UA-779 digital 123 monitor, a blood sample was drawn, and a saliva sample was taken using salivettes (Sarstedt, Leicester, 124 UK). Two behavioral tasks designed to induce mental stress were then administered in random order 125 (21,23). Both tasks were performed for 5 min. One was a computerized version of the Stroop color-126 word interference task which involved successive presentation of target color words (e.g. red, blue) 127 printed in another color. Four names of colors printed in incongruous colors at the bottom of the 128 computer screen, and participants were requested to press the computer key that corresponded to the 129 position at the bottom of the screen of the name of the color in which the target word was printed. The 130 rate of presentation of stimuli was adjusted to the performance of the participant in order to ensure 131 sustained demands. The second task was mirror tracing, which involved tracing with a metal stylus a 132 star that could only be seen in mirror image. Each time the stylus came off the star a mistake was 133 registered and a loud beep was emitted by the apparatus (Lafayette Instruments Corp., Lafayette, IN, 134 USA). Participants were told that the average person could complete five circuits of the star in the 135 available time. These tasks were selected because they have been shown to stimulate similar appraisals 136 of involvement and engagement from participants across the social gradient. A second saliva sample 137 was taken immediately after tasks, with further samples at 20, 45 and 75 min after tasks. 138 139

Biological measures 140
Saliva samples were analyzed for cortisol concentration using a time resolved immunoassay with 141 fluorescence detection, at the Technical University Dresden, as described previously (24,25). The intra-142 and inter-assay coefficients of variation were less than 8%. Total and high density lipoprotein (HDL) 143 cholesterol were measured in serum stored at 4 o C within 72 h using enzymatic colometric methods. 144 Glycated hemoglobin was measured using Tosoh G7 HPLC analyzer calibrated to Diabetes Control 145 and Complications Trial (DCCT) standards. An adaptation of the method first described by Cawthon 146 (26)  Reactions containing serial dilutions of a reference DNA standard were included in each polymerase 156 chain reaction plate to generate the telomere (T) and β-globin gene (S) standard curves required for 157 quantitation, and relative mean TL, expressed as a T/S ratio, was derived. The coefficient of variation 158 of these assays was 2.3%. 159 160

Data reduction and statistical analysis 161
The mental stress protocol in this study did not generate large cortisol responses, with many 162 respondents not showing an increase following tasks. Cortisol stress responsivity was therefore 163 quantified by calculating differences scores between the baseline cortisol concentration and the samples 164 obtained both immediately after tasks and 20 minutes later. Individuals who showed a ≥ 1 nmol/L 165 increase (equivalent to a 20% increase) between baseline and either sample were defined as cortisol 166 responders, and the remainder as non-responders. Differences between the responder groups at baseline 167 were analyzed using analysis of variance and χ 2 methods for continuous and categorical variables 168 respectively. The cortisol profiles across the mental stress testing session of responder and non-169 responder groups were compared using repeated measures analysis of variance with sample as the 170 within-person factor and responder status as the between-person factor. Associations between cortisol 171 stress responsivity and telomere length at baseline were analyzed using multivariable regression, 172 including age, gender, grade of employment, smoking status and time of stress testing (morning or 173 afternoon) as covariates. A similar method was used to analyze associations between cortisol stress 174 responsivity and follow-up telomere length, except in this case baseline telomere length was included 175 as a covariate. Results are presented as standardized regression coefficients (β) with standard errors. 176 In a sensitivity analysis, we added cardiovascular risk factors (systolic BP, BMI, total and HDL 177 cholesterol, and glycated hemoglobin) to the model; these factors were not included in the main model 178 since missing data on some variables reduced the sample size. 179 Absolute measures of telomere length can vary across laboratories, but rankings of relative 180 length are highly correlated (28). In view of the different systems used at baseline and follow-up, we 181 therefore computed standardized telomere length scores for the two time points. However, repeating 182 the analyses with standardized as opposed to absolute values generated identical statistical findings, so 183 the latter are presented in the Results section. 184 185

186
The 411 participants included 156 cortisol responders and 255 non-responders. The characteristics of 187 these two groups are summarized in Table 1. Participants generally had favorable risk profiles, with 188 few smokers, blood pressure and glycated hemoglobin in the healthy range, and no marked elevation of 189 BMI or cholesterol. There were no differences in any sociodemographic or physiological factors 190 between the two groups. There was a non-significant tendency of cortisol responders to be more likely 191 to have undertaken mental stress testing in the afternoon compared with non-responders (p = 0.096), so 192 time of day was included as a covariate in the analyses. 193 Cortisol concentrations in the responders and non-responders to behavioral challenge are shown 194 in Figure 2. There was a robust interaction between responder group and trial (p < 0.001). It can be 195 seen that cortisol concentrations were similar in the two groups at baseline. But while the responder 196 group showed an average 47% increase in salivary cortisol after tasks, values declined steadily in the 197 non-responder group. Even 75 min after mental stress tests had been completed, cortisol concentration 198 remained more than 30% higher in the responder than non-responder groups. 199 The mean T/S ratio averaged 0.992 ± 0.07 at baseline, and 0.894 ± 0.15 at follow-up. This 200 indicates a significant decrease in telomere length over the 3 year interval (p < 0.001). Telomere 201 lengths at the two time points were moderately correlated (r = 0.31, p <0.001). There was a small 202 positive association between baseline telomere length and change over time (r = 0.20), indicating that 203 participants with longer telomeres showed greater shortening. Telomere length on follow-up was 204 inversely associated with age (p < 0.001), and was shorter in men than women (p < 0.001). 205 The relationship between cortisol stress responsivity and telomere length at baseline was 206 negative, though not significant ((β = -0.061, SE = 0.049, p = 0.22). But we found that cortisol stress 207 responsivity was associated with shorter telomere length on follow-up after adjustment for baseline 208 telomere length, age, gender, grade of employment, smoking status and time of stress testing (β = -209 0.10, SE = 0.046, p = 0.029). The other independent predictors of shorter telomeres on follow-up were 210 older age, male sex, and shorter telomere length at baseline. Figure 3 illustrates the pattern of change in 211 telomere length over time in cortisol responders and non-responders to stressors, showing the greater 212 shortening over time in stress responders. There was no interaction between time of stress testing and 213 cortisol responsivity in predicting telomere length on follow-up. 214 The association was unchanged in the sensitivity analysis which included baseline systolic BP, 215 BMI, total and HDL cholesterol, glycated hemoglobin, and time interval between baseline and follow-216 up; the regression coefficient for cortisol responsivity was (n = 378, β = -0.11, SE = 0.049, p = 0.031). 217 218 Discussion 219 In this study, we tested the notion that cortisol responses to mental stress would be associated with the 220 rate of telomere attrition over time. We found that healthy late middle-aged men and women who 221 responded to standardized behavioral challenges with larger increases in salivary free cortisol showed 222 greater shortening of leukocyte telomeres over a 3 year period. This association was independent of 223 baseline telomere length, age, gender, socioeconomic status (SES) defined by grade of employment, 224 smoking, cardiovascular risk factors (blood pressure, cholesterol, BMI, glycated hemoglobin) and 225 length of follow-up. The difference in telomere attrition between cortisol responders and non-226 responders corresponded to 107 base pairs on follow-up, indicating a difference of approximately two 227 years in aging (29). 228 The cortisol responses during mental stress testing in this study were small. A major purpose of 229 the study from which these data were drawn was to evaluate SES differences in stress reactivity and 230 recovery (23). Consequently, the task protocol was designed to be perceived as equally stressful across 231 the SES spectrum, and was selected after pretesting on this criterion. It did not involve socially 232 evaluative tasks such as the Trier Stress Test (TSST) that are known to elicit large cortisol responses 233 (30), since such tasks are often appraised differently by higher and lower social status individuals, 234 compromising any differences in physiological responsivity. The range of individual differences as 235 well as absolute magnitude of cortisol responses was therefore smaller than in some other 236 investigations. However, the value of the cortisol responder categorization adopted here has been 237 endorsed by evidence that individuals classified as cortisol responders show an increased risk of 238 incident hypertension (22) as well as more rapid progression of subclinical coronary artery disease as 239 indexed by coronary artery calcification (21). Brief cortisol responses to short-term tasks are of little 240 significance in themselves. However, the magnitude of acute cortisol responses is positively associated 241 with cortisol output in everyday life (31). If these responses are representative of people's habitual 242 profile of cortisol when confronted by the challenges of everyday life, they may contribute to chronic 243 neuroendocrine activation that could have deleterious health consequences. 244 Research relating telomere length with measures of cortisol output at rest have produced mixed 245 results (32,33), suggesting that relating individual differences in cortisol responses to standardized 246 mental stress with telomere length may be a valuable strategy. Epel et al (16) found that urinary cortisol 247 concentration collected over a night following a behavioral stress battery was inversely associated with 248 telomere length in healthy women. A study of older female caregivers of partners with dementia 249 showed relationships between telomere length and cortisol responses to behavioral challenge (19), 250 while work with children as young as 5 to 6 years has demonstrated that cortisol reactivity to mildly 251 stressful tasks is inversely correlated with telomere length (17,18). By contrast, a study of older men 252 and women in Finland showed no associations between telomere length and cortisol responses to acute 253 stress exposure, but is difficult to interpret since stress testing took place an average 2.1 years after 254 telomere assays (20). Our study builds on these findings by establishing a longitudinal relationship, 255 since cortisol responsivity predicted telomere shortening over time. The results are also consistent with 256 longitudinal clinical studies indicating that telomere length is shorter during active Cushing's syndrome 257 than when patients are in remission (34). 258 A puzzling feature of our results is that no association was present between cortisol responsivity 259 and telomere length at baseline. There was a negative association between cortisol responsivity and 260 baseline telomere length, but it was not significant. It is potentially relevant is that the studies of adults 261 that have shown associations between cortisol responsivity and telomere length have focused on 262 individuals exposed to chronic stressors such as caregiving or having children with severe disabilities 263 (16,19). No association has previously been observed in general population samples of the type 264 involved in the present study (20). It is possible that in our sample of relatively healthy older men and 265 women, these associations only emerged after several years. 266 We found a positive correlation between baseline telomere length and the magnitude of the 267 change in length over time. Regression to the mean has been put forward as the explanation of this 268 phenomenon (35). However, regression to the mean is unlikely to be the explanation for the association 269 with cortisol stress responsivity, since if anything, cortisol responders had slightly shorter telomeres at 270 baseline. Regression to the mean would therefore operate against the effects observed here. 271 The mechanisms underlying these associations have yet to be defined in detail. Telomere length 272 is regulated dynamically and does not decrease monotonically with advancing age (1). Faster telomere 273 attrition over time may result from several causes, including the expansion of leukocyte subsets that 274 occurs during inflammation and immunological responses, a decrease in telomerase activity, and 275 oxidative stress (27). Although cortisol responses might be expected to inhibit inflammation, 276 simultaneous heightened inflammation and cortisol is common in response to behavioral stress. A 277 reason for this might be because glucocorticoids have proinflammatory effects under some 278 circumstances. In vitro administration of glucocorticoids induces cytokine overexpression and NF-κB 279 activation in isolated macrophages (36), while pre-treatment with cortisol has been found to enhance 280 interleukin 6 responses to endotoxin (37). Cortisol administration in vitro also appears to reduce 281 telomerase activity (38). Frank, Watkins and Maier (39) have proposed that glucocorticoid responses to 282 stress may be neuroendocrine warning signals to the innate immune system, sensitizing 283 neuroinflammatory processes even after the corticosteroid response has dissipated. The combined 284 effect of reduced telomerase activity and oxidative stress would impinge negatively on the maintenance 285 of telomere length, particularly in the context of chronic inflammation, thus providing a plausible 286 explanation for the current findings. 287 This study has a number of limitations. The participants were middle-aged and older white 288 European men and women with no serious chronic illness, and results may not generalize to other 289 groups. Telomere length was measured in PBMCs, and values may differ in lymphocyte 290 subpopulations. Measures were also made with two different PCR machines at the two time points; 291 although this might affect comparisons of absolute values on the two occasions, it does not affect the 292 relative changes that are central to these results, so findings were the same with standardized measures 293 of telomere length. The cortisol responses were less substantial than those recorded with socially-294 evaluative stress testing, reducing the variability in responsivity profiles. We did not include a no stress 295 control group in this study, since we have previously found that the measurement protocol itself does 296 not induce physiological responses (40). 297 A strength of the study is that our findings were obtained in a well characterized longitudinal 298 population cohort, with a rather larger sample than has previously evaluated cortisol responses to acute 299 mental stress and telomere length. The results may have implications for understanding the pathways 300 through which social-environmental factors and mental ill-health impact cellular aging. If associations 301 between stress exposure and mental distress and telomere length are mediated through cortisol 302 responsivity, it is possible that the effects of mental stress on cellular aging might be reduced not only 303 by modifying stress exposure (which is not necessarily practical), but also by attenuating the 304 physiological components of the stress response. 305 In conclusion, the results of this study strongly suggest that heightened cortisol responsivity to 306 psychological stress is associated with accelerated cellular aging as indexed by leukocyte telomere 307 length. This indicates that heightened cortisol responsivity is not simply a consequence of more 308 advanced cellular aging, but may contribute to the cellular aging process.