Introduction
The hypothalamic–pituitary–adrenal (HPA) axis secretes the hormone ‘cortisol’ throughout the day to ensure everyday functioning of the body, and following emotionally stressful experiences (Dickerson & Kemeny, Reference Dickerson and Kemeny2004; Lupien et al. Reference Lupien, Maheu, Tu, Fiocco and Schramek2007; Wolkowitz et al. Reference Wolkowitz, Burke, Epel and Reus2009). Cortisol is thought to facilitate adaptation to stress (Lupien et al. Reference Lupien, McEwen, Gunnar and Heim2009; Wolkowitz et al. Reference Wolkowitz, Burke, Epel and Reus2009), but it has also been associated with poorer cognitive functioning (Lee et al. Reference Lee, Glass, McAtee, Wand, Bandeen-Roche, Bolla and Schwartz2007), accelerated cognitive aging (Seeman et al. Reference Seeman, McEwen, Singer, Albert and Rowe1997), cognitive impairments (Karlamangla et al. Reference Karlamangla, Singer, Chodosh, McEwen and Seeman2005) and Alzheimer's disease (Rothman & Mattson, Reference Rothman and Mattson2010) in older age populations. A longitudinal study of 194 older adults from the MacArthur Studies of Successful Aging found that increasing overnight urinary cortisol over the 2.5-year study period was associated with decline in memory performance in women, although not in men (Seeman et al. Reference Seeman, McEwen, Singer, Albert and Rowe1997). One larger population-based study on 1154 participants aged 65–88 years found no association between elevated morning cortisol and change in memory function over a period of 6 years, although elevated cortisol was associated with poorer memory at baseline (Comijs et al. Reference Comijs, Gerritsen, Penninx, Bremmer, Deeg and Geerlings2010). Lee et al. (Reference Lee, Glass, McAtee, Wand, Bandeen-Roche, Bolla and Schwartz2007) found that higher secretion of cortisol over 1 day [as reflected by greater area under the curve (AUC)] correlated with a broad range of cognitive domains, including speed of processing, executive functioning and verbal fluency, uniformly across different age groups (e.g. 50–54 to 65–73 years) and sex. One study failed to show a relationship of elevated cortisol with cognition (Lupien et al. Reference Lupien, King, Meaney and McEwen2001), whereas another reported associations of lower morning cortisol levels with worse cognitive performance (Evans et al. Reference Evans, Fredhoi, Loveday, Hucklebridge, Aitchison, Forte and Clow2011). Mostly, these studies of cortisol and cognitive functions have been limited by small unrepresentative study samples (Lupien et al. Reference Lupien, Lecours, Lussier, Schwartz, Nair and Meaney1994; Seeman et al. Reference Seeman, McEwen, Singer, Albert and Rowe1997; Li et al. Reference Li, Cherrier, Tsuang, Petrie, Colasurdo, Craft, Schellenberg, Peskind, Raskind and Wilkinson2006; Beluche et al. Reference Beluche, Ritchie and Ancelin2010) and cross-sectional design (Lee et al. Reference Lee, Glass, McAtee, Wand, Bandeen-Roche, Bolla and Schwartz2007; Gomez et al. Reference Gomez, Posener, Keller, DeBattista, Solvason and Schatzberg2009; Franz et al. Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011).
Associations of cortisol with cognitive functions during childhood have also been reported. For instance, in the 1958 British Birth Cohort, hypo-secretion of post-waking (i.e. 45 min after waking) cortisol at 45 years in men was preceded by lower cognitive ability in childhood (Power et al. Reference Power, Li and Hertzman2008). Similarly, in a cross-sectional sample of 53 Spanish children aged 9 to 12 years, lower early morning cortisol was found to be negatively associated with speed of memory, independently of daily stress perception (Maldonado et al. Reference Maldonado, Fernandez, Trianes, Wesnes, Petrini, Zangara, Enguix and Ambrosetti2008). Furthermore, children of mothers with elevated cortisol during their pregnancy were more likely to exhibit poorer cognitive development (LeWinn et al. Reference LeWinn, Stroud, Molnar, Ware, Koenen and Buka2009; Davis & Sandman, Reference Davis and Sandman2010). It is possible, therefore, that effects of cortisol on cognitive abilities observed at later ages might reflect earlier associations with cognitive function, although with few exceptions (Franz et al. Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011) studies to date have neglected potential developmental HPA axis/cognition links. If cognitive function in childhood plays a role in the foundation of the development of HPA axis function (or vice versa), such associations should be taken into account in investigations of cortisol and cognition at later stages of life.
The 1958 British Birth Cohort provides a unique opportunity to examine associations between cortisol at 45 years and cognitive function at 50 years in a large prospective birth cohort, while taking account of factors at earlier life stages. The availability of data on childhood cognition enables us to test whether any association of cortisol with adult cognitive functions is attributable to associations with cognitive ability in the first two decades of life. This study adopted a life-course perspective to address three questions:
-
(1) Is cortisol associated with different cognitive functions (i.e. verbal memory and fluency and speed of processing) in mid-life?
-
(2) To what extent do these associations (if any) reflect associations with childhood cognition?
-
(3) Are associations of cortisol with cognitive functions in mid-life, but not explained by childhood cognition, due to other lifetime covariates such as birthweight and depression/anxiety?
Method
Participants
Data for this study were collected for the 1958 British Birth Cohort (or the National Child Development Study), a prospective population-based cohort consisting of 18 558 individuals: 17 638 participants enrolled in the Perinatal Mortality Survey all born 1 week in March 1958 in England, Scotland and Wales, and 920 immigrants with the same birth dates included to age 16 years. Information was collected in childhood (7, 11 and 16 years) and in adulthood (23, 33, 42, 45 and 50 years). Ethical approval for saliva collection and clinical interview at 45 years was given by the South East Multi-Centre Research Ethics Committee, and all cohort members gave written informed consent. Sampling and data collection procedures and attrition have been described in detail elsewhere (Power & Elliott, Reference Power and Elliott2006; Atherton et al. Reference Atherton, Fuller, Shepherd, Strachan and Power2008).
Salivary cortisol measurement
At 44–45 years, participants underwent a biomedical examination by a trained research nurse visiting their home. Participants were requested to collect two saliva samples on the next convenient day, the first at 45 min after awakening (time 1) and the second 3 h later (time 2) on the same day. A reminder was sent to 53% of those consenting to return a sample, if they had not done so within 2 weeks. Participants were asked to chew on a salivette until it was soaked, to record the date and time of collection, and to store the sample at room temperature until mailed to the laboratory. Although salivary cortisol is stable at room temperature for up to 30 days, samples were frozen after reaching the laboratory to reduce microbial growth. Cortisol levels were analyzed at the University of Dresden with a commercial immunoassay kit with chemiluminescence detection (CLIA; IBL-Hamburg, Germany). The lower sensitivity of this assay is 0.44 nmol/l, with intra- and interassay precision of <10% for a wide range of cortisol concentrations. High levels (>50 nmol/l) were rerun in a second assay for confirmation.
Cortisol indicators
Extreme outliers for time 1 (t1) and time 2 (t2) cortisol were truncated [at 2 nmol/l for <2 nmol/l (n=24 at t1, n=122 at t2) and at 100 nmol/l for >100 nmol/l (n=22 at t1, n=20 at t2)] so that extreme values did not exert a disproportionate influence on analyses. Not all samples were collected at the specified period after waking and after an interval of 3 h, leading to variation around the target time for t1 [mean (s.d.)=49 (15) min] and t2 [mean (s.d.)=3 h 5 min (27 min)]. Cortisol level was influenced by both the time of awakening and the time since awaking, and to take account of this variability the truncated cortisol values for each individual were centered at 08:08 h (45 min after a mean waking time of 07:22 h) and t2 at 11:08 h (3 h 45 min after mean awakening time) based on predictions from linear regression models at these time points (Power et al. Reference Power, Li and Hertzman2008). The time-adjusted cortisol values were used to derive three cortisol indicators: (1) t1; (2) t2 cortisol levels; (3) the t2–t1 delta calculated as the difference between t2 and t1 cortisol levels divided by 3 h (higher values reflect flatter, that is less negative, slopes). Pearson correlation between t1 and t2 cortisol was 0.28 and delta correlated at −0.79 with t1 and at 0.38 with t2, all p<0.001.
Cognitive functions at 50 years
The cognitive measures included in the 1958 British Birth Cohort have been widely used in other longitudinal studies [e.g. the 1946 British Birth Cohort in Wadsworth et al. (Reference Wadsworth, Kuh, Richards and Hardy2006) and the English Longitudinal Study of Aging in Llewellyn et al. (Reference Llewellyn, Lang, Langa and Huppert2008) ]. The cognitive assessment was conducted by the interviewer visiting the participant's home at the 50-year survey. Participants were requested to wear their spectacles, if needed, and interviewers recorded any contextual factors that might impair performance during tests (i.e. presence of others in the room, interruption/distraction, noisy environment, deaf or hard of hearing, too tired, illness or physical impairment, impaired concentration, blind or poor eyesight, very nervous or anxious, other mental impairment, problem with laptop, others). The cognitive tasks were administered to participants as follows: immediate and delayed word lists examined how many words a participant could recall from a list of 10 common words (e.g. child, book, tree) immediately after the word list was read (immediate memory; range 0–10), and after a short delay of about 5 min (delayed memory; range 0–10), during which other cognitive tasks were completed. For most (98%) participants, the words were read by a computer voice at the rate of one word every 2 s; for the remaining 2%, words were read by the interviewer at the same pace. The latter group remembered fewer words; hence the method of word delivery was controlled in the analyses. Four randomly assigned word lists (a/b/c/d) were used and differences observed in both immediate and delayed recall across the four lists were taken into account in the analyses. In the animal naming task, participants were asked to name as many different animals as possible in 60 s (range 0–65). This test provides a measure of verbal fluency based on the ability to access mental vocabulary rapidly. For the letter cancellation test, developed for the 1946 Birth Cohort (Richards et al. Reference Richards, Kuh, Hardy and Wadsworth1999), participants were presented with a page incorporating 125 upper-case letters of the alphabet, arranged in 26 rows and 39 columns. Of these, 65 were target letters (P and W). Participants were instructed to cross out as many target letters as possible within 1 min. The total number of letters searched (range 84–780), that is the sum of all items processed whether correctly or incorrectly, assessed processing speed. Pearson correlations between adulthood cognitive tests ranged from 0.08 to 0.65; all p's<0.001 (see online Supplementary Table S1).
Childhood cognitive ability
Cognitive ability was assessed at 7, 11 and 16 years using appropriate tests for mathematics and reading and a standardized test assessing general ability. At each age, tests were administered in the participant's school by their teacher. The arithmetic test at age 7 comprised 10 problems with graded levels of difficulty (range 0–10); teachers read the questions to poor readers. At age 11, the mathematics test was constructed by the National Foundation for Educational Research in England and Wales (range 0–40). At 16 years, a mathematics comprehension test was constructed at Manchester University (range 0–31). The Southgate test (range 0–30) was used to detect poor readers at age 7 (Southgate, Reference Southgate1962): children selected from several words the one corresponding to a picture, teachers also read out words that the children identified from a list. Reading tests at ages 11 and 16 years were parallel to the Watts Vernon comprehension test (range 0–35). A general ability test at age 11 approximated the conventional intelligence test, with verbal (range 0–40) and non-verbal (range 0–40) components (Douglas, Reference Douglas1964). The seven tests were standardized for month and year of assessment, and averaged to obtain a summary score for childhood cognitive ability. Pearson correlations between childhood tests ranged from 0.48 to 0.81; all p's<0.001 (online Supplementary Table S1).
Covariates
Testing conditions
Factors specific to cortisol measurements were: medication at 45 years known to influence cortisol (yes/no), saliva collection day (weekday/weekend), and sex. Factors specific to cognitive tests at 50 years were: testing time of week and day (weekday am/pm/evening/weekend am/pm/evening), presence of others in the room (yes/no), the way in which the word list was administered (computer voice/interviewer), word list (a/b/c/d), and other contextual factors impairing performance during test.
Other covariates
Covariates were identified from the literature and included socio-economic position (SEP) in childhood and adulthood (Li et al. Reference Li, Power, Kelly, Kirschbaum and Hertzman2007; Geoffroy et al. Reference Geoffroy, Côté, Giguère, Dionne, Zelazo, Tremblay, Boivin and Séguin2010), smoking habits (Rohleder & Kirschbaum, Reference Rohleder and Kirschbaum2006; Power et al. Reference Power, Li and Hertzman2008), body mass index (BMI; Corley et al. Reference Corley, Gow, Starr and Deary2010), depression/anxiety (Bremmer et al. Reference Bremmer, Deeg, Beekman, Penninx, Lips and Hoogendijk2007; Power et al. Reference Power, Li, Atherton and Hertzman2010), heavy alcohol drinking (Anttila et al. Reference Anttila, Helkala, Viitanen, Kåreholt, Fratiglioni, Winblad, Soininen, Tuomilehto, Nissinen and Kivipelto2004; Badrick et al. Reference Badrick, Bobak, Britton, Kirschbaum, Marmot and Kumari2008) and menopausal status (Richards et al. Reference Richards, Kuh, Hardy and Wadsworth1999).
SEP in childhood was based on father's occupation at birth (or at 7 years if missing) using the Registrar General's Social classification, grouped as I or II (professional/managerial), IIINM (skilled non-manual), IIIM (skilled manual) and IV and V (semi-skilled and unskilled manual, including single mother households). SEP in adulthood was based on the participant's current or most recent occupation at 42 years and categorized as above. Smoking habits at 42 years (current or former/never) and heavy drinking at 45 years (yes/no) were reported by the participants. Highest qualification attained by 42 years (none/some qualifications/O-level or equivalent/A-level or equivalent/higher degree). A survey nurse administered the Clinical Interview Schedule (Lewis et al. Reference Lewis, Pelosi, Araya and Dunn1992) to assess depression/anxiety in the past week at 45 years (e.g. those with ⩾2 symptoms of depression and/or anxiety were identified). BMI at 45 years (or at 50 years if missing) was defined as a participant's weight in kilograms divided by the square of his height in meters (kg/m2). Post-menopausal women at 45 years (12 months of amenorrhea with no other reason reported to explain cessation) were distinguished from those who were pre-menopausal (3 months of amenorrhea), peri-menopausal or whose periods stopped because of surgery or some other reason.
Statistical analyses
Linear regression models were used to examine associations between cortisol at 45 years and cognitive functions at 50 years. Both response and explanatory variables were standardized Z scores, so β parameters represent change in standard deviation (s.d.) units in the cognitive outcomes (e.g. immediate/delayed word list, animal naming, letter cancellation) for 1 s.d. increase on the continuous cortisol indicators (i.e. t1, t2, delta). The analytical approach consisted of four steps: (1) examination of associations between the three cortisol indicators and the four cognitive tests, adjusting for sex and factors related to measurement of cortisol (i.e. medication, day of week) and cognitive functions (i.e. testing date and time of day, months and years, computer voice, word list, presence of other persons, other factors), (2) adjustment for childhood cognitive ability, (3) adjustment for other covariates (e.g. birthweight, SEP in childhood and adulthood, education, smoking and drinking habits, BMI, menopausal status, depression/anxiety), and (4) testing the interaction between each cortisol indicator and sex on 50 years cognitive measures. No consistent interaction was found; hence we present results for men and women combined with adjustment for sex. For each cortisol indicator, when its unadjusted association with cognitive function at 50 years was significant, the percentage reduction of the association after the adjustment of childhood cognition or other covariates was calculated using the formula: [(A – B)/A]×100, where A is the initial β coefficient and B is the adjusted β coefficient.
Eligible participants for this study were those who completed at least one of the four cognitive tests at 50 years (n=9649). Information on cortisol was available for 5771 participants (60% of the 9649), of whom 1116 had incomplete data on covariates. The study sample varies from 4502 for cancellation letter to 4581 for immediate word list and animal naming. We applied inverse probability weighting to adjust for potential selection bias (that is, missing data due to missing covariates and losses to follow-up) with respect to the surviving cohort in 2008–2009 (n=17 089). The probability of being in the study sample was estimated from logistic regressions using factors associated with sample attrition including sex, social class at birth, mathematics scores and internalizing and externalizing behavior problems at 7 years (Atherton et al. Reference Atherton, Fuller, Shepherd, Strachan and Power2008). Sensitivity analyses were conducted, first to test whether results differed when participants not adhering to the cortisol collection protocol were excluded. We classified participants as not adhering to the protocol if their first cortisol measurement deviated from the prescribed time by >15 min (i.e. cortisol collection within 0–28 or 61–355 min after waking). We compared adherent versus non-adherent participants on key variables using χ2 and t tests. As a second sensitivity analysis, we investigated whether the lowest 5% extreme of t1 cortisol distribution (using a cut-off at 7.24 for men and 7.94 nmol/l for women) was associated with cognitive tests at 50 years. This was done because we found previously that men with lowest 5% t1 cortisol performed significantly poorly on math tests at ages 11 and 16 years (Power et al. Reference Power, Li and Hertzman2008). Finally, we used multiple imputations by chained equations methods in SPSS version 19 (SPSS Inc., USA) to test whether the results would have differed if there were no missing data on covariates. We created 30 complete datasets, and conducted regression analyses that combined the results from the 30 datasets. The results based on imputation were similar to those presented here based on inverse probability weighting alone; weak and inconsistent interactions are not reported.
Results
Table 1 presents sample characteristics on key variables. Table 2 shows β coefficients for associations of cortisol indicators at 45 years with cognitive tests at 50 years. Individuals with higher t2 cortisol scored slightly lower on the immediate and delayed memory and on the animal naming tasks by approximately 0.05 s.d. for 1 s.d. increase in t2 cortisol. Adjustments for childhood cognition reduced these associations by 30–31%, but the associations remained (p<0.05). Controlling for other covariates (including birthweight, lifetime SEP, education, smoking and drinking habits, BMI, menopausal status, and depression/anxiety) further reduced (by 10–17%), but did not abolish, the associations (p<0.05). Similarly, individuals with flatter delta cortisol scored approximately 0.04 s.d. lower than their counterparts on immediate verbal memory. After adjustment for childhood cognition, the association was reduced by 12% and continued to be significant, even after controlling for other covariates. However, this association was not replicated in sensitivity analyses, excluding non-adherent participants. Furthermore, elevated t1 cortisol was associated with slower letter cancellation, although this association was not evident in analysis excluding non-adherent participants. Participants not adhering to the saliva collection protocol for t1 (8.5%) were more likely than others (p⩽0.001 for all) to be male (9.9% v. 7.2% female) and had an earlier wake-up time (07:12 v. 07:24 h), lower t1 cortisol (20.21 v. 21.49 nmol/l), poorer immediate (6.42 v. 6.66) and delayed (5.26 v. 5.57) verbal memory and poorer cognitive ability in childhood (–0.09 v. 0.18) but did not differ by t2 cortisol, depression/anxiety, verbal fluency and speed of processing. Exclusion of non-adherent participants had little effect on the results with t2 cortisol, although individuals with less negative delta scored lower on the animal naming task by about −0.036 s.d. for 1 s.d. unit increase in cortisol delta. This association remained largely unchanged after childhood cognition and other covariates were taken into account. As shown in Table 3, individuals with t1 cortisol in the lowest 5% of the distribution performed more poorly on immediate and verbal word-list tests than their counterparts, but there were no consistent associations with animal naming and processing speed (data not shown). Associations were partially explained by childhood cognition, with reductions after adjustment of 39% for immediate word list and 40% for delayed word list. This pattern of results was not altered after non-adherent participants were excluded, although β values were smaller.
Table 1. Description of the 1958 British Birth Cohort on key characteristics (n=4655)
BMI, Body mass index; SEP, socio-economic position; s.d., standard deviation.
a Calculated using centered t1 and t2 cortisol values, that is allowing for measurement time.
b n=4621 for delayed word list and 4575 letter cancellation.
Table 2. Weighted regressions predicting cognitive outcomes at 50 years with cortisol indicators at 45 years (z scores)
s.e., Standard error; n.a., not applicable.
Model 1 adjusted for sex and testing conditions: medication at 45 years, cortisol testing day, cognitive testing date and time of day, computer voice, word list, presence of other persons, other contextual factors.
Model 2 adjusted for summary score for childhood cognitive ability between 7 and 16 years.
Model 3 adjusted for other covariates: birthweight, socio-economic position (SEP) in childhood and adulthood, educational attainment by 42 years, heavy drinking at 45 years, body mass index (BMI) at 45 years, smoking habits at 42 years, menopausal status at 45 years, depression/anxiety at 45 years.
Table 3. Weighted regressions predicting verbal memory at 50 years with lowest 5% t1 cortisol at 45 years
s.e., Standard error; n.a., not applicable.
Model 1 adjusted for sex and testing conditions: medication at 45 years, cortisol testing day, testing date and time of day, cognitive testing months and years, computer voice, word list, presence of other persons, other contextual factors.
Model 2 adjusted for summary score for childhood cognitive ability between 7 and 16 years.
Model 3 adjusted for other covariates: birthweight, socio-economic position (SEP) in childhood and adulthood, educational attainment by 42 years, heavy drinking at 45 years, body mass index (BMI) at 45 years, smoking habits at 42 years, menopausal status at 45 years, depression/anxiety at 45 years.
Discussion
In the 1958 British Birth Cohort, we found consistent associations whereby elevated cortisol in the late morning (t2) measured in mid-adulthood (45 years) was associated with worse verbal memory and verbal fluency at 50 years. Effect sizes were small in magnitude, with a deficit of approximately 0.05 s.d. in verbal memory and verbal fluency functions for each s.d. unit higher cortisol level in late morning. About 30% of these associations were attributable to an earlier association with childhood cognition. Nonetheless, there remained an association between t2 cortisol at 45 years and verbal memory and verbal fluency functions at 50 years after accounting for childhood cognition and a broad range of lifetime covariates (i.e. birthweight, SEP in childhood and adulthood, education, smoking habits, BMI, heavy alcohol drinking, depression/anxiety symptoms, and menopausal status). Early morning cortisol (t1) and cortisol morning delta showed inconsistent associations with cognitive outcomes. In addition, individuals with extremely low t1 performed about 0.20 s.d. poorer than their counterparts on the verbal memory tests. Nonetheless, about 40% of these associations were attributable to childhood cognition.
Methodological considerations
Our results need to be interpreted in light of the study limitations. First, this study, like many others, relies on motivation of participants for collecting cortisol samples at the prescribed time and reporting the exact time at which samples were taken (Kudielka et al. Reference Kudielka, Broderick and Kirschbaum2003; Kunz-Ebrecht et al. Reference Kunz-Ebrecht, Kirschbaum, Marmot and Steptoe2004). A total of 8.5% of participants were ‘non-adherent’ because their first sample (i.e. 45 min after waking) was collected at least 15 min before or after the prescribed time. Typically, cortisol rises up to 50–60% of its value at waking in the 30–45 min after awaking, and declines rapidly in the following few hours (Wolkowitz et al. Reference Wolkowitz, Burke, Epel and Reus2009), and the absence of a morning cortisol rise has been associated with negative mental health outcomes (Adam & Kumari, Reference Adam and Kumari2009). There is a danger of missing the cortisol rise if the sample is taken too early or too late (Kudielka et al. Reference Kudielka, Broderick and Kirschbaum2003; Kunz-Ebrecht et al. Reference Kunz-Ebrecht, Kirschbaum, Marmot and Steptoe2004). We found that exclusion of non-adherent participants did not modify the general pattern of results, mostly important for extremely low t1. However, some bias may remain because some participants may have reported incorrect times of saliva collection. Second, our cortisol indicators might not adequately capture an individual's secretion level. Our protocol includes only two morning cortisol samples during 1 day from which three cortisol indicators were derived. More cortisol samples during the day or night, collected over several years, may have provided a better indication of cumulative exposure to cortisol and shown stronger associations with cognitive outcomes. Furthermore, it was not possible to assess relationships of cortisol awakening response (CAR) and cognitive functions because cortisol was not measured upon awakening. Third, if there is an effect of cortisol on cognition in childhood that persisted into adulthood, such an effect might be underestimated because childhood cognitive measures differed from those in adulthood. Fourth, with regard to cognitive decline associated with aging, participants of the 1958 British Birth Cohort are relatively young (50 years) and it is possible that a negative impact of cortisol on cognition could manifest more severely at later ages, given the increased vulnerability of the brain in the elderly (Lupien et al. Reference Lupien, McEwen, Gunnar and Heim2009.). Finally, because of sample attrition, the study sample might not completely represent the original population. Although analyses were weighted to minimize the impact of potential bias on the results, it is still possible that subgroups of the population may be unrepresented. Our study also included several methodological strengths: (1) an unselected, non-clinical sample to make inference to the general population; (2) a large sample size, which enabled us to detect small, but significant, associations, (3) detailed measures of cognitive ability in childhood (7–16 years) and adulthood (50 years), and (4) the inclusion of life-course covariates.
Interpretation of findings
We are not aware of other epidemiological studies that have investigated associations of naturally occurring cortisol and cognitive functions in the 5-year interval studied here, and, more generally, there is only limited information on the longitudinal course of the HPA axis and links with cognition. Nonetheless, our finding of associations between late morning cortisol and verbal memory and verbal fluency at 50 years is consistent with previous observational studies showing effects of elevated basal cortisol on worse memory functions (Sauro et al. Reference Sauro, Jorgensen and Teal Pedlow2003; Lee et al. Reference Lee, Glass, McAtee, Wand, Bandeen-Roche, Bolla and Schwartz2007; Evans et al. Reference Evans, Fredhoi, Loveday, Hucklebridge, Aitchison, Forte and Clow2011; Franz et al. Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011). It is currently unknown whether our 5-year interval from adult cortisol and cognition assessment (i.e. 45 to 50 years) is particularly important relative to other life stages, but for cognitive function there is evidence of age-related decline over this period. For instance, decline in verbal memory and fluency has become apparent in mid-life (i.e. 43 to 53 years) for some individuals (Richards et al. Reference Richards, Shipley, Fuhrer and Wadsworth2004), although a steeper rate of decline is mostly seen with advancing age (Hedden & Gabrieli, Reference Hedden and Gabrieli2004). Thus, our study time-frame of 45–50 years for adult cortisol and cognitive function informs understanding of the influence of cortisol on cognitive functions in a period where age-related decline is just becoming apparent. For the purpose of preventive interventions, it is important to study influences that precede much of the decline in cognitive function, as others have argued elsewhere (Salthouse, Reference Salthouse2009).
Given previous literature showing associations between cortisol and cognition in childhood (Maldonado et al. Reference Maldonado, Fernandez, Trianes, Wesnes, Petrini, Zangara, Enguix and Ambrosetti2008; Power et al. Reference Power, Li and Hertzman2008; LeWinn et al. Reference LeWinn, Stroud, Molnar, Ware, Koenen and Buka2009; Davis & Sandman, Reference Davis and Sandman2010) an important objective of our study was to establish the extent to which associations of cortisol with cognition in mid-life were attributable to earlier associations with childhood cognition. Associations of elevated cortisol 3 h 45 min after awaking with verbal memory and fluency attenuated by about 30% when childhood cognition was taken into account. Similarly, associations of extremely low cortisol levels 45 min after awaking with verbal memory attenuated by 40%. Our interpretations of these findings are twofold. First, because the relationships of elevated late morning cortisol and worse verbal memory and fluency were not completely abolished when taking account of early childhood cognition, this study supports the hypothesis that elevated late-morning cortisol at 45 years predicts worse verbal memory and fluency at 50 years. Second, the attenuation of adult cortisol levels/verbal memory and fluency associations with adjustment for childhood cognition suggests that the latter could play a role in the establishment of healthy HPA functioning, which is then reflected in cortisol levels more than 30 years later. The biological plausibility of such developmental HPA axis/cognitive function links is supported by the presence of glucocorticoid receptors in the hippocampus and frontal lobes, and the role of these receptors and brain regions in glucocorticoid down-regulation (Lupien et al. Reference Lupien, Maheu, Tu, Fiocco and Schramek2007). The developmental links between cortisol and cognition are supported by our previous study with the 1958 cohort that showed an association of childhood cognition with cortisol patterns at 45 years (Power et al. Reference Power, Li and Hertzman2008). In line with this argument, Franz et al. (Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011) recently examined cross-sectional associations of salivary cortisol and cognitive functioning at 51–55 years in the Vietnam Era Twin Study of Aging (VETSA), while adjusting for prior cognitive functions at 20 years. Even though mid-life cortisol–cognition associations in adulthood were generally not altered by prior cognition, cognitive ability at 20 years predicted higher cortisol AUC 35 years later.
We did not detect a consistent association of cortisol with the letter cancellation task, which may measure different cognitive skills [general intelligence factor (g)] compared to the three other tests, to which it is only weakly correlated (r<0.10). Nonetheless, it is possible that associations may emerge at later ages, given that slower speed of processing in the presence of elevated cortisol has been reported (Lee et al. Reference Lee, Glass, McAtee, Wand, Bandeen-Roche, Bolla and Schwartz2007; Beluche et al. Reference Beluche, Ritchie and Ancelin2010; Evans et al. Reference Evans, Fredhoi, Loveday, Hucklebridge, Aitchison, Forte and Clow2011; Franz et al. Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011), although not by all (Hinkelmann et al. Reference Hinkelmann, Moritz, Botzenhardt, Riedesel, Wiedemann, Kellner and Otte2009). Moreover, we found no consistent pattern of association of early morning cortisol (t1) with cognitive functions. It is unclear why late mid-morning cortisol (t2) predicted later cognitive functions and early morning cortisol (t1) did not, although such inconsistency for early morning cortisol has been reported by others (Evans et al. Reference Evans, Fredhoi, Loveday, Hucklebridge, Aitchison, Forte and Clow2011; Franz et al. Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011). Yet it is notable that, although the present study found no associations for continuous t1 cortisol, individuals with extremely low t1 performed poorly on immediate and delayed word-list tests. Two other studies detected associations between small cortisol awaking response and poorer cognitive performance in childhood (Maldonado et al. Reference Maldonado, Fernandez, Trianes, Wesnes, Petrini, Zangara, Enguix and Ambrosetti2008) and in adulthood (Evans et al. Reference Evans, Fredhoi, Loveday, Hucklebridge, Aitchison, Forte and Clow2011), whereas previously for men in our cohort, lowest t1 cortisol was found to be associated with lower scores on child school tests (Power et al. Reference Power, Li and Hertzman2008). The demonstration in the present study that adjustments for childhood cognitive ability weakened associations between lowest t1 cortisol at 45 years and immediate and delayed verbal memory by 40% suggests that early life cognition/cortisol associations contribute substantially to the association in adulthood.
In general, effect sizes for elevated basal cortisol and worse memory functions in past observational studies have ranged from small to large (Sauro et al. Reference Sauro, Jorgensen and Teal Pedlow2003). For instance, in one small cross-sectional sample of middle-aged men and women, the correlation (r) of higher cortisol with poorer verbal memory was −0.28, which corresponds to a small–moderate effect (Gomez et al. Reference Gomez, Posener, Keller, DeBattista, Solvason and Schatzberg2009). In another longitudinal study of older adults, increasing cortisol over a 5-year period correlated at −0.60 with poorer memory function (Lupien et al. Reference Lupien, Lecours, Lussier, Schwartz, Nair and Meaney1994). Comijs et al. (Reference Comijs, Gerritsen, Penninx, Bremmer, Deeg and Geerlings2010) reported that for 1 s.d. unit greater cortisol, individuals learned 1.29 fewer words (out of 15), when no adjustment for covariates was made. In the present study, effect sizes linking elevated cortisol in late morning at 45 years with verbal memory and fluency functions at 50 years were small in magnitude (i.e. cognitive deficits of 0.05 s.d. unit for 1 s.d. unit higher cortisol in late morning). However, even when effect sizes are small, the age equivalence of elevated cortisol on cognition could have meaningful public health implications (Lee et al. Reference Lee, Glass, Wand, McAtee, Bandeen-Roche, Bolla and Schwartz2008; Franz et al. Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011). For instance, with cross-sectional correlations of −0.09 to −0.14 and a narrow age range sample (i.e. 51–60 years), Franz et al. (Reference Franz, O'Brien, Hauger, Mendoza, Panizzon, Prom-Wormley, Eaves, Jacobson, Lyons, Lupien, Hellhammer, Xian and Kremen2011) calculated that the age equivalency effect of a cortisol AUC increase from the 25th to the 75th percentile was comparable to an increase in age from 1.76 to 3.53 years.
Conclusions
Although effect sizes were small in magnitude, this study supports the hypothesis that certain cortisol secretion patterns are associated with poorer memory and verbal fluency in mid-life, but that contributions from early cognition to these associations are evident. Repeated measurements of cognition and cortisol collected throughout life are needed to clarify further the evolution of associations across the lifespan.
Note
Supplementary material accompanies this paper on the Journal's website (http://journals.cambridge.org/psm).
Acknowledgements
M.-C. Geoffroy was support by a fellowship from the Canadian Institutes of Health Research. L. Li was supported by a Medical Research Council (MRC) Career Development Award in Biostatistics. Data collection at 45 years was funded by the MRC, grant G0000934, including measurement of cortisol levels under the direction of Professor Kirschbaum (Biological Psychology, Department of Psychology, University of Dresden, Germany). The Great Ormond Street Hospital (GOSH)/University College London (UCL) Institute of Child Health was supported in part by the Department of Health's National Institute for Health Research (NIHR) Biomedical Research Centre. The Centre for Paediatric Epidemiology and Biostatistics was supported in part by the MRC in its capacity as the MRC Centre of Epidemiology for Child Health. C. Hertzman was supported by the Canada Research Chairs Program.
Declaration of Interest
None.
