The increase in life expectancy and the exponential increase in the number of elderly people give new importance to the development of preventive strategies to reduce or delay the onset of cognitive impairment, a major component of age-related diseases. One leading cause of cognitive impairment is an increase in brain oxidative stress.1,2 Selenium, an antioxidant and a main constituent of brain selenoproteins, may be particularly important for the maintenance of brain functions.3 Seleno-glutathione peroxidase (GSH-Px) constitutes an important line of defense against free radicals acting against hydrogen peroxide and lipid peroxidation.4 However, the most important selenoprotein for cerebral functions is probably selenoprotein P.5 This protein is synthesized at the cerebral level and protects the brain against oxidative damage, particularly peroxynitrite.6 Furthermore, selenium status decreases with old age.7–10 Therefore, marginal or deficient selenium status may be a risk factor for a decline of cognitive functions. Selenium and cognition changes could also both reflect the aging process.
Only one study has investigated the relationship between longitudinal cognitive decline and baseline selenium level;11 the highest declines in cognitive functions were associated with the lowest plasma selenium concentrations at baseline. Limited data are available from selenium supplementation studies in the elderly. In the EPESE cohort (n = 2082), Gray et al12 showed that subjects who currently used antioxidant supplements (vitamins C, E, and A and selenium or zinc) had a lower risk of cognitive decline than nonusers. However, it is impossible to isolate the specific effect of selenium with multiantioxidant supplementation.
The present study considered selenium change during follow-up rather than just considering baseline selenium level. We investigated the relationships between short-term selenium changes (the first 2-year) and long-term selenium changes (9-year) with cognitive changes during the 9-year follow-up of the EVA study (“Etude du Vieillissement Artériel”).
METHODS
Subject Cohorts
The EVA study is a 9-year longitudinal study with 6 waves of follow-up.13 During the first 2 years (1991–1993; EVA0), 1389 volunteers (574 men and 815 women) born between 1922 and 1932 and residing in the town of Nantes (Western France) were recruited from electoral rolls and, to a lesser extent, through information campaigns. The subsequent follow-up waves were EVA2 (1993–1995), EVA3 (1995–1997), EVA4 (1997–1999), and EVA5 (1999–2000); the sixth and last follow-up of the EVA study (EVA6) was conducted between June 2000 and December 2001. The numbers of subjects who completed a general questionnaire and a cognitive evaluation at each wave were 1272 at EVA2, 1188 at EVA3, 1042 at EVA4, 792 at EVA5, and 702 at EVA6. The study protocol was approved by the Ethical Committee of the University Center Hospital of Kremlin-Bicêtre, Paris. Signed informed consent was obtained from all participants at enrollment.
Data Collection
Questionnaire and General Medical Examination
The general questionnaire at baseline allowed us to obtain information on sociodemographic factors such as sex, age, educational achievement (no school or primary school/high school or university), and consumption habits such as smoking (current/ former/ nonsmokers) and alcohol consumption (≥20 mL/<20 mL per day). Alcohol intake was determined from the subject's estimated average amount of alcoholic beverages ingested weekly and expressed in milliliters of alcohol per day. In addition, height and weight were measured and body mass index (BMI) was calculated (kg/m2). Two independent measures of systolic and diastolic blood pressure were made with a digital electronic tensiometer after a 10-minute rest.
Cognitive Evaluation
At each wave, trained neuropsychologists evaluated cognition with a neuropsychologic battery of tests, including assessment of a range of cognitive domains and a global test, the Mini-Mental Status Examination (MMSE).14 Visual attention was assessed with the Trail Making Test part B (TMTB).15 The Digit Symbol Substitution (DSS) from the Wechsler Adult Intelligence Scale–Revised measured sustained attention and logical reasoning.16 Psychomotor speed was evaluated with the Finger Tapping Test (FTT).
Biologic Parameters
At EVA0, EVA2, and EVA6, selenium was determined in serum using electrothermal atomic absorption spectrometry (Perkin Elmer 5100 ZT, Norwalk, CT) as previously described17 A selenium electrodeless discharge lamp and a Zeeman longitudinal background correction were used. Serum was diluted in a solution containing 0.1 M nitric acid and 0.2% (wt/vol) Triton X 100. Ten microliters of this dilution and 5 μL of matrix modifier were introduced onto the platform of a pyrolytic graphite furnace. Concentration was obtained using an addition calibration. Seronorm trace element serum was used as internal quality control (Sero, Billingstad, Norway). We considered the long-term change of plasma selenium by the difference between the plasma selenium measured at the 9-year follow-up and at baseline (n = 751). We also assessed the short-term plasma selenium variability by the difference between the 2-year measurement and baseline (n = 1064). Total plasma cholesterol and plasma glucose level were also measured using routine methods.
At each wave of follow-up, a new general questionnaire was completed by participants; clinical examination, neuropsychologic tests, and blood sampling were also repeated. We thus updated all variables during the study except for alcohol consumption, height, and weight, which were obtained only at EVA0, EVA2, and EVA3.
Statistical Methods
The characteristics of subjects at inclusion were described in 2 groups: those who did and did not complete the follow-up. To test the differences between these 2 groups, χ2 test and the Student t test were used. Percentile distribution and means with standard deviations (SDs) were described for both cognitive and selenium change variables. Classic linear regressions were used to assess the association between 9-year selenium change and cognitive change, and the association between 2-year selenium change and cognitive changes during the 2, 4, 6, and 9 years of follow-up. Analyses were controlling for age, sex, education, and selenium level at baseline.
To simultaneously take into account cognitive changes at each wave of the study and the within-subject correlation of measurements, we used mixed models (MIXED procedure in SAS; SAS Institute, Cary, NC) to analyze associations between cognitive and selenium changes. We also dichotomized cognitive decline by using 2 cutoffs. For each subject and at each wave, we calculated the cognitive score difference between that wave and baseline. Cutoff points corresponded to the 25th and 10th percentile of the distributions of the mean of these differences. For MMSE, the cognitive decline variable was first defined by a loss of 2 points and then by a loss of 3 points (25th and 10th percentiles, respectively). The same cutoffs were used for all cognitive continuous variables and corresponded to differences of −3 and −6 points for DSS, +2 and −7 taps for FTT, and a time difference of 7.8 and 32.4 seconds for TMTB. To analyze these dichotomous cognitive variables, we performed a mixed logistic model with Gaussian random effect (NLMIXED procedure in SAS).18
Selenium changes were analyzed as continuous variables. Analyses were, first, adjusted for time and selenium level at baseline. Time was an explanatory variable and considered to be a combination of follow-up duration and age at inclusion; this combination allowed us to test whether the effect of time is more pronounced in older subjects by introducing an interaction term between these 2 covariables. Second, analyses were adjusted for other potential confounding factors (which could be fixed or updated at each wave of the study) associated with cognition or selenium level such as sociodemographic factors, consumption habits, but also health factors or indicators such as BMI (in kg/m2), hypertension (systolic or diastolic blood pressure ≥140 or ≥90 mm Hg, respectively, or use of hypertensive drugs or report of hypertension medical history), diabetes (plasma glucose level ≥7.80 mmol/L or use of antidiabetic drugs or report of diabetes medical history), dyslipidemia (total cholesterol ≥6.2 mmol/L or use of lipid-lowering drugs or report dyslipidemia medical history), and history of cardiovascular diseases (self-reported history of myocardial infarction, angina pectoris, stroke, or use of vascular drugs).
Results of mixed linear models are expressed by linear regression coefficient (β) with their 95% confidence interval (CI). Results of mixed logistic models are expressed by odds ratio (OR) with their 95% CI. All statistical analyses were performed using SAS software version 9.1 (SAS Institute, Inc.).
RESULTS
Mean (±SD) age was 65.0 years (±3.0 years) for both genders. The other main characteristics of the study population at baseline have been previously described.13 Characteristics of the 702 subjects who completed the 9-year follow-up were compared with the 687 who did not (including 101 deaths) and are reported in Table 1. Subjects who did not complete the whole study were more likely to be men, current or former smokers, and persons with hypertension, a history of cardiovascular disease, or a higher BMI (Table 1). We also showed that cognitive performances at baseline were lower in subjects who did not complete the whole follow-up. Although plasma selenium level was associated with mortality,19 there was no association between loss to follow-up and plasma selenium level (Table 1).
;)
TABLE 1: Baseline Characteristics of Subjects According to 9-Year Follow-up Status
Classic Linear Regression Analyses
In the EVA study population, baseline mean plasma selenium level (±SD) was 1.09 μmol/L (±0.20 μmol/L). We observed a decrease of plasma selenium over the entire follow-up period. Means of selenium declined −0.055 (SD ± 0.20) μmol/L at 2 years and −0.096 (SD ± 0.21) μmol/L at the 9-year follow-up (Table 2). We noted an increase in cognitive performances means (Table 2). To assess the short- and long-term association between selenium and cognitive changes, we performed linear regression models between 2-year and 9-year selenium change and cognitive changes during 2, 4, 6, and 9-year follow-up (Table 3). Associations were adjusted for age, sex, education, and selenium level at baseline. We showed that 2-year selenium change was not associated with cognitive change during 2, 4, 6, or 9-year follow-up for any of the cognitive tests. Nine-year selenium change was associated with 9-year cognitive change for MMSE, but not for the other cognitive tests.
TABLE 2: Means and Percentile Distributions of Selenium and Cognitive Change From Baseline
TABLE 3: Linear Regressions Models for the Association Between 2-, 4-, 6-, and 9-Year Cognitive Change and 2- and 9-Year Selenium Change
Mixed Models
Factors Associated With Cognitive Change and Selenium Change
To take into account all cognitive changes during the follow-up and the within-subject variability, we performed mixed linear models. We used these models mainly to determine which factors were associated with cognitive change by restricting analyses to the MMSE test. Cognitive change was associated with time of follow-up (β = 0.09; 95% CI = 0.07 to 0.11) but not with age at inclusion. The interaction term between time of follow-up and age at inclusion was not significant, suggesting that the effect of time on cognitive decline was no more pronounced in older subjects than in younger ones for the narrow age range studied (60–70 years at baseline). Diabetes and hypertension were modestly associated with a higher decrease of cognitive performances (β = 0.31 [0.004 to 0.62] and β = 0.14 [−0.02 to 0.30], respectively). Sex, education, tobacco status, alcohol consumption, BMI, history of cardiovascular diseases, dyslipidemia, and baseline plasma selenium level were not associated with cognitive performances change during the follow-up (data not shown). During the follow-up, occurrence of cardiovascular events, as well as obesity, were associated with greater declines in plasma selenium (data not shown). No association was found between plasma selenium and other factors.
Association Between 2-Year Selenium Change and 9-Year Cognitive Change
Using crude mixed linear models, modeling cognitive change (2-, 4-, 6-, 8-, and 9-year) by the first 2-year selenium change, there was no association with any of the 4 cognitive tests (Table 4). These results were confirmed in the multivariate models (for the MMSE, β = 0.01 [−0.52 to 0.50]; for DSS, β = −0.27 [−1.6 to 1.1]; for TMTB, β = −8.4 [−20.7 to 3.9]; for FTT, β = −2.8 [−6.7 to 1.1]).
TABLE 4: Association Between Selenium Change During the First 2 Years and Cognitive Change Up to 9 Years of Follow-up: Results of Mixed Linear Models
To understand the relationship between cognitive decline and 2-year selenium change, we used a mixed logistic model with Gaussian random effects (Table 5). In these models, cognitive decline was considered as a dichotomous variable. After controlling for time and plasma selenium level at baseline, the short-term plasma selenium decrease was only weakly associated with cognitive decline. There results were not changed after controlling for potential confounding factors (Table 5).
TABLE 5: Associations Between 2-Year or 9-Year Plasma Selenium Change Decrease and 9-Year Cognitive Decline
Association Between Selenium Change and Cognitive Change During the 9 Years of Follow-up
For these analyses, we considered cognitive and selenium variables simultaneously measured at inclusion, 2 years, and 9 years. In these mixed linear models, cognitive performances (as continuous variables) were related to selenium levels and time of follow-up. Interaction terms between selenium and time of follow-up expressed the change of selenium according to cognitive change during follow-up (Table 6). In the crude analyses, selenium changes were associated with DSS and FTT but not with TMTB or MMSE. Analyses adjusted for time, sex, education, diabetes, hypertension, dyslipidemia, and history of cardiovascular diseases gave similar results (Table 6). However, for these models to be applicable, distributions of cognitive variables should be Gaussian; this condition was met for DSS and FTT but not for MMSE and TMTB. In sensitivity analysis, we applied the same models limited to subjects who had all 3 selenium measurements and cognitive evaluation (inclusion, 2- and 9-year). Results were quite similar although with lower power because analyses were carried out on 570 subjects instead of the initial 1371.
TABLE 6: Associations Between Plasma Selenium Changes and Cognitive Changes During the 9 Years of Follow-up: Results of Mixed Linear Models
Associations between selenium change and cognitive decline were studied by using a mixed logistic model with Gaussian random effect (Table 5). After controlling for time, sex, education, plasma selenium level at baseline, diabetes, hypertension, dyslipidemia, and history of cardiovascular diseases, we observed that probability of cognitive decline increased with the decrease of plasma selenium change over time. Among subjects who had a decrease of their plasma selenium levels, the higher the decrease of plasma selenium, the higher probability of cognitive decline. Among subjects who had an increase of their plasma selenium levels, the probability of cognitive decline was higher in subjects with the smallest selenium increase. Plasma selenium change was associated with MMSE decline at the 2-point cutoff (OR = 2.31; CI = 1.12 to 4.77) but not at the 3-point cutoff (1.41; 0.52 to 3.83). We observed an association for the DSS at the 25th cutoff (2.60; 1.22 to 5.57) and for TMTB at the 10th percentile (3.18; 1.14 to 8.88). For FTT, decline increased when selenium decreased for the 2 cutoffs we considered (for the 10th percentile, 2.40 [1.21 to 4.77] and for the 25th percentile, 4.33 [1.60 to 11.72]).
DISCUSSION
The EVA study is a longitudinal study of cognitive and vascular aging in an elderly general population. The study provided plasma selenium measurements at 3 points in time and assessments of cognitive functioning at 6 points in time. Declines in selenium were associated with cognitive decline as measured by 4 neuropsychologic tests—MMSE, DSS, FTT, and TMTB—after controlling for potential confounders. There were weaker associations between short-term selenium change and cognitive changes during the follow-up.
The EVA cohort is a free-living community-dwelling population with normal neuropsychologic test results at inclusion and a selenium status similar to that recently reported in French adults.20 All covariables for every model were updated at each waves of the study, except BMI and alcohol consumption for which we only took into account the baseline data. Alcohol consumption use may be associated with both cognitive decline and changes in selenium. However, correlations between alcohol consumption of the first 3 waves were very strong, and we made the assumption regarding these correlations and the change in tobacco status (which is linked to alcohol behavior) that, in the EVA cohort, alcohol consumption remained constant during the follow-up.21
In epidemiologic studies of cognitive decline, biologic status is generally considered as a static exposure associated with health changes. In this study, we extended our approach to dynamic exposure considering selenium changes during a time period rather than as a single measurement. However, this approach requires appropriate statistical analyses. Classic linear regressions are well suited to study selenium change and cognitive change between 2 given moments. For our study, however, this method is too restrictive when considering our initial objective, the wealth and complexity of our data, and sample selection. Mixed models allow us to simultaneously take into account measurements of the whole follow-up, and results are thus more powerful because sample selection is not as strong. We performed mixed linear or logistic models considering cognitive change as continuous or categorical. Because cognitive variables are not strictly normally distributed, conditions for computing mixed linear models are not optimal. The associations between selenium and cognitive change remained strong when cognitive change variables were dichotomized.
The various statistical models applied to our data offer both advantages and limits. However, we think the mixed logistic model is best suited to study long-term cognitive decline assessed by neuropsychologic tests in a healthy elderly population.
The study's obvious main limit is sample selection throughout follow-up. Subjects who did not complete the whole follow-up had the lowest cognitive performances at inclusion, and we could hypothesize that these subjects also had the highest cognitive decline during the follow-up. However, they did not differ with regard to selenium level at baseline or plasma selenium decrease for the first 2 years. Moreover, with analyses limited to subjects who had the 3 plasma selenium measurements and complete cognitive assessment (n = 570), results were quite similar although power was reduced. Finally, a relationship between plasma selenium and cognition change is found in subjects with weak cognitive decline. Sample selection may limit interpretation of our results but should not affect the relationship itself.
In previous works conducted on the same cohort at baseline13 and after the 4-year follow-up,11 we showed that, at baseline, plasma selenium and cognitive functions were positively associated. Baseline selenium plasma was lower in the subjects who develop cognitive decline as measured by 4-year decline of 3 points in MMSE scores. We therefore hypothesize that plasma selenium may be one of the factors associated with cognitive decline. This hypothesis is supported by the role of selenium in redox reactions and in thyroid hormone synthesis, both being involved in cognitive impairment.22,23 In addition, the brain contains large amounts of selenium and represents a target organ with respect to selenium supply and retention.5 So far, genes for at least 25 selenoproteins24 have been identified in the human genome and most of them are expressed in the brain (although their specific roles for normal brain functions and neurologic diseases have not been completely elucidated).
We found an association between plasma selenium change and 9-year cognitive decline only with 9-year selenium change, not 2-year change. These results are in agreement with the effect of antioxidant supplementation observed in some long-term studies12,25 and the lack of effect after a 6-month period in patients with mild Alzheimer disease.26 Nevertheless, these intervention studies were conducted with multivitamin and mineral supplements, thus making it impossible to identify a specific selenium effect. It is also possible that health changes, including poor cognitive function, may lead to dietary or behavioral changes that affect selenium levels.
Taken as a whole, our results suggest that plasma selenium decrease is associated with cognitive decline. The real importance of selenium in the brain and the capacity of the brain to manage selenium depletion is just beginning to be explored.5 Molecular biology has recently contributed to the recognition of selenium and selenium-dependent enzymes as modulators of brain function. Our results are in agreement with this approach even if a recent experimental work on knockout mice suggests that plasma selenium cannot reflect brain selenium status due to the maintenance of brain selenoprotein P synthesis.27
Our results, together with information on involvement of selenoproteins in brain functions, support possible relationships between selenium status and neuropsychologic functions in aging people. In this context, the preventive effect of selenium supplementation at a nutritional level needs to be evaluated with large-scale studies. This dynamic approach could shed new lights on the potential benefits of supplementation.
REFERENCES
1. Finkel T, Holbrook NJ. Oxidants, oxidative stress and the biology of ageing.
Nature. 2000;408:239–247.
2. Harman D. Aging; a theory based on free radical and radiation chemistry.
J Gerontol. 1956;2:298–300.
3. Chen J, Berry MJ. Selenium and selenoproteins in the brain and brain diseases.
J Neurochem. 2003;86:1–12.
4. Rayman MP. The importance of selenium to human health.
Lancet. 2000;356:233–241.
5. Schweizer U, Schomburg L, Savaskan NE. The neurobiology of selenium: lessons from transgenic mice.
J Nutr. 2004;134:707–710.
6. Mostert V. Selenoprotein P. properties, functions, and regulation.
Arch Biochem Biophys. 2000;376:433–438.
7. Berr C, Nicole A, Godin J, et al. Selenium and oxygen-metabolizing enzymes in elderly community residents: a pilot epidemiological study.
J Am Geriatr Soc. 1993;41:143–148.
8. Ducros V, Faure P, Ferry M, et al. The sizes of the exchangeable pools of selenium in elderly women and their relation to institutionalization.
Br J Nutr. 1997;78:379–396.
9. Olivieri O, Stanzial AM, Girelli D, et al. Selenium status, fatty acids, vitamins A and E, and aging: the Nove Study.
Am J Clin Nutr. 1994;60:510–517.
10. Savarino L, Granchi D, Ciapetti G, et al. Serum concentrations of zinc and selenium in elderly people: results in healthy nonagenarians/centenarians.
Exp Gerontol. 2001;36:327–339.
11. Berr C, Balansard B, Arnaud J, et al. Cognitive decline is associated with systemic oxidative stress: the EVA study. Etude du Vieillissement Arterielqq.
J Am Geriatr Soc. 2000;48:1285–1291.
12. Gray SL, Hanlon JT, Landerman LR, et al. Is antioxidant use protective of cognitive function in the community-dwelling elderly?
Am J Geriatr Pharmacother. 2003;1:3–10.
13. Berr C, Coudray C, Bonithon-Kopp C, et al. Demographic and cardiovascular risk factors in relation to antioxidant status; the EVA study.
Int J Vitam Nutr R. 1998;68:26–35.
14. Folstein M, Anthony JC, Parhad I, et al. The meaning of cognitive impairment in the elderly.
J Am Geriatr Soc. 1985;33:228–235.
15. Robins Wahlin TB, Backman L, Wahlin A, et al. Trail Making Test performance in a community-based sample of healthy very old adults: effects of age on completion time, but not on accuracy.
Arch Gerontol Geriatr. 1996;22:87–102.
16. Wechsler D.
The Wechsler Adult Intelligence Scale–Revised. New York; 1981.
17. Arnaud J, Prual A, Preziosi P, et al. Selenium determination in human milk in Niger: influence of maternal status.
J Trace Elem Electrolytes Health Dis. 1993;7:199–204.
18. Diggle P, Heagerty P, Liang K-Y, et al.
Analysis of Longitudinal Data. Oxford Stastistical Science Series, 2nd ed. Oxford University Press; 2002.
19. Akbaraly NT, Arnaud J, Hininger-Favier I, et al. Selenium and mortality in the elderly: results from the EVA study.
Clin Chem. 2005;51:2117–2123.
20. Galan P, Viteri FE, Bertrais S, et al. Serum concentrations of beta-carotene, vitamins C and E, zinc and selenium are influenced by sex, age, diet, smoking status, alcohol consumption and corpulence in a general French adult population.
Eur J Clin Nutr. 2005;59:1181–1190.
21. Dufouil C, Tzourio C, Brayne C, et al. Influence of apolipoprotein E genotype on the risk of cognitive deterioration in moderate drinkers and smokers.
Epidemiology. 2000;11:280–284.
22. Basun H, Forssell LG, Wetterberg L, et al. Metals and trace elements in plasma and cerebrospinal fluid in normal aging and Alzheimer's disease.
J Neural Transm Park Dis Dement Sect. 1991;3:231–258.
23. Smorgon C, Mari E, Atti AR, Dalla Nora E, et al. Trace elements and cognitive impairment: an elderly cohort study.
Arch Gerontol Geriatr Suppl. 2004;9:393–402.
24. Kryukov GV, Castellano S, Novoselov SV, et al. Characterization of mammalian selenoproteomes.
Science. 2003;300:1439–1443.
25. Mendelsohn AB, Belle SH, Stoehr GP, et al. Use of antioxidant supplements and its association with cognitive function in a rural elderly cohort: the MoVIES Project. Monongahela Valley Independent Elders Survey.
Am J Epidemiol. 1998;148:38–44.
26. Planas M, Conde M, Audivert S, et al. Micronutrient supplementation in mild Alzheimer disease patients.
Clin Nutr. 2004;23:265–272.
27. Schweizer U, Streckfuss F, Pelt P, et al. Hepatically derived selenoprotein P is a key factor for kidney but not for brain selenium supply.
Biochem J. 2005;386:221–226.
