Abstract
Objective: To investigate whether plasma interleukin-6 (IL-6) is cross-sectionally related to poorer cognitive function and whether a baseline plasma IL-6 measurement can predict risk for decline in cognitive function in longitudinal follow-up of a population-based sample of nondisabled elderly people.
Methods: A prospective cohort study of 779 high-functioning men and women aged 70 to 79 from the MacArthur Study of Successful Aging was conducted. Regression modeling was used to investigate whether baseline IL-6 levels (classified by tertiles) were associated with initial cognitive function and whether IL-6 levels predicted subsequent declines in cognitive function from 1988 to 1991 (2.5-year follow-up) and from 1988 to 1995 (7-year follow-up).
Results: Subjects in the highest tertile for plasma IL-6 were marginally more likely to exhibit poorer baseline cognitive function (i.e., scores below the median), independent of demographic status, social status, health and health behaviors, and other physiologic variables (odds ratio [OR] = 1.46; 95% CI: 0.97, 2.20). At 2.5 years, those in both the second tertile of IL-6 (OR = 2.21; 95% CI: 1.44, 3.42) and the third tertile (OR = 2.03; 95% CI: 1.30, 3.19) were at increased risk of cognitive decline even after adjusting for all confounders. At 7 years of follow-up, only those in the highest IL-6 tertile were significantly more likely to exhibit declines in cognition (OR = 1.90; 95% CI: 1.14, 3.18) after adjustment for all confounders.
Conclusions: The results suggest a relationship between elevated baseline plasma IL-6 and risk for subsequent decline in cognitive function. These findings are consistent with the hypothesized relationship between brain inflammation, as measured here by elevated plasma IL-6, and neuropathologic disorders.
Interleukin-6 (IL-6) has been linked to progressive inflammatory neuropathologic disorders including AD, viral and bacterial meningitis, AIDS dementia complex, and stroke.1 Blood levels of IL-6 are significantly elevated in individuals with AD,2 and polymorphisms in the IL-6 gene that decrease plasma IL-6 may be associated with a lower risk of developing AD.3 Evidence from transgenic mice4 that overexpress brain IL-6 also suggests that IL-6 could play a role in the neuropathophysiologic processes that lead to deficits in memory and learning. Based on this evidence, we hypothesized that plasma IL-6 might have a negative impact on cognitive functioning in humans and could be used as a biomarker for individuals at increased risk for cognitive decline. We hypothesized that high plasma levels of IL-6 would be associated with low cognitive function and that baseline plasma levels of IL-6 would be negatively associated with changes in cognitive function over both 2.5-year and 7-year follow-up time periods.
Methods.
Study population.
The MacArthur Study of Successful Aging was a longitudinal cohort study of high-functioning men and women aged 70 to 79, designed to identify factors associated with successful aging.5 Subjects aged 70 to 79 were selected in 1988 from the East Boston, MA, Durham, NC, and New Haven, CT, sites of the Established Populations for the Epidemiologic Study of the Elderly based on criteria designed to identify those in the top third of the population with respect to physical and cognitive function.6
Selection criteria for cognitive performance included scores of ≥6 correct on the nine-item Short Portable Mental Status Questionnaire7 and ability to remember three or more of six elements on a delayed recall of a short story. Selection criteria for physical performance included reporting no disability on a seven-item scale of activities of daily living and no more than one disability on eight items tapping gross mobility, able to hold a semitandem balance for at least 10 seconds, and able to stand from a seated position five times within 20 seconds without using their arms.
Of the 4,030 age-eligible subjects, 1,313 (32.6%) met selection criteria and 1,189 (90.8%) consented to be enrolled. Each of these subjects was asked to complete the MacArthur Battery, a 90-minute face-to-face interview designed to be given in the respondent’s home and covering detailed assessments of physical and cognitive functioning and performance, productive activities, social networks, social support, other psychosocial characteristics, and biomedical and health status measurements. As part of this battery, subjects were also asked to provide blood samples. Among those who responded to the survey, 955 (80.3%) agreed to have blood drawn and 880 (74% of cohort) provided sufficient blood to have plasma stored. Samples were processed and frozen at −80 °C within 4 hours. All surviving participants were reinterviewed at 2.5-year (1991) and 7-year (1995 to 1996) periods; the same assessments were completed at each follow-up.
Measurement of IL-6.
Levels of IL-6 were determined from stored baseline plasma samples (n = 880). Samples were sent to the University of Vermont at Burlington in 1996 for measurement of IL-6 by ELISA (High Sensitivity Quantikine Kit; R&D Systems, Minneapolis, MN). The detectable limit for IL-6 was 0.10 pg/mL with an interassay coefficient of variation of 7%. Values were measured in duplicate, with averages being reported. The average plasma IL-6 level was 4.35 pg/mL (SD = 7.08). For the analyses reported here, a categorical measure representing tertiles of IL-6 was used to evaluate possible nonlinear associations between levels of IL-6 and cognitive function. The tertiles were derived based on the distribution of IL-6 in the cohort. The highest tertile included subjects with IL-6 values of ≥3.8 pg/mL, the bottom tertile included those with values <2.13 pg/mL, and the middle tertile includes those with values between 2.13 and 3.8 pg/mL. The distribution of this cohort was similar to other cohorts of older adults.8
Measurement of cognitive function.
Cognitive function was assessed at baseline and follow-up based on five tasks that measured naming, verbal memory, spatial recognition, abstraction, and spatial ability. The first task was the evaluation of confrontation naming by means of the Boston Naming Test, a test of language.9 The second task was a delayed verbal memory test based on the incidental recall of naming items from the Boston Naming Test.10 The third task evaluated spatial memory by means of the Delayed Recognition Span Test.11 The fourth test utilized four items from the Similarities Subtest of the Wechsler Adult Intelligence Scale–Revised,12 evaluating the ability of subjects to form abstract concepts. The final cognitive task was the copying of geometric figures, evaluating the subjects’ ability to perceive and reproduce spatial relationships.13 A summary measure of total cognitive function was developed; this score is the sum of all subtest scores and has a range of 0 to 89, where 89 represents the highest cognitive function possible.14
The primary outcomes examined in these analyses reflect dichotomous measures that focus explicitly on those who exhibited either poor versus better cognitive function at baseline or declines in cognitive performance during the follow-up. In all cases, the criterion cut-points used to define the outcome groups were chosen a priori. Our choice of dichotomous outcomes (rather than continuous cognitive scores) was motivated by our hypothesis that higher IL-6 values would be associated specifically with the risk of showing “poor/low” cognitive performance at baseline and, most importantly, would be associated with risks for declines over time (versus change per se). Cognitive function at baseline was examined by classifying subjects into “low” versus “high” cognitive function groups based on a median (50th percentile) split. The average total cognitive score in 1988 was 52.93 (SD = 9.91). Change in cognitive function was calculated for each follow-up (i.e., 2.5 and 7 years) by subtracting baseline (1988) total cognitive function scores from comparable scores from the 2.5-year (1991) and 7-year (1995) follow-ups, resulting in negative change scores for those whose cognitive performance had declined. The average change in cognitive function was 0.35 (SD = 7.07) at the 2.5-year follow-up and −3.86 (SD = 8.82) at the 7-year follow-up, though the ranges of change scores were reasonably large for both intervals including both negative (declines) and more positive changes: −37 to 20 for the 2.5-year interval and −49 to 21 for the 7-year interval. Based on these change scores, we created dichotomous outcomes for both the 2.5- and the 7-year follow-ups that defined decline in cognitive performance as having a change score that fell into the bottom tertile of the distribution of change scores for that follow-up. This approach was taken to permit an explicit focus on those individuals who exhibited the largest (and perhaps most functionally important) changes. The analyses of these dichotomous outcomes examine risk for “decline” versus “no decline” where the latter group included those whose changes were either sufficiently small as to reflect minimal change and those whose change was positive. As in previous analyses from our group,15 the tertile cut-point was selected a priori in order to identify a group with the strongest evidence of decline (i.e., largest decline scores) and simultaneously create a group of “decliners” of sufficient size to provide stable estimates of risk. For the 2.5-year interval, subjects who declined by ≥3 points were classified as having declined in that interval. For the 7-year interval, those who declined by ≥7 points were classified as having declined. For each interval, those classified as having declined were compared with those who did not meet criteria for decline. Previous analyses from the MacArthur Study have indicated that examination of such dichotomous measures, focusing on those in the lowest tertile of the change score distribution, can be particularly effective in identifying factors associated with increased risk for decline as compared with those who show either no change or even improvements.15 Information on results from parallel analyses based on the continuous cognitive scores (and changes in those scores from baseline to each follow-up assessment) is also provided.
Covariates.
Potential confounders considered included standard sociodemographic and health status measures that previous research suggested were related to cognitive functioning16 and IL-6.17 Demographic and social status variables examined as potential confounders included 1) age, measured as the respondent’s age at the time of the baseline interview (range 70 to 79 years); 2) reported annual household income, measured in $10,000 increments (<$2,000 to $50,000 or more); 3) education, measured as years completed (range 0 to 17 years); 4) marital status, classified as “currently married” versus “not currently married”; and 5) ethnicity, classified as “nonwhite” (predominantly African American) versus “white.”
Health behaviors examined included 1) smoking status, categorized as “ever” versus “never” smokers on the basis of self-report; 2) alcohol intake, assessed by self-report and characterized as “any consumed in the last month” versus “none”; and 3) physical activity, assessed by a summary measure, adapted from the Yale Physical Activity Survey,18 focusing on frequency and intensity levels of current leisure- and work-related activity.
Measures of health status included 1) body mass index (BMI; weight [kg]/height [m2]); 2) hemoglobin A1c (HBA1c), assayed by affinity chromatography methods19; 3) total cholesterol, measured by standard enzymatic methods; and 4) systolic and diastolic blood pressure, reported as the average of three seated readings. The presence of six major chronic conditions was assessed based on self-reports of doctor-diagnosed diabetes mellitus, cancer, myocardial infarction, stroke, hip fracture, and bone fracture other than hip.
Statistical analysis.
All analyses were performed using the SAS System for Windows (version 6.12; Cary, NC). Associations between IL-6 and baseline cognitive function as well as associations with cognitive decline were assessed via linear regression analyses for continuous outcomes and via logistic regression analyses for dichotomous outcomes. Risks for poor cognitive function at baseline or decline in cognitive function during follow-up were assessed comparing those in the middle and top tertile of IL-6 values with those in the bottom tertile of IL-6 values (reference group).
Analysis-of-variance assessments were used to evaluate potential continuous confounders, and cross-tabulation analyses were used to evaluate categorical confounders, comparing means and/or proportions across tertiles of IL-6 and dichotomous cognitive outcomes. Covariates included in the final multivariate models were those found to be associated with cognitive function at p < 0.10; they included age, race, sex, yearly income, education level, alcohol intake, activity level, BMI, self-reported history of cancer or diabetes, and HBA1c levels. Baseline cognitive performance scores were also included in all longitudinal analyses. Because of the variable time interval between baseline and follow-up interviews for cohort members, we also considered a measure of the actual time lapse between baseline and follow-up (e.g., ranges of 22 to 46 months [1.8 to 3.8 years] for the 1991 follow-up and 60 to 104 months [5 to 8.8 years] for the 1995 follow-up). However, these measures were found to be unrelated to measures of cognitive change, and their inclusion in multivariate models did not affect estimated IL-6 effects. Thus, measures of time lapsed between assessments were not retained in the final models.
Results.
Of the original 1,189 individuals enrolled in the MacArthur Study, 779 (66%) had complete data for the analyses of baseline cognitive functioning and IL-6. Not unexpectedly, those with missing data tended to be the less healthy and more disadvantaged cohort members. The most common reason for exclusion was missing data on IL-6 (n = 309; owing to refusal to provide a blood sample). Compared with those with complete baseline data, individuals excluded because of missing IL-6 data tended to be women and nonwhite and to report lower income and less alcohol use in the last month; they also had lower cognitive function at baseline (table 1). As shown in table 1, these same characteristics were more generally associated with the presence of missing data on other covariates as well. By the time of the 2.5-year follow-up, 71 subjects (6% of original cohort) had died and 434 (36.5%) had incomplete data on one or more of the measures included in our analyses. As shown in table 2, those who had died by the time of the 2.5-year follow-up were more likely to be men; they were also more likely to have had higher IL-6 values and tended to have lower cognitive function and higher glycosylated hemoglobin at baseline. Similar patterns were seen for those who had died by the time of the 7-year follow-up (n = 273; 23%), though this latter group also exhibited lower baseline activity scores and a history of having been a smoker (table 3). Those known to be alive but excluded from the longitudinal analyses owing to missing data were similar to those excluded owing to mortality in that they too had lower baseline cognitive function. However, this group was more likely to be women and to report lower annual household income and was less likely to report any alcohol consumption in the last month (see table 3). In general, subjects included in the analyses tended to be those who were somewhat more advantaged at baseline with respect to health and functioning as well as health behaviors and socioeconomic status. Those included in the analyses had an average age of approximately 74 years. They reported an average of 10.7 years of education and an average yearly income between $16,000 and $17,000. The sample was 18% nonwhite and 45% men.
Table 1 Descriptive statistics for subjects included in analyses versus those excluded owing to missing data at baseline
Table 2 Descriptive statistics for subjects included in analyses versus those excluded owing to missing data at 1991 follow-up
Table 3 Descriptive statistics for subjects included in analyses versus those excluded owing to missing data at 1995 follow-up
IL-6 and baseline cognitive function.
Figure 1 presents box plots of the distributions of baseline cognitive performance scores for each of the three IL-6 tertile groups. Consistent with the hypothesis that higher IL-6 level would be associated with poorer cognitive performance, the distributions of these scores show a shift toward somewhat lower median scores (solid lines in middle of each box indicating 50th percentile point) as one moves from the lowest to the middle and highest tertiles of Il-6; the 75th and 25th percentiles (i.e., top and bottom of each box) also show a trend toward lower scores moving from the lowest to the highest IL-6 group. The horizontal line crossing all three of the box plots at the value of 53 reflects the cut-point used to define “poor” cognitive function (i.e., median for the cohort as a whole). Again, moving from the lowest to the highest IL-6 group, the box plots reveal that a growing percentage of individuals have scores that fall below this criterion value (i.e., 38% in the lowest tertile of IL-6, 48% in the middle tertile, and 57% in the highest IL-6 tertile; χ2 = 16.7, p < 0.001).
Figure 1. Box plots illustrating the distributions of baseline cognitive scores for each of the three interleukin-6 tertile groups (1 = bottom tertile, 2 = middle tertile, 3 = top tertile). Horizontal line at value 53 indicates the median cut-point used to define the “poor/good” cognitive outcome for logistic regression analyses.
Though these trends were consistent with the hypothesis that higher IL-6 would be associated with poorer cognitive performance, logistic regression analyses of poor versus better cognitive performance based on the median split revealed no significant differences in the likelihood of poor versus better cognitive performance across tertiles of IL-6 after multivariate adjustments.
Results for linear regression analyses of the continuous data revealed parallel nonsignificant differences.
IL-6 and decline in cognitive function risk.
Figures 2 and 3⇓ present box plots of the distributions of change scores for cognitive performance from 1988 to 1991 (2.5 years) and from 1988 to 1995 (7 years) for each of the three IL-6 groups. As can be seen in figure 2, the 2.5-year change distributions for the middle and highest IL-6 tertiles show a somewhat greater proportion of individuals with larger declines as shown by the somewhat lower median values (solid line in middle of each box) and the lower values for the 25th percentile of the distribution (i.e., solid line at bottom of each box). Figure 3 illustrates a similar pattern for the 7-year change distributions, showing somewhat lower median (50th percentile) values in the middle and highest IL-6 tertile groups and a more obvious shift toward greater decline scores for the highest IL-6 group such that their median and 25th percentile values are the lowest of any group.
Figure 2. Box plots illustrating the distributions of 2.5-year change scores for cognitive performance for each of the three interleukin-6 tertile groups (1 = bottom tertile, 2 = middle tertile, 3 = top tertile). Horizontal line at value −3 indicates the cut-point used to define decline versus no decline outcome for logistic regression analyses.
Figure 3. Box plots illustrating the distributions of 7-year change scores for cognitive performance for each of the three interleukin-6 tertile groups (1 = bottom tertile, 2 = middle tertile, 3 = top tertile). Horizontal line at value −7 indicates the cut-point used to define decline versus no decline outcome for logistic regression analyses.
The horizontal lines that cross all three box plots in figures 2 and 3⇑ show the cut-points used to define our outcome classifications of “decline” (i.e., having a change score in the bottom tertile of the overall change distribution for the full sample: −3 points at 2.5 years and −7 points at the 7-year follow-up). At each follow-up, greater proportions of those in the middle and highest IL-6 tertiles had decline scores that fell into this bottom tertile. At 2.5 years, the figures were 25% of the lowest IL-6 tertile as compared with 39.4% of the middle IL-6 tertile (p = 0.002) and 37.5% of the top tertile (p = 0.006). Comparable figures for the 7-year follow-up were 28% of the lowest IL-6 tertile as compared with 32% for the middle IL-6 tertile (p = 0.45) and 40% of the highest IL-6 tertile (p = 0.01).
Multivariate logistic analyses of the relationship between baseline IL-6 and risk of cognitive decline during the initial 2.5-year follow-up period (i.e., defined as a decline of ≥3 points) revealed that subjects in both the highest and the middle tertiles of plasma IL-6 were significantly more likely to experience declines in cognitive functioning. For the highest tertile group, the fully adjusted odds ratio (OR) was 2.03 (95% CI: 1.30, 3.19); for the middle tertile group, the fully adjusted OR was 2.21 (95% CI: 1.44, 3.42). Over the 7-year follow-up, those in the highest tertile of baseline IL-6 continued to experience significantly greater risk for decline in cognitive function (defined has a loss of ≥7 points) (fully adjusted OR = 1.90; 95% CI: 1.14, 3.18). Subjects in the middle tertile did not differ significantly from those in the first tertile (fully adjusted OR = 1.18; 95%CI: 0.71, 1.95).
Similar multivariate linear regression analyses for continuous measures of change revealed parallel negative, though largely nonsignificant, associations between baseline IL-6 and subsequent changes in cognitive performance measured in terms of raw change scores. The only significant effect was for change in cognitive performance between baseline and 1991 (i.e., the initial 2.5-year interval), with those in the middle tertile of IL-6 showing greater declines (b = −1.46, p = 0.02); the coefficient for those in the top tertile of IL-6 was also negative though nonsignificant (b = −0.50, p = 0.4). Analyses of changes in cognitive performance over the 7-year interval also revealed negative though nonsignificant associations (middle versus bottom tertile IL-6, b = −0.22, p = 0.8; top versus bottom tertile IL-6, b = −1.02, p = 0.3).
Discussion.
High levels of IL-6 were associated with poorer baseline cognitive function and predicted increased risks for cognitive decline at the 2.5- and 7-year follow-ups for a cohort of older men and women who were relatively high functioning at the inception of the study. These effects were independent of demographic characteristics, social status, as well as measures of health status and health behaviors.
There are several possible reasons why parallel, though nonsignificant, trends were seen for the analyses of IL-6 in relation to the continuous measures of cognitive change (as compared with the significant findings for the dichotomous measures of “declines”). IL-6 levels may be related to risks for change in cognitive performance across the full range of cognitive change scores (as suggested by the negative coefficients in our linear models), but our sample size was not sufficient to detect this effect in the face of possible confounding of measured change with measurement error. To the extent that such measurement error is more likely reflected in smaller changes from one time to another, the logistic analyses focusing on only the largest measured declines may be less subject to such effects and may therefore more clearly reflect underlying relationships between IL-6 and risk of cognitive decline. Alternatively, it is possible that higher IL-6 values are, in fact, not related to cognitive declines in a linear fashion—being associated more strongly perhaps with larger (and perhaps more functionally important) declines but not related to smaller changes in performance. Further research is needed to elucidate more clearly the relationship between IL-6 and cognitive aging.
Our findings linking IL-6 to risk for cognitive decline are consistent with hypotheses suggesting negative impacts of inflammatory cytokines on long-term potentiation.20,21⇓ Some of the same pathophysiologic processes involving IL-6 that have been proposed with respect to cognitive diseases such as AD could also underlie the relationships observed in our data between IL-6 and cognitive decline. One current theory postulates a self-amplifying pathophysiologic cascade wherein β-amyloid protein-induced secretion of IL-6 is augmented (along with nitric oxide and reactive oxygen species), resulting in neuronal injury.22 Other AD models also support this idea, including the complement-based/membrane attack complex23 and IL-1 theories.24
Though the data reported here reflect plasma levels of IL-6, several lines of evidence suggest that plasma levels may be at least partially of brain-borne origin. At the neurocellular level, IL-6 and IL-6 mRNA are found in many areas of the human brain, including the pyramidal and granular neurons of the hippocampus,25 and there is evidence that cytokines can cross the blood–brain barrier,26 suggesting the possibility that higher circulating IL-6 may also contribute to cognitive risks if higher circulating levels contribute to higher brain levels through plasma to brain passage of IL-6.
There are several limitations of the data presented here that should be acknowledged. The data reflect a cohort selected at baseline to represent high-functioning older adults, and the generalizability of our findings to the broader population of older adults remains to be seen. Indeed, it is possible that an even stronger association might be seen in a more broadly representative population with possibly wider ranges of both IL-6 and cognitive function scores.
The less consistent association with cognitive risk seen for those in the middle tertile of IL-6 (i.e., significant for the 2.5-year follow-up but not at 7 years) was not predicted. It is possible that this reflects the fact that IL-6 classification was based on only a single baseline measure, which may have become a less accurate marker of ongoing IL-6 exposure over time. To the extent that fluctuations in IL-6 levels during follow-up resulted in declines from the middle to the lower tertile, this could contribute to weaker associations between baseline classification and longer-term cognitive risks. The consistently elevated risks for cognitive decline seen for those initially classified into the top tertile may reflect the fact that their IL-6 levels, even if they likewise decline to somewhat lower levels, may still have left individuals at levels associated with increased cognitive risk.
Despite these caveats, the data reported here are noteworthy in that they show longitudinal evidence that high plasma IL-6 levels are associated with increased risks for decline in cognitive function over time in a group of older individuals characterized by relatively high initial cognitive function. Further research is needed to elucidate mechanisms of neurodegeneration and inflammation that likely contribute to these observed associations so that we can better identify those at greatest risk for cognitive decline and develop therapies to prevent, or at least delay, the onset of such declines.
Acknowledgments
Supported by the MacArthur Research Network on Successful Aging and the MacArthur Research Network on SES and Health through grants from the John D. and Catherine T. MacArthur Foundation, the John A. Hartford Foundation/American Federation for Aging Research Medical Student Geriatrics Scholars Program, the Epidemiology, Demography, and Biometry Office and the Behavioral and Social Research Program of the National Institute on Aging (grants AG-17056 and AG-17265), and the Alzheimer’s Association.
Acknowledgment
The authors thank Dr. Russell Tracy and his colleagues at the University of Vermont who performed the assays for IL-6.
- Received November 2, 2000.
- Accepted April 5, 2002.
References
Disputes & Debates: Rapid online correspondence
NOTE: All authors' disclosures must be entered and current in our database before comments can be posted. Enter and update disclosures at http://submit.neurology.org. Exception: replies to comments concerning an article you originally authored do not require updated disclosures.
- Stay timely. Submit only on articles published within the last 8 weeks.
- Do not be redundant. Read any comments already posted on the article prior to submission.
- 200 words maximum.
- 5 references maximum. Reference 1 must be the article on which you are commenting.
- 5 authors maximum. Exception: replies can include all original authors of the article.
- Submitted comments are subject to editing and editor review prior to posting.