Ten-Year Change in Plasma Amyloid β Levels and Late-Life Cognitive Decline

Background  Plasma levels of amyloid β peptide (Aβ) are potential biomarkers of early cognitive impairment and decline and of Alzheimer disease risk.

Objective  To relate midlife plasma Aβ measures and 10-year change in plasma Aβ measures since midlife to late-life cognitive decline.

Design  Prospective study of a population-based sample.

Setting  Academic research.

Participants  Plasma Aβ40 and Aβ42 levels were measured in 481 Nurses' Health Study participants in late midlife (mean age, 63.6 years) and again 10 years later (mean age, 74.6 years). Cognitive testing also began 10 years after the initial blood draw. Participants completed 3 repeated telephone-based assessments (mean span, 4.1 years). Multivariable linear mixed-effects models were used to estimate relations of midlife plasma Aβ40 to Aβ42 ratios and Aβ42 levels to late-life cognitive decline, as well as relations of 10-year change in Aβ40 to Aβ42 ratios and Aβ42 levels to cognitive decline.

Main Outcome Measures  The 3 primary outcomes were the Telephone Interview for Cognitive Status (TICS) findings, a global score averaging the results of all tests (TICS, immediate and delayed verbal recall, category fluency, and attention), and a verbal memory score averaging the results of 4 tests of verbal recall.

Results  Higher midlife plasma Aβ40 to Aβ42 ratios were associated with worse late-life decline on the global score (P = .04 for trend). Furthermore, increase in Aβ40 to Aβ42 ratios since midlife predicted greater decline in the global score (P = .03 for trend) and in the TICS (P = .02 for trend). There was no association of cognitive decline with midlife plasma Aβ42 levels alone or with change in Aβ42 levels since midlife.

Conclusion  In this large community-dwelling sample, higher plasma Aβ40 to Aβ42 ratios in late midlife and increases in Aβ40 to Aβ42 ratios 10 years later were significantly associated with greater decline in global cognition at late life.

Alzheimer disease (AD) is generally diagnosed at old ages; however, pathologic changes begin many years earlier. Therefore, identification of easily measurable biomarkers at midlife that can predict dementia is a priority for AD prevention.¹ Moreover, because subtle cognitive decline is associated with higher risk of subsequent AD,² biomarkers of decline in “young old” persons may be particularly valuable. Plasma levels of circulating amyloid β peptide (Aβ) that end at amino acid 40 (Aβ40) or 42 (Aβ42) have increasingly been explored as such biomarkers.

Specifically, decreases in plasma Aβ may reflect decline of soluble Aβ in the periphery, as it accumulates in insoluble brain plaques in patients with AD.³ However, data on plasma Aβ have been mixed about prediction of dementia.^3-9 Investigations have found associations between absolute plasma Aβ40 or Aβ42 levels and dementia, but directions of these associations have varied. Results have been more consistent with consideration of the plasma Aβ40 to Aβ42 ratios at older ages.^5,7,8,10 For example, Graff-Radford et al⁷ observed that healthy elderly persons with Aβ42 to Aβ40 ratios in the lower quartiles had a higher relative risk of the development of mild cognitive impairment and AD (P = .04) (ie, higher Aβ40 to Aβ42 ratios associate with higher risk). The first large study⁵ of change in plasma Aβ ratios demonstrated that decreases (assessed over 4.5 years) in Aβ42 to Aβ40 ratios (ie, increases in Aβ40 to Aβ42 ratios) at older ages predicted higher rates of incident AD.

However, there have been few large-scale prospective studies^7,11 that relate plasma Aβ to the outcome of cognitive decline. Furthermore, previous work focused on participants who were older.^7,11,12 Consequently, there is limited knowledge about whether plasma Aβ levels reflect existing pathology or can predict decline in younger persons. Furthermore, change in plasma Aβ levels from midlife to late life may be relevant in the identification of trajectories of cognitive decline. Therefore, we measured plasma Aβ40 and Aβ42 levels at late midlife and 10 years later in 500 women from a population-based sample. We related midlife plasma Aβ40 to Aβ42 ratios and Aβ42 levels to cognitive decline, as well as change in Aβ40 to Aβ42 ratios and Aβ42 levels over the subsequent 10 years to cognitive decline.

The Nurses' Health Study¹³ included 121 700 female US registered nurses aged 30 to 55 years when the study began in 1976. Since then, participants have completed biennial mailed questionnaires that update health and lifestyle information. Between 1989 and 1990, blood samples were provided by 32 826 women, and 18 672 of these provided blood samples again from 1999 to 2001. Characteristics of those who gave blood twice were similar to the entire blood cohort. For example, the mean alcohol intake was 5.3 g/d; the prevalence of past smoking was 40% among women who gave blood twice and 39.7% among the entire blood cohort; the mean body mass index (BMI [calculated as weight in kilograms divided by height in meters squared]) was 25.2 among women who gave blood twice and 25.4 among the entire blood cohort.

In addition, from 1995 to 2001, all Nurses' Health Study participants aged 70 years or older without diagnosed stroke were invited to participate in a telephone-based study of cognition, and 19 395 women (93.3% of those eligible) completed an initial assessment. Two additional assessments were performed approximately 2 years apart. Follow-up exceeds 90% for the cognitive study.

Venous whole-blood samples were obtained in heparinized tubes, shipped on ice to a central facility, processed (centrifuged and aliquotted as plasma, buffy coat, and red blood cells), and then stored at −130°C. Most samples arrived within 26 hours of being drawn; precautions were taken to prevent thawing of specimens during storage.

By means of stored blood samples, we assayed plasma Aβ40 and Aβ42 by sandwich enzyme-linked immunosorbent assay (ELISA). Plates (384-well MaxiSorp; Nunc, Naperville, Connecticut) were coated with capture antibodies (2G3 for Aβ40 and 21F12 for Aβ42) in phosphate-buffered saline (PBS) and incubated for 4 hours at room temperature (RT) and then blocked overnight at 4°C. Plates were washed 3 times with PBS–polysorbate 20 (Tween 20; ICI Americas Inc, Wilmington, Delaware), and samples were loaded into the wells and incubated with detector antibody (biotinylated 266 to the midregion of Aβ) for 2 hours at RT. Samples were then reincubated in a solution of this detector antibody for 2 hours at RT. Finally, samples were incubated with streptavidin–alkaline phosphatase (Promega, Madison, Wisconsin) in PBS for 1 hour at RT and washed 3 times with a Tris-buffered saline solution. The signal was amplified (AttoPhos, Promega) and measured with a fluorescent plate reader (Victor2; PerkinElmer, Boston, Massachusetts).

Because we measured Aβ collected at 2 time points and were concerned that between-plate variation might interfere with the assessment of within-subject change in Aβ levels over time, the samples were paired on a single plate and measured simultaneously. All sample pairs, such as blinded quality control (QC) pairs, were distributed randomly across plates.

We assessed the plasma Aβ assays before and during the study. Seven plates were used for the ELISAs. Blinded duplicate QC pairs were included on each plate. Overall coefficients of variation (CVs) for the 100 QC samples were high (45.0% for Aβ40 and 34.7% for Aβ42). However, median within-pair CVs (across 50 QC pairs) were low (7.1% for Aβ40 and 7.6% for Aβ42); therefore, we had excellent ability to consider within-person change in Aβ levels. In addition, high between-plate variability seemed to explain the high overall CVs. For example, after separation of between-plate and within-plate variability, the mean within-plate CV was 10.2%. High between-plate variability but low within-plate variability have been reported previously in plasma Aβ ELISAs and have led to the recommendation of comparison of samples loaded on the same plate.¹⁴

Earlier work established the stability of Aβ40 and Aβ42 levels in specimens with varying processing times.¹⁵ Although processing delays for Nurses' Health Study blood samples are typically no longer than 24 hours, intraclass correlations of greater than 0.95 for Aβ40 and Aβ42 values were demonstrated after processing delays of up to 48 hours.¹⁵ Furthermore, we addressed the reliability of Aβ measures in long-archived plasma samples. The median (range) CVs for replicates of 12 plasma samples that had been in frozen storage (−130°C) for a mean of 17 years were 9.7% (0.2%-16.1%) for Aβ40 and 14.8% (9.9%-17.3%) for Aβ42. Therefore, we established that Nurses' Health Study blood collection and storage conditions were adequate for the yielding of valid results.

Testing included the following: Telephone Interview for Cognitive Status¹⁶ (TICS), a test of general cognition similar to the Mini-Mental State Examination¹⁷; immediate and delayed recall trials of the East Boston Memory Test¹⁸; category fluency (the naming of as many different animals as possible during 1 minute); delayed recall of the TICS 10-word list; and digit span backward (repeating backward increasingly long series of digits). Reliability and validity of this method have been established.¹⁹ Test-retest (r = 0.7, P < .001) reliability was high. In 61 highly educated women, the global score from the telephone battery correlated strongly (r = 0.81) with a global score from 21 in-person neuropsychological tests. Finally, in a small clinical validation study, poor performance on our telephone battery was significantly associated with an 8-fold risk of dementia diagnosis.

General cognition and verbal memory were the primary outcomes; verbal memory, in particular, is a strong predictor of eventual AD.²⁰ To assess general cognition, we considered the TICS findings, as well as a global score calculated by averaging of the z scores of all 6 tests. The verbal memory score was calculated by averaging of the z scores of the immediate and delayed recalls of the East Boston Memory Test and the TICS 10-word list.^21,22 Global and verbal memory scores were only calculated for those who completed all component tests.

To maximize efficiency, we obtained the present study sample (n = 500) by first oversampling participants from the top and bottom 20% of the distributions of cognitive decline in our population and then by selecting random participants from the remainder of the distribution. This sampling strategy ensured adequate power to detect differences in cognitive change across plasma Aβ groups without the measurement of Aβ levels in the entire cohort. Finally, we excluded 19 women from analysis because their Aβ40 or Aβ42 levels were below the limit of detection. Therefore, the final sample included 481 women. Health and lifestyle characteristics were similar between this sample (mean alcohol intake, 5.3 g/d; mean activity, 17.2 metabolic equivalent tasks [MET]/wk; and mean BMI, 25.0) and all participants who returned blood samples (mean alcohol intake, 5.3 g/d; mean activity, 17.4 MET/wk; and mean BMI, 25.2). This study was approved by the institutional review board of Brigham and Women's Hospital.

Because of the between-plate variation already discussed, we created batch-specific z scores of midlife plasma Aβ40 to Aβ42 ratios and Aβ42 levels. Therefore, the unit of analysis was a batch-specific difference of 1 SD. For analyses of change in Aβ measures since midlife, we calculated the percentage change in each. Midlife and late-life samples for each nurse were always assayed on the same plate, and within-pair CVs were low (as already discussed); therefore, batch correction was unnecessary in analyses that used percentage change. We calculated relations of percentage change in Aβ40 to Aβ42 ratios and Aβ42 levels to cognitive decline by means of a 1-SD increment for each. To address possible nonlinear relations (eg, threshold effects), we also performed categorical analyses by means of quartiles of percentage change in Aβ measures.

We used linear mixed-effects models²³ to examine relations of midlife Aβ40 to Aβ42 ratios and Aβ42 levels, as well as change in Aβ40 to Aβ42 ratios and Aβ42 levels, to cognitive decline across 3 repeated assessments. The model included the following fixed effects: time since initial cognitive assessment (years), age at initial cognitive assessment, education (associate, bachelor, masters, or doctoral degree), Aβ40 to Aβ42 ratio or Aβ42 level, and interaction terms of time-×-age and time-×–Aβ40 to Aβ42 ratio (or time-×–Aβ42 level). The model also included the following potential confounders: BMI, history of hypertension (yes or no), history of dyslipidemia (yes or no), history of heart disease (yes or no: any history of myocardial infarction, chronic angina, angiography-confirmed coronary disease, coronary angioplasty or coronary artery bypass grafting), cigarette smoking (current, past, or never), postmenopausal hormone use (current, past, or never), physical activity (METs/wk), and alcohol use (in grams per day), all of which were determined as of blood draw, as well as history of depression (yes or no: determined either by meeting the validated cutoff on the Medical Outcomes Short-Form 36 Mental Health index²⁴ or regular antidepressant use), which was ascertained as of cognitive assessment. Added to these fixed effects were 2 person-specific random effects: baseline cognitive level (random intercept) and rate of change (random slope).

Because many participants completed initial cognitive testing just before their second blood draw, in a secondary analysis we evaluated cognitive change between the second and third assessments (ie, change after the second blood draw). This procedure guaranteed a strict prospective analysis, although most participants (60.5%) provided their second blood sample no later than 12 months after their initial cognitive assessment, and 85.2% provided the sample within 18 months. We used linear regression analysis to estimate the mean differences in cognitive change (mean interval between the second and third assessments, 2.4 years) associated with intraindividual change in plasma Aβ40 to Aβ42 ratios and Aβ42 levels. However, results were identical to analyses that included all assessments; therefore, we only present the data for all 3 assessments.

We conducted a key secondary analysis to address concerns that relations of Aβ measures to vascular factors or subclinical vascular disease could explain associations between Aβ and cognitive decline. Rather than considering confounders as of the first blood draw, we considered history of vascular factors at any time as of the initial cognitive testing. Vascular factors included in the models were smoking, BMI, physical activity, hypertension, dyslipidemia, diabetes mellitus, and heart disease. Furthermore, we adjusted for history of transient ischemic attack during follow-up and removed from analysis participants who developed stroke during the course of cognitive testing or who underwent carotid endarterectomy at any point (n = 12). Although the influence of impaired renal function on plasma Aβ is also of concern,⁴ there were no women in this sample with a history of renal disease. Finally, although addressing late-life predictors was not the primary objective of this study, we examined late-life plasma Aβ measures in relation to cognitive outcomes in separate analyses. All statistical analyses were performed by means of commercially available software (SAS version 9.1; SAS Institute, Cary, North Carolina).

Table 1 summarizes participant characteristics at midlife by quartile of Aβ40 to Aβ42 ratio. Overall, characteristics were similar across quartiles. However, there was a trend toward increased prevalence of depression with increasing Aβ40 to Aβ42 ratios. Women with the lowest Aβ40 to Aβ42 ratios tended to have lower prevalence of heart disease and current smoking, and there was a suggestion of higher physical activity in this group. Women in the lowest quartile also seemed to have higher prevalence of current postmenopausal hormone use.

Distributions of plasma Aβ measures at midlife and late life are summarized in Table 2. Overall, the range of late-life Aβ42 levels seemed marginally lower than midlife levels; late-life Aβ40 to Aβ42 ratios seemed higher than midlife ratios.

Age- and education-adjusted models showed significantly faster rates of decline in the global score associated with higher midlife Aβ40 to Aβ42 ratios, with borderline findings in the TICS (Table 3). Estimates from multivariable-adjusted models were identical. For example, each 1-SD increment in midlife Aβ40 to Aβ42 ratios was associated with a −0.02 U/y decrease in the global score. To help interpret this estimate, we compared it with the effect of age. In our population, each additional year of age was associated with a –0.01 U/y decline in the global score; therefore, each 1-SD increment in midlife Aβ40 to Aβ42 ratios was cognitively equivalent to 2 years of aging. In analyses of Aβ42 levels alone, there were no associations between midlife plasma Aβ42 levels and any outcome.

In analyses of temporal change in plasma Aβ levels, we observed significantly faster multivariable-adjusted rates of decline in the TICS and the global score in participants with higher percentage increases in Aβ40 to Aβ42 ratios (Table 4). For example, each 1-SD increment of percentage change in Aβ40 to Aβ42 ratios was associated with a −0.08 points-per-year greater decline in the TICS, cognitively equivalent to 2 years of aging. The association between change in Aβ40 to Aβ42 ratios since midlife and cognitive decline in the TICS was slightly stronger than that observed with midlife Aβ40 to Aβ42 ratios alone. Each 1-SD increment of percentage change in Aβ40 to Aβ42 ratios was associated with a −0.02 U/y greater decline on the global score, cognitively equivalent to 2 years of age and identical to the effect observed for midlife Aβ40 to Aβ42 ratios. In categorical analyses, there was no evidence of threshold effects of higher percentage change in Aβ40 to Aβ42 ratios for the TICS or global score, as there were linear trends across the quartiles for both. For example, the multivariable-adjusted mean differences in decline for the TICS were −0.15, −0.16, and −0.21 points per year, respectively, in the second, third, and fourth quartiles of percentage increase in Aβ40 to Aβ42 ratios; the effect of being in the highest quartile compared with being in the lowest quartile was cognitively equivalent to almost 5 years of age (data not shown). Finally, there were no associations between any outcome and temporal change in plasma Aβ42 levels alone.

All findings were unchanged after addressing confounding by vascular factors at the time of initial cognitive testing and after exclusion of those who developed stroke or underwent carotid endarterectomy. For example, each 1-SD increment in midlife Aβ40 to Aβ42 ratios was associated with a −0.02 U/y greater decline on the global score (P = .02 for trend), and each 1-SD increment of percentage change in Aβ40 to Aβ42 ratios was associated with a −0.08 points-per-year greater decline in the TICS (P = .02 for trend) (data not shown).

In separate analyses, late-life Aβ40 to Aβ42 ratios predicted subsequent decline in general cognition. For example, each 1-SD increment of percentage change in late-life Aβ40 to Aβ42 ratios was associated with a −0.02 U/y (P = .02) greater decline on the global score (data not shown).

Although not the focus of the present analyses, we also examined relations of midlife plasma Aβ40 levels alone, as well as temporal change in Aβ40 levels, to late-life cognitive decline. There were no associations between plasma Aβ40 levels alone and any outcomes (data not shown).

After adjustment for multiple potential confounders, midlife plasma Aβ40 to Aβ42 ratios, but not plasma Aβ42 levels alone, were associated with significantly worse late-life decline in global cognition. Similarly, greater temporal increase in Aβ40 to Aβ42 ratios, but not in Aβ42 levels alone, from midlife to late life predicted a significantly faster rate of cognitive decline.

When comparing estimates associated with midlife plasma Aβ40 to Aβ42 ratios vs the change in plasma Aβ40to Aβ42 ratios since midlife, temporal change since midlife seemed to be a slightly stronger predictor of decline on one of the cognitive measures. Furthermore, the association between accelerated cognitive decline and the observed temporal increases in plasma Aβ40 to Aβ42 ratios is compelling biologically because it is compatible with a plausible mechanism. Specifically, decreases in cerebrospinal fluid and plasma Aβ levels are expected over time as soluble Aβ gradually accrues in insoluble Aβ plaques in the brain, a probable early event in AD pathogenesis.²⁵ Indeed, cerebrospinal fluid Aβ42 levels decline during development of amnestic mild cognitive impairment and AD. To the extent that peripheral decline in Aβ42 levels may be greater than that in Aβ40 levels,²⁶ a temporal change in Aβ40 to Aβ42 ratios may provide a stronger indication of this pathology condition than Aβ42 levels alone or a single measure of the Aβ40 to Aβ42 ratio.

To our knowledge, no prior studies that involve large cohorts have addressed the predictive ability of midlife plasma Aβ levels and change in plasma Aβ levels since midlife with regard to decline on repeated cognitive measures. Therefore, our findings contribute uniquely to the literature. Nevertheless, results from recent investigations that involve older subjects seem consistent with our findings about Aβ40 toAβ42 ratios, which indicate that this factor may be the most valuable predictor in terms of plasma Aβ measures. Graff-Radford et al⁷ observed an association (P = .02) between lower late-life Aβ42 to Aβ40 ratios (ie, higher Aβ40 to Aβ42 ratios) and subsequent decline on the Mattis Dementia Rating Scale²⁷ among 379 persons (median age, 77 years at blood draw) who were administered cognitive tests approximately 5 years apart. Similarly, Sun et al¹⁰ reported a cross-sectional association between higher late-life Aβ40 to Aβ42 ratios and poorer cognition among depressed elders (mean age, 73.8 years). With one exception,⁴ most studies^5,7,8 of dementia that have addressed plasma Aβ40 to Aβ42 ratios have identified significant associations. Overall, it seems that longitudinal studies that measure Aβ40 to Aβ42 ratios may hold the most promise for delineation of the use of plasma Aβ measures to identify persons at risk for late-life cognitive dysfunction.

In addition to the measurement of change in Aβ measures over 10 years, the present study has several strengths. Measurement of midlife Aβ values likely yields less confounding owing to age (and the accompanying variability in levels with aging^3,9) or other related health variables (eg, vascular disease). In addition, we adjusted for various potential confounders, such as depression, heart disease, hypertension, and dyslipidemia. Although we did not collect data on some vascular measures (eg, white matter lesions), it is reassuring that we found no change in estimates after we controlled for lifetime history of a wide array of cardiovascular and cerebrovascular factors. This consistency was especially important because vascular disease may affect plasma Aβ levels^4,28 and because cardiovascular disease may represent an independent path to cognitive decline.²⁹ Finally, the use of repeated cognitive measures allowed evaluation of differences in paths of change.

Limitations should also be considered. Overall CVs for plasma Aβ40 and Aβ42 measures were high because of between-plate variation. Measurement variation could result in underestimation of relations between plasma Aβ measures and cognition. However, within-plate measurement error was low, and analyses corrected for the between-plate variation. Moreover, there was excellent within-pair reliability; therefore, analyses of intraindividual change in Aβ levels would be less affected by measurement variability. Finally, generalizability is a concern in our population of largely white female health professionals. Although biologic mechanisms among these women are likely similar to those in the general population, research addressing diversity is needed. For example, studies may examine differing effects of midlife plasma Aβ measures on age at onset of cognitive decline among racial/ethnic minorities.

In conclusion, this prospective study provides preliminary evidence that higher plasma Aβ40 to Aβ42 ratios at midlife and late life, as well as increases in Aβ40 to Aβ42 ratios between midlife and late life, may predict cognitive decline. These associations require confirmation in other large-scale longitudinal studies. Midlife Aβ40 to Aβ42 ratios alone predicted late-life cognitive decline, which suggests that this ratio may prove valuable for early identification of those at high risk of cognitive impairments. Nonetheless, more work is needed to address whether changes in plasma Aβ40 to Aβ42 ratios since midlife are ultimately more sensitive predictors than measurement of the Aβ40 to Aβ42 ratio at a single time point. The benefits of such work are clear. Plasma biomarkers could aid in targeting prevention within large populations by the identification of high-risk individuals years before clinical symptoms are evident.

Correspondence: Olivia Okereke, MD, MS, Division of Aging and Channing Laboratory, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, 181 Longwood Ave, Third Floor, Boston, MA 02115 (olivia.okereke@channing.harvard.edu).

Accepted for Publication: April 13, 2009.

Author Contributions:Study concept and design: Okereke, Selkoe, and Grodstein. Acquisition of data: Okereke, Xia, and Grodstein. Analysis and interpretation of data: Okereke, Selkoe, and Grodstein. Drafting of the manuscript: Okereke, Selkoe, and Grodstein. Critical revision of the manuscript for important intellectual content: Okereke, Xia, Selkoe, and Grodstein. Statistical analysis: Okereke. Obtained funding: Grodstein. Administrative, technical, and material support: Okereke, Xia, and Grodstein. Study supervision: Grodstein.

Financial Disclosure: None reported.

Funding/Support: This study was supported by grants AG24215, CA49449, CA87969, and R37AG006173 (Dr Selkoe) from the National Institutes of Health.

Additional Contributions: Pankaj D. Mehta, PhD, Wei Q. Qiu, MD, PhD, and Xiaoyan Sun, MD, PhD, help pilot test the Aβ assays, and Helena Judge Ellis, BA, and Shelley Tworoger, PhD, managed the Nurses’ Health Study blood laboratory.