Plasma β Amyloid and the Risk of Alzheimer Disease and Dementia in Elderly Men: A Prospective, Population-Based Cohort Study

Background  β Amyloid (Aβ) protein accumulates in the brains of individuals with Alzheimer disease (AD) and is detectable in cerebrospinal fluid and plasma.

Objective  To examine plasma levels of Aβ peptides Aβ₄₀ and Aβ₄₂ as predictors of incident AD and other types of dementia.

Design  Prospective, population-based cohort study.

Setting  The Uppsala Longitudinal Study of Adult Men.

Participants  Plasma Aβ₄₀ and Aβ₄₂ levels were analyzed as predictors of incident AD in 1045 men at age 70 years and 680 men at age 77 years using Cox proportional hazards analyses. Alzheimer disease and other types of dementia were diagnosed by standardized screening, clinical evaluation, and medical record review.

Main Outcome Measures  Hazard ratios of AD (primary outcome) and vascular dementia or other dementia (secondary outcomes) according to baseline levels of plasma Aβ₄₀ and Aβ₄₂.

Results  From the age of 77 years at baseline, 46 individuals developed AD at follow-up (median, 5.3 years). A low plasma Aβ₄₀ level at age 77 years was associated with higher incidence of AD. The multivariate-adjusted hazard ratio was 4.87 (95% confidence interval, 1.63-14.6) for the lowest Aβ₄₀ tertile compared with the highest tertile. On follow-up from age 70 years at baseline (median, 11.2 years), 82 individuals developed AD. Plasma Aβ₄₀ and Aβ₄₂ levels measured at age 70 years were not significantly associated with incident AD.

Conclusions  Low plasma Aβ₄₀ levels predicted incident AD in elderly men independently of potential confounders. Plasma Aβ₄₂ levels were not significantly associated with AD incidence. The clinical value of Aβ measurement in plasma remains to be established in future studies.

Alzheimer disease (AD) is a progressive neurodegenerative disease and the most common form of dementia. β Amyloid (Aβ) protein deposition in senile plaques in the brain is a pathologic hallmark of AD.^1-3 β Amyloid occurs in 2 prominent forms: one containing 40 (Aβ₄₀) and the other containing 42 (Aβ₄₂) amino acids^4,5; Aβ₄₀ and Aβ₄₂ can be detected and quantified in cerebrospinal fluid (CSF) and plasma and are potential predictors of AD.^6-10

In CSF, low levels of Aβ₄₂ are strongly associated both with manifest AD and with future development of AD.^6-9 In plasma, results from previous studies are conflicting. Plasma Aβ is increased in familial AD with mutations in the presenilin or APP genes as well as in people with Down syndrome (trisomy 21).^11-13 In cross-sectional studies in patients without any known mutations, plasma Aβ levels in individuals with AD or mild cognitive impairment were either higher,^14-16 lower,¹⁷ or unchanged.^11,18 Only 3 longitudinal studies have addressed this issue. Of these, one study¹⁹ reported an association between high plasma Aβ₄₂ levels and risk of AD, while the other studies^20,21 reported an association between risk of AD and increased levels of Aβ₄₀ and a low Aβ₄₂:Aβ₄₀ ratio, but not between AD risk and Aβ₄₂ levels. Given the variability of the findings in previous studies and the potential implications of Aβ levels as a marker of disease risk, we investigated the relationship between plasma Aβ levels and incident AD, vascular dementia (VaD), and all types of dementia in a large, prospective, population-based cohort study of elderly men.

The Uppsala Longitudinal Study of Adult Men (ULSAM) was initiated in 1970 when all 50-year-old men living in Uppsala, Sweden, were invited to participate in a health survey, initially focused on identifying factors for cardiovascular disease (described in detail at http://www.pubcare.uu.se/ULSAM). Our analyses are based on the second and third reinvestigations of the ULSAM cohort, when the participants were aged approximately 70 (1990-1994, n = 1221) and 77 (1998-2001, n = 838) years.

Plasma Aβ₄₀ and Aβ₄₂ levels were measured in 1082 (88.6%) participants at age 70 years and in 733 (87.5%) at age 77 years. For our study, all participants with a diagnosis of any type of dementia at baseline (3 at age 70 years and 33 at age 77 years) and those who did not agree to have their medical records reviewed (34 at age 70 years and 20 at age 77 years) were excluded. Thus, our study samples comprise 1045 participants at age 70 years and 680 participants at age 77 years. Two serial samples (at ages 70 and 77 years) were available from 630 individuals. Fifty participants provided samples only at age 77 years. The study was approved by the regional ethical committee at Uppsala University. Informed consent from all participants was obtained.

Information on definite and possible risk factors for dementia were collected from examinations at ages 70 and 77 years. Fasting serum cholesterol concentrations were assayed by enzymatic techniques. The apolipoprotein E gene (APOE) was genotyped by minisequencing.²² Smoking status (current smokers or nonsmokers) was assessed by questionnaires, and education level was assessed by interviews at age 72 years. Education level was stratified as low (elementary school only, 6-7 years), medium (high school), or high (college studies). Systolic and diastolic blood pressure was measured in the supine position after a 10-minute rest to the nearest 2 mm Hg. Hypertension at baseline was defined as systolic blood pressure at or above 140 mm Hg, diastolic blood pressure at or above 90 mm Hg, and/or use of antihypertensive medication.²³ The presence of diabetes at baseline was defined as a fasting plasma glucose level of 126.1 mg/dL (to convert to mmol/L, multiply by 0.0555) or more and/or the use of oral hypoglycemic agents or insulin. Body mass index was calculated as weight in kilograms divided by height in meters squared. Plasma samples were collected and stored at − 70°C until thawing. We analyzed Aβ₄₀ and Aβ₄₂ levels in plasma using a well-characterized enzyme-linked immunosorbent assay method with BNT77 (IgA mouse anti-Aβ 11-28; Takeda, Osaka, Japan) and horseradish peroxidase–conjugated detector antibodies (BA27, IgG2 mouse anti-Aβ₄₀; and BC05, IgG1 mouse anti-Aβ₄₂; Takeda) as previously described.^11,24 We analyzed Aβ₄₀ in duplicate and Aβ₄₂ in triplicate samples. Assays were performed blinded to all clinical data.

Participants were invited to participate in cognitive testing at ages 72 (n = 999), 77 (n = 804), and 82 (n = 523) years. Participants with low test performance (Mini-Mental State Examination score < 26; or, at age 82 years, MMSE < 26 and/or a high-risk result in a 7-minute screen)^25,26 were referred to the Geriatric Memory Clinic at the Uppsala University Hospital for a thorough clinical assessment. With the aim to identify all cases of AD and other types of dementia, all available medical records from the Uppsala University Hospital, the general practice offices in Uppsala, the community nursing homes, and the dementia group homes were reviewed from 1990 until December 31, 2005. Most medical care is provided in these settings. The records of all possible cases of dementia or cognitive impairment were reviewed and 2 experienced geriatricians, working independently of each other, assigned diagnoses. In case of disagreement, a third experienced geriatrician reviewed the case, and the diagnosis was determined by majority decision. The diagnosis of AD was defined according to the National Institute of Neurological and Communication Disorders and Stroke–Alzheimer's Disease and Related Disorders Association criteria²⁷ and the Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition) criteria.²⁸ Most of the AD cases (76 of 82 [93%] at age 70 years and 42 of 46 [92%] at age 77 years) were primarily assessed and diagnosed at the Geriatric Memory Clinic and the Geropsychiatric Clinic, Uppsala University Hospital. Ninety-three percent of the AD cases underwent a computed tomography scan, the results of which showed consistency with AD, ie, normal atrophy or mild to moderate white matter changes. Vascular dementia was defined according to the Chui criteria.²⁹ Other dementia disorders included Parkinson disease with dementia and frontotemporal dementia.³⁰ Cases of dementia with insufficiently recorded medical data were diagnosed as dementia not otherwise specified. Mild cognitive impairment was defined as subjective and objective evidence of a cognitive decline that did not interfere with activities of daily life.³¹ All cases were validated according to current guidelines.^27-31

Analysis plans were defined a priori. Separate analyses were performed at 70 years and 77 years to explore the association between plasma Aβ₄₀ and Aβ₄₂ levels and AD incidence at different points in the same cohort. Kaplan-Meier curves with log-rank tests were used to examine the AD-free survival by cohort-specific tertiles of Aβ₄₀ and Aβ₄₂ levels, as well as the Aβ₄₂:Aβ₄₀ ratio. Cox proportional hazard regression models were used to estimate crude and multivariate-adjusted estimates of the hazard ratio (HR) of AD according to Aβ levels (Stata, version 8; Stata Corp, College Station, Texas). To control for confounders, and at the same time to avoid over-fitted models, forward selection was applied to identify covariates associated with AD to include in the full models. Candidate covariates (APOE ε4 allele status, hypertension, diabetes, body mass index, serum cholesterol, serum creatinine, smoking, and education level) were incorporated into Cox models by forward selection (P < .05, by likelihood ratio test for retention). For both the 70-year-old cohort and the 77-year-old cohort, only age and APOE genotype (coded as presence or absence of APOE ε4 allele) remained significant and were included in the multivariate models of Aβ level. We then tested whether introducing dummy variables for tertiles of Aβ₄₀ or Aβ₄₂ levels were significant in the optimal forward selection model by using likelihood ratio tests. To determine if there was residual confounding, we compared the β coefficient of the Aβ₄₀ and Aβ₄₂ variables in the final forward selection model with that of the maximal model incorporating all covariates. Observations were censored at death, date of migration/emigration, date of diagnosis of AD or any other type of dementia, or end of follow-up (December 31, 2005). Dates of deaths and of migration/emigration were obtained from the continually updated Swedish National Population Register.

The proportionality of hazards was confirmed using Schoenfeld tests. Sensitivity analysis was performed to assess for the possible effect of inclusion of prevalent cases of dementia in the baseline cohorts. To assess for the effects of participation on the follow-up examinations, we also included dummy variables for being alive but not participating in the examinations at age 77 or 82 years. In secondary analyses in the 70-year-old cohort, VaD and all types of dementia (including AD) were defined as outcomes and Cox models contained age, APOE genotype, and diabetes covariates chosen in forward selection models, as described previously.

The clinical characteristics of the participants at baseline for the 70-year-old and the 77-year-old cohorts are presented in Table 1. In the 70-year-old cohort, the participants were followed up for a median time of 11.2 years (maximum of 14.4 years), totaling 10 208 person-years at risk. One hundred forty-six men had a diagnosis of any type of dementia, of whom 82 developed AD and 25 had diagnosed VaD. Participants in the 77-year-old cohort had a median follow-up of 5.3 years and a maximum follow-up of 7.9 years, totaling 3420 person-years at risk. Seventy-four participants developed any type of dementia, of whom 46 had diagnosed AD and 10 had diagnosed VaD (Table 2). Presence of the APOE ε4 allele was significantly associated with AD in the 70-year-old (HR, 2.8; 95% confidence interval [CI], 1.7-4.7) and the 77-year-old (HR, 2.1; 95% CI, 1.1-3.8) cohorts.

Kaplan-Meier curves on the cumulative HR of AD by tertiles of plasma Aβ₄₀ and Aβ₄₂ levels in the 2 cohorts are shown in the Figure. In unadjusted Cox proportional hazards analyses in the 77-year-old cohort, low Aβ₄₀ levels were significantly associated with higher AD incidence, both as a continuous variable and when split into tertiles (model A, Table 3). Men in the lowest tertile of Aβ₄₀ level had a 5-fold higher risk of AD relative to the highest tertile (95% CI, 1.63-14.6; P = .006). A low plasma Aβ₄₀ level remained significantly associated with increased risk of AD when adjusting for age and APOE genotype (model B, Table 3). There was a similar trend for Aβ₄₂ level and risk of AD. In the unadjusted models, the lowest tertile of Aβ₄₂ level was significantly associated with increased risk of AD (P = .03). In the adjusted models, men in the lowest tertile of Aβ₄₂ level had a more than 2-fold increase of AD risk relative to the highest tertile, which did not reach significance (95% CI, 0.95-5.56; P = .06). In the 70-year-old cohort, neither Aβ₄₀ nor Aβ₄₂ level was associated with risk of AD (Table 4). Thus, there appears to be a temporal trend in the influence of Aβ₄₀ level and even Aβ₄₂ level in AD risk; Aβ₄₀ and Aβ₄₂ levels are not associated with AD at age 70 years, but low levels tend to be associated with AD at age 77 years. There was no association between a change in Aβ₄₀ and Aβ₄₂ levels between ages 70 and 77 years and AD risk among the 630 individuals with 2 plasma samples (data not shown). At age 70 years, the middle but not the lowest tertile of the Aβ₄₂:Aβ₄₀ ratio was significantly associated with reduced risk of AD incidence.

The incidence rates of VaD and all types of dementia according to baseline plasma Aβ levels at ages 70 and 77 years are presented in Table 3 and Table 4. In adjusted models in the 70-year-old cohort, 1 SD increase in Aβ₄₂:Aβ₄₀ ratio was associated with increased risk of VaD (HR, 1.78; 95% CI, 1.18-2.94; P = .006). The association between Aβ₄₂ level and risk of VaD was borderline significant (P = .06). In the 77-year-old cohort, the lowest tertile of Aβ₄₀ level was significantly associated with increased risk of all type of dementia (HR, 2.40; 95% CI, 1.24-4.50; P = .008).

To rule out the possibility of residual confounding, HRs from the maximal model of all covariates (education and vascular risk factors) were compared with those from the association between Aβ₄₀ level and incident AD in the 77-year-old cohort; the association of the Aβ₄₂:Aβ₄₀ ratio with VaD remained significant. None of the results were affected when excluding cases diagnosed within 2 years after baseline or cases of mild cognitive impairment at baseline, after adjustments for participation or exclusion of participants who died and who might have had undiagnosed AD. Interactions between Aβ level and the covariates APOE ε4 allele status and age were investigated in the 70-year-old and 77-year-old cohorts and no significant interactions were found.

In this study, low plasma Aβ₄₀ levels at approximately 77 years of age were associated with an increased risk of incident AD, independent of age and APOE genotype. Furthermore, low plasma Aβ₄₂ level was associated with AD incidence, but this association did not remain statistically significant after adjustments for APOE genotype. There was no link between plasma Aβ concentrations (measured 7 years earlier when the participant was aged approximately 70 years) and risk of AD, indicating that this association is only present in older ages. The middle, but not the lowest, tertile of the Aβ₄₂:Aβ₄₀ ratio in the 70-year-old cohort was significantly associated with AD incidence. Furthermore, in the tertile analysis of the participants in the 77-year-old cohort, the data suggest a threshold effect for Aβ₄₀ level and a dose effect for Aβ₄₂ level. It is unclear whether this reflects true differences in the biology of these proteins, the particular parameterization in the survival models, or the sensitivity of the assay at very low levels of Aβ.

There are 3 previous longitudinal studies on plasma Aβ levels and risk of AD. In one study, van Oijen et al²⁰ recently reported an association between dementia and a high level of plasma Aβ₄₀ and a low Aβ₄₂:Aβ₄₀ ratio, but not a high Aβ₄₂ level, in the case-cohort Rotterdam Study. This study had a mean follow-up time similar to ours (8.6 years), but its participants were about 9 years younger than the participants in the 77-year-old cohort in ULSAM. Graff-Radford et al²¹ also report an association between a low Aβ₄₂:Aβ₄₀ ratio and AD and MCI incidence. Mayeux et al^19,32 found that those who developed AD during follow-up (mean, 5 years) had significantly higher Aβ₄₂ but not Aβ₄₀ levels at baseline. Although the results of these studies appear inconsistent, their differences may be related to the temporal changes in plasma Aβ levels associated with age and timing relative to incident AD. Plasma Aβ levels appear to decline in individuals who are likely to develop AD.¹⁹ Plasma Aβ levels may be higher years before the AD diagnosis, reflecting either genetic predisposition or the balance between Aβ production and clearance, but decrease closer to diagnosis. In addition, plasma Aβ has been analyzed with different methods, and these studies may be measuring different Aβ species depending on the antibody affinity.^10,11,15 Assays also vary in the ability to detect N-terminally truncated Aβ or Aβ incorporated into soluble oligomers. The assay we employed detects free and protein-bound monomeric, but not oligomeric, Aβ.^11,24 Third, misclassification is an issue for any clinical study of dementia but is likely to be random with respect to Aβ levels, and false negatives due to misclassification would probably drive results toward the null hypothesis. Results were unchanged after excluding all those who died (who might have had undiagnosed AD), minimizing the possibility of competing risks. Adjustments for potential confounders did not influence the results. Associations between plasma Aβ levels and AD incidence may also differ in different racial subpopulations.

The mechanism, whether peripheral and/or central, by which low Aβ₄₀ levels in plasma are associated with a higher risk of AD remains to be clarified.^19,33-35 It is now well established that AD is associated with a decline of Aβ₄₂ levels in CSF.^6,8,36,37 We believe that low plasma Aβ levels mirror a decline of Aβ in CSF owing to increased aggregation in the brain.^38,39 In the Tg2576 transgenic mouse model of AD, progressive amyloid deposition in the brain was associated with parallel declines of Aβ levels in CSF and plasma.³⁸ In humans, previous studies have reported no association between CSF and plasma levels in manifest AD.⁴⁰ However, a case-control study of ours⁴¹ suggests a correlation between Aβ₄₀ and Aβ₄₂ in CSF and plasma in healthy individuals but not in people with AD, indicating that plasma levels of Aβ₄₀ and Aβ₄₂ might mirror the aggregation process in the central nervous system before clinically manifest AD. It should be noted that it is not possible to establish causality in the present study. The associations between brain, CSF, and plasma levels of Aβ in those with AD and healthy elderly individuals need further study.

Interestingly, plasma Aβ levels seem to be associated not only with AD but with other types of dementia. This is the first prospective longitudinal study reporting a significant association between the plasma Aβ₄₂:Aβ₄₀ ratio and risk of VaD. This is in contrast with the Rotterdam Study, in which high levels of Aβ₄₀ increased the risk of VaD. Owing to the small number of VaD cases, this association needs to be further investigated in other longitudinal studies.

The strengths of our study include its prospective longitudinal design, population-based setting, large number of participants, nearly complete follow-up, validation of the cases (limiting the inclusion of false-positive cases), and the repeated measurements of plasma Aβ levels. Moreover, the sample handling and enzyme-linked immunosorbent assay were standardized to reduce variability and measurement error. Finally, the study was restricted to men of the same age in an ethnically homogeneous geographic region, reducing confounding by other genetic and disease-, age-, and sex-related factors. However, these same factors can also limit the generalizability.

In conclusion, this study shows that low plasma Aβ₄₀ levels are associated with a higher risk of developing AD in elderly men. The association between low plasma Aβ and incidence of AD might be a reflection of an increased deposition of Aβ in the central nervous system prior to the onset of clinically manifest cognitive decline. This is also the first prospective longitudinal population-based study to report a significant association between a high Aβ₄₂:Aβ₄₀ ratio and risk of VaD. Further prospective longitudinal studies of different types of dementia and with repeated measurements of Aβ levels in both plasma and CSF are needed to clarify the role of plasma Aβ as a predictor of different types of dementia.

Correspondence: Lars Lannfelt, MD, PhD, Uppsala University, Department of Public Health and Geriatrics, Uppsala Science Park, Dag Hammarskölds väg 14B, 751 85 Uppsala, Sweden (lars.lannfelt@pubcare.uu.se).

Accepted for Publication: April 27, 2007.

Author Contributions: Drs Sundelöf and Giedraitis contributed equally. Study concept and design: Sundelöf, Sundström, Hyman, Basun, and Kilander. Acquisition of data: Sundelöf, Giedraitis, Irizarry, Rönnemaa, Gunnarsson, Lannfelt, and Kilander. Analysis and interpretation of data: Sundelöf, Giedraitis, Irizarry, Sundström, E. Ingelsson, Ärnlöv, Hyman, M. Ingelsson, and Kilander. Drafting of the manuscript: Sundelöf, Giedraitis, Irizarry, Gunnarsson, M. Ingelsson, Lannfelt, and Kilander. Critical revision of the manuscript for important intellectual content: Sundelöf, Giedraitis, Irizarry, Sundström, E. Ingelsson, Rönnemaa, Ärnlöv, Hyman, Basun, M. Ingelsson, Lannfelt, and Kilander. Statistical analysis: Giedraitis, Irizarry, Sundström, and Ärnlöv. Obtained funding: Sundelöf, Giedraitis, Irizarry, Hyman, and Lannfelt. Administrative, technical, and material support: Sundelöf, Irizarry, Hyman, and Lannfelt. Study supervision: Irizarry, Sundström, E. Ingelsson, Hyman, Basun, M. Ingelsson, Lannfelt, and Kilander.

Financial Disclosure: Dr Basun worked for AstraZeneca when the study was performed.

Funding/Support: This study was supported by Wallenberg Consortium North, Hjärnfonden, Bertil Hållstens forskningsstiftelse, Alzheimerfonden, grant 2003-5546 from the Swedish Research Council, the European Union 6th Consortium APOPIS (contract No. LSHM-CT-2003-503330), Stiftelsen Gamla Tjänarinnor, Capios Forskningsstiftelse, Gun och Bertil Stohnes forskningsstiftelse, Swedish Lions Research Foundation, grant 1T32NS048005-01 from the National Institutes of Health, grant AG05134 from the Massachusetts Alzheimer Disease Research Center, AFAR Beeson Award (Dr Irizarry), J.D. French Alzheimer's Foundation, and an unrestricted grant from AstraZeneca.