Age but Not Diagnosis Is the Main Predictor of Plasma Amyloid β-Protein Levels FREE

AMYLOID β-PROTEIN (Aβ) is a major component of amyloid plaques in brain of patients with Alzheimer disease (AD). Amyloid β-protein is derived from the β-secretase pathway of amyloid precursor protein (APP) processing by the enzymatic activity of the β-site APP cleaving enzyme—which releases the N-terminus of Aβ from APP—and a presenilin-dependent γ-secretase activity that releases the C-terminus of Aβ from the membrane.¹ The most common forms of Aβ contain 40 (Aβ40) or 42 (Aβ42) amino acids. The Aβ42 is more fibrillogenic and deposits early in amyloid plaques.^2- 3 In addition to being deposited in the brain, Aβ can be detected in cerebrospinal fluid (CSF) and plasma, leading to the analysis of Aβ levels in these fluids as biomarkers of the cerebral amyloidosis in AD.

Cerebrospinal fluid Aβ42 level is reduced in AD^4- 7 and is inversely proportional to dementia severity in some studies.⁸ Plasma Aβ42 level is increased in patients with familial AD mutations.^9- 10 Studies of Aβ40 and Aβ42 in plasma of patients with sporadic AD have been equivocal, some suggesting increased Aβ40 or Aβ42 levels in AD or preclinical AD,^11- 12 but others showing no change.^{9- 10,13- 14} Sensitive measurement of plasma Aβ levels in a large patient group is required to clarify the clinical, demographic, and genetic factors that influence plasma Aβ levels, and as a prerequisite for proposing plasma Aβ as a biomarker for diagnosis, progression, and treatment effects. The principal goal of this study, therefore, was to determine the sensitivity and specificity of plasma Aβ40 and Aβ42 levels for the diagnosis of AD. A related goal was to examine the relationship of plasma Aβ measures with disease severity, medication use, apolipoprotein E (APOE) genotype, and other demographic variables in a prospectively characterized outpatient clinic population.

Plasma samples were collected from patients in the Memory and Movement Disorders Units of Massachusetts General Hospital, Boston, with a diagnosis of AD,¹⁵ mild cognitive impairment (MCI),¹⁶ nondemented Parkinson disease (PD), and no dementia. Informed consent was obtained from the patient and caregiver by a staff physician. The study was approved by the Massachusetts General Hospital Institutional Review Board. The following anonymized data were available for each case: (1) subject demographics, including date of birth, age, sex, race, education, family history of AD (defined as first-degree relative with AD), and family history of dementia; (2) clinical characteristics, including diagnosis, onset of disease, disease duration, Blessed Dementia Scale–Information-Memory-Concentration (BDS-IMC) score,¹⁷ the Clinical Dementia Rating Scale score,¹⁸ and Hoehn and Yahr PD severity scale score¹⁹; (3) current medication use, including cholinesterase inhibitors, estrogen, carbidopa-levodopa, dopamine agonists, anticholinergics, anti-inflammatory medications, hypoglycemic agents, antioxidants, aspirin, and statins; and (4) protocol notes, including last meal, processing details, and protocol violations.

From each patient, 22.5 mL of blood was collected in three 7.5-mL polypropylene sterile plunger tubes (S-Monovette; Sarstedt, Newton, NC), containing potassium EDTA, by a trained phlebotomist. The blood samples were cooled to 4°C for 15 minutes. A serum-plasma separator was added (Sure-Sep II; Organon, West Orange, NJ). In rapid succession, the samples were centrifuged at 3300 rpm (1380g) for 15 minutes and aliquoted in 960-µL quantities into polypropylene tubes containing 40 µL of a protease inhibitor cocktail (Complete, 1 tablet in 2 mL of phosphate-buffered saline; Roche, Indianapolis, Ind), then frozen on dry ice. The samples were stored at −80°C until ready for use.

To block cross-reaction of unidentified components of human plasma with the enzyme-linked immunosorbent assay (ELISA), plasma was precleared with mouse IgG1 κ (Sigma-Aldrich Corp, St Louis, Mo) cross-linked to agarose beads (CNBr-activated Sepharose 4B; Amersham Biosciences, Piscataway, NJ).¹⁰ Preclearing was performed by diluting 300 µL of each plasma sample with 525 µL of sample buffer (20mM phosphate, 400mM sodium chloride, 2mM EDTA, 10% blocking agent [Block Ace Liquid; Dainippon Pharmaceutical, Osaka, Japan], 0.2% bovine serum albumin, 0.0765% 3-{[3-cholamidopropyl]dimethylammonio}-1-propanesulfonate [CHAPS], pH 7.2), and 75 µL of the agarose beads covalently cross-linked to nonspecific mouse IgG1 κ. After incubation for 2 hours at 4°C, the beads were removed by centrifugation.

For this assay,¹⁰ 96-well microtiter plates (Maxisorp Black; Nalge Nunc, Rochester, NY) were coated with the capture antibody—5-µg/mL BNT77 (mouse IgA anti–Aβ 11-28; Takeda Chemical Industries, Osaka, Japan)—and blocked with blocking buffer (25% Block Ace Liquid in phosphate-buffered saline) for 6 hours. Pretreated plasma samples (100 µL, in triplicate) were incubated in BNT77-coated wells containing 50 µL of sample buffer overnight at 4°C. The plates were washed 4 times with phosphate-buffered saline, then reacted with horseradish peroxidase–conjugated detector antibodies (BA27 mouse IgG2 anti-Aβ40, 1:1000; BC05 mouse IgG1 anti-Aβ42, 1:1000, 0.5 µg/mL; Takeda Chemical Industries) in 75 µL of sample buffer for 4 hours at room temperature. After 6 washes with phosphate-buffered saline, horseradish peroxidase enzyme activity was measured with a fluorogenic substrate (Quanta Blu; Pierce, Rockford, Ill) on a fluorometer (Wallac Victor² 1420 Multilabel Counter; Perkin-Elmer, Boston, Mass) with a 320-nm excitation filter and 400-nm emission filter. Each plate contained known concentrations of human synthetic Aβ 1-40 and Aβ 1-42 (Bachem, King of Prussia, Pa) in sample buffer to construct a log-log standard curve. These ELISAs can detect N-terminally truncated β-site APP cleaving enzyme–cleaved Aβ species (Aβ 11-40/42) as well as full-length Aβ (Aβ 1-40/42), but not α-secretase cleaved products (p3; Aβ 17-40/42).²⁰

Within groups, Aβ variables were regressed on age, sex, duration of illness, and BDS-IMC score. Significant factors (age and sex) were included in an analysis of covariance with Aβ measures as dependent variables comparing diagnostic groups with nondemented controls, as well as other demographic and clinical variables. Most of the analyses in this study were well powered. For continuous factors, with the sample sizes of 92 to 146 in the control, PD, and AD cases, and a 2-tailed test at P = .05, the power was 80% to detect a population correlation of approximately r = 0.25 to 0.29. Only for the MCI group with a sample size of 37 was the power weaker, at 70% to detect a correlation of r = 0.4. Power for between-group comparisons was 80% to detect differences of approximately 0.4 SD (0.5 SD for MCI). There were few or no missing values for medication use and demographic variables, so power was similarly strong for analyses involving them.

The sensitivity and specificity of the antibodies and the ELISA have been published.¹⁰ In our hands, the ELISA had a sensitivity of 1 pmol/L for Aβ40 and Aβ42. The recovery of exogenous Aβ40 and Aβ42 added to plasma was greater than 90%, irrespective of the presence or absence of the IgG1 κ resin, indicating that these ELISAs can detect both free Aβ and Aβ bound to plasma proteins.^10,21 Repeated measures of frozen aliquots of the same sample yielded SDs less than 10%, and correlation of repeated measures of samples showed r²>0.96.

Plasma samples were collected from 371 outpatients with a diagnosis of sporadic AD (n = 146), MCI (n = 37), nondemented control cases (n = 92), and PD (n = 96) (Table 1). Relative to the control group, the patients with AD were significantly older (P<.001), had fewer years of education (P<.001), had a greater family history of AD (P = .03), and had a greater APOE ϵ4 allele frequency (0.38 vs 0.11). Relative to the control group, the MCI group had a greater APOE ϵ4 allele frequency (0.39 vs 0.11), and the PD group had a significantly higher proportion of men (P<.001).

By regression analysis, we found that the most robust determinant for Aβ40 and Aβ42 levels in each diagnostic group was age (Figure 1). Other effects were seen only in single diagnostic groups. Within the AD and MCI groups, there was no association of Aβ measures with duration of illness or severity of dementia, as estimated by the BDS-IMC scores (Figure 2).

For the control group, the Aβ variables (Aβ40, Aβ42, and the ratio of Aβ42 to total Aβ [Aβ42/Aβ]) were regressed on age and sex. Age had a significant positive relation to Aβ40 (P<.001) and Aβ42 (P = .005). Sex had a significant relation to Aβ40 (P = .02), with women having a higher mean than men. For the AD group, Aβ variables were regressed on age, sex, duration of illness, and the BDS-IMC scores. The only significant effects were that age had a positive relation to Aβ40 (P = .001) and to Aβ42 (P = .01). For the MCI group, as for the AD group, Aβ variables were regressed on age, sex, duration of illness, and the BDS-IMC. The only significant effect was that age had a positive relation with Aβ40 (P<.01). For the PD group, as for other diagnostic groups, Aβ variables were regressed on age, sex, duration of illness, and the BDS-IMC. The only significant effects were that age had a positive relation with Aβ40 (P<.001) and with Aβ42 (P = .01), and duration of PD had a weak positive relation for Aβ40 (P<.05). Pursuant to these findings, age and sex were included as covariates in the group comparisons that follow.

After covarying for age, there was no significant difference in Aβ measures between diagnostic groups (Table 1). Analyses of covariance were run with the Aβ variables as dependent variables comparing AD vs control groups crossed with a sex factor and including age as a covariate. The only significant effects involving group comparisons were significant interactions between sex and diagnostic group for Aβ40 (P = .04) and for Aβ42 (P = .04). In both cases, the interaction was due to the control group having a higher mean than the AD group among women, with the reverse situation occurring among men. The MCI and control groups as well as the PD and control groups were compared with the same analysis of covariance used to compare the AD and control groups. No significant effects involving diagnostic group were found.

To determine whether other genetic or clinical features affect Aβ measures, we evaluated the number of APOE ϵ4 alleles, family history of dementia, family history of AD, and medication use.

Number of APOE ϵ4 alleles (0, 1, or 2) was crossed with sex and diagnostic group (AD and MCI only; PD and control subjects were not included because there were too few individuals who were homozygotes for APOE ϵ4), and age was covaried. Dependent variables were Aβ40, Aβ42, and Aβ42/Aβ. No significant effects involving APOEϵ4were found (Figure 3).

Presence or absence of family history of dementia and of family history of AD were crossed with sex and diagnostic group (AD, PD, MCI, and controls), and age was covaried. Dependent variables were Aβ40, Aβ42, and Aβ42/Aβ. No effects involving family history were significant except for complex higher-order interactions involving sex and diagnostic group (Figure 4).

Whether or not participants were taking various medications was analyzed in relation to Aβ40, Aβ42, and Aβ42/Aβ. In separate analyses, the medications were cholinesterase inhibitors, anti-inflammatory drugs, antioxidants, estrogen, and statins. Only data for women were analyzed in the case of estrogen. Medication use was crossed with diagnostic group and sex, and age was covaried. In each analysis, only diagnostic groups with sufficient numbers of participants taking the medication were included. No effects involving medications were found to be significant (Figure 5).

Since amyloid plaques are a fundamental feature of AD neuropathology, and Aβ can be detected in CSF and plasma, Aβ measures in biological fluids are compelling candidate biomarkers for AD diagnosis and progression.²² The combination of low Aβ42 level and elevated tau protein in CSF has modest sensitivity and specificity for diagnosing AD.⁶ Plasma Aβ or Aβ42 is increased in familial AD with presenilin or APP mutations as well as in Down syndrome with APP triplication,^9- 10,23 but, on the basis of our study and others, these plasma measures do not reliably differentiate sporadic AD from control cases.^{9- 10,13- 14}

We collected plasma samples from a cohort of 371 patients, and specifically studied patients with MCI and a neurodegenerative control group of nondemented patients with PD, in addition to AD and neurologically normal controls. This large and diverse sample allowed us to examine which genetic, demographic, and clinical factors were significantly associated with the variance in plasma Aβ levels. The results of our study indicate that the primary influence on plasma Aβ40 and Aβ42 levels is age rather than diagnosis, with higher Aβ40 and Aβ42 levels in older patients regardless of diagnostic category. This effect of age is consistent with the findings of Younkin et al²⁴ and Mayeux et al.¹¹ After controlling for age, there was no significant difference in Aβ levels among the diagnoses. Studies using similar antibodies to our assay (either BAN50 or BNT77 capture antibodies and BA27/BC05 detector antibodies) and others (3D6 capture antibody and 21F12 anti-Aβ42 detector antibody) also found no significant differences between AD and control cases.^{9- 10,13- 14} In contrast, ELISAs using 6E10 capture with R162/R164 or R165 detector antibodies have detected elevated plasma Aβ measures in AD or incipient AD, with a large overlap with non-AD cases.^11- 12 Our study did not detect elevated Aβ measures in MCI cases, which could be considered preclinical AD; however, it is important to note that in the study by Mayeux et al,¹¹ elevated Aβ levels were present before any cognitive impairment in those who subsequently became demented.

In secondary analyses, we investigated other factors associated with AD risk and therapy, including education, sex, family history of dementia, family history of AD, APOE genotype, and use of classes of medications. When age was covaried, no significant effects were found for medication use, APOE genotype, or family history of dementia. Within the AD and MCI groups, plasma Aβ level did not correlate with duration or severity of memory impairment. These results indicate that the variance in plasma Aβ levels in late-onset AD is largely related to age, although we cannot rule out other genetic factors besides APOE, PS-1, and APP, since there is evidence that plasma Aβ levels behave like heritable traits (independent of diagnosis or family history of AD).^25- 26

Few published studies have correlated plasma Aβ levels with medication use. Our cross-sectional results with statins are consistent with another cross-sectional study finding that plasma levels of Aβ were not associated with statin use,²⁷ and with a study indicating no association of CSF Aβ42 levels with statin use²⁸; however, lovastatin reduced serum Aβ levels in a dose-dependent manner during 3 months in a placebo-controlled study of hypercholesterolemic patients,²⁹ and simvastatin treatment for 26 weeks showed a trend toward reduced CSF Aβ40 levels.³⁰ We did not detect significant effects on plasma Aβ by commonly used current medication classes for AD—cholinesterase inhibitors and antioxidants (eg, vitamin E)—as well as by putative preventive agents against the development of AD—estrogen, nonsteroidal anti-inflammatory drugs, and statins.³¹ While large class effects on plasma Aβ were not found in this analysis, we cannot rule out individual medication effects. Specific medications within the nonsteroidal anti-inflammatory drug and statin classes may differ in effects on APP processing and AD risk. For instance, in a study of the nonsteroidal anti-inflammatory drugs, ibuprofen, sulindac sulfide, and indomethacin were more effective than naproxen, aspirin, and celecoxib in reducing Aβ42 production in cell culture.³² Among the statins, a reduced prevalence of AD was associated with lovastatin and pravastatin but not simvastatin.³³ However, insufficient numbers of our sample population took any single medication to allow subclass analysis of this sort. Some classes of medications may affect AD risk without affecting APP metabolism and Aβ levels, such as cholinesterase inhibitors and antioxidants.

Besides these clinical, demographic, and genetic factors, the physiologic processes that affect plasma Aβ levels are unknown, in particular where Aβ in plasma is synthesized and metabolized. Studies in APP transgenic mice suggest an equilibrium between Aβ deposited in brain, soluble Aβ in CSF, and Aβ in plasma: in aging Tg2576 APP KN670-1ML mice, age-related Aβ deposition in the brain is associated with a reduction in CSF and plasma Aβ levels.³⁴ The Aβ injected intraventricularly in rats is cleared into the blood,³⁵ intravenously administered Aβ can enter mouse brain,³⁶ and peripheral administration of Aβ antibodies in APP transgenic mice can bind Aβ from the CSF-brain compartment.³⁷ Studies in humans have failed to demonstrate, however, correlation of CSF Aβ levels and plasma Aβ levels.³⁸ Alternatively, extracerebral sources such as platelets are a source of Aβ in plasma.³⁹ The age-related increase of Aβ species in plasma may be a peripheral reflection of increases in Aβ production or reduction in Aβ clearance in the brain leading to increased Aβ deposition and AD with aging; changes in the central or peripheral activity of Aβ synthetic enzymes (eg, β-secretase or γ-secretase) or Aβ catabolic enzymes (eg, insulin-degrading enzyme or neprilysin) with aging remain to be clarified.

This study demonstrates that age is the principal correlate of plasma Aβ levels, rather than diagnosis, medication use, or APOE genotype. Therefore, plasma Aβ is not a reliably sensitive or specific biomarker of AD or MCI diagnosis in cross-sectional study. Longitudinal analysis of plasma in the course of a double-blind placebo-controlled study of specific drugs could detect more sensitive effects of medications on plasma Aβ measures in individual patients. Clinical follow-up of individuals in our study is also under way to determine whether these baseline levels of Aβ40, Aβ42, or Aβ42/Aβ predict future cognitive decline, as suggested by the results of Mayeux et al¹¹ and the studies of familial AD with PS-1 and APP mutations.^9- 11 Serial measurement of population-based samples could also determine whether the pattern of change in Aβ levels in plasma is predictive of conversion to AD or progression of established AD.⁴⁰

Corresponding author and reprints: Michael C. Irizarry, MD, Alzheimer Disease Research Unit, Massachusetts General Hospital–East, B114-2010, 11416th St, Charlestown, MA 02129 (e-mail: mirizarry@partners.org).

Accepted for publication November 14, 2002.

Author contributions: Study concept and design (Drs Fukumoto, Hyman, Growdon, and Irizarry and Ms Tennis); acquisition of data (Drs Fukumoto, Hyman, and Irizarry and Ms Tennis); analysis and interpretation of data (Drs Fukumoto, Locascio, Hyman, Growdon, and Irizarry); drafting of the manuscript (Drs Fukumoto, Locascio, Hyman, and Irizarry); critical revision of the manuscript for important intellectual content (Drs Fukumoto, Locascio, Hyman, Growdon, and Irizarry and Ms Tennis); statistical expertise (Drs Locascio, Hyman, and Irizarry); obtained funding (Drs Hyman, Growdon, and Irizarry); administrative, technical, and material support (Drs Fukumoto, Growdon, and Irizarry and Ms Tennis); study supervision (Drs Fukumoto, Hyman, Growdon, and Irizarry and Ms Tennis).

This study was supported by grants AG00793 and AG05134 from the National Institutes of Health, Bethesda, Md, and the Lawrence J. and Anne Cable Rubenstein Foundation. Dr Fukumoto's salary is supported by Takeda Chemical Industries, Osaka, Japan.

We thank Marisa Dreisbach, Kerri Anne Giglio, Bonnie Cheung, Sarah McKenzie Hallen, Lue Davis, and Ellen Valentine for phlebotomy collection, sample processing, and administrative support.