Amyloid-Associated Depression: Title and subTitle BreakA Prodromal Depression of Alzheimer Disease? FREE

Two large population studies, the Rotterdam and the Mayo Clinic cohorts, have shown that a low concentration of amyloid-β peptide 42 (Aβ₄₂) combined with a high concentration of Aβ₄₀ in plasma increases the risk of developing Alzheimer disease (AD).^{1 - 2} Depression also increases the risk or is an early symptom of AD in the elderly population.³ We recently reported that depressed elderly individuals without cardiovascular disease (CVD) have a lower concentration of Aβ₄₂, leading to a high Aβ₄₀:Aβ₄₂ ratio, in plasma compared with both depressed elderly individuals with CVD and those without depression in a homebound elderly population.⁴ This result led us to hypothesize that depression characterized by a high Aβ₄₀:Aβ₄₂ ratio may represent a distinct depression subtype, which we have termed amyloid-associated depression.

Although plasma Aβ peptides cannot be used to diagnose AD,⁵ plasma Aβ₄₂ levels decline significantly at a preclinical or early stage of AD,⁶ suggesting a biomarker for a prodrome of the disease. In amyloid precursor protein (APP) transgenic mice, plasma Aβ₄₂ level declines significantly before the pathological changes of AD are formed in the brain.^{7 - 8} Elevated plasma Aβ₄₀ level is correlated with cerebral microvascular pathological features, which are linked with both late-life depression and AD.^{9 - 10} Therefore, it is possible that the combination of high Aβ₄₀ and low Aβ₄₂ levels in plasma may be used as a biomarker of pathological change in the brain at a preclinical stage of AD.

Studies have demonstrated that there is an increased risk of the development of AD in some, but not all, individuals with late-life depression, suggesting a prodromal state of AD.^{11 - 12} We hypothesized that a high plasma Aβ₄₀:Aβ₄₂ ratio plus clinical symptoms of depression may represent a prodromal depression of AD and may lead to a more imminent cognitive deterioration than in those who have a high plasma Aβ₄₀:Aβ₄₂ ratio without depression. If our assumption is correct, at the cross-sectional level, patients with amyloid-associated depression should present with a pattern of cognitive impairment consistent with prodromal AD, ie, prominent memory dysfunction, compared with those with nonamyloid depression or those without depression. To test this hypothesis, we investigated the relationships among plasma Aβ peptide levels, depression, and cognitive function in a homebound elderly population.

We studied a group of 995 subjects, all of whom had been tested for depression status and plasma Aβ peptide levels in the Nutrition, Aging, and Memory in the Elderly study, an ongoing, population-based study supported by the National Institute on Aging. Subjects included homebound elderly clients who were enrolled in 1 of 4 home care agencies in the Boston area between 2003 and 2006. Anyone receiving home care services was registered with 1 of these agencies if he or she lived in the city of Boston, had an annual income less than $18 890, and needed home care service. All of the homebound elderly subjects 60 years or older from each of the 4 agencies were invited to participate in the study.

Eligibility for enrollment required that the participants speak English, be physically able to participate in the study home visits, and have sufficient vision and hearing to read and hear the content of the neuropsychological tests. Of all 1803 eligible subjects, 1190 individuals (66.0%) enrolled and gave informed consent to participate in the study. The population was screened for significant cognitive impairment by means of the Mini-Mental State Examination (MMSE)¹³ and for estimated verbal IQ by means of the North American Adult Reading Test.¹⁴ Those with MMSE scores of 10 or less or verbal IQ less than 75 were not eligible to continue in the study, and eligible subjects were subsequently examined. Each subject engaged in 3 home visits conducted by a research assistant, who drew a fasting blood sample and collected data on depression and medical conditions.¹⁵ Among them, 995 subjects had both blood samples available for plasma Aβ measurements and data on depression.

The blood samples were centrifuged immediately after the blood was drawn. The sandwich Aβ enzyme-linked immunosorbent assay was used. Plates were coated with 2G3 (anti-Aβ₄₀) and 21F12 (anti-Aβ₄₂) antibodies overnight at 4°C. Samples were then loaded and incubated overnight at 4°C followed by incubation with a biotinylated monoclonal anti–N-terminus Aβ antibody (3D6B) for 2 hours. Finally, streptavidin-conjugated alkaline phosphatase (Promega Corp, Madison, Wisconsin) was added and incubated, and the signal was amplified by adding alkaline phosphatase fluorescent substrate (Promega Corp), which was then measured. The lowest detection for both Aβ peptides was 1.6 pg/mL in the standard curves, with percentage coefficient of variation between 1.1 and 7.2, and these were used as the cutoff points for comparison and regression if levels were below the cutoff point of detection (6 samples). The samples with higher levels than the standard curve were repeated with dilutions for measurement. The intracorrelations with 2 other laboratories that have published the results of their Aβ measurements^{16 - 17} showed r = 0.63 and 0.84 for Aβ₄₀ and r = 0.90 and 0.96 for Aβ₄₂.

Depressive symptoms were assessed by means of the Center for Epidemiological Studies Depression Scale (CES-D)¹⁸ ; a CES-D score of 16 or greater was used as the cutoff point for clinical depression.¹⁹ In 106 subjects in our study, this CES-D cutoff point had a sensitivity of 0.90 and a specificity of 0.83 for the DSM-IV diagnosis of major depression by a board-certified psychiatrist.

Subjects with a CES-D score of 16 or greater and a plasma Aβ₄₀:Aβ₄₂ ratio greater than the median (7.1) were defined as having amyloid-associated depression (n = 177). Those with a CES-D score of 16 or greater and a plasma Aβ₄₀:Aβ₄₂ ratio less than or equal to the median were defined as having nonamyloid depression (n = 171). The other subgroup included those without depression (CES-D score < 16; n = 647).

Research assistants, trained by a board-certified neuropsychologist, administered the following cognitive tests.

Subjects were classified as having CVD according to whether they had been previously informed by a physician that they had congestive heart failure, coronary heart disease, angina pectoris, or a heart attack. Diabetes mellitus was defined as the use of antidiabetic medication or fasting glucose level greater than 126 mg/dL (to convert to millimoles per liter, multiply by 0.0555) (available on 96% of the samples). Stroke history was recorded. Current hypertension was defined as the average of systolic blood pressure greater than 140 mm Hg or diastolic blood pressure greater than 90 mm Hg at 2 determinations or was considered present if the subject was taking antihypertensive medications.

A 244–base pair segment of the APOE gene (OMIM 107741), which included the 2 polymorphic sites, was amplified by polymerase chain reaction with a robotic thermal cycler (ABI 877; Applied Biosystems, Foster City, California). The polymerase chain reaction products were digested with 5 U of HhaI, and the fragments were separated by electrophoresis on 8% polyacrylamide nondenaturing gel. The specific allelic fragments were E2, E3, and E4. APOE4 was defined by E4/4, E3/4, or E2/4.²⁰ Renal function, which is associated with plasma Aβ,²¹ was assessed through measurements of serum creatinine.

Statistical analysis was performed with SAS (version 9.1; SAS Institute Inc, Cary, North Carolina). Mean with standard deviation and t test or analysis of variance were used for the variables with a normal distribution, and median (quartile 1–quartile 3 [Q1-Q3]) and Wilcoxon rank sum test or Kruskal-Wallis test were used for the variables with a skewed distribution. The χ² test was used to compare proportions for binary end points. Subjects' data were also divided into plasma Aβ quartiles and stratified by depression. Linear regression was used to examine associations between different cognitive domains as an outcome and the interaction between depression and plasma Aβ₄₀:Aβ₄₂ ratio or amyloid-associated depression or nonamyloid depression while adjusting for potential confounders including age, race, sex, school, creatinine level, APOE4, vascular diseases, and antidepressant medication use. All Aβ₄₀, Aβ₄₂, and Aβ₄₀:Aβ₄₂ ratio values were transformed to log₁₀ (logAβ). These analyses were also stratified by the cognitive scores (MMSE score ≤ 23 and > 23; LM Delayed Recall score ≤ 14.6 and > 14.6) to observe the relationships in subjects with and without significant cognitive impairment. The correlations between logAβ₄₂ and the different cognitive scores were analyzed by Spearman correlation in each quartile of plasma Aβ₄₀ stratified by depression status. Because Bonferroni correction was applied for regression analyses, the 2-sided significance level of P < .0167 was used.

Nine hundred ninety-five subjects with depression status and plasma Aβ measurements from the Nutrition, Aging, and Memory in the Elderly study were included in this analysis. The average age of this sample was 75.3 (SD, 8.4) years, and 757 subjects (76.1%) were female. The sample was multiethnic, with 607 (61.0%) white, 375 (37.7%) African American, and 13 (1.3%) other ethnicities. Of 989 subjects who provided information on education, 641 (64.8%) had completed high school.

Depression, defined as a CES-D score of 16 or greater, was observed in 348 of the subjects (35.0%). Subjects with depression were younger (mean [SD] age, 73.8 [8.5] vs 76.0 [8.3]; P < .001) and tended to have less education than those without depression, whereas there were no differences in sex and ethnicities between the groups (Table 1). Medically, the depression subgroup had similar APOE4 frequencies, lipid profile, and rates of hypertension, stroke, and diabetes mellitus. Those with depression had higher rates of CVD (51.5% vs 38.6%; P < .001) and were more likely to be taking antidepressants (40.1% vs 21.8%; P < .001) than were those without depression. In addition, subjects with depression tended to have a slightly higher level of creatinine than did those without depression.

Distributions of plasma Aβ peptides were skewed, showing the following levels: Aβ₄₀: median, 132.7 pg/mL; minimum, 1.6 pg/mL; maximum, 1324.9 pg/mL; Aβ₄₂: median, 18.7 pg/mL; minimum, 1.6 pg/mL; maximum, 780.8 pg/mL; and Aβ₄₀:Aβ₄₂ ratio: median, 7.1; minimum, 0.04; maximum, 86.0. Subjects with depression had a lower concentration of plasma Aβ₄₂ (median, 17.1 vs 19.4 pg/mL; P = .02) than did those without depression, but there were no differences in plasma Aβ₄₀ level and Aβ₄₀:Aβ₄₂ ratio between those with and without depression (Table 1). When the subjects with CVD and antidepressant use were removed, those with depression had significantly higher plasma Aβ₄₀:Aβ₄₂ ratios (median, 8.9 vs 6.4; P < .001), lower concentrations of plasma Aβ₄₂ (median, 14.1 vs 19.2 pg/mL; P = .006), and a tendency toward higher concentrations of plasma Aβ₄₀ (median, 138.7 vs 125.9 pg/mL; P = .06) than did those without depression. This result was consistent with our previously published study.^{4 ,22}

In this study sample, subjects with depression showed significantly lower scores in every cognitive domain than did those without depression (data not shown). With multivariate linear regression analysis after adjusting for potential confounders including age, education, the APOE4 allele, and medical conditions, the interaction between depression and a high plasma Aβ₄₀:Aβ₄₂ ratio was found to be associated with cognitive impairment (Table 2), especially with memory (LM Delayed Recall: β = −1.9, SE = 0.7, P = .006) and tended to be associated with language (Verbal Fluency: β = −1.8, SE = 0.9, P = .05) and executive function (Trails B: β = 9.0, SE = 5.7, P = .12). Because plasma Aβ is associated with the preclinical stage of AD^{1 - 2} but the association disappears once the disease occurs,⁵ we expected that these relationships should be different between those with and without significant cognitive impairment if these associations truly represent a prodromal stage of AD. As shown in Table 2, the interaction between depression and plasma Aβ₄₀:Aβ₄₂ ratio remained associated with cognitive impairment among those with an MMSE score greater than 23 or with an LM Delayed Recall score greater than 14.6 (the norm −1.5 SDs), cutoffs used to exclude cases with severe cognitive impairment including potential cases with dementia in the study sample²³ ; however, the relationships disappeared among those with an MMSE score of 23 or less or with an LM Delayed Recall score of 14.6 or less. Adding cholesterol and high-density lipoprotein cholesterol levels into the regression models did not influence the relationships (data not shown).

These relationships were further visualized by dividing subjects into 4 quartiles by plasma Aβ₄₀:Aβ₄₂ ratio (Figure). Among subjects who had current depression, LM Delayed Recall scores decreased with every increase of plasma Aβ₄₀:Aβ₄₂ quartile, with a P value of .03 (Figure, A), and this relationship became more significant (mean [SE]: Q1, 18.6 [1.4]; Q2, 18.6 [1.3]; Q3, 14.3 [1.4]; Q4, 13.7 [1.0]; P = .007) when the subjects who were using antidepressants were removed (Figure, B). Word Learning List Delayed Recall, which assessed not only memory but also verbal learning, and the other cognitive scores tended to decrease with the increase in each Aβ₄₀:Aβ₄₂ quartile but did not reach statistical significance among subjects with depression (Figure, C-E). In contrast, at the cross-sectional level, mean cognitive scores were similar across all quartiles of Aβ₄₀:Aβ₄₂ in those without depression (Figure).

To investigate the interplaying role of both peptides, the relationship between plasma Aβ₄₂ and cognition was further analyzed within each Aβ₄₀ quartile. Among the depressed subjects in the highest plasma Aβ₄₀ quartile (Table 3), plasma Aβ₄₂ concentration was positively correlated with LM Delayed Recall scores (r = 0.29, P = .005) and tended to correlate with MMSE scores (r = 0.22, P = .04) and inversely with Trails B scores (r = −0.23, P = .03), but was not correlated with verbal fluency. In contrast, all of these correlations between plasma Aβ₄₂ and cognitive functions were not observed among nondepressed subjects, even though they also had the highest quartile of plasma Aβ₄₀ (Table 3). There was no correlation between plasma Aβ₄₂ levels and cognitive function among those in the lower quartiles of plasma Aβ₄₀ regardless of depression status (data not shown).

While subjects in the fourth quartile had an Aβ₄₀:Aβ₄₂ ratio with a median equal to 14.2, which is near the levels (> 12 to 16) shown to increase the risk of AD in the 2 cohorts,^{1 - 2} both the fourth and third quartiles showed significantly lower LM Delayed Recall scores than the first or second quartile of Aβ₄₀:Aβ₄₂ among subjects with depression (Figure). We therefore used the Aβ₄₀:Aβ₄₂ median of the whole sample to define amyloid-associated depression as the following: (1) no depression: CES-D score less than 16; (2) amyloid-associated depression: CES-D score of 16 or greater and Aβ₄₀:Aβ₄₂ greater than the median; and (3) nonamyloid depression: CES-D score of 16 or greater and Aβ₄₀:Aβ₄₂ ratio less than or equal to the median.

While no differences were found in any of the other variables between the 2 depression subgroups, subjects with amyloid-associated depression were older (mean [SD] age, 74.9 [8.5] vs 72.7 [8.3] years; P = .01), had a slightly higher level of creatinine (mean [SD], 1.3 [1.2] vs 1.1 [1.0] mg/dL; P = .01) (to convert creatinine to micromoles per liter, multiply by 88.4), and were less likely to take antidepressants (32.4% vs 46.2%; P = .009) than were those with nonamyloid depression (Table 4).

Although there was no difference in CES-D scores between the 2 subgroups, those with amyloid-associated depression had poorer memory scores in the LM Delayed Recall (mean [SD], 15.9 [9.6] vs 18.9 [9.2]; P = .004) and tended to have lower Word Learning List Delayed Recall scores (mean [SD], 3.0 [2.7] vs 3.6 [2.6]; P = .03), lower language ability (Verbal Fluency mean [SD], 23.9 [10.9] vs 26.5 [11.7]; P = .03), and lower executive function (Trails B test mean [SD], 236.0 [77.8] vs 214.1 [84.3]; P = .01) than did those with nonamyloid depression (Table 4). In contrast, scores evaluating general cognition measured by either the MMSE or Digit Symbol and scores on visuospatial function measured by Block Design were similar between those with amyloid-associated depression and those with nonamyloiddepression.

To further test the existence of the foregoing depression subtypes, multivariate linear regression analysis was applied to investigate the relationship between each cognitive domain and amyloid-associated depression vs nonamyloid depression relative to those without depression, after adjusting for potential confounders. Among the subjects with amyloid-associated depression and those without depression combined, amyloid-associated depression was associated with poor memory (LM Delayed Recall: β estimate = −1.9, SE = 0.4; P < .001), visuospatial dysfunction (Block Design: β estimate = −1.4, SE = 0.4; P < .001) (Table 5), and executive dysfunction (Trails B: β estimate = 16.0, SE = 3.3; P < .001) after adjusting for the confounders including antidepressant use and others. Although subjects with nonamyloid depression had a significantly higher level of plasma Aβ₄₂ (median, 30.7 vs 19.2 pg/mL; P < .001), but not Aβ₄₀, than those without depression, nonamyloid depression was not associated with memory performance (LM Delayed Recall: β estimate = −0.9, SE = 0.8; P = .30) but was related to visuospatial (Block Design: β estimate = −3.0, SE = 0.8; P < .001) (Table 5) and executive (Trails B: β estimate = 21.2, SE = 6.8; P = .002) dysfunction after adjusting for these confounders. Adding covariables of lipids and statins did not affect these relationships (data not shown).

The relationships between plasma Aβ level, cognitive function, and depression remain complex. The present study suggests that there may be at least 2 depression subtypes from the cognitive perspective: (1) amyloid-associated depression, which is associated with poor memory and other cognitive dysfunction (Table 2 and Figure), and (2) nonamyloid depression, which is associated with only visuospatial and executive dysfunction (Tables 4 and 5). Because a high plasma Aβ₄₀:Aβ₄₂ ratio increases the risk of AD prospectively,^{1 - 2} we hypothesize that amyloid-associated depression is more likely to be a prodromal stage of AD than is a high Aβ₄₀:Aβ₄₂ ratio without depression. Antidepressants, specifically selective serotonin reuptake inhibitors, have been shown to be associated with lower plasma Aβ₄₀ levels.²² Consistently, this study showed that fewer subjects with amyloid-associated depression took antidepressants than did those with nonamyloid depression (Table 4).

There is growing evidence in the literature that depression may be either a risk factor for AD^{3 ,24 - 27} or an early symptom of AD.^{12 ,28 - 32} A neuropathology study has shown that history of depression is associated with increased amyloid plaques and neurofibrillary tangles, which are the neuropathological hallmarks of AD.³³ The pattern of cognitive impairment, prominently poor memory, in amyloid-associated depression (Table 4 and 5) was consistent with mild cognitive impairment (MCI),²³ which is presumed to be a prodromal stage of AD.^{27 ,34} Although MCI is associated with depression, depression is also found to increase the risk of developing MCI.^{32 ,35 - 36}

The severity of depressive symptoms was correlated with lower concentration of plasma Aβ₄₂, a determining factor of Aβ₄₀:Aβ₄₂ ratio, only when CVD cases were excluded from the study sample.⁴ Most likely there are different subtypes of depression in the elderly population, including (1) early-onset depression, (2) poststroke depression,³⁷ (3) vascular depression related to CVD and other vascular risk factors that lead to executive dysfunction,³⁸ (4) preclinical depression of AD, and (5) comorbid depression in AD. Because these depression subtypes have different underlying pathological characteristics and prognoses, not all should be related to plasma Aβ levels, or they may be related to plasma Aβ levels differently. In contrast to amyloid-associated depression, nonamyloid depression was not found to be associated with memory (Table 5) even though these subjects had higher levels of plasma Aβ₄₂ than did those without depression. Although the rates of CVD (92 of 171 [53.8%] vs 82 of 177 [46.3%]; P = .32) and stroke (31 of 171 [18.1%] vs 35 of 177 [19.8%]; P = .89) were similar between subjects with nonamyloid depression and those with amyloid-associated depression, cerebral microvascular pathological data in these cases is unknown. Regardless, nonamyloid depression was strongly associated with a cognitive pattern of vascular depression and visuospatial and executive dysfunction (Table 5). This and other factors may explain the seemingly conflicting result in another study in which depressed patients had a higher, not lower, level of plasma Aβ₄₂ than the controls.^{39 - 40} In fact, our own study has also shown that vascular depression related to CVD is not associated with lower plasma Aβ₄₂ levels.⁴

It is intriguing to find that the combination of 2 peptides in plasma, a high level of Aβ₄₀ and a low level of Aβ₄₂, was associated with poor memory, whereas each peptide alone was not found to have this relationship in the regression analysis (Table 3). The peptides Aβ₄₂, a major component of AD pathological findings in the brain,⁴¹ and Aβ₄₀, a component of cerebral amyloid angiopathy,⁴² are produced by the processing of APP. Although plasma Aβs reflect dynamic levels governed by both peripheral^{21 ,43 - 46} and central⁴⁷ nervous system origins, these 2 peptides in plasma have been linked to pathological conditions in the brain. First, high plasma Aβ₄₀ level is associated with cerebral microvascular pathological changes, white matter hyperintensities, and lacunar infarct.^{9 - 10} Both white matter hyperintensities and lacunar infarcts are linked to cognitive impairment,^{48 - 49} dementia incidence,⁵⁰ and depression⁴⁹ in the elderly population. Second, studies using APP transgenic mice have demonstrated that plasma Aβ₄₂ level declines significantly before Aβ₄₂ is deposited in the brain to form the AD changes.^{7 - 8} Therefore, a high Aβ₄₀:Aβ₄₂ ratio in plasma may be a biomarker to indicate cerebral microvascular pathological changes, which are associated with high plasma Aβ₄₀ level, coexisting with the AD pathological features, which may be linked to plasma Aβ₄₂ decline at the preclinical stage. These 2 pathological conditions in the brain additively or synergistically result in more severe cognitive dysfunction than either condition alone.^{51 - 52}

Plasma Aβ level correlates with cerebrospinal fluid Aβ concentration in APP transgenic mice when they are young and pathological changes of AD have not yet formed. However, this correlation disappears as the animals age and Aβ is deposited in the brain.⁴⁷ Similarly in humans, although plasma Aβ level cannot be used for the diagnosis of AD,^{5 ,18 ,53 - 54} it is still possible that plasma Aβ level can be used as a screening tool for the risk of AD. In this study, the relationship between poor memory and a high plasma Aβ₄₀:Aβ₄₂ ratio in depression was observed among subjects with normal cognition or MCI, but the relationship disappeared among those with severe cognitive impairment (Table 2). Two large population studies, mainly containing patients with late-onset AD, have shown that a high plasma Aβ₄₀:Aβ₄₂ ratio, determined by both low Aβ₄₂ and high Aβ₄₀ levels, increases the risk of AD prospectively.^{1 - 2} Although another population study reports that high plasma Aβ₄₂ level is associated with the risk of AD,⁵⁵ plasma Aβ₄₂ level declines in these subjects before the onset of the disease, leading to low levels of plasma Aβ₄₂ at the prodromal stage. Unlike late-onset AD, both early-onset AD⁵⁶ and Down syndrome^{57 - 59} present with high plasma Aβ₄₂ level at the preclinical stage. Another study shows that patients with MCI have higher levels of plasma Aβ₄₂ than do controls but only in women.⁶⁰ It is not yet known whether plasma Aβ status is different at the preclinical stage of AD with and without depression. Because only a subset of the patients with late-onset AD present with depression at the start of the disease, our data suggest that the combination of depressive symptoms and plasma Aβ level may more specifically inform the prodromal stage of AD than does either one status alone.

Although our study found that amyloid-associated depression presented with poor memory, and others hypothesized that elevated plasma Aβ₄₂ level in depression may contribute to AD,⁶¹ without a longitudinal study we cannot yet conclude that amyloid-associated depression vs nonamyloid depression is a precursor of AD. We have used different analytical methods and applied Bonferroni correction to prevent against type I error, but the cross-sectional design is still the major limitation of our study. Other limitations include the following: (1) depression was based on the CES-D score rather than the DSM-IV criteria, and we had no information about the onset and the course; (2) some variables, such as CVD and stroke, were self-reported; (3) the complexity of plasma Aβ^{62 - 63} requires brain investigation to validate amyloid-associated depression as a unique depression subtype; (4) although our assays in measurements of Aβ₄₀ vs Aβ₄₂ in plasma had high sensitivity and specificity, the field lacks standard assays for interlaboratory comparisons; and (5) the correlation coefficients shown in Table 3 were significant, but probably not large enough to be applied in clinical practice. Nevertheless, our discoveries have found the relationship between depression severity and low plasma Aβ₄₂ level⁴ and further have linked a high plasma Aβ₄₀:Aβ₄₂ ratio in depression with poor memory, especially among those with normal or mildly impaired cognition. Our findings warrant prospective studies to examine whether amyloid-associated depression is related to pathological changes in the brain and predict the onset of cognitive decline and AD in this and other populations.

Correspondence: Wei Qiao Qiu, MD, PhD, Department of Psychiatry, Tufts–New England Medical Center, Campus Box 1007, 750 Washington St, Boston, MA 02111 (wqiu@tufts-nemc.org).

Submitted for Publication: August 15, 2007; final revision received November 29, 2007; accepted November 30, 2007.

Financial Disclosure: None reported.

Funding/Support: This study was supported by grants AG-022476 (Dr Qiu) and AG-21790 (Dr Folstein) from the National Institute on Aging. Support was also provided through the General Clinical Research Center funded by the National Center for Research Resources of the National Institutes of Health under grant MO1-RR00054. Dr Steffens was supported by grant MH70027 from the National Institutes of Mental Health.

Additional Contributions: We thank Dennis J. Selkoe, MD, for his support and collaboration and the Nutrition, Aging, and Memory in Elders study staff and the Boston home care agencies for their hard work and acquisition of subjects.