Introduction

Age-related cognitive decline is a complex convergent phenotype that likely derives in part from brain senescence and in part from prodromal dementing illnesses, most commonly Alzheimer's disease (AD). Brain-derived neurotrophic factor (BDNF) is an activity-dependent secreted protein that, along with its receptors, is expressed widely in the central nervous system and is critical to organization of neuronal networks and synaptic plasticity, especially in the hippocampus, in a variety of animal models and apparently in humans [1], [2]. Although controversy remains, these human data derive mostly from investigations of an allelic variant of BDNF (Val66Met or Met-BDNF), inheritance of which has been associated with poorer cognitive performance in healthy older adults [3], impaired memory in patients with schizophrenia and healthy controls [1], [2], abnormal hippocampal activation as assessed by fMRI [2], and an approximately 4% to 11% smaller hippocampal volume as determined by MRI in healthy adult volunteers [3], [4]. The mechanisms that underlie these associations of functional and structural differences with inheritance of Met-BDNF are not clear; however, one study has shown diminished depolarization-induced secretion of Met-BDNF compared to Val-BDNF, and failure of Met-BDNF to localize to secretory granules or synapses in transfected neurons [2]. Together, these findings have led to the hypothesis that reduced BDNF secretion is one mechanism of age-related cognitive decline. We are unaware of any published study that has yet tested this hypothesis.

Methods

All procedures were approved by the institutional review boards of the University of Washington; all subjects were recruited from University of Washington Alzheimer's Disease Research Center and provided written informed consent. Subjects underwent detailed clinical and laboratory evaluation and were classified as no cognitive impairment (Normals), amnestic mild cognitive impairment (MCI) [5], or probable AD [6]. Inclusion and exclusion criteria, method of CSF collection, and analysis were exactly as previously described [7].

CSF concentration of BDNF, total tau and Aβ₄₂ was measured using an X-MAP-based assay [7]. APOE genotype was determined by a restriction digest method [8]. BDNF SNP rs6265 was genotyped using TaqMan allelic discrimination detection, as previously described [9].

Neuropsychological tests included: (i) Paragraph Recall - a test of declarative memory [10], [11]. Paragraphs were modeled after the Logical Memory subtests I and II of the WMS-R [12]. Total score for immediate recall and delayed recall (each with possible range 0–25) were used. (ii) Category Fluency - a test of semantic memory [13]. Total number of unique animals named in 60 seconds was used. (iii) Trail Making Test, Parts A and B - test of ability to adapt to shifting task demands. Time taken to complete Part B (upper bound of 300 sec), a measure of executive function [14], was used.

One-way analysis of variance (ANOVA) was used with post hoc Bonferroni multiple pair-wise comparisons for continuous variables and Fisher's exact test for categorical variables to assess differences between diagnostic and genetic groups. Linear regression models were used for associations of demographic characteristics with CSF BDNF concentration, and cross-sectional relationships between CSF BDNF concentration and coincident cognitive test performance. Raw scores were used for each test except log-transformed times for Trails B to remove skewness. We used two-stage regression (least squares slope for each test in each individual over time, then weighted regression model with slope as response variable) to assess association of baseline CSF BDNF concentration with subsequent longitudinal changes of cognitive test performance [15]. Weights were based on subjects having different numbers of follow-up visits at different times after baseline. Statistical analyses were performed using S-PLUS version 8.0 [16] and R version 2.7.1 [17].

Results

Age-matched Normals, MCI, and AD

Table 1 presents demographics and baseline CSF BDNF levels in 128 normal controls (Normals), 9 MCI and 21 AD patients. To compare measures between these three groups after adjusting for age, a subset of the Normals (n = 76) age≥50 years (Older Normals, ON) was used as a comparison group. CSF BDNF level was lower in AD patients compared to ON with a mean difference of 25 pg/ml (post-hoc Bonferroni test, p = 0.02). CSF BDNF level for the nine MCI patients was not different from either group (both p>0.05). The frequencies of Met-BDNF genotype (G/G vs. G/A or A/A) were not different between ON, MCI, and AD subjects (Fisher's Exact Test p = 0.36). Moreover, CSF BDNF concentration in AD subjects was not different among Met-BDNF genotype G/G (70%, 217±38 pg/ml), G/A (25%, 214±39 pg/ml), or A/A (5%, 213 pg/ml, n = 1) (F = 0.02, df = 2, p = 0.99).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Subject characteristics and CSF BDNF levels at baseline.

https://doi.org/10.1371/journal.pone.0005424.t001

CSF BDNF concentration in all cognitively normal subjects

Our focus was on 128 cognitively normal volunteers (Normals) in whom we measured CSF BDNF, Aβ₄₂, and total tau, and who underwent neuropsychological evaluation (Table 1). CSF BDNF concentration was not different among rs6265 genotype in Normals: G/G (60%, 242±37 pg/ml), G/A (36%, 253±26 pg/ml), or A/A (5%, 251±28 pg/ml) (F = 1.52, df = 2, p = 0.22).

CSF BDNF concentration decreased with age among Normals and was higher in women than men (both p<0.001; Figure 1). In a multivariate linear regression with CSF BDNF as the dependent variable and age, gender, presence of the APOE ε4 allele, and Met-BDNF (G/G vs G/A or A/A) as independent predictors, both age and gender were significantly associated with CSF BDNF level (p<0.001). Neither presence of APOE ε4 allele (p = 0.33) nor Met-BDNF (p = 0.09) was related to CSF BDNF levels. In addition, there was no interaction between APOE ε4 allele and Met-BDNF genotype (p = 0.25), nor between age and gender (p = 0.45). Removing the three lowest values of BDNF for males did not change the results. Finally, after adjusting for age and gender, CSF BDNF levels were related inversely (slope = −0.03) to CSF levels of Aβ₄₂ (p = 0.05) but not to CSF total tau (p = 0.34).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Cross-sectional relationships between age and CSF BDNF concentration by gender.

https://doi.org/10.1371/journal.pone.0005424.g001

Association between CSF BDNF concentration, memory, and cognitive ability in cognitively normal subjects at baseline.

Most (n = 121) Normals underwent extensive neuropsychological testing at baseline (Table 1). Table 2 shows the regression coefficients and p-values associated with baseline CSF BDNF levels for each cognitive test in multivariate regression models after adjusting for age, gender, and years of education (Model 1); adjusting for the previous variables as well as concentration of CSF Aβ₄₂ and total tau (biomarkers of latent and prodromal AD [18], [19], Model 2); and adjusting for the previous variables as well as presence of APOE ε4 and Met-BDNF (Model 3). There was an association between higher CSF BDNF concentrations and better performance on Paragraph Recall - Delayed that was independent of CSF Aβ₄₂ and total tau concentrations (Model 2) as well as presence of APOE ε4 and Met-BDNF (Model 3); correlation with the Paragraph Recall - Immediate was significant when all predictor variables were in the model (Model 3). An association between lower CSF BDNF and poor performance on Trail Making Test Part B also was observed. There was no significant correlation between CSF BDNF concentration and performance on the Category Fluency test.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Relationships between baseline CSF BDNF concentration with cross-sectional and longitudinal cognitive performance.

https://doi.org/10.1371/journal.pone.0005424.t002

Association between baseline CSF BDNF concentration and longitudinal changes in cognitive performance.

Fifty Normals (27 women), aged 40–100 years (mean±SD = 71.5±8.9) had at least one clinical follow-up visit (mean±SD 4.0±1.1, Range = 2–6) with an average length of follow-up of 3.3 years (SD = 1.2 years, Range = 0.7–5.8 years) (Table 1). Figure 2 shows unadjusted annual changes in cognitive test scores versus baseline CSF BDNF concentration. After adjusting for age, gender, years of education, and baseline test score, lower values of baseline BDNF concentration were significantly associated with greater annual decline in Immediate and Delayed Recall scores on the Paragraph Recall test, and a greater annual decline in the Category Fluency test (Model 4). These relationships were unchanged after further adjusting for baseline CSF Aβ₄₂ and total tau concentrations (Model 5), suggesting that prediction by baseline CSF BDNF concentration for subsequent cognitive changes is independent of preclinical AD. These relationships also were unchanged after further adjusting for presence of APOE ε4 and Met-BDNF (Model 6). All results were unaffected by removing the subject who had the largest annual decline for Immediate Recall, Delayed Recall, and Category Fluency (Figure 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Annual change in cognitive test scores versus baseline CSF BDNF concentration.

https://doi.org/10.1371/journal.pone.0005424.g002

Discussion

We made three novel observations in cognitively normal individuals. (i) CSF BDNF decreased across the human life-span in the absence of dementia or MCI and was independent of inheritance of Met-BDNF or the APOE ε4 allele. (ii) Women had higher average CSF BDNF concentrations than men. (iii) Lower CSF BDNF concentration was associated strongly with poorer memory and less so with diminished executive function; importantly, these associations were independent of CSF biomarkers of preclinical AD, suggesting that the mechanisms that contribute to early AD and to age-related decline in CSF BDNF might be independent. We must acknowledge that this study has a relatively small sample size and short duration of follow-up and further studies are needed to validate our results.

While ours is the first study of which we are aware to investigate age-related changes in CSF BDNF, other studies have investigated disease-associated changes in CSF BDNF in patients with idiopathic Parkinson disease [7], [20] or in patients with AD [7], [21], [22]. We confirmed that CSF BDNF is further reduced beyond age-related decline in patients with probable AD. CSF and serum BDNF concentrations do not correlate [23]; yet, increased and decreased serum BDNF levels are related to early and late stages, respectively, of AD [24].

The gender difference in CSF BDNF observed in this study is especially interesting, and we speculate that this may be due to hormonal effects. Animal studies have shown that estrogen receptors colocalize to cells that express BDNF and its receptor trkB, and estrogen regulates the expression of BDNF [25]. The relationship between estrogen levels and BDNF expression and secretion, and their potential effect on the cognition in humans requires further studied.

Our data showed that progressive decline in CSF BDNF concentration was a feature of advancing age independent of preclinical AD or dementia, and was associated strongly with reduced performance in declarative memory tests and less strongly with performance on tests of executive function, perhaps a reflection of the special role of BDNF in hippocampal function. Age-related reduction in CSF BDNF was independent of inheritance of Met-BDNF or APOE ε4. Further reduction in CSF BDNF occurred in AD. Our data suggest that reduced secretion of BDNF in the central nervous system is one mechanism that may contribute to age-related cognitive decline.