Introduction

Approximately 46.8 million people suffer from dementia worldwide, with an accompanying total estimated global cost at 818 billion dollars. Because the pathological changes related to Alzheimer’s disease (AD) are supposed to begin from 10 to 15 years before the onset of memory decline or dementia1, establishment of peripheral biomarkers that enable early detection of dementia or memory decline are urgently needed2,3.

Brain-derived neurotrophic factor (BDNF), a member of the neurotrophin family, has various important roles in neuronal differentiation and survival, neurite outgrowth, gene expression and synaptic plasticity in the rodent brain4,5,6. BDNF is also important for memory formation in rodents7. Moreover, BDNF is first synthesized as a precursor proBDNF protein. ProBDNF is cleaved to mature BDNF in the intracellular endoplasmic reticulum or extracellularly by proteases such as plasmin and matrix metalloproteases8. Interestingly, the binding of BDNF to its cognate tropomysin related kinase B (TrkB) receptors supports neuronal survival, whereas binding of proBDNF to p75 neurotrophin receptor (NTR) leads to apoptosis, indicating that proBDNF and mature BDNF elicit opposing neuronal responses in the rodent brain9,10. In aged mice, the expression of proBDNF is increased in the hippocampus relative to young mice and intra-hippocampal infusions of proBDNF lead to a progressive and significant impairment of memory function11.

In rodents, BDNF can cross the blood–brain barrier (BBB)12 and serum levels of BDNF are shown to well correlate with brain-tissue BDNF levels13. Although some reports showed serum levels of BDNF did not correlate with CSF levels of BDNF in AD patients14, BDNF is supposed to be one of peripheral biomarkers that enable early detection of dementia or memory decline15. In humans, accumulating evidence shows that hypofunction of BDNF signaling is associated with neurodegenerative diseases, including Parkinson's disease16 and AD17,18,19. However, it remains unclear how both proBDNF and BDNF are involved in memory decline and/or the onset of dementia among older adults17. To our knowledge, this is the first report to examine whether peripheral levels of both BDNF and proBDNF are associated with memory function in older adults with or without dementia. Our research might be useful for testing the hypothesis that proBDNF and mature BDNF elicit opposing roles on memory function in older adults.

Results

Background characteristics

Because the distribution in levels of BDNF and proBDNF were highly skewed, the log-transformed BDNF (log10 BDNF) and proBDNF (log10 proBDNF) values were used for statistical analysis. Log10 proBDNF and log10 BDNF correlated well (r = 0.290, p < 0.001, Fig. 1) in the same populations. The characteristics of the study population among tertiles of BDNF values are provided in Table 1. The lowest tertile of BDNF was associated with higher age, fewer cases of metabolic syndrome, lower levels of RBMT score, diastolic blood pressure, albumin, Hemoglobin A1c, low density lipoprotein, and estimated glomerular filtration rate.

Figure 1
figure 1

Log10 proBDNF and log10 BDNF correlated well (r = 0.290, p < 0.001) in our populations.

Full size image
Table 1 Characteristics of the study population.
Full size table

The RBMT score was negatively correlated with hippocampal atrophy (i.e., higher ZAdvance score) (Pearson correlation coefficient r = 0.380, p < 0.001), while the RBMT score was positively correlated with BDNF (r = 0.203, p = 0.001) (Fig. 2A, B). Hippocampal atrophy and low BDNF levels were synergistically correlated with memory dysfunction (Fig. 2C).

Figure 2
figure 2

Hippocampal atrophy and lower BDNF levels were independently associated with memory impairment determined by the Rivermead Behavioral Memory Test (RBMT). (A) The RBMT score was negatively correlated with hippocampal atrophy (i.e., higher ZAdvance score) (Pearson correlation coefficient r = 0.380, p < 0.001). (B) The RBMT score was positively correlated with BDNF (r = 0.203, p = 0.001). (C) Hippocampal atrophy and low BDNF levels were synergistically correlated with memory dysfunction. The different colors—blue, red, and green—were used to indicate the groups with low, medium, and high tertiles of BDNF levels, respectively.

Full size image

Logistic regression analysis

Multivariate analysis was carried out with logistic regression analysis for cognitive function as the dependent variable and the log10 BDNF value, age, sex, education, physical activity, sport index, and MRI findings as the independent variables. When possible confounders were entered into the binary logistic regression model (the forward stepwise method), the independent predictors of memory dysfunction (RBMT < 17) were age (OR = 2.575/10 years; 95% CI 1.540–4.305; p < 0.001), log10 BDNF (OR = 0.546; 95% CI 0.299–0.999; p = 0.050), sport activity (OR = 0.346; 95% CI 0.134–0.895; p = 0.029), and hippocampal atrophy (OR = 0.546/z score; 95% CI 0.299–0.999; p = 0.050); MRI findings such as silent brain infarction, white matter lesions, and cerebral microbleeds did not enter into the equation via the forward stepwise procedure, while the independent predictors of executive dysfunction (modified Stroop test ≥ 27 s) were age (OR = 4.272; 95% CI 2.604–7.009; p < 0.001), and sport activity (OR = 0.391; 95% CI 0.154–0.989; p = 0.047) (Table 2).

Table 2 Potential correlating factors for cognitive function.
Full size table

Structural equation modeling

The findings mentioned above led us to the hypothesis that age, sport activity, hippocampal atrophy and lower BDNF levels caused memory dysfunction, while hippocampal atrophy and low BDNF levels were not associated with executive dysfunction. We investigated the relationship between cognitive function tests, hippocampal atrophy, sport activity, and the log10 BDNF value, using a graphical multivariate analysis SEM. Path analysis based on SEM indicated that the direct paths from age (β = 0.290, p < 0.001), sport (β = 0.186, p = 0.001), hippocampal atrophy (ZAdvance) (β = 0.142, p = 0.019), and log10 BDNF (β = 0.138, p = 0.015) to memory dysfunction (Rivermead) were significant (Fig. 3). Although the direct path from age to executive dysfunction (Stroop) were significant (β =  − 0.426, p < 0.001), the direct paths from log10 BDNF to hippocampal atrophy and executive dysfunction were not significant. These results were quite similar to those from the logistic regression analysis shown in Table 2. The direct path from log10 BDNF to hippocampal atrophy was not significant, while the direct paths from age and education to hippocampal atrophy were significant (β = 0.321, p < 0.001 and β =  − 0.188, p = 0.004, respectively), as had been shown in our previous study20. The measures of model fitness were as follows: chi-square N.S., GFI = 0.986, AGFI = 0.961, CFI = 0.992, and RMSEA = 0.026. Thus, the presented model reasonably fit the data.

Figure 3
figure 3

Path analysis showing that age, sport activity, hippocampal atrophy and BDNF were individually associated with Rivermead Behavioral Memory Test scores. Path analysis based on structural equation modeling indicated that the direct paths from age (β =  −0.321, p < 0.001), sport (β = 0.191, p < 0.001), hippocampal atrophy (ZAdvance) (β =  −0.220, p < 0.001), and log10 BDNF (β =  −0.150, p < 0.01) to memory dysfunction (Rivermead) were significant.

Full size image

Discussion

In the present cross-sectional study, logistic regression analysis revealed that age, sport activity, hippocampal atrophy, and BDNF were independently associated with memory function determined by RBMT. Path analysis based on SEM indicated that age, sport activity, hippocampal atrophy and BDNF but not proBDNF were individually associated with RBMT. These findings suggested that physical inactivity, hippocampal atrophy and hypofunction of BDNF were individually associated with age-related memory impairment; these may all be strategic targets for dementia prevention.

To our knowledge, this is the first report to examine whether peripheral levels of both BDNF and proBDNF are associated with memory function in older adults without dementia.

ProBDNF modulates synaptic plasticity, including the induction or enhancement of long-term depression21, through the inhibition of dendritic outgrowth in rodent hippocampus22. In addition, proBDNF increases the amount of amyloid β (Aβ) deposition and induces memory dysfunction in mice models of AD23. In this study, however, SEM revealed that proBDNF was not directly associated with memory dysfunction in community-dwelling older adults. BDNF modulates synaptic plasticity, including the induction or enhancement of long-term potentiation24, through the promotion of neurogenesis25 and/or dendritic outgrowth26,27 in rodent hippocampus. These findings are consistent with our observation showing that hypofunction of BDNF was associated with age-related memory impairment. In addition, exercise enhances memory function in both rodents28 and human29. These effects are accompanied by increased neuronal proliferation or survival and enhancement of dendritic outgrowth in the hippocampus30, mainly mediated by BDNF signaling31,32. Thus, we could not dissociate the potential effects of lower BDNF and lower physical exercise on memory impairments in the present study. In healthy young college students, exercise (a short period of high-intensity cycling) results in the enhancement of hippocampus-related memory function (face–name matching) without affecting the Stroop word–color test, which was accompanied by increased serum BDNF levels33. Likewise, Frerris et al. reported that exercise (graded exercise test) improved the Stroop word–color test, but they could not find any significant correlation between the improvement of the Stroop test and an increase in BDNF34. These reports are compatible with our findings that BDNF was associated with RBMT but not the Stroop test. In contrast, Giacobbo et al. found a significant correlation between scores of the Stroop test and serum concentration of BDNF in young adults who were deprived of sleep35. The Stroop word–color test is shown to recruit the anterior cingulate cortex and other frontal cortical regions36. We need further investigations with a larger sample size to clarify the relationship between BDNF and the Stroop test in older adults.

Apathy is supposed to be a risk factor37 or prodromal symptom38 of AD. Alvarez et al. have reported that apathy is associated with lower levels of BDNF in AD patients; in that study, apathy was evaluated using the Neuropsychiatric Inventory39. In the present study, however, SEM analysis revealed that both BDNF and proBDNF were not associated with the apathy scale in community-dwelling older adults (unpublished observation). These discrepancies might be due to differences in the evaluation methods for apathy; we used Starkstein’s apathy scale in this study. In addition, we recently reported that low-grade inflammation was associated with apathy indirectly via confluent DWMLs40. Thus, low-grade inflammation rather than BDNF signaling might be associated with apathy in older adults.

We noticed some limitations 1) cross-sectional study design limited our interpretation of results regarding the cause and effect and 2) we could not exclude the possible residual confounding factors related to the SEM analysis. In addition, we used maximum likelihood estimation method by converting values of sport activity and ZAdvance to logarithm and by dividing scores of Rivermead and Stroop tests to quintile. As a result, we could reach multivariate kurtosis value to 4.143, although it did not mean normality (< 1), but moderate non-normality (< 10) by excluding proBDNF from the SEM analysis. Although the strength of our study is its measurements of both proBDNF and BDNF in each subject, the newly identified BDNF pro-peptide—proBDNF is cleaved to BDNF and this BDNF pro-peptide41—was not measured, leaving the possibility that BDNF pro-peptide could interfere with the statistical effects of proBDNF on memory function.

In conclusion, we suggest that physical inactivity, hippocampal atrophy and hypofunction of BDNF signaling individually associate with age-related memory dysfunction and they might all be targets to prevent dementia.

Methods

Participants and protocol approval

Between 2010 and 2016, we performed a cross-sectional observational study in the rural community of Sefuri village (Saga, Japan), which had a total population of 1739 people as of April 201440. We examined 297 consecutive volunteers aged 60–89 years, who were independent in their daily life without apparent dementia. Subjects apparently not eligible for this study were excluded; these were cognitive impairment (n = 7); psychiatric disorders, including depression (n = 5); claustrophobia or contraindications for MRI (n = 8); a history of stroke (n = 9); brain tumor (n = 1); chronic subdural hematoma (n = 1); a history of head trauma (n = 3); chronic renal failure (n = 3); and insufficient clinical information (n = 4). Finally, we analyzed 256 subjects in the present study. This study was approved by the National Hospital Organization Hizen Psychiatric Center Institutional Review Board approved the study (approval numbers: 15–1 and 24–4) and procedures were carried out in accordance with the approved guidelines. Written informed consent was obtained from all participants.

Clinical assessments

The participants underwent a structured clinical interview, general hematology and biochemical tests. Blood pressure was measured in the sitting position using the standard cuff method; beginning in 2013, simultaneous blood pressure measurements were recorded from both arms using a pair of automated sphygmomanometers (Omron model HEM-1020, Omron, Japan). Vascular risk factors were defined as previously described40. Briefly, arterial hypertension was considered to be present in participants with a history of repeated blood pressure recordings ≥ 140/90 mmHg, and in those being treated for hypertension. Diabetes mellitus was defined as a fasting plasma glucose level of ≥ 6.99 mmol/L (126 mg/dL) and/or HbA1c of ≥ 6.5%, or a previous diagnosis of diabetes mellitus. Hyperlipidemia was considered to be present in participants with a total serum cholesterol concentration of ≥ 5.69 mmol/L (220 mg/dL), and in those being treated for hyperlipidemia. Metabolic syndrome was defined by the presence of central obesity and a minimum of two of three factors: a blood pressure of ≥ 130/85 mmHg, a fasting blood glucose level of ≥ 6.1 mmol/L (110 mg/dL), and a triglyceride level of ≥ 1.69 mmol/L (150 mg/dL) and/or HDL cholesterol level of < 1.03 mmol/L (40 mg/dL). The estimated glomerular filtration rate (eGFR) was calculated using the Modification of Diet in Renal Disease equation for the Japanese modification as previously described37. Smoking was defined as the participant smoking an average of at least 10 cigarettes per day, while former smokers were considered nonsmokers. Alcohol use was defined as the participant reporting drinking one or more alcoholic beverages (10 g of ethanol) per week.

Physical activity

Physical activity was assessed with a questionnaire modified from the Baecke questionnaire on habitual physical activity42. The questionnaire consisted of three components: leisure time, work, and sport activities. Items concerning leisure time and work activities were coded on five-point scales. The leisure time index was assessed with four questions on comparison with others, sweat, sport, and walking. The work index was assessed through four questions on standing, walking, heavy loads, and sweat. The sport index was expressed as reported hours per week in each category multiplied by the metabolic equivalent of task (MET)43. Sport activity was defined as sport index 4.0 MET× hours/week or over.

Assessment of cognitive function

Participants were tested individually in a quiet room. All participants underwent the Rivermead Behavioral Memory Test (RBMT) for memory function assessment and the modified Stroop test for executive or frontal lobe function assessment as previously described20,44. Briefly, the RBMT consists of 11 subtests—first and second names, belonging, appointment, picture recognition, story (immediate, delay), face recognition, route (immediate, delay), messages (immediate, delay), orientation and date—with four parallel forms designed to identify memory deficits that might be encountered during daily living45. The test does not adhere to any particular theoretical model of memory; instead, it attempts to mimic the demands made on memory by normal daily life. The standard profile score of RBMT yields scores between 0 and 24 (higher scores indicate better performance); subjects with a standard profile score of RBMT less than 17 (the lowest quintile) were operationally defined as having memory impairment as previously described20. The modified Stroop test comprises two parts; subjects were asked to name the colors of dots in Part I and colors of incongruent words (Chinese characters) in Part II. The difference in time between the two parts was considered to be due to interference effects, and the subject was considered to have executive dysfunction if the total score was above the most prolonged fifth quintile (i.e., ≥ 27 s in the present study).

Apathy scale

Each item of the Starkstein apathy scale46 was quantified on a visual analog scale, where one end of a 60-mm long line is ‘absolutely correct’ and the other end is ‘completely wrong’, as previously described40,47,48. Because the item–total correlations of questions 3 (Are you concerned about your condition?) and 11 (Are you unconcerned with many things?) had been the weakest of the 14 original questions, we excluded the scores of these two questions from the analysis. This apathy scale yields total scores of 0–720, with lower scores indicating apathetic behavior. Depressed mood and insomnia were rated as ‘none’, ‘sometimes’, ‘frequent’, and ‘always’; depressed mood and insomnia were defined as an always or frequent presence of these symptoms. Patients who had been previously diagnosed with clinical depression or were taking medication for depression were excluded from the study.

Measurement of serum proBDNF and BDNF levels

Serum specimens were stored at − 80 °C until the measurements were performed. Serum levels of proBDNF and BDNF were measured using the human proBDNF enzyme-linked immunosorbent assay (ELISA) Kit (Adipo Bioscience, Santa Clara, CA, USA) and the human BDNF ELISA Kit (Adipo Bioscience), respectively. We selected these ELISA kits because some human BDNF ELISA kits frequently used in previous reports recognized not only BDNF but also proBDNF because of limited specificity49. To minimize assay variance, serum levels of proBDNF and mature BDNF from each subject were measured on the same day. All samples were analyzed in duplicate. Measurements were performed according to the manufacturer's instructions and based on our experiences as previously reported50,51,52,53.

Assessment of MRI findings

A combination of T1-weighted, T2-weighted, and fluid attenuated inversion recovery images (FLAIR) is required to accurately detect both silent brain infarction and white matter lesions54. Imaging was performed on a 1.5 T MRI scanner (Achieva, Philips, the Netherlands) using the T1- and T2-weighted, fluid-attenuated inversion recovery, and T2*-weighted images. Silent brain infarction was defined by low signal intensities on T1-weighted images, and high signal intensity areas on T2-weighted images, and a diameter of ≥ 3 mm, as previously described. We differentiated enlarged perivascular spaces from silent brain infarction based on their location, shape, and size. Lesions of < 3 mm in diameter are more likely to be perivascular space than lacunes, and the presence of moderate to severe basal ganglia perivascular space was recorded55. The white matter lesions were defined as isointense with normal brain parenchyma on T1-weighted images, and high signal intensity areas on T2-weighted images. We used the validated rating scale of DWMLs by Fazekas et al.: grade 0, absent; grade 1, punctate foci; grade 2, beginning confluence of foci; and grade 3, large confluent areas56. For periventricular hyperintensities, we determined the presence and severity (grade 0, absent; grade 1, pencil thin; and grade 2, smooth halo lining) using FLAIR images. Two researchers who were blinded to all clinical data, independently reviewed all scans. We evaluated the degree of hippocampal atrophy, using a free software program—the Voxel-based Specific Regional analysis system for Alzheimer’s Disease (VSRAD) advance version based on statistical parametric mapping 8 (SPM8) plus Diffeomorphic Anatomical Registration Through an Exponentiated Lie algebra (DARTEL)57. To preserve gray matter volume within each voxel, Matsuda et al. modulated the images by the Jacobean determinants derived from the spatial normalization by DARTEL. We checked the segmentation process according to the VSRAD manual (https://www.vsrad.info/index2.html); a qualified technician (K. Kawakami) confirmed good contrast between gray and white matter, no unacceptable irregularity in the images, no apparent artifacts, no unacceptable low intensities in T1-weighted images, and no unacceptable ventricular enlargements. We determined the extent of atrophy as the averaged value of positive voxel-by-voxel z-scores, where z-score = ([control mean] − [individual value]) / (control SD) (i.e., the higher the value, the higher the extent of atrophy). We used three indicators—the severity of atrophy obtained from the averaged positive values of z-score in the target (hippocampus) volume of interest (VOI) (hereafter referred to as ZAdvance), and the percentage rates of the coordinates with the z-score exceeding the threshold value of 2.00 in the target (hippocampus) VOI and in the whole brain VOI—for characterizing atrophy of the hippocampus and the whole brain. Although the normative data set was an external one created by Matsuda et al.57, age and sex were well balanced between our subjects and the external normative data set (i.e., 37 men and 43 women with a mean age of 70.4 ± 7.8 (SD) years; we examined 256 subjects (120 men and 136 women) with a mean age of 68.4 ± 7.0 (SD) years with a compatible MRI protocol.

Statistical analysis

All clinical variables are presented as the mean ± standard deviation. All tests were two-sided, and the level of statistical significance was set at p < 0.05. The data were analyzed using IBM SPSS Statistics version 18 for Windows (SPSS Japan Inc., Tokyo, Japan). For the univariate analysis, the chi-square test or Fisher’s exact test were used to investigate between-group differences in categorical variables, while unpaired t-tests were used to investigate differences in continuous variables. Pearson’s correlation coefficients were used to assess the relationship between the log10 BDNF value or hippocampal atrophy and RBMT score. Multiple comparisons were performed using ANOVA, followed by Bonferroni testing. Multivariate analysis was carried out with the forward stepwise method of logistic regression analysis. The association between the cognitive function tests and the BDNF values was tested using binary logistic regression analysis, adjusted for age, sex, education, physical activity, sport index, and MRI findings. We investigated the relationship between cognitive function tests, age, education, physical activity, sport index, hippocampal atrophy, and BDNF using structural equation modeling (SEM)58. We performed SEM analysis on 256 subjects, excluding three subjects whose proBDNF levels exceeded the maximum concentration measurable by ELISA. The SEM was described as path diagrams, wherein the square boxes represented measured observations and circles represented latent constructs. Single-headed arrows represented a simple regression relationship and double-headed arrows represented correlations. The parameters estimation was done with maximum likelihood estimation (the default setting of AMOS). To construct the SEM model, we investigated the variables associated with memory dysfunction, and logistic regression analysis revealed that older age, less sport activity, hippocampal atrophy, and lower BDNF levels were independently associated with memory impairment determined with the RBMT. Executive function determined with the Stroop test was added to this model, because executive function is a distinctive ability from memory. Consequently, model fit for SEM analysis did not lead to an inflated chi-square test value. We also examined several indices of model fit for SEM analysis with their acceptable thresholds: low chi-square values relative to degrees of freedom with an insignificant P value (p > 0.05); values > 0.95 for goodness of fit index (GFI), adjusted goodness of fit index (AGFI), and comparative fit index (CFI); values < 0.07 for root mean square error of approximation (RMSEA).