Midlife Adiposity predicts earlier onset of Alzheimer’s dementia, neuropathology and presymptomatic cerebral amyloid accumulation

Abstract

Understanding how midlife risk factors influence age-at-onset (AAO) of Alzheimer’s disease (AD) may provide clues to delay disease expression. Although midlife adiposity predicts increased incidence of AD, it is unclear whether it affects AAO and severity of Alzheimer’s neuropathology. Using a prospective population-based cohort, the Baltimore Longitudinal Study of Aging (BLSA), this study aims to examine the relationships between midlife body mass index (BMI) and (1) AAO of AD (2) severity of Alzheimer’s neuropathology and (3) fibrillar brain amyloid deposition during aging. We analyzed data on 1,394 cognitively normal individuals at baseline (8643 visits; average follow up interval 13.9 years), among whom, 142 participants developed incident AD. In two sub-samples of BLSA, 191 participants underwent autopsy and neuropathological assessment, and 75 non-demented individuals underwent brain amyloid imaging. Midlife adiposity was derived from BMI data at 50 years of age. We find that each unit increase in midlife BMI predicts earlier onset of AD by 6.7 months (P=0.013). Higher midlife BMI was associated with greater Braak neurofibrillary tangle but not CERAD neuritic plaque scores at autopsy overall and with greater fibrillar amyloid measured by global mean cortical distribution volume ratio and within the precuneus. In conclusion, midlife overweight predicts earlier onset of AD and greater burden of Alzheimer’s neuropathology. A healthy BMI at midlife may delay the onset of AD.

Introduction

Reducing the burden of Alzheimer’s disease (AD) is a global public health imperative¹ and the impact of an intervention that delays the onset of AD even modestly will be enormous. For example, an intervention that delayed onset of AD by just two years would decrease the worldwide prevalence of the disease by an estimated 22.8 million cases in 2050². In the United States alone, an intervention that offered only a one-year delay in AD onset would result in annual savings of $10 billion in each of the following 10 years.³. This degree of delay in AD onset may be achieved by changes in societal behaviors such as dietary or exercise habits known to modify AD risk^{4, 5}. Moreover, to achieve maximum effect, such changes should logically be instituted in midlife or earlier. However, despite the promise of altering the trajectory of AD by modifying midlife risk factors, the effect of such risk factors on the age at onset (AAO) of AD has received little attention to date.

Midlife obesity is a risk factor for AD,^6–12 and the World Health Organization estimates that the worldwide prevalence of overweight people today is already 1.4 billion¹³. If present trends continue, this figure will increase substantially¹⁴. The problem is especially acute in the United States where almost 90% of American adults will be overweight by 2030¹⁵. Although previous studies have established that midlife obesity is associated with increased risk and age-specific incidence of AD^{7, 8, 10, 12}, it is unclear how obesity and overweight at midlife may influence the age-at-onset (AAO) of AD. Equally importantly, it is not known whether midlife BMI affects the extent of AD neuropathology. We therefore used longitudinal assessment data of older individuals (N=1394; mean follow up interval 13.9 years but ranging to 35.2 years) within the Baltimore Longitudinal Study of Aging (BLSA)¹⁶ to ask whether midlife adiposity was related to the AAO of incident AD. Using neuropathological data from the autopsy cohort of the BLSA, we also examined the relationship of midlife adiposity with measures of both neuritic plaque and neurofibrillary tangle pathology. Finally, we used in vivo brain amyloid imaging with ¹¹C-Pittsburgh Compound-B (PiB) PET to ask whether midlife adiposity was related to the magnitude of fibrillar brain amyloid burden, even in non-demented older individuals.

Methods

Participants

The BLSA is a prospective, ongoing cohort study of community-dwelling volunteer participants in Baltimore begun in 1958. As such, it is among the largest and longest-running longitudinal studies of aging in the United States^{16, 17}. Its community-dwelling unpaid volunteer participants are predominantly white, of upper-middle socioeconomic status, and with an above average educational level. Initially limited to men, enrollment of women started in 1978. In general, at the time of entry into the study, participants had no physical or cognitive impairment. Detailed examinations, including neuropsychological assessments and neurological, laboratory, and radiological evaluations, were conducted every 2 years. Since 2003, participants older than 80 years have received yearly assessments. Written informed consent was obtained at each visit, and the study was approved by the local Institutional Review Board and the National Institute on Aging.

Midlife Body Mass Index (BMI)

BMI at 50 years of age was used to define ‘midlife BMI.’ At each visit, height and weight were measured using calibrated scales by trained technicians and BMI at age 50 years was estimated for participants in the study sample using linear mixed effects model (details described below). To avoid confounding with behavioral abnormalities associated with cognitive impairment, BMI was considered only if measured at visits before onset of dementia or mild cognitive impairment (MCI).

Age at onset of Alzheimer’s disease

After each visit, cognitive status was considered at consensus diagnosis conferences using established procedures described previously,¹⁸ relying on information from neuropsychological tests as well as clinical data. Diagnoses of dementia and Alzheimer’s disease (AD) were based on DSM-III-R¹⁹ and the NINCDS-ADRDA criteria²⁰ respectively. Our sample included all participants who entered the BLSA from 1978 onwards when recruitment of females into the study began. We excluded participants whose last visit was before age 50 (N=377). We also excluded individuals with diagnoses of MCI²¹ or non-AD dementia (N=134). A total of 142 individuals received a clinical diagnosis of incident AD among the sample of 1394 participants who were followed over a mean interval of 13.9 years (sample 1 in Fig. 1).

Fig. 1.

Schematic summary of the three study samples used in the present analyses derived from the Baltimore Longitudinal Study of Aging (BLSA). Dashed lines indicate reasons for exclusion of participants from a sample.

Neuropathological studies

The autopsy program of the BLSA was initiated in 1986. We have previously described details of this sample²². Briefly, the mean age at death in the autopsy sample is 88.3 ± 7.3 years (range 69.3–103.2), and the mean interval between last evaluation and death is 8.7± 6.7 months²². As reported previously, the autopsy subsample is not significantly different from the BLSA cohort as a whole in terms of the rates of dementia and clinical stroke²³. As of September 2009, 197 BLSA participants had undergone autopsy and neuropathological assessment following death. We excluded 4 African Americans and 2 subjects who did not have BMI measurements while they were cognitively normal. The final sample included 191 participants (sample 2 in Fig. 1). Postmortem brain examinations were performed by an experienced neuropathologist (J.C.T.). Neuritic plaques and neurofibrillary tangles were assessed as described previously²⁴. AD pathology was assessed on silver stained material and graded according to the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD)²⁵ and Braak criteria²⁶.

[¹¹C] PiB PET Imaging

In vivo amyloid imaging by ¹¹C-PiB PET study was initiated in 2005 in a subset of BLSA Neuroimaging substudy participants who agreed to annual neuroimaging assessments and were free of central nervous system disease (dementia, stroke, bipolar illness, epilepsy), severe cardiac disease (myocardial infarction, coronary artery disease requiring angioplasty or coronary artery bypass surgery), severe pulmonary disease, or metastatic cancer²⁷. As of 2014, 80 participants had undergone at least one PiB PET imaging. Five participants who developed mild cognitive impairment or dementia were excluded from the current analysis, yielding 75 participants who remained cognitively normal through follow up (sample 3 in Fig. 1). Data from the first ¹¹C-PiB PET imaging visit were used in the current analysis.

¹¹C-PiB PET Acquisition

PET scans were conducted on a GE Advance scanner. Subjects were fitted with a thermoplastic mask in order to reduce motion artifacts. Dynamic [¹¹C]-PiB PET data acquisition was performed in 3D mode immediately following an intravenous bolus injection of [¹¹C]-PiB with an acquisition protocol of 4×0.25 minutes, 8×0.5 minutes, 9×1 minutes, 2×3 minutes, 10×5 minutes (total 70 minutes, 33 frames). All PET images were corrected for attenuation using ⁶⁸Ge transmission scans, and were reconstructed to have 2×2 mm pixel size and 4.25 mm slice thickness.

MRI Acquisition

MR imaging was performed at the same study visit as PET imaging. Magnetization prepared rapid acquisition with gradient echo (MPRAGE) scans were conducted on a 3T Philips scanner using the following parameters: TR=6.8ms, TE=3.2ms, α=8°, 256×256 matrix, 170 slices, pixel size=1×1mm, slice thickness=1.2mm.

Image Processing

Cortical reconstruction and volumetric segmentation of the MPRAGE images for regional definitions were performed using the FreeSurfer image analysis suite (version 5.1; http://surfer.nmr.mgh.harvard.edu/), with manual editing to fix the cortical surfaces. The details have been described previously²⁸.

Remaining image analysis steps were carried out using an automated in-house pipeline developed using Medical Image Processing, Analysis and Visualization software²⁹ and the Java Image Science Toolkit³⁰. The frames of the dynamic PiB PET image were aligned using rigid registration with a least-squares cost function onto the 2-minute mean of the PiB PET image³¹. For each subject, the inhomogeneity-corrected MPRAGE was rigidly registered onto the 20-minute mean of the PiB PET image using normalized mutual information as cost function³². FreeSurfer labels were mapped onto the PiB PET space using the resulting transformation for the MPRAGE.

Parametric images of regional β-amyloid load (as given by the distribution volume ratio, or DVR) were generated using a 2-compartment simplified reference tissue model (SRTM)³³. The subject-specific cerebellar gray matter segmentation from FreeSurfer was used as the reference region for nonspecific radiotracer binding³⁴. Mean cortical distribution volume ratio (DVR) and regional DVRs of the precuneus (left and right) were the principal metrics for these analyses. Mean cortical DVR is the weighted average of values for the superior, middle and inferior frontal and orbitofrontal, superior parietal, supramarginal and angular gyrus, precuneus, superior, middle and inferior occipital, superior, middle and inferior temporal, anterior, middle and posterior cingulate regions, which provides a global index of cortical fibrillar amyloid burden. Precuneus DVR was examined because this region has been reported to show the earliest accumulation of fibrillar amyloid in the preclinical stages of AD³⁵.

Other covariates

Both sex and race were dummy coded in the analyses (male coded as 1; female coded as 0; White coded as 1; African-American coded as 0). Education was coded as the ‘number of years.’ Cardiovascular comorbidities were computed by counting the total number of the following self-reported medical conditions: hypertension, hypercholesterolemia, diabetes mellitus, smoking status, history of angina, myocardial infarction, stroke and transient ischemic attack. In addition, hypertension was also defined as recorded systolic blood pressure >= 140 mmHg or diastolic blood pressure >= 90 mmHg, diabetes as recorded fasting blood glucose >= 126 mg/dl and hypercholesterolemia as recorded cholesterol >= 200 mg/dl. Smoking status was defined as ‘non’- or ‘ever’ smoker.

Statistical analysis

We applied a linear mixed effects model to the entire BLSA longitudinal sample using body mass index (BMI) as the dependent variable to derive BLUE (Best Linear Unbiased Estimates)³⁶ of midlife BMI. Fixed effects of the model included intercept, age, age-squared and their interactions with sex and race. Random effects included intercept, age, and, age-squared. This model is flexible and powerful in that it allows non-linear BMI trajectories both at the individual and population levels. Furthermore, the model utilizes all available BMI data, including longitudinal assessments, to derive estimates of an individual’s BMI at age 50 years with the corresponding standard errors. To account for the precision of predicted midlife BMI values, we derived weights for these estimates as the inverse of square of standard errors (i.e., for ith subject the estimated weight was $w_i=1/{\sigma }_i^2$) in all subsequent regression analyses. Estimated midlife BMI values with higher precision were thus given greater weights in the regression analyses. The weights depended on several factors including the number and frequency of repeat measurements, proximity of the actual measurement to age 50 years and within-individual variability relative to between-individual variability. Supplementary Figure S1 uses two BLSA participants to illustrate the relationship between number of repeated BMI measurements, proximity of the actual measurement relative to age 50 years, and the estimated midlife BMI with corresponding standard errors. Since our main goal was to examine the precise relationship between midlife BMI and AAO of AD, we used accelerated failure time (AFT) models from survival analyses which allowed us to express the effects of covariates on the time- to-event (incident AD) in terms of an acceleration/decceleration factor. On the other hand, the most commonly used Cox proportional hazard model expresses the effects of covariates in terms of relative risk of AD. The AFT model is formulated as logT_i = X_iβ + σε_i, where outcome T_i is the time to event for ith subject, defined here as the age at onset of AD (‘event age’) for participants with AD and age at most recent visit as ‘censored age’ for a normal participant. σ is a scale parameter and ε_i is the random disturbance term following a certain distribution. The predictors (X_i) included midlife BMI, sex, race, year of birth, number of years of education and cardiovascular comorbidities. We fitted a series of AFT models assuming different distributions for T_i that included: exponential, log normal, log logistic, Weibull, and generalized gamma. Maximum likelihood estimation was used for each model, and Akaike Information Criterion (AIC)/Bayesian Information Criteria (BIC) were used to select the model that best fit the data. The precision of midlife BMI estimates was taken into account in the model by weighting the log likelihood function for each subject. In separate analyses, we also estimated the effect of midlife BMI on the risk of AD using Cox proportional hazard model (Supplementary Table S1).

Separate weighted least square (WLS) regression was used with pathology scores (CERAD and Braak) and DVRs (PET-PiB amyloid burden) as dependent variables and midlife BMI as the main predictor. Other covariates included sex, race, years of education, and cardiovascular comorbidities. For pathology outcomes, age at death was also included as a covariate.

In all the regression analyses, linearity and additivity of the effect of midlife BMI on the age at onset, extent of AD pathology and DVRs were tested by including the natural spline of BMI and interaction between BMI and sex term separately in the models. As they were not significant, they were dropped from the final model. Thus, all 3 models meet the assumptions of linearity and additivity. All analyses were performed using SAS 9.3 (Cary, NC).

Results

Demographic characteristics

Table 1 shows the demographic characteristics of the study samples from the BLSA cohort. In the sample for the accelerated failure time analysis, among 1394 participants, 142 (10.2%) developed incident AD with an average age of onset of 83.3 years. In the autopsy sample, clinical diagnoses prior to autopsy included AD (71), mild cognitive impairment (26), non-AD dementia (24), cognitive impairment – not MCI (2) and cognitively normal (68).

Table 1.

Demographic characteristics of participants in study samples from the BLSA

	Accelerated failure time sample (n=1394)	Autopsy sub-sample (n=191)	[11C] PiB PET sub-sample (n=75)
Age at baseline, years, mean(SD) (range)	59.5 (15.6) (22.9 – 94.9)	64.1 (14.8) (25.0–87.3)	55.1 (14.3) (28.6–86.9)
Male, No.(%)	512 (36.7)	130 (68.1)	38 (49.3)
White, No. (%)	1,040 (74.6)	191 (100)	57 (76.0)
Education, years, mean(SD) (range)	16.0 (2.7) (8–21)	17.0 (2.4) (8–20)	17.1 (2.2) (12–21)
Follow-up years, mean(SD) (range)	13.9 (9.5) (0–35.2)	20.3 (11.6) (3.6–48.3)
Follow-up visits, mean(SD) (range)	6.2 (4.2) (1–24)	12.0 (6.6) (3–32)
Cardiovascular comorbidities, mean(SD) (range)	1.06 (0.99) (0–4)	2.0 (1.3) (0–6)	1.6 (1.0) (0–4)
BMI at age 50, kg/m2, mean(SD) (range)	25.8 (4.1) (17.4–44.2)	24.8 (3.6) (17.7–44.2)	25.7 (3.4) (20–36.9)
Age at death, years, mean(SD) (range)		88.3(7.3) (69.3–103.2)
Age at PiB PET, mean (SD) (range)			75.5 (8.1) (55.7–92.4)
AD, No. (%)	142 (10.2)	71 (37.2)

BLSA, Baltimore Longitudinal Study of Aging; AD, Alzheimer’s disease; [¹¹C] PiB PET, Carbon 11-labeled Pittsburg Compound-B positron emission tomography; SD, standard deviation

Midlife BMI and age-at-onset of AD

Among AFT models with different time-to-event distributions, log normal distribution fitted the data best. After adjustment for sex, race, years of education, birth year and cardiovascular comorbidities, midlife BMI had a significant effect on the AAO of AD with an acceleration factor of −0.0067 (P=0.013), indicating that each unit increase in midlife BMI was associated with about 0.67% acceleration or about a 6.7 months decrease in age-at-onset of AD (Table 2). Figure 2 shows the relationship between midlife BMI and 10^th percentile time-to-event of AD (i.e. the age when 10% of the sample would have developed AD) from the final AFT model. In the figure, each unit increase in midlife BMI is associated with an earlier AAO of AD by an average of 6.7 months. When we examined the effect of BMI on risk of AD using a Cox proportional hazards model, we found that higher mid-life BMI was associated with an increased risk of AD (HR = 1.07, P=0.024) (Supplementary Table S1).

Table 2.

Association between midlife BMI and age at onset of Alzheimer’s Disease (AD); results from the accelerated failure time model (N=1394)

Predictors	Acceleration factor (SE) P-value
Cardiovascular comorbidities	0.026 (0.017) 0.12
Education 0.0009	(0.0046) 0.85
Race (1=White, 0 = AA)	0.13 (0.057) 0.021
Birth year	0.031 (0.015) 0.033
Sex (1=male, 0=female)	0.098 (0.034) 0.0038
Midlife BMI	−0.0067 (0.0027)* 0.013

^*Interpreted as each unit increase in midlife BMI being associated with 0.67% decrease in age-at-onset of AD (equivalent to approximately 6.7 months earlier in age-at-onset of AD).

Fig 2.

Relationship between 10^th percentile survival time for Alzheimer’s disease (AD) and midlife BMI with 90% confidence intervals from the accelerated failure time model using the log normal distribution. The 10^th percentile survival time for AD is defined as the age when 10% of the subjects in the sample will develop AD. It shows that a unit increase in midlife BMI is associated with an earlier age-of-onset of AD by an average of 6.7 months.

Midlife BMI and severity of Alzheimer’s neuropathology

In a model adjusted for sex, age at death, number of years of education and cardiovascular comorbidities, midlife BMI was significantly associated with Braak score (p=0.0087) with higher midlife BMI predicting greater neurofibrillary tangle burden. However, we did not find significant associations between midlife BMI and neuritic plaque burden as measured by the CERAD score (p=0.27) (Table 3).

Table 3.

The Association between midlife BMI and AD pathology (Braak and CERAD scores) (N=191)*

β (SE) P-value	Sex (1=male, 0=female)	Midlife BMI
Braak	−0.64 (0.34) 0.062	0.071 (0.027) 0.0087
CERAD	−0.57 (0.31) 0.063	0.026 (0.024) 0.27

Abbreviations: SE, standard error;

^aAdjusted for sex, age of death, education, and cardiovascular comorbidities.

Midlife BMI and in vivo fibrillar brain amyloid burden

Higher BMI at age 50 years was marginally associated with an increase in mean cortical DVR (P=0.075). Similarly, higher BMI at age 50 years was also marginally associated with increased fibrillar amyloid deposition in the left and right precuneus (p = 0.061, 0.079 respectively) (Supplementary Table S2).

Discussion

Our results show a robust association between higher BMI at age 50 years and earlier onset of AD, greater severity of Alzheimer’s neuropathology at autopsy and higher fibrillar amyloid deposition in non-demented elderly. Our findings clearly indicate that higher adiposity at midlife is associated with a long-lasting effect on accelerating the clinical course of AD. We found a strong association between higher midlife BMI and greater severity of neurofibrillary pathology. Finally, we extended our study to ask if midlife BMI was related to measures of fibrillar amyloid deposition during life. We found that higher midlife BMI was indeed associated with both greater global fibrillar amyloid burden and increased deposition within the precuneus, a brain region especially vulnerable to early accumulation of amyloid even in non-demented older individuals³⁵.

Our study has several strengths. First, the BLSA is among few longitudinal studies with serial cognitive assessments of initially healthy individuals and prospective consensus diagnoses of incident dementia that enable accurate estimates of AAO of AD. The frequency of serial assessments (every two years on average) and long follow up duration are particular strengths of our study. The paucity of studies that have explicitly tested the association between midlife adiposity and AAO of AD may be due in part, to the lack of accurate midlife BMI or AAO data in these studies. Furthermore, our use of an accelerated failure time model allowed us to directly examine the relationship between midlife BMI and AAO of AD. This method provides an intuitive and easy to interpret summary statistic in terms of an acceleration/deceleration factor from survival curves which is especially useful in aging research³⁷. In contrast, the traditional Cox proportional hazards model estimates the hazard function and summarizes the effects of midlife BMI in terms of the ratio of age-specific incidence (i.e., the hazard ratio). Finally, the availability of an autopsy sample with detailed assessment of neuropathology as well as a neuroimaging substudy with in vivo estimates of amyloid deposition make our analyses especially unique in suggesting possible biological mechanisms underlying the relationship between midlife adiposity and the AAO of AD. A potential limitation of our study is that the BLSA is a predominantly Caucasian sample of highly educated and healthy individuals. It would be important, therefore to test whether our findings are also valid in larger, more diverse samples.

We focused our analyses on midlife BMI in relation to AAO of AD for several reasons. Firstly, identification of an environmental risk factor in midlife may allow for early lifestyle interventions exerting long lasting effects to delay the onset of AD^38–40. Secondly, a growing body of evidence links healthy behaviors during midlife with various metrics of healthy aging⁴¹. Lastly, measures of BMI in late-life are more likely to be confounded by comorbid medical conditions as well as by perturbations in feeding behavior in the early stages of AD^{42, 43} whereas midlife BMI may be less prone to such confounding variables.

While we did not directly assess the mechanisms underlying the association between midlife BMI and onset of AD, this relationship could be mediated in part by an increase in both neurofibrillary tangle pathology as well as greater accumulation of fibrillar amyloid in vulnerable brain regions such as the precuneus. Interestingly, we have reported previously that midlife obesity is also associated with greater rates of regional brain atrophy in individuals progressing to cognitive impairment⁴⁴. Together with our current results, these findings suggest that increasing amyloid accumulation, accelerated brain atrophy and neurodegeneration may be pathological events underlying the association between midlife BMI and earlier AAO of AD. It may be important to also consider the role of chronic inflammation associated with obesity as a trigger for the initiation or worsening of AD neuropathology^{45, 46}.

In summary, we found that greater midlife adiposity was associated with earlier age at onset of AD and greater severity of Alzheimer’s neuropathology. Our findings suggest that maintaining a healthy body mass index at midlife could have a long-lasting impact in delaying the onset of AD. Equally importantly, they relate the age at onset of AD to a modifiable risk factor that can be easily measured at a single time point at home, and reveal a long interval of opportunity during which lifestyle changes might delay the onset of AD. Our findings raise the possibility that inexpensive, non-invasive interventions targeting midlife obesity and overweight could substantially alter the trajectory of Alzheimer’s disease, reducing its global public health and economic impact.