Materials and Methods

Subject Recruitment and Brain Collection
The design of GCS, objectives, protocols of subject recruitment, and brain collection have been previously described in detail (34, 35). Briefly, the GCS was a population-based study conducted in 44 counties in northern Georgia. The GCS was primarily designed to identify biological, psychological, and social factors contributing to survivorship and successful aging. Brain tissues from frontal (FC) and temporal cortices (TC) were collected from 47 subjects who were a subset of centenarians enrolled in the phase III of the GCS (2001–2007) and gave consent to donate brain tissue upon death. After tissue collection, all samples were coded and stored at −80°C until the measures of nutrient concentration. All protocols were performed with an approval from the University of Georgia Institutional Review Board. Separate approval for using de-identified data for the present analyses was obtained from the Tufts University/Tufts Medical Center Institutional Review Board.

Nutrient Concentration Measures in Brain Tissues
Protocols for brain lipid extraction, separation, quantification, and concentrations have been previously and separately described for carotenoids, vitamin A (retinol), vitamin E (α- and γ-tocopherol [TP]), vitamin K (phylloquinone [PK] and menaquinone-4 [MK-4]), and individual FAs (27, 28, 36–39). In short, separation and quantification of five major dietary carotenoids (lutein, zeaxanthin, cryptoxanthin, β-carotene, and lycopene), retinol, and TPs were performed using high-performance liquid chromatography (HPLC) coupled with a photodiode array detector. The limit of detection (LOD) was 0.2 pmol for carotenoids, 2.0 pmol for retinol, 2.7 pmol for TPs per injection. Only levels of the all-trans isomer of each carotenoid, which is the most predominant isomer in human brain tissues, are reported (26, 40). Separation and detection of PK and MK-4 were performed using HPLC coupled with a fluorescence detector. The LOD was 0.03 pmol for both vitamin K vitamers. Separation and detection of individual FAs were performed using a gas chromatography coupled with a flame ionization detector, and expressed as molar percentage (mol%). Total saturated FAs (SFAs) represent the sum of 10:0, 12:0, 14:0, 15:0, 16:0, 18:0, 20:0, 22:0, and 24:0. Total monounsaturated FAs (MUFAs) represent the sum of 16:1 (n-9), 16:1 (n-7), 18:1 (n-9), 18:1 (n-7), 20:1 (n-9), 22:1 (n-9), and 24:1 (n-9). Total omega-3 polyunsaturated FAs (n-3 PUFAs) represent the sum of 18:3, 18:4, 20:3, 20:5, 22:5, and 22:6. Total omega-6 polyunsaturated FAs (n-6 PUFAs) represent the sum of 18:2, 18:3, 20:2, 20:3, 20:4, 22:2, 22:4, and 22:5. Total trans-FAs represent the sum of 16:1 (n-9), 16:1 (n-7), trans-6-octadecenoic acid (18:1, n-10 to 12), 18:1 (n-9), 18:1 (n-7), trans-9, trans-12- octadecenoic acid (18:2 TT/TCTX), and conjugated linoleic acid (18:2, CLA).

Cognitive Assessment and Cognitive Domain Composite Scores
After enrollment in the GCS, cognitive assessment was performed every 6 months at the subject's residence as reported earlier (34). Cognitive data were obtained from the visit closest to death (<1 year for all subjects). Dementia status was assessed by geriatric psychiatrists using Global Deterioration Scale (GDS) and subjects were grouped based on GDS score. A score of 1–2 on GDS was clinically defined as no dementia; a score of 3 represented mild cognitive impairment; and a score of 4–7 represented increasing severity of dementia from mild to severe (41). Cognitive tests included Mini-Mental State Examination (MMSE, 24–30 = normal cognition; 19–23 = mild; 10–18 = moderate; or ≤ 9 = severe cognitive impairment) (42), Severe Impairment Battery (SIB, < 63 = very severely impaired cognition) (43), Fuld Object Memory Evaluation (FOME) (44), Controlled Oral Word Association Test (COWAT) (45), Wechsler Adult Intelligence Scale Third Edition (WAIS-III) Similarities (46), Behavioral Dyscontrol Scale (BDS) (47), and the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) battery which included Verbal Fluency (VF), Boston Naming Test (BNT), Constructional Praxis (CP), and Word List Memory Test (WLMT) (48, 49). Depression was assessed using Geriatric Depression Scale Short Form (GDSSF) (50), and activities of daily living were assessed using Direct Assessment of Function Status (DAFS) (51). All subtests have been validated and are considered reliable measures of cognition in normal aging and in AD (52).
To calculate cognitive domain composite scores, scores from each cognitive test were normalized using z-scoring as previously reported (53). Composite scores of five cognitive domains (memory, executive function, language, visuospatial function, attention), depression, and activities of daily living were then calculated by averaging the z-scores of tests based on the method adapted from Bowman et al. (15). The calculation method has also previously been reported and shown in Supplementary Table 1. Global cognition composite scores were also derived by combining total cognitive testing z-scores, MMSE, and SIB. Missing test scores were excluded and the denominator changed accordingly for the calculation of composite scores.

Statistical Analysis
Values are presented as mean (SD). All statistical tests were performed in R 3.5.1 with a significance level set at α = 0.05. Findings with p < 0.1 but > 0.05 were reported as borderline significant. Comparisons of subject characteristics between non-demented (GDS 1–3, n = 23) and demented subjects (GDS 4–7, n = 24) were performed using Student's two-sample t-test and Fisher's exact test for continuous and categorical variables, respectively.
NPs were derived from concentrations of carotenoids, vitamins A, E, K, SFAs, MUFAs, n-3 PUFAs, n-6 PUFAs, and trans-FAs averaged from FC and TC (vitamin K only from FC due to limited brain sample availability) using PCA with a function pca in the R package “pcaMethods” (54). Concentrations of all nutrients were log transformed prior to PCA. Nutrient concentration matrix was unit-variance scaled and centered. We chose non-linear iterative partial least squares algorithm to calculate principal components, or NPs in our case, which is an iterative approach for estimating independent principal components by extracting them one at a time (55). This variation of PCA can handle small amount of missing values, which in our case were PK and MK-4 concentrations from two individuals due to insufficient brain tissues for vitamin K analysis. Only NPs with eigenvalue >1 were reported.
To investigate the relationship between brain nutrient concentrations or NPs and cognitive domain composite scores, Pearson's correlation test was performed with an adjustment for covariates sex, education, hypertension, diabetes, and presence of APOE ε4 allele. Additional adjustment for antithrombotic use was performed for vitamin K, and antidepressant use for depression score. Sub-analyses in non-demented (GDS 1–3) were also performed. Heat maps aided the visualization of correlations.