Mediterranean vs. Western diet effects on the primate cerebral cortical pre-synaptic proteome: Relationships with the transcriptome and multi-system phenotypes

Abstract

INTRODUCTION

Diet quality mediates aging-related risks of cognitive decline, neurodegeneration, and Alzheimer's disease (AD) through poorly defined mechanisms.

METHODS

The effects of diet on the presynaptic proteome of the temporal cortex were assessed in 36 female cynomolgus macaques randomized to Mediterranean or Western diets for 31 months. Associations between the presynaptic proteome, determined by synaptometry by time-of-flight (SynTOF) mass spectrometry, adjacent cortex transcriptome, and multi-system phenotypes were assessed using a machine learning approach.

RESULTS

Six presynaptic proteins (DAT, Aβ42, calreticulin, LC3B, K48-Ubiquitin, SLC6A8) were elevated in the presynaptic proteome in Mediterranean diet consumers (p < 0.05). Transcriptomic data and multi-system phenotypes significantly predicted SynTOF markers. Selected SynTOF markers were correlated with changes in white matter volumes, hepatosteatosis, and behavioral and physiological measures of psychosocial stress.

DISCUSSION

These observations demonstrate that diet composition drives cortical presynaptic protein composition, that transcriptional profiles strongly predict the presynaptic proteomic profile, and that presynaptic proteins were closely associated with peripheral metabolism, stress responsivity, neuroanatomy, and socio-emotional behavior.

Highlights

• Mediterranean and Western diets differentially altered the cortical presynaptic proteome, which is strongly associated with neurodegeneration and cognitive decline.
• Presynaptic proteomic markers were predicted by transcriptomic profiles in the adjacent cortex, and by multi-system anatomical, physiologic, and behavioral phenotypes.
• The data demonstrate that brain phenotypes and brain-body interactions are influenced by common dietary patterns, suggesting that improving diet quality may be an effective means to maintain brain health.

1 BACKGROUND

Diet and dietary constituents have the potential to mediate the risk of cognitive decline and neurodegeneration. For example, consumption of a Western diet high in saturated animal fats, simple sugars, and sodium increases the risk of obesity, diabetes, and cardiovascular disease,^1-3 all of which increase the risk of dementia and Alzheimer's disease (AD).⁴ Western diet consumption is also associated with smaller hippocampal volume,⁵ which has been associated with reduced cognitive ability. In contrast, consumption of a Mediterranean diet high in plant mono- and polyunsaturated fats, plant proteins, and antioxidants has been associated with reduced risk of cardiovascular disease,³ diabetes,⁶ cognitive impairment,⁷ and AD dementia,⁸ although some studies have failed to find relationships between Mediterranean diet and cognitive health.^{9, 10} These observational findings have important limitations: they derive mostly from self-reported data that undermine accurate measurement of diet composition and consumption, and they are confounded by other features associated with cognitive health, such as regular exercise, educational attainment, and smoking behavior. Experimental data on the neurobiological impacts of these two common and distinct dietary patterns are critical to advancing our understanding of the impacts of diet composition on brain health.

We conducted a randomized, controlled preclinical trial of middle-aged, female cynomolgus macaques (Macaca fascicularis) to determine the impacts of long-term consumption (31 months, equivalent to ∼9 years for humans) of either a Western or Mediterranean diet on multiple, interrelated body systems.^11-18 We have demonstrated that Mediterranean diet consumption protected study subjects from higher caloric intake, adiposity, hyperinsulinemia, and hepatosteatosis that occurred in monkeys consuming the Western diet.¹⁹ Consumption of the Mediterranean diet also yielded a healthier gut microbiome profile,^{17, 18} lower skeletal muscle mitochondrial respiration,²⁰ better ovarian function,¹³ lower sympathetic activity, faster and higher heart rate responses to acute stress with more rapid recovery, lower cortisol responses to acute psychological stress and adrenocorticotropin (ACTH) challenge,²¹ and a less inflammatory transcriptome of circulating monocytes.¹⁴ For the brain, the Mediterranean diet resulted in lower total and gray matter volumes and cortical thicknesses and higher white matter volumes,¹² as well as a less inflammatory cortical gray matter transcriptional profile.²² These phenotypes were associated with an anti-inflammatory and neuroprotective transcriptomic profile relative to animals that consumed the Western diet. Subjects consuming the Western diet also exhibited brain region-specific disruptions of mitochondrial bioenergetics.²³ Importantly, behavioral divergences accompanied these anatomical, physiological, and transcriptomic differences in the brain. Consumption of the Mediterranean diet mitigated social isolation and anxiety,¹⁵ whereas those consuming the Western diet exhibited increased social isolation and higher anxiety.

Building on these observations, here we report contrasting effects of long-term consumption of Western versus Mediterranean diets that recapitulate human diet patterns on the molecular composition of single presynapses of the lateral temporal cortex. We used synaptometry by time-of-flight (SynTOF) mass spectrometry, a powerful technology, to quantify single presynapse molecular diversity of approximately 3.6 million synapses.^{24, 25} Furthermore, we demonstrate via machine learning (ML) models that the transcriptome from the adjacent cortex predicts the presynaptic proteome identified with SynTOF. Finally, we examined relationships between SynTOF markers and multi-system phenotypes relevant to health status. Our robust and novel randomized trial in middle-aged, female nonhuman primates (NHPs) shows profound differences between the Mediterranean and Western diet groups in cerebral cortical presynaptic proteome and transcriptome, along with significant associations with physiologic, metabolic, neuroanatomic, and behavioral phenotypes, that have far-reaching implications for brain health.

2 METHODS

2.1 Subjects

Thirty-six middle-aged (mean = 9.0, range = 8.2–10.4 years at the initiation of the study, estimated by dentition) female cynomolgus macaques (M. fascicularis; SNBL USA SRC, Alice, TX) were studied. Following a 1-month quarantine, the monkeys were housed in small social groups (n = 4–5) in indoor enclosures (3 m × 3 m × 3 m) and maintained on a 12/12 light/dark cycle, with exposure to daylight. All experimental protocols complied with state and federal regulations and guidelines from the US Department of Health and Human Services. This work was conducted with the approval of the Wake Forest University School of Medicine Institutional Animal Care and Use Committee.

2.2 Design

The study was designed as a preclinical randomized trial (Figure 1) 39 months in length, which is approximately equivalent to 10–12 years for human beings. During an 8-month pre-treatment phase, the animals consumed a standard monkey diet with ad libitum access to water. Following the pre-treatment phase, individuals were assigned to either Mediterranean (N = 17; MED) or Western (N = 19; WEST) diet consumption for 31 months using stratified randomization, resulting in the two groups balanced on pre-treatment markers of health, including body weight, body mass index (BMI), circulating plasma cortisol, and plasma triglyceride concentrations.¹⁹ Throughout the experiment, each monkey was offered a 120 kcal diet/kg body weight/day.

FIGURE 1

Experimental design. Design of this randomized, controlled preclinical trial of 36 middle-aged, female cynomolgus macaques (Macaca fascicularis) to determine the impacts of long-term consumption (31 months, equivalent to ∼9 years for humans) of either a Western or Mediterranean diet. The phenotypes collected across the course of the study shown here are described in Table S2 and more detail in the referenced publications. CSF, cerebrospinal fluid; CT, computed tomography; HPA, hypothalamic-pituitary-adrenal; IVGTT, intravenous glucose tolerance test.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature with PubMed and found numerous publications that are cited that suggest that Alzheimer's disease (AD) risk and pathogenesis are influenced by dietary constituents, almost all are derived from self-reported data or short-term studies.

2. Interpretation: Our findings support the hypotheses that diet composition drives cortical presynaptic protein composition, that transcriptional profiles predict the presynaptic proteomic profile, and that presynaptic proteins were closely associated with peripheral metabolism, stress responsivity, neuroanatomy, and socio-emotional behavior.

3. Future directions: This article represents a comprehensive assessment of diet effects on brain biology and brain-body interactions during middle age in a non-human primate model of AD-like neuropathology. Future studies are exploring single nucleus multiomic profiles in the same brain region to help delineate specific cell types involved in these interactions.

2.3 Diets

Experimental diets were carefully designed to recapitulate the Western²⁶ and Mediterranean²⁷ diet patterns of human beings. We chose to feed diet patterns rather than individual nutrients because individual nutrients are consumed, absorbed, and utilized by humans in the broader context of their whole diet pattern. Additionally, diet patterns, rather than single nutrients, better predict human morbidity and mortality.²⁸

The experimental diets were prepared at the Wake Forest University School of Medicine, details have been previously described^11-19 and are shown in Table 1^29-34 with specific ingredients shown in Table S1. The Western diet was a semi-purified diet prepared from foods humans consume and recapitulated many major components of human Western diets. The Mediterranean diet was a carefully designed and prepared semi-purified diet using foods that humans consume and was assessed in pilot studies before the study reported here. Diets were identical in cholesterol content (∼320 mg/2000 kcal/day) and percent of calories from proteins, fats, and carbohydrates. However, the sources of the macronutrients differed. Protein and fat in the Mediterranean diet were primarily plant-derived, whereas proteins and fats were primarily animal-derived in the Western diet. Thus, the Mediterranean diet was lower in saturated fat, and the Western diet was lower in monounsaturated fats. Additionally, the ratio of omega-6 to omega-3 fatty acids was low in the Mediterranean diet (3:1), whereas the ratio in the Western diet was relatively high (approximately 15:1). The Mediterranean diet also contained nuts, fruits, and vegetables, and was higher in antioxidants primarily from extra virgin olive oil and walnuts, which are rich in polyphenols and other antioxidants.

TABLE 1. Experimental diets.

	Human^29-34		Nonhuman primate^19
Parameter	Western	Mediterranean	Western^f	Mediterranean^f	Chow^g
% of calories
Protein	15^a	17^b	16	16	18
Carbohydrate^h	51^a	51^b	54	54	69
Fat	33^a	32^b	31	31	13
% of total fats
Saturated	33^a	21^b	36	21	26
Monounsaturated	36^a	56^b	36	57	28
Polyunsaturated	24^a	15^b	26	20	32
Other nutrients
ω6:ω3 fatty acids	15:1^c	2.1–3:1^d	14.8:1	2.9:1	12:01
Cholesterol mg/Cal^g	0.13^a	0.16^b	0.16	0.15	trace
Fiber g/Cal	0.01^a	0.03^e	0.02	0.04	0.01
Sodium mg/Cal	1.7^a,f, ^a,f	1.3^b,e, ^b,e	1.7	1.1	0.25

• Note: Experimental diet compositions¹⁹ and the human dietary patterns^29-34 the study aimed to reflect. Values have been rounded to the nearest whole number.
• ^a Ref.26
• ^b Refs.30, 31
• ^c Ref.32
• ^d Ref.33
• ^e Ref.34
• ^f Developed and prepared at Wake Forest University School of Medicine.¹⁹
• ^g LabDiet Chemical Composition Diet 5037/8. Type of fat known in 86% of total fat. Omega-6 from corn and pork fat.
• ^h Human carbohydrate calories include alcohol.

2.4 Multi-system phenotyping

Study subjects were extensively phenotyped during the last half of the experimental period (13–31 months on experimental diet; Figure 1). We  assessed 70 phenotypes (Table S2), representing 14 categories of anatomy, physiology, and behavior: socioemotional behavior, stress physiology (hypothalamic-pituitary-adrenal function), heart rate variability (HRV),²¹ body composition, morphometrics, insulin resistance, physical activity levels, and gait speed (activity),³⁵ calories consumed, hepatosteatosis,¹⁹ neuroanatomy,¹² ovarian functioning,¹³ cerebrospinal fluid (CSF) biomarkers of neuropathology,³⁶ and coronary artery atherosclerosis.³⁷ Characteristics of sleep include latency to sleep onset, total sleep time, and number of awakenings, which were determined from actigraphy using Actisleep software.^{38, 39} The descriptions of the collection methods and individual dependent variables have been extensively described in the literature (see Table S2).

2.5 Lateral temporal cortex samples

After the 31-month exposure to the experimental diets, subjects were anesthetized and euthanized according to the guidelines established by the Panel on Euthanasia of the American Veterinary Medical Association. The post mortem interval for tissue collection was short (1-h or less) to ensure the acquisition of high-quality specimens for subsequent transcriptomic and proteomic analyses. Four-millimeter coronal sections were cut, and samples from alternating left and right lateral temporal cortex were dissected and preserved for downstream analyses. Tissue for transcriptomic analyses was covered with finely crushed dry ice and frozen for at least 20 min, then placed in a vacuum-sealed sample bag and stored in a −80°C freezer until processing. A systematic and detailed stepwise protocol from the entire tissue preservation and synaptosome processing is published by us.²⁵ For synaptosome preparations, the tissue was quickly minced, and transferred to a tube containing 0.32 mol/L sucrose with 10% dimethyl sulfoxide solution, and then slowly frozen in a cryo box overnight at −80°C. Samples were stored in a −80°C freezer until further processing. See Figure 1 for sample location. To obtain the crude synaptosome fraction, the minced tissue was thawed and homogenized in 0.32 mol/L sucrose in 10 mmol/L Tris buffer with proteinase and phosphatase inhibitors using a glass/Teflon homogenizer, exactly as done previously with human samples.^{40, 41} The homogenate was centrifuged at 1000 × g at 4°C for 10 min, then the supernatant was removed and centrifuged again at 10,000 × g at 4°C for 20 min. The resulting pellets were resuspended in 0.32 mol/L sucrose/Tris in 10% dimethyl sulfoxide solution and stored at −80°C to maintain synaptosome integrity.⁴²

2.6 SynTOF mass spectrometry

Following our established and published protocol for a SynTOF panel of 38 antibodies,⁴³ synaptosome samples were prepared, barcoded, and stained. Mass synaptometry data were acquired, debarcoded, and sequentially gated for the percentage of negative and positive events. Positive events were selected by CD56 and SNAP25, and nonpresynaptic events selected by CD11b, gephyrin, PSD95, and MBP. Nonpresynaptic events were removed from downstream analyses. Synaptosome processing and analysis pipelines were previously described by us at.^{24, 25, 40, 41} Antibodies included in the analysis are in (Table S3) The mean pseudo bulk expression of the 30 SynTOF markers are shown in Figure S1A.

2.7 Cortical transcriptome

Tissue from the lateral temporal cortex immediately adjacent to the synaptosome preparation was used for transcriptomic analyses as previously described.²² Briefly, RNA isolation and sequencing were conducted by the Wake Forest University School of Medicine Cancer Genomics Shared Resource, with the researchers blinded to treatment conditions. RNA was isolated using the Zymo Quick-DNA/RNA Miniprep Plus Kit (Zymo, Irvine, CA, USA) and purified using the RNA Clean and Concentrator-5 kit (Zymo). All RNA integrity number (RIN) scores were > 7.7, indicating high integrity. Bulk RNA-seq was then performed using an Illumina NovaSeq 6000 (single end, 100 bp). We subsequently aligned sequenced reads to the cynomolgus macaque genome (Macaca_fascicularis_6.0) using STAR alignment software.⁴⁴ Quasi-constant transcripts showing the same values for most of the samples were removed for multivariate analysis.

2.8 Analysis

Statistical analyses first considered the following covariates: social status (dominant vs. subordinate), hemisphere from which brain tissue was collected (right vs. left), and time of day (morning vs. afternoon) of necropsy. The two-sided Mann–Whitney U test did not reveal any significant differences between the distributions of these covariates across the two groups, so they were eliminated from subsequent analyses (Figure S1B). We used the framework from Berson et al.²⁵ to analyze and visualize the single pre-synapses. Differentially expressed synaptic markers were identified using multi-testing corrected Wilcoxon's test (p < 0.05) using the Benjamini-Hochberg correction false discovery rate (FDR).⁴⁵ All statistical tests and correction algorithms were performed using the SciPy open-source software (v1.14.1).⁴⁶

Decision Tree and L1-penalized linear models were used within a regression framework to predict SynTOF mean expression (Figure S2). The models were trained to minimize the Mean Square Error loss. The input variables—that is, features—to the ML models were (1) the transcriptomic data and (2) the normalized multi-system phenotypes. A feature selection process was conducted to identify the variables most relevant to model prediction. A two-loop nested cross-validation framework was utilized, using mean absolute error as the evaluation criterion, to mitigate overfitting and ensure unbiased model performance evaluation.⁴⁷ The hypermeters we considered included the model type (decision tree or linear model) and the number of selected features. Specifically, the dataset was first split into outer training and test sets using the RepeatedStratifiedFold cross-validation function from scikit-learn to ensure balanced splitting across folds. It was configured with 10 splits, 50 repeats, and an initial random state = 1. The outer training set was then further split into an inner training and test set using the StratifiedKFold strategy with 10 splits. The inner training set was used to train the model to predict the SynTOF markers, and the inner test was used to evaluate their performance. The best-performing model (decision tree vs. linear model), minimizing error on the inner test set, was selected and validated in the outer loop to ensure the robustness of the results. Reported correlations were quantified using SciPy Spearman methods. The best-performing model is shown in Figure 3A.

An output feature was deemed predictable if the Spearman correlation between its ground truth and predicted expression was positive and significant (p < 0.05). For downstream analysis, the best minimal model (i.e., the model with the lowest number of selected features) that passed these criteria was chosen.

Feature importance was determined using the SHAP algorithm⁴⁸ as it has been shown to outperform conventional feature selection methods and lead to better model interpretability. All the models were run on Python (3.12.8) using scikit-learn packages (v 1.6.0) with default hyperparameters.⁴⁹ All the p-values reported were adjusted for multi-testing correction using the Benjamini-Hochberg algorithm³⁹ from the SciPy software.⁴⁶ All the figures were generated using Matplotlib (v3.9.4)⁵⁰ and Seaborn (v0.13.2).⁵¹

3 RESULTS

3.1 Diet altered the presynaptic proteome in the NHP cerebral cortex

We first investigated the impact of the Western versus Mediterranean diet on presynaptic molecular composition using 30 SynTOF markers with non-zero expression (Figure S1A). Six presynaptic proteins were elevated in the Mediterranean compared to the Western diet group: dopamine transporter (DAT), beta-amyloid₁₄₂ (Aβ42)_, calreticulin, light chain 3B (LC3B), K48-ubiquitin, and the creatine transporter solute carrier family six member 8 (SLC6A8) (all p ≤ 0.01) (Figure 2A,B). Proteomic expression at the synapse level enabled accurate prediction of membership within the diet groups (cross-validated area under the receiver operating characteristic curve [AUROC]: 0.87) (Figure 2C). The content of specific synaptosomal proteins did not reflect the corresponding steady-state transcript levels for each individual target in bulk transcriptomics in the adjacent temporal cortex (p > 0.1).

FIGURE 2

Effects of diet on NHP presynaptic proteome. Results of a randomized, controlled preclinical trial of middle-aged, female cynomolgus macaques to determine the impact of long-term consumption of Western vs. Mediterranean diets. Data presented here represent presynaptic proteomic phenotypes in the lateral temporal cortex determined by SynTOF mass spectrometry. (A) Differential effects of diet were assessed using pseudo-bulk presynaptic relative contents of SynTOF targets. The SynTOF-detected levels that were significantly different between the two diet groups are annotated (multi-testing adjusted Wilcoxon's p < 0.05, fold-change > 0.1). (B) Box plot illustrating effects of diet on SynTOF-detected presynaptic levels. **p < 0.01 ***p < 0.001. (C) SynTOF markers predicted the diet group with a cross-validated AUROC of 0.87 using a linear model (Ridge regression). (D) UMAP visualization of single-presynapse organization in NHP brain illustrating 15 individual clusters based on results from 30 SynTOF probes. (E) Heat map illustrating relative concentrations of SynTOF peptides in individual presynaptic clusters. AUROC, area under the receiver operating characteristic curve; NHP, nonhuman primate; SynTOF, synaptometry by time-of-flight.

We also assessed diet-induced differences in single presynapses using a comparative framework previously published.²⁵ Consistent with our prior SynTOF studies of multiple regions of the human and NHP brain, including the cerebral cortex, we identified 15 presynaptic clusters which could be subdivided into “high expressor” clusters (e.g., 4, 7, and 9), and “low expressor” clusters (e.g., 3 and 13) (Figure 2 D,E).^{25, 40} No diet effects were observed on the SynTOF total mean event frequency—that is, the frequency of occurrences in which a particle was detected by the flow cytometer—across presynaptic clusters (Figure S3A). However, presynaptic levels of EAAT1, a marker of astrocytic reaction, were significantly higher in clusters 4 and 7, while levels of K-48 ubiquitin, a marker of proteasomal protein degradation, were significantly lower in cluster 11 in the Western compared to the Mediterranean group (Figure S3B).

3.2 Tissue transcriptomic profiles broadly predict diet-induced changes in presynaptic molecular composition

The SynTOF markers selected represent a complex mix of proteins, proteolytic fragments, and post-translational modifications. As noted above, presynaptic peptide concentrations were not correlated with the overall transcript level for the parent protein, not surprising considering post-translational processing was responsible for generating many of the presynaptic proteins/peptides, and that bulk RNAseq measures the steady state of mRNA across the tissue and cannot distinguish cell type or cell region origin. In addition, neurons are structurally complex with significant spatial distribution of function, including local mRNA and protein distribution and regulation.⁵² We investigated the extent to which overall transcriptional profiles in the adjacent temporal cortex predicted the relative concentrations of SynTOF markers using the same cross-validated machine-learning pipeline (see the Methods section). In these models, diet was included as a covariate. Global transcriptomic profiles significantly predicted all SynTOF markers with a minimal number of features (30/30 SynTOF markers p < 0.05, 22/30 p < 0.001, [Figure 3A,B and Figure S4]). Spearman rho correlations between each of the 30 SynTOF markers and transcriptomic data are given in Figure S4. The strongest associations between SynTOF markers and cortical gene expression are shown in Figure 3C. We observed strong positive correlations between SPATA22 transcript levels and three SynTOF markers, including TMEM230, Aβ40, and LRRK2 (adjusted p < 0.05), and a weaker correlation with GAMT (an enzyme responsible for creatine synthesis) (raw p < 0.05, adjusted p < 0.08). Strong positive correlations were also observed between TFAP2C (Transcription Factor AP-2 Gamma) and pTau, CD47, PARKIN, and GAD65 (adjusted p < 0.02) (Table S4).

FIGURE 3

Transcript correlations with SynTOF markers. Results of a randomized, controlled preclinical trial of middle-aged, female cynomolgus macaques to determine the impact of long-term consumption (31 months, equivalent to ∼9 years for humans) of either a Western or Mediterranean diet. Data presented are from the lateral temporal cortex using SynTOF mass spectrometry and RNAseq transcriptomics. (A) Cross-validated performance of diet-adjusted machine learning models predicting SynTOF marker expression from the transcriptional profile (p < 0.05) using a feature selection algorithm. Only the best models per SynTOF marker are shown here (see the Methods section). (B) Distribution of SynTOF marker model prediction p values by the number of selected transcripts. Only a few transcripts are needed to enable the prediction of all the SynTOF markers (30/30 SynTOF markers p < 0.05, 22/30 p < 0.001). (C) Associations between individual transcripts and SynTOF markers. Transcript contribution to the model prediction was quantified using the SHAP algorithm. Node sizes are proportional to the importance of the transcript in SynTOF marker prediction. Node colors represent the Spearman correlation between SynTOF markers and corresponding gene expression. SynTOF, synaptometry by time-of-flight.

3.3 Cerebral cortical presynaptic protein associations with diet-induced changes in multi-system phenotypes, including lipid metabolism, cardiovascular function and atherosclerosis, stress reactivity, and socioemotional behavior

We used a cross-validated machine-learning pipeline to assess correlations of the presynaptic proteome with multi-system phenotypes, including body composition, neuroimaging, behavioral, and physiological characteristics, defined in Table S2, with diet included as a covariate. The multi-system phenotypes predicted 26 out of 30 total SynTOF markers (p < 0.05) (Figure 4A and Figure S5). A correlation network between multi-system phenotypes and SynTOF markers (Figure 4B) showed that other than intra-synaptosome correlations with each other (dark blue dots), the SynTOF markers were most strongly correlated with hepatosteatosis (light purple/gray dots) and neuroanatomical volumes and thicknesses (red dots).

FIGURE 4

Associations of SynTOF markers with multi-system phenotypes. Data presented are multi-system phenotypes previously described (see Table S2), and lateral temporal cortex proteomic data determined by SynTOF mass spectrometry. (A) Cross-validated performance of diet-adjusted machine learning models predicted 26 of 30 SynTOF marker expressions from the multi-system phenotypes (p < 0.05), with and without feature selection. Only the best nested cross-validated models per marker are shown (see the Methods section). (B) Correlation network between multi-system phenotypes and SynTOF markers. Node size represents the strength of the univariate association of the covariate with diet (−log(p-value)). Edges represent the top 10% of the correlation between different covariates. (C) Association of SynTOF markers with behavioral, physical, neuroimaging, and physiologic characteristics. Horizontal dotted lines represent major clusters of SynTOF markers; vertical dotted lines delineate classes of phenotypes (e.g., behavior, stress, CSF biomarkers etc.). Node sizes are proportional to the importance of the phenotype in SynTOF prediction determined using the SHAP algorithm (39). Node colors represent the Spearman univariate correlation between SynTOF markers and phenotypes. Clustering was performed on the univariate correlation expression matrix using pair-wise Euclidian distance. SynTOF, synaptometry by time-of-flight.

We explored multi-system phenotype associations with SynTOF markers using univariate Spearman correlations and their contribution to model prediction using the SHAP algorithm (Figure 4C). Selected multi-system phenotypes were grouped into classes by similarity (e.g., neuroanatomy, stress physiology, socioemotional behavior) along the horizontal axis, and major clusters of SynTOF markers are indicated along the vertical axis of Figure 4C. MRI-determined neuroanatomy phenotypes were important predictors of the SynTOF marker glial fibrillary acidic protein (GFAP). GFAP was positively correlated with percent change (between baseline and study end) in white matter (Rho = 0.62, adjusted p < 0.01). Also notable, hepatosteatosis was positively associated with the SynTOF markers DAT, Aβ42, Aβ40, and K48-Ubiquitin (Rho > 0.55, adjusted p < 0.05). These relationships were buttressed by significant associations with calories consumed, body composition, and carbohydrate metabolism.

Individual associations of the 26 SynTOF markers with each phenotype are presented in Table S5. Several SynTOF markers were correlated with percent changes in brain volumetrics, including total brain (a-synuclein_pS129, LC3B), deep gray matter (a-Synuclein_pS129, LRRK2), caudate (VMAT2, ApoE), and cortical thicknesses of the metaROI (LRRK2) (|Rho| > 0.30, p-value before adjustment < 0.05). Likewise, several SynTOF markers were also associated with socioemotional behavior. Social isolation measured as percent time spent alone showed a high correlation with 3 of the markers (b-Amyloid_X40, DAT, LRRK2) (Rho ← 0.40, p-value before adjustment < 0.05) and percent time spent in body contact with conspecifics was positively correlated with 5 of the markers (b-Amyloid_X42, b-Amyloid_X40, DAT, SLC6A8, GAMT) (Rho > 0.35, p-value before adjustment < 0.05). Percent time spent exhibiting depressed behavior positively correlates with 3 SynTOF markers (LC3B, DAT, GAMT) (Rho > 0.33, p-value before adjustment < 0.05). Furthermore, related measures of the hypothalamic-pituitary-adrenal response to stress—that is, the plasma cortisol response to an adrenocorticotropic hormone (ACTH) challenge (b-Amyloid_X40, DAT, TMEM230_C20orf30) and to an acute social stress test (3NT, Casp, PrP) were each correlated with 6 SynTOF markers (|Rho| > 0.30, p-value before adjustment < 0.05).

4 DISCUSSION

The primary goal of this experiment was to determine the impact of two commonly consumed dietary patterns—the Western and Mediterranean diets—on the presynaptic proteome in the primate lateral temporal cortex and secondarily to assess the relationships of those proteomes with other key phenotypes in this preclinical study. Here, we utilized SynTOF, a powerful technology for quantifying single presynapse molecular diversity^{24, 25} along with a validated 30-plex SynTOF panel to assess presynapse proteins from lateral temporal cortex.²⁵ These presynaptic markers were deliberately selected to sample mechanisms broadly thought to contribute to neurodegenerative diseases. Notably, six presynaptic proteins exhibited elevated levels in the Mediterranean diet group compared to the Western diet group, including dopamine transporter (DAT), beta-amyloid1-42 (Aβ42), calreticulin, light chain 3B (LC3B), K48-ubiquitin, and the creatine transporter SLC6A8. These presynaptic protein alterations were not mirrored by the steady-state transcript levels within the adjacent temporal cortex, a finding that is not unexpected as numerous targets are products of transcript-independent post-translational processing and events, as noted below. Differences in Aβ42 levels between the Mediterranean and Western diet presynapses suggest there are alterations in the endoproteolytic processing of presynaptic amyloid precursor protein (APP), a key feature of AD. Likewise, increased presynapse DAT expression indicates a response of the dopaminergic system to the Mediterranean diet, and others have observed that adherence to a Mediterranean diet in mid-life lowers the risk of Parkinson's disease.⁵³ Finally, genetic variants in SLC6A8 that lead to reduced creatine transporter activity are a common cause of X-linked neurobehavioral changes, including autism.^{54, 55} Our results may provide a presynapse molecular rationale for the Mediterranean diet as an adjuvant therapy.

We also identified distinct presynaptic clusters characterized by differential expression patterns, with individuals falling into high versus low expression groups. While there were no significant diet-induced effects on overall presynaptic event frequency across the clusters, the presynaptic markers EAAT1 and K-48 ubiquitin exhibited differential levels between diet groups within specific clusters. The Western group exhibited increased presynaptic EAAT1, a phenotype previously observed in high-expressing presynapses from cerebral cortical regions of the human brain impacted by AD (Figure S3B). EAAT1 is an astrocytic protein tightly coupled to synaptic glutamate transport, which has been shown by others to be elevated in the brain of mouse models of AD, perhaps related to neuroinflammation.⁵⁶ Furthermore, global transcriptomic profiles in the adjacent cortex predicted the relative concentrations of SynTOF markers, highlighting the interplay between transcriptional regulation and presynaptic protein expression.

To our knowledge, this is the first study to explore the relationships between the presynaptic proteome and the transcriptome in adjacent areas of the cortex of primates. We observed strong positive correlations between two transcripts—SPATA22 (spermatogenesis associated 22) and TFAP2C (Transcription Factor AP-2 Gamma)—and several SynTOF markers. SPATA22 was associated with TMEM230, Aβ40, and LRRK2 (p < 0.05) and with GAMT (an enzyme responsible for creatine synthesis) (p < 0.05, adjusted p < 0.08). SPATA22, originally discovered for its role in meiosis, is highly expressed in the human brain and involved in protein deubiquitination, regulation of necroptosis, and tumor necrosis factor (TNF) -mediated signaling.⁵⁷ SPATA22 may be important in oligodendrocyte physiology, and further study is needed to better understand its potential implications for central nervous system health. TFAP2C was positively correlated with presynaptic pTau, CD47, PARKIN, and GAD65. TFAP2C is a DNA-binding transcription factor involved in retinoic acid-mediated differentiation, which has been associated with the expression of the estrogen .receptor.⁵⁸ Taken together, these relationships exhibit the complex interplay between the transcriptome and proteome in the temporal cortex.

Multivariate correlation analyses revealed several associations between presynaptic protein profiles and multi-system phenotypes, with neuroanatomical changes and liver hepatosteatosis showing particularly strong correlations with synaptic protein markers. These findings underscore the multifaceted impact of diet on presynaptic molecular composition and its implications for brain health and function across various physiological domains.

Of the associations between SynTOF markers and neuroanatomical phenotypes, the strongest were between GFAP and change in white matter, caudate, and superior temporal gyrus thickness over the course of the experiment. Known functions of GFAP include astrocyte cytoskeletal structure and strength and support for adjacent neurons and the blood–brain barrier.⁵⁹ GFAP is a protein marker for astrocyte activity, including neuroinflammation. Elevated GFAP in CSF is predictive of the progression of neurodegenerative diseases (e.g., AD, Parkinson's disease) and cognitive decline.^{60, 61} Previously, we observed relationships between a neuroinflammatory cortical transcriptome and brain volumetrics in this group of NHPs and hypothesized that changes in brain volumetrics were due to neuroinflammation.²² The relationships observed here between GFAP and neuroanatomy support the connection between neuroanatomic changes and neuroinflammation.

Hepatosteatosis, as measured by reduced computed tomography (CT) attenuation due to lipid accumulation, was associated with > 60% of the measured SynTOF markers, suggesting potential relationships between diet-induced changes in hepatic lipid metabolism, fat accumulation, and presynapse protein content. The Western diet increased hepatosteatosis in a subset of subjects in this study.¹⁹ Metabolic-associated fatty liver disease (MAFLD) (a.k.a. non-alcoholic fatty liver disease [NAFLD]) is a common chronic liver condition affecting a large population. MAFLD has been associated with cognitive impairment, including compromised executive function and global cognitive decline. These cognitive decrements are associated with the insulin resistance, lipotoxicity, and systemic inflammation that accompanies MAFLD.⁶² Furthermore, mid-life MAFLDM predicts later-life AD and related dementias.⁶³ Processes associated with MAFLDM and neurobiological transitions warrant further study to better understand the pathophysiologic mechanisms underlying this relationship.

A notable strength of this study lies in its growing body of evidence demonstrating the connections between the brain, body, and behavior. Indeed, several of these medial temporal cortical SynTOF markers were associated with social isolation, social contact, and depression. Cynomolgus macaques—like human beings—are highly gregarious and rely on their social relationships. This species has been used in numerous studies of factors that impact socioemotional behaviors, which, in turn, impact health. In cynomolgus macaques, social isolation in the presence of a Western diet increases coronary artery atherosclerosis and related cardiovascular risk factors.^{64, 65} Depression has also been extensively studied in cynomolgus macaques, generating 32 peer-reviewed papers between 1997 and 2023, including descriptions of related temporal lobe neuroanatomy and neurophysiology (for reviews, see Ref.66-68). The limbic system, which runs through the temporal lobe, is a prime mediator of socioemotional behavior. These behaviors are often related to physiological stress responsivity (cortisol response to ACTH, cortisol response to acute social isolation), which were also associated with several of the SynTOF markers. Our findings underscore the complexity of dietary influences on an individual's health, including affecting how an individual engages with their social environment. Our previous work has demonstrated a strong overlap between human and macaque presynaptic composition in the cerebral cortex and neostriatum, which was significantly divergent from the mouse,²⁵ emphasizing the critical importance of the primate model in this study.

Weaknesses of this trial include a lack of cognitive assessments, which would have been valuable as phenotypic comparators to the presynaptic proteome but beyond the scope of this trial. Budgetary constraints limited this study to females, so we were unable to explore potential sex differences in responses. Finally, associations between the presynaptic proteome and multi-system phenotypes are presented primarily for hypothesis generation, and any causal relationships remain to be determined.

In summary, this study of middle-aged, female cynomolgus macaques combines the strengths of a controlled randomized preclinical trial, a NHP model, transcriptomics, and multi-system phenotyping with the powerful technology of SynTOF. An important strength of this trial was the examination of human-like diet patterns. We chose to examine human diet patterns rather than single nutrients because it is now widely recognized that foods begin to interact as soon as they are consumed. Therefore, single-component analysis may reveal little about what is happening in situ with a complete diet. Likewise, we specifically did not compare human translational diet patterns to monkey chow because monkey chow is not like any human diet or any monkey diet in the wild. The study reported here builds on a growing body of data supporting the broad impact of diet composition on brain, behavior, and disease susceptibility. We elucidated single presynaptic proteomic diversity in response to two commonly consumed human diet patterns, Mediterranean and Western, which may provide a presynapse molecular rationale for the Mediterranean diet as an adjuvant therapy for neurological diseases. Relationships between SynTOF markers and multi-system phenotypes highlight the importance of neuroinflammation in changes in brain volumetrics. Numerous associations between SynTOF markers and hepatosteatosis suggest a potential role for diet-induced changes in lipid metabolism and presynapse molecular composition. SynTOF-marker associations with socioemotional behavior and physiological responses to social, environmental stress provide new areas for exploration. These observations will likely have great translational significance due to the rigor of the design and implementation of the preclinical trial, the similarity of the NHP model to human biology, and the significance of diet composition to lifelong health. Overall, this work provides important insights into the neurobiological effects of diet and extends our knowledge regarding the complexity that environmental influences such as diet have on the body and the brain.

ACKNOWLEDGMENTS

This work was supported by grants from NIH: RF1 AG077443 (T.J.M.), R01 HL087103 (C.A.S.), R01 HL122393 (T.C.R.), U24DK097748 (T.C.R.), RF1AG058829 (C.A.S. and S.C.), P30AG072947, and T32AG033534. E.B. is funded by the Phil & Penny Knight Initiative for Brain Resilience at the Wu Tsai Neurosciences Institute, Stanford University.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to disclose. Author disclosures are available in the Supporting information.

CONSENT STATEMENT

No human subjects were involved.

Open Research

DATA AVAILABILITY STATEMENT

The code is freely available at https://github.com/elo-nsrb/Brain-study-NHP-2024. Raw data can be downloaded from Dryad (10.5061/dryad.stqjq2cc8).