Medium-chain triglycerides improve cognition and systemic metabolism in mouse models of Alzheimer's disease

Abstract

Lifestyle-based interventions, including dietary modifications, can reduce dementia risk. In this regard, dietary supplementation with medium-chain triglycerides (MCT) has shown potential therapeutic benefits in individuals with Alzheimer's disease. These effects are widely presumed to be mediated by hepatic conversion of MCT into circulating ketones. However, the physiological and cellular mechanisms underlying the benefits of MCT remain understudied, particularly in the context of Alzheimer's disease. Here, we investigated the cellular and molecular changes occurring in the brain and systemically in response to dietary supplementation with MCT versus a ketogenic diet. The experimental design consisted of comparing a 70% carbohydrate control diet to either a control diet supplemented with 10% MCT or a carbohydrate-free high-fat ketogenic diet. Diets were tested in two Alzheimer's disease mouse models, slow-progressing 3xTg-AD mice that model pre-symptomatic/early stages and rapidly progressing 5xFAD mice that model late stages of the disease. We found that MCT supplementation and a ketogenic diet both improved hippocampal-dependent spatial learning and memory, increased dendritic spine density of hippocampal neurons and modulated hippocampal expression of genes associated with mitochondrial functions, synaptic structure and insulin signalling in Alzheimer's disease mouse models. However, unlike the ketogenic diet, MCT supplementation did not elevate circulating ketones, suggesting different mechanisms. Indeed, MCT supplementation enhanced the peripheral insulin response of Alzheimer's disease mice, while the ketogenic diet conversely unveiled their latent metabolic vulnerability, increasing their hyperglycaemia, body weight gain and adiposity. The systemic metabolic disturbances of Alzheimer's disease mice correlated with transcriptomic alterations in hepatic lipid metabolism and ketogenesis genes and increased lipid droplet accumulation. These liver metabolic abnormalities were partially reversed by both MCT supplementation and the ketogenic diet, but in distinct ways. Notably, the ketogenic diet selectively triggered hepatic neutral lipid depletion and prominent proinflammatory gene expression, while MCT downregulated expression of cholesterol-related genes. Collectively, these findings reveal that MCT supplementation in the context of Alzheimer's disease improves cognition and systemic metabolism without elevating circulating ketone levels.

Keywords: Alzheimer’s disease; cognition; ketogenic diet; ketones; medium-chain triglycerides; peripheral metabolism.

Figure 1

One month of either MCT or KD improves cognition in symptomatic Alzheimer's disease mice, but MCT does so without inducing hyperketonaemia. (A) One month of dietary intervention in 6-month-old female 5xFAD and their wild-type (WT) littermate control mice. Created in BioRender. Mbra, P. (2025) https://BioRender.com/g0o387c. Mice received either a Control diet (C), a medium-chain triglyceride-supplemented diet (MCT/M) or a carbohydrate-free, high fat ketogenic diet (KD/K) for 1 month and the Morris water maze was performed at the end of the interventions (n = 7–11 animals/genotype/diet). (B) Macronutrient proportion in the three diets (detailed in Supplementary Table 1). (C) Longitudinal measures of blood β-hydroxybutyrate (BHB). Genotype and diet effects are evaluated by two-way ANOVA for each time point. Baseline: not significant; Day (D) 14: significant diet effect [P < 0.0001, F(2,56) = 16.15]; D35: significant diet effect [P < 0.0001, F(2,56) = 27.38]. Data are expressed as median and min/max, with box plot identifying the 25%–75% range. (D–G) Morris water maze. (D) Average escape latency of four trials/day during the learning phase. Experimental group (combination of genotype and diet as a single variable) and time effects are evaluated by a mixed-effects model and show a significant time effect [P < 0.0001, F(2.747,141.9) = 27.67]. Data are expressed as mean ± SEM. (E) Learning rate is the average slope of escape latency of each mouse from Day 1 to Day 3. Genotype and diet effects are evaluated by two-way ANOVA and are not significant. Data are expressed as median and min/max, with box plot identifying the 25%–75% range. (F) Ability to discriminate target quadrant (TQ) from opposite quadrant (OQ). Preference (left y-axis) was defined as the difference between the total time spent in the quadrants and the expected baseline duration of 25% (right y-axis). Experimental group (combination of genotype and diet as a single variable) and quadrant discrimination factors are evaluated by two-way ANOVA, showing a significant quadrant discrimination factor [P < 0.0001, F(1,90) = 53.33]. Results of Fisher's post hoc test comparing TQ versus OQ for each group are presented on the graph as follows: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Data are expressed as median and min/max, with box plot identifying the 25%–75% range. (G) Merged heat maps per group indicating the preferred location of the mice. The pool is separated into four quadrants: TQ contained the platform during the training phase and OQ is 180° from TQ. Pairwise comparisons were conducted using Fisher's post hoc test. In C–E, asterisks indicate a significant diet effect (difference with the Control diet) within the same genotype; number signs indicate a significant genotype effect (difference with the WT group) within the same diet. Statistical significance is presented on the graphs as follows: *^/#P < 0.05; **^/##P < 0.01; ***^/###P < 0.001; ****^/####P < 0.0001.

Figure 2

Long-term administration of MCT and KD in pre-symptomatic Alzheimer's disease mice increases spine density in hippocampal dentate gyrus. (A) Dietary interventions in pre-symptomatic Alzheimer's diseasemice: 3-month intervention in 5xFAD mice (n = 6 animals/genotype/diet) starting at 2 months of age, 5- to 6-month intervention in 3xTg-AD mice starting at 3 months of age (n = 6 animals/genotype/diet for 5-month intervention, n = 17–28 animals/genotype/diet for 6-month intervention). For each experiment, genotype and diet effects are evaluated by two-way ANOVA. Created in BioRender. Mbra, P. (2025) https://BioRender.com/g0o387c. (B) Blood β-hydroxybutyrate (BHB) levels in 3xTg-AD mice (n = 17–22/genotype/diet), two-way ANOVA for each time point. Baseline: genotype effect [P = 0.0234, F(1,108) = 5.29]; Day 14 (D14): diet effect [P < 0.0001, F(2,115) = 117.30]; and interaction between diet and genotype [P < 0.0001, F(2,115) = 20.47]; Day 171 (D171): diet [P < 0.0001, F(2,112) = 20.54] and genotype effects [P = 0.0480, F(1,112) = 3.40] with interaction [P < 0.0001, F(2,112) = 10.07]. (C–F) Representative pictures of neurons stained with Golgi-Cox staining in a brain section (C, ×4 magnification, scale bar = 250 μm) and within the regions of interest: (D) the dentate gyrus (DG, ×100 magnification, scale bar = 5 μm) and (E) the cornu ammonis 1 (CA1, ×100 magnification, scale bar = 25 μm). The spines (F, scale bar = 5 μm) were quantified on secondary (2°, represented by II) and tertiary (3°, represented by III) dendrites (n = 50–90 neurons/genotype/diet). (G–J) Quantification of spine density in 5xFAD mice. (G–H) Spine density in DG. (G) Secondary dendrites {two-way ANOVA identifies diet effect [P = 0.0003, F(2,447) = 8.178] and interaction between diet and genotype [P = 0.0008, F(2,447) = 7.31]}. (H) Tertiary dendrites {two-way ANOVA identifies diet effect [P < 0.0001, F(2,446) = 12.91]}. (I and J) Spine density in CA1. (I) Secondary dendrites {two-way ANOVA identifies genotype effect [P = 0.0040, F(1,390) = 8.40]}. (J) Tertiary dendrites {two-way ANOVA identifies genotype effect [P = 0.0001, F(1,390) = 14.71]}. (K–N) Quantification of spine density in 3xTg-AD mice. (K and L) Spine density in DG. (K) Secondary dendrites {two-way ANOVA identifies genotype effect [P = 0.0010, F(1,442) = 11.07] and interaction between diet and genotype [P = 0.0038, F(2,442) = 5.66]}. (L) Tertiary dendrites {two-way ANOVA identifies diet effect [P = 0.0060, F(2,440) = 5.18] and a trend in genotype factor [P = 0.069, F(1,440) = 3.31]}. (M and N) Spine density in CA1. (M) Secondary dendrites (two-way ANOVA: not significant). (N) Tertiary dendrites {two-way ANOVA identifies genotype effect [P = 0.0031, F(1,405) = 8.88]}. All data are expressed as median and min/max, with box plot identifying the 25%–75% range; pairwise comparisons were conducted using Fisher's post hoc test. Asterisks indicate significant diet effect (difference with the Control diet) within the same genotype; number signs indicate significant genotype effect (difference with the WT group) within the same diet. Statistical significance is presented on the graphs as follows: *^/#P < 0.05; **^/##P < 0.01; ***^/###P < 0.001; ****^/####P < 0.0001. C = Control diet; KD/K = carbohydrate-free, high fat ketogenic diet; MCT/M = medium-chain triglyceride supplemented diet.

Figure 3

MCT and KD prevent transcriptomic changes related to energetic, neurogenic and synaptic processes within the hippocampus of 3xTg-AD mice. Bulk RNA sequencing was performed on the hippocampus of randomly chosen mice within the 6-month intervention cohort of 3xTg-AD mice (Fig. 2A, n = 3–5/genotype/diet). (A) Heat map of differentially expressed genes (DEGs) between wild-type (WT) mice and 3xTg-AD mice on the Control diet (601 Alzheimer's disease-related DEGs (AD-DEGs) between WT_C versus 3xTg_C including 274 upregulated and 326 downregulated, DESeq2 cut-off P < 0.01) with their average expression throughout experimental groups (n = 3–5/genotype/diet). (B) Venn diagram of the following comparisons: AD-DEGs (WT_C versus 3xTg_C), KD effect (WT_C versus 3xTg_K), MCT effect (WT_C versus 3xTg_M); 385 AD-DEGs (sum of the asterisks) no longer significantly different in WT_C versus 3xTg_K and/or WT_C versus 3xTg_M. (C) Top 10 Gene Ontology (GO): biological processes [false discovery rate (FDR) < 0.05] associated with AD-DEGs that are either upregulated (left) or downregulated (right). (D) Networks of significant pathways associated with AD-DEGs (GO: molecular function, FDR < 0.05). Each pathway is represented by a green circle, connected pathways share at least 30% genes, brighter green circles are more significantly enriched pathways, larger green circles are pathways with larger gene sets and thicker lines represent more overlapped genes between connected pathways. The three largest networks of pathways are encircled by dash lines: (1) energy metabolism and transmembrane transporters; (2) synaptic structure; (3) cell growth. (E–H) Heat maps of AD-DEGs associated with main networks of pathways. (E) Energy metabolism and transmembrane transporters for network (1); (F) synapse structure for network (2); (G) cell growth: tyrosine-kinase receptors signalling for network (3); and (H) cell growth: IEGs (immediate early genes) in the pathway ‘transcription coactivator’ associated with network (3). C = Control diet; KD/K = carbohydrate-free, high fat ketogenic diet; MCT/M = medium-chain triglyceride supplemented diet.

Figure 4

MCT, unlike KD, mitigates abnormal glucose tolerance and reinforces insulin response in pre-symptomatic Alzheimer's disease mice. (A) Timeline of metabolic measures in 3xTg-AD mice subjected to 6-month dietary interventions: glycaemia and ketonaemia, body composition analysis, indirect calorimetry and intraperitoneal (IP) glucose/insulin tolerance test (GTT, ITT). Peripheral tissues (liver, spleen, adipose tissue) were collected for post-mortem analysis. For each experiment genotype and diet effects are evaluated by two-way ANOVA. When time is a factor, time and experimental group (combination of genotype and diet as a single variable) are evaluated by a mixed-effects model. Created in BioRender. Mbra, P. (2025) https://BioRender.com/g0o387c. (B) Glucose/ketone index (GKI) is the ratio of blood glucose levels to blood β-hydroxybutyrate (BHB). GKI < 10 indicates a ketosis state. Two-way ANOVA for each time point shows: baseline: not significant, Day (D)14: significant diet effect [P < 0.0001, F(2,115) = 27.42] and significant interaction between diet and genotype factors [P = 0.0235, F(2,115) = 3.88], D171: significant diet effect [P < 0.0001 F(2,112) = 20.89] and genotype effect [P = 0.0212, F(1,112) = 5.47] with interaction [P = 0.0016, F(2,112) = 6.86]. Data are expressed as median and min/max, with box plot identifying the 25%–75% range. (C and D) Calorimetry. (C) Respiratory exchange ratio (RER) average over 24 h (n = 8–12/genotype/diet). Two-way ANOVA shows diet effect [P < 0.0001, F(2,60) = 62.43]. (D) Cumulative measures of RER over 24 h showing variation from the dark phase to the light phase (n = 5–7/genotype/diet). Mixed-effects model shows significant time effect [P < 0.0001, F(5.937,198.4) = 8.41] and experimental group effect [P < 0.0001, F(5,29) = 18.96]. (E–G) Caloric intake and body fat. (E) Average daily food intake over 6 months on diet (n = 22–28/genotype/diet). Two-way ANOVA shows diet effect [P < 0.0001, F(2,148) = 71.68] and genotype effect [P < 0.0001, F(1,148) = 39.5] with interaction [P = 0.0314, F(2,148) = 3.54]. (F) Body weight gain per caloric intake. Two-way ANOVA shows diet effect [P < 0.0001, F(2,148) = 21.83] and genotype effect [P < 0.0001, F(1,148) = 29.54] with interaction [P < 0.0001, F(2,148) = 10.36]. (G) Fat pad dissection (n = 17–24/genotype/diet). Abdominal fat: two-way ANOVA shows diet effect [P < 0.0001, F(2,124) = 37.39] and genotype effect [P < 0.0001, F(1,124) = 72.76] with interaction [P < 0.0001, F(2,124) = 25.70]. Subcutaneous fat: two-way ANOVA shows significant diet effect [P < 0.0001, F(2,125) = 35.02] and genotype effect [P = 0.0046, F(1,125) = 8.32] with interaction [P < 0.0001, F(2,125) = 13.50]. Brown back fat: two-way ANOVA shows diet effect [P < 0.0184, F(2,125) = 4.13]. (H and I) Intraperitoneal insulin tolerance test (ITT) after 4 months on diet (n = 11–17/genotype/diet). (H) Curves of blood glucose change after IP injection of insulin. Mixed-effects model shows significant time effect [P < 0.0001, F(3.202,254.9) = 22.36] and experimental group effect [P = 0.0119, F(5,80) = 3.15] with interaction [P < 0.0001, F(25,398) = 2.57]. (I) Rate of glucose clearance between T = 0 min and T = 45 min post-insulin injection (rate of insulin-induced hypoglycaemia). Two-way ANOVA shows significant diet effect [P = 0.0033, F(2,82) = 6.13]. (C–I) Data are expressed as mean ± SEM, pairwise comparisons were conducted using Fisher post hoc test. Asterisks indicate significant diet effect (difference with the Control diet) within the same genotype; number signs indicate significant genotype effect (difference with the WT group) within the same diet. Statistical significance is presented on the graphs as follows: *^/#P < 0.05; **^/##P < 0.01; ***^/###P < 0.001; ****^/####P < 0.0001. C = Control diet; KD/K = carbohydrate-free, high fat ketogenic diet; MCT/M = medium-chain triglyceride supplemented diet.

Figure 5

MCT supplementation and KD partially reverse hepatic lipid dysregulation of Alzheimer's disease mice. (A and B) Hepatic lipid content. (A) Hepatocellular lipid droplet accumulation observed with haematoxylin-eosin (HE, magnification ×10), Oil Red O (ORO, magnification ×10) and boron dipyrromethene (BODIPY) staining (magnification ×20). CV = central vein, PV = portal vein. Scale bars = 250 μm. (B) Quantification of lipid vacuoles from HE images with ImageJ (n = 3–4/genotype/diet). Two-way ANOVA shows significant diet effect [P = 0.0007, F(2,19) = 10.92] and genotype effect that approaches significance [P = 0.066, F(1,19) = 3.82]. Data are expressed as mean ± SEM; pairwise comparisons were conducted using Fisher post hoc test: **P < 0.01 (C–J) Liver RNA sequencing. (C) Volcano plot of Liver Alzheimer's disease-related differentially expressed genes [AD-DEGs: DEGs between wild-type (WT) and 3xTg-AD mice on the Control diet (WT_Control = 4, 3xTg_Control = 4, DESeq2, P < 0.01)]. (D) Network of top 20 pathways involving up- and downregulated AD-DEGs [using Kyoto Encyclopedia of Genes and Genomes (KEGG), false discovery rate (FDR) < 0.05]. The circles highlight pathways associated with lipid metabolism (‘Lipids’) and xenobiotic processes (‘Xenobiotic’). (E and F) Heat maps of Liver AD-DEGs (mean of gene expression per group, n = 4/genotype/diet). (E) Liver AD-DEGs associated with lipid metabolism (108 genes); (F) Liver AD-DEGs associated with ketogenesis (6 genes). (G) Pie charts showing the proportion of Liver AD-DEGs restored by MCT and KD (Liver AD-DEGs no longer significantly different between WT_C and 3xTg_M or 3xTg_K). (H) Distribution of ‘Lipids’ Liver AD-DEGs (Liver AD-DEGs associated with lipid metabolism) according to their expression fold change in 3xTg-AD mice on Control, MCT or KD, each compared with baseline (WT mice on the Control diet). The higher the proportion of genes with a fold change near to 0, the closer the 3xTg-AD mice are to WT baseline. (I) Variation rate of average fold change of ‘Lipids’ Liver AD-DEGs restored by MCT (45/108 genes) and KD (61/108 genes). A rate of −100% indicates a complete recovery of expression to WT_C levels while 0 indicates no change in the expression versus 3xTg_C levels. Unpaired t-test shows ***P = 0.0003. (J) Liver AD-DEGs (26/843 genes) significantly inversely deregulated by KD when compared with WT mice on the Control diet. C = Control diet; KD/K = carbohydrate-free, high fat ketogenic diet; MCT/M = medium-chain triglyceride supplemented diet.

Figure 6

Novel liver transcriptomic alterations induced by MCT and KD in Alzheimer's disease mice are both shared and unique. (A) Venn diagram showing how novel diet-induced liver differentially expressed genes (DEGs) were defined. DEGs induced in 3xTg-AD mice by either MCT (3xTg-C versus 3xTg-M) or KD (3xTg-C versus 3xTg-KD) were overlapped with Alzheimer's disease related DEGs (AD-DEGs:WT-C versus 3xTg-C), which were subtracted. (B) Heat map of the 72 novel DEGs significantly modulated by both MCT and KD in 3xTg-AD mice. (3xTg_C versus 3xTg_M, 3xTg_C versus 3xTg_K, DESeq2 cut-off P < 0.01, mean expression per group, n = 4/genotype/diet). (C–E) MCT-specific novel DEGs. (C) Heat map of the 183 genes modulated by MCT but not KD in 3xTg-AD mice (mean expression per group, n = 4/genotype/diet, 3xTg-C versus 3xTg-M, DESeq2, cut-off P < 0.01). (D) Top 10 pathways associated with MCT-specific novel DEGs [Gene Ontology (GO) Biological Process, false discovery rate (FDR) < 0.05]. (E) Heat map of the subset of MCT-specific novel DEGs that are associated with cholesterol metabolism. (F–H) KD-specific novel DEGs. (F) Heat map of genes modulated by KD but not MCT in 3xTg-AD mice (mean expression per group, n = 4/genotype/diet, 3xTg_C versus 3xTg_K, DESeq2, cut-off P < 0.01). (G) Top 20 GO Biological Processes involving KD-specific novel DEGs and highlighting ‘Immune system process’ (cut-off FDR < 0.05). (H) Heat map of the subset of KD-specific novel DEGs that are upregulated immune genes (n = 4/genotype/diet, DESeq2, cut-off P < 0.01, Log2FC > 2). C = Control diet; KD/K = carbohydrate-free, high fat ketogenic diet; MCT/M = medium-chain triglyceride supplemented diet; WT = wild-type.