Gut microbial-derived indole-3-propionate improves cognitive function in Alzheimer's disease

Abstract

Intermittent fasting (IF) offers a potential strategy to counteract Alzheimer's disease (AD) progression. In our 16-week study on AD transgenic mice, IF positively affected cognitive function and reduced amyloid-β (Aβ) accumulation, verifying the IF's role in modulating neuroinflammation. Multiomics integration revealed strong links between IF-induced hippocampal gene expression, gut microbiota, and serum metabolites beneficial for cognition. Indole-3-propionic acid (IPA) emerged as a pivotal microbial metabolite. Blocking its neuronal receptor, pregnane X receptor (PXR), abolished IF's effects. Human data paralleled these findings, showing lower IPA levels in patients with mild cognitive impairment and AD than in controls. IPA supplementation and IPA-producing Clostridium sporogenes reproduced IF's cognitive benefits, whereas PXR blockade in neurons or disruption of IPA synthesis abrogated them. IPA crossed the blood-brain barrier, exhibited potent anti-inflammatory activity, and suppressed Aβ accumulation, essential for neuroprotection. These results underscore microbial metabolites regulated by IF, particularly IPA, as therapeutic candidates for AD, highlighting the critical role of the gut-brain axis in neurodegeneration.

Fig. 1.. IF alleviates cognitive impairment and Aβ deposition and protects the synapse ultrastructure in AD mice.

(A) Schematic of the treatment with IF or ad libitum in each group (WT + ad libitum: n = 20, WT + IF: n = 20, AD + ad libitum: n = 20, AD + IF: n = 20). (B) Body weight change (n = 20). (C) Food intake (n = 20). (D) Escape latency (n = 20). Cognitive function was measured by the Morris water maze test (see Materials and Methods). (E) Representative immunofluorescence images of Aβ plaques in the piriform cortex (n = 6 slices per group). Scale bars, 200 μm. (F) Aβ plaque density (number of plaques per square micrometer) in the piriform cortex (n = 6). (G) Representative images of the ultrastructure of synapse in the hippocampus of female mice (n = 6). (H and I) The length and width of PSD in the hippocampus of mice (n = 6). (J) A z-score scaled heatmap of 175 differentially expressed genes between AD + ad libitum and AD + IF with P < 0.05. The top 25 genes are presented. (K) Gene Ontology (GO) and KEGG pathway annotations for 175 genes whose expression differed between AD + ad libitum and AD + IF. Wnt, wingless-related integration site; mTOR, mammalian target of rapamycin; MAPK, mitogen-activated protein kinase. In (D), * indicates a significant difference between WT + ad libitum and AD + ad libitum, while # indicates a significant difference between AD + ad libitum and AD + IF. Data are mean ± SEM. **P < 0.01 and ^#P < 0.05. Significant differences between means were determined by two-way ANOVA with Tukey’s multiple comparisons test. DAPI, 4′,6-diamidino-2-phenylindole; mTOR, mammalian target of rapamycin; MAPK, mitogen-activated protein kinase.

Fig. 2.. IF restructured the gut microbiome and metabolome of AD mice.

(A) Differences in a selection of six genera and 28 OTUs differed between AD + ad libitum and AD + IF groups. The least-square means ±95% confidence intervals obtained from ANOVA (P < 0.05) are presented. (B) A z-score scaled heatmap of top 10 bacteria that differed between AD + ad libitum and AD + IF with P < 0.05. (C) A z-score scaled heatmap of differential metabolites between AD + ad libitum and AD + IF with P < 0.05. The top 20 differential metabolites are presented. (D) Metabolic pathways influenced by IF treatment in AD mice (AICAR, 5-Aminoimidazole-4-carboxamide1-β-D-ribofuranoside). (E) Plasma levels of IPA (n = 18). (F) DNA level of fldC in colon content (n = 8). Data are mean ± SEM. *P < 0.05 and **P < 0.01. Statistical significance for (E) and (F) was determined by two-way ANOVA followed by Tukey’s multiple comparisons test.

Fig. 3.. Multiomics integration for IF treatment.

(A) The performance of predictive models for signatures of IF status. Omic signatures included 175 differential AD + ad libitum versus AD + IF genes expressed in the hippocampus, 34 gut bacteria, and 87 plasma metabolites. For each dataset, multivariate predictive modeling was conducted using partial least squares discriminant analysis incorporated into a repeated double cross-validation framework. Prediction performance is shown in downstream figures: Each swim lane represents one mouse. For each sample, class probabilities were computed from 200 double cross-validations. Class probabilities are color-coded by class, presented per repetition (smaller dots), and averaged over all repetitions (larger dots). Misclassified samples are circled. Predictive accuracy was calculated as the number of correctly predicted samples/total number of measured samples. (B) Model performance of DIABLO integrative analysis of omic signatures concerning IF. The use of DIABLO maximized the correlated information between genes, bacteria, and metabolites. Scatterplots depicting the clustering of groups, i.e., AD + ad libitum and AD + IF, based on the first component of each dataset from the model, showed significant separation between groups. A scatterplot displays the first component in each dataset (top diagonal plot) and Pearson correlation between components (bottom diagonal plot). (C to E) Lollipop plot of contributions of key omic signatures identified by DIABLO integrative modeling for discriminating AD + IF from AD + ad libitum. Loading of DIABLO integrative modeling for each predictor is presented. Blue bars indicate IF-induced improvements in predictors. Predictors that were lower in AD + IF compared with AD + ad libitum are in red. TES, N-Tris(hydroxymethyl)methyl-2-aminoethanesulfonic acid sodium salt hydrate; NAAG, N-acetylaspartylglutamate. (F) The Circos plot shows the positive (negative) correlation, denoted as brown (gray) lines, between selected multiomics features.

Fig. 4.. The neuroprotective effects of IF are mediated by gut microbiota.

For antibiotic administration experiment: (A) Schematic of the treatment with ABx and IF in each group (n = 4 to 5). (B) Body weight change (WT + ad libitum, WT + ad libitum + ABx, AD + ad libitum + ABx, and AD + IF: n = 5; AD + ad libitum and AD + IF + ABx: n = 4). (C) Primary latency in acquisition trials of Barnes maze [same group sizes as in (B)]. (D) Hippocampal TNFα mRNA levels (n = 3). (E) Representative immunofluorescence of Aβ plaques in the hippocampal CA1 region and piriform cortex (n = 3 slices per group). Scale bar, 100 μm. (F and G) Aβ plaque density (number of plaques per square micrometer) in the hippocampal CA1 region and piriform cortex (n = 3 mice per group). For ketoconazole (intraperitoneal) injection experiment: (H) Schematic of the treatment with KCZ (intraperitoneal) and IF in each group (n = 7). (I) Primary latency in Barnes maze acquisition trials (n = 7). (J) Primary latency in probe trial of Barnes maze (n = 6). (K) Aβ plaque density (number of plaques per square micrometer) in the CA1 region of the hippocampus (AD + IF: n = 5, AD + IF + KCZ: n = 4). (L) Representative immunofluorescence images of Aβ plaques in the piriform cortex and CA1 region of the hippocampus [same group sizes as in (K)]. Scale bar, 100 μm. (M) Aβ plaque density (number of plaques per square micrometer) in the piriform cortex [same group sizes as in (K)]. (N) Hippocampal TNFα mRNA levels (n = 4). In (C), * indicates a significant difference between WT + ad libitum and AD + ad libitum, # indicates a significant difference between AD + ad libitum and AD + IF, and & indicates a significant difference between AD + IF + ABx and AD + IF. Data are mean ± SEM. *P < 0.05, **P < 0.01, ^##P < 0.01, and ^&&P < 0.01. For the antibiotic experiment [(B) to (G)], statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparisons test. For the KCZ experiment [(I) to (N)], comparisons between the two groups were performed using a Student’s t test.

Fig. 5.. IPA mimics the neuroprotective effects of IF.

Human data: (A) Workflow for collection and processing of samples. (B) Fecal IPA concentrations (n = 9 to 11). (C) Serum IPA concentrations (n = 9 to 11). Correlation between serum IPA and (D) MoCA or (E) MMSE scores. For IPA (gavage) experiment: (F) Schematic of the IPA treatment (n = 6). (G) Primary latency in Barnes maze acquisition trials (n = 6). (H) Discrimination index in the novel object recognition (NOR) test (n = 6). (I) Hippocampal TNFα mRNA expression (n = 4). (J) Representative immunofluorescence of Aβ plaques in the piriform cortex and CA1 region of the hippocampus (n = 3 slices per group). Scale bar, 100 μm. (K and L) Aβ plaque density (number of plaques per square micrometer) in the CA1 region of the hippocampus and piriform cortex. For the C.S. WT or C.S. △fldC recolonization experiment: (M) Schematic of the treatment with vancomycin and C.S. WT or C.S. △fldC (n = 10). The content of IPA in the (N) culture medium (n = 3) and (O) mouse serum (n = 9 to 10). (P) Discrimination index in the NOR test (n = 9 to 10). (Q) Representative immunofluorescence images of Aβ plaques in the piriform cortex and CA1 region of the hippocampus (n = 3 slices per group). Scale bar, 200 μm. (R) The spontaneous alternation of the Y-maze test (n = 9 to 10). (S and T) Aβ plaque density (number of plaques per square micrometer) in the CA1 region of the hippocampus and piriform cortex. Data are mean ± SEM. *P < 0.05 and **P < 0.01. For human data [(B) and (C)], significance was determined by one-way ANOVA with Tukey’s test. For the IPA gavage experiment [(G) to (L)], a two-way ANOVA with Tukey’s test was used. For the C. sporogenes colonization experiment [(N) to (T)], a Student’s t test was used for comparisons between the two groups.

Fig. 6.. IPA attenuates neuroinflammation via activating PXR.

For PXR-shRNA (i.c.v.) experiment: (A) Schematic of the treatment with PXR-shRNA (i.c.v.) and IPA in each group (n = 8). (B) Discrimination index in the NOR test and (C) spontaneous alternation in the Y-maze test (n = 7 to 8), SCM-shRNA, mice were treated with AAV-scrambled shRNA (negative control). (D and F) Quantification of Aβ plaque density in the CA1 region and piriform cortex (plaques per square micrometer). (E) Representative immunofluorescence images of Aβ plaques in CA1 (n = 3 slices per group). Scale bars, 200 μm. (G) Quantification of the Western blot of Bace-1 protein levels relative to glyceraldehyde phosphate dehydrogenase (GAPDH) in the hippocampus (n = 3). (H) Representative Western blot of Bace-1, Cyp3a11, Mdr1, p-NFκB, NFκB, and GAPDH protein levels in the hippocampus (n = 3). (I) The mRNA levels of Bace-1 in the hippocampus (n = 4 to 5). (J and K) Quantification of the Western blot of Cyp3a11 and Mdr1 protein levels relative to GAPDH in the hippocampus (n = 3). (L and M) The mRNA levels of Cyp3a11 and Mdr1b in the hippocampus (n = 4 to 5). (N) Quantification of the Western blot of p-NFκB protein levels relative to NFκB in the hippocampus (n = 3). (O) The mRNA levels of TNFα in the hippocampus (n = 4 to 5). For TNFα (i.c.v.) experiment: (P) Schematic of the treatment with TNFα (i.c.v.) and IPA in each group (n = 7 to 8). (Q) The spontaneous alternation of the Y-maze test (n = 7 to 8). (R) The mRNA levels of Cyp3a11 in the hippocampus (n = 6). (S) TNFα content in the hippocampus (n = 6). Data are mean ± SEM. *P < 0.05 and **P < 0.01. For the PXR-shRNA and TNFα injection experiments, which involved comparisons across four groups [(B) to (D), (G), (I) to (O), and (Q) to (S)], statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparisons test.