Protective exercise responses in the dentate gyrus of Alzheimer's disease mouse model revealed with single-nucleus RNA-sequencing

Abstract

Exercise's protective effects in Alzheimer's disease (AD) are well recognized, but cell-specific contributions to this phenomenon remain unclear. Here we used single-nucleus RNA sequencing (snRNA-seq) to dissect the response to exercise (free-wheel running) in the neurogenic stem-cell niche of the hippocampal dentate gyrus in male APP/PS1 transgenic AD model mice. Transcriptomic responses to exercise were distinct between wild-type and AD mice, and most prominent in immature neurons. Exercise restored the transcriptional profiles of a proportion of AD-dysregulated genes in a cell type-specific manner. We identified a neurovascular-associated astrocyte subpopulation, the abundance of which was reduced in AD, whereas its gene expression signature was induced with exercise. Exercise also enhanced the gene expression profile of disease-associated microglia. Oligodendrocyte progenitor cells were the cell type with the highest proportion of dysregulated genes recovered by exercise. Last, we validated our key findings in a human AD snRNA-seq dataset. Together, these data present a comprehensive resource for understanding the molecular mediators of neuroprotection by exercise in AD.

Extended Data Fig. 1. Neurogenic niche response to exercise and AD at the single-nuclei level.

a, MWM latency to reach the escape platform in acquisition (Three-way ANOVA, Exercise n.s. P = 0.0964, Genotype ****P <0.0001, Exercise × genotype n.s. P = 0.2664, Exercise × genotype × time *P = 0.0263). b, Acquisition 24h probe trial (Two-way ANOVA, Exercise **P = 0.0074, Genotype n.s. P = 0.2919, Exercise × genotype n.s. P = 0.8698), and c, 24h probe trial in reversal (Two-way ANOVA, Exercise **P = 0.0067, Genotype n.s. P = 0.3812, Exercise × genotype n.s. P = 0.904). d, Daily running activity (Two-way repeated measures ANOVA, Time ****P < 0.0001, Genotype n.s. P = 0.2706, Time × genotype n.s. P = 0.3457). e, Open field (OPF) (Two-way ANOVA, Exercise n.s. P = 0.8558, Genotype n.s. P = 0.2011, Exercise × genotype n.s. P = 0.6738), f, Spontaneous alternation behavior (SAB) (Two-way ANOVA, Exercise n.s. P = 0.1004, Genotype n.s. P = 0.1187, Exercise × genotype n.s. P = 0.2082), and g, Contextual fear conditioning (CFC) test in all mice (Two-way repeated measures ANOVA, Context ****P < 0.0001, Group n.s. P = 0.438, Context × group n.s. P = 0.4465, followed by Sidak’s multiple comparisons WT-Sed AvsB ***P = 0.0002, WT-Run AvsB ***P = 0.0004, APP/PS1-Sed AvsB n.s. P = 0.0519, APP/PS1-Run AvsB n.s. P = 0.1497). h, Number of cells per cell cluster within each group. i, PCA analysis of pseudobulk data from all samples. j, UMAP representation of marker genes expression in different clusters. Color represents expression level according to the scale bar on the right. k, Percentage of cells per cell cluster within each group. l, The scatter plot shows regulator genes based on GeneWalk analysis observed in WSvsAS. Each dot represents a regulator gene, and the color represents the cell cluster. For all behavior experiments WT-Sed n = 12, WT-Run n = 12, APP/PS1-Sed n = 9, APP/PS1-Run n = 9 (a-g), for snRNAseq WT-Sed, WT-Run, APP/PS1-Run n = 5, APP/PS1-Sed n = 4 (h and k). Data represent the mean ± s.e.m. of biologically independent samples.

Extended Data Fig. 2. Representative enriched pathways across different cell types.

GSEA pre-ranked on gene ontology results for all cell types between the ASvsAR and WSvsAS in neuronal cells (a) and glia cells (b). The representative enriched pathways were selected based on FDR < 0.25.

Extended Data Fig. 3. Remodeling of adult hippocampal neurogenesis in exercise and AD.

a, Quantification of BrdU+NeuN+ adult-born neurons in the dorsal and ventral DG and representative higher magnification confocal images with anti-BrdU (green) and NeuN (red) from WT and APP/PS1, sedentary or running, mice. Scale bar, 50μm. n = 6 per group. Two-way ANOVA, Dorsal DG: Exercise ****P < 0.0001, Genotype *P = 0.0269, Exercise × genotype n.s. P = 0.058, Ventral DG: Exercise *P < 0.0101, Genotype n.s. P = 0.1629, Exercise × genotype n.s. P = 0.305. b, Quantification of DCX+ cells in the dorsal and ventral DG from WT and APP/PS1, sedentary or running, mice. n = 6 per group. Two-way ANOVA, Dorsal DG: Exercise n.s. P = 0.0914, Genotype **P = 0.0013, Exercise × genotype n.s. P = 0.1552, Ventral DG: Exercise **P = 0.0069, Genotype ****P < 0.0001, Exercise × genotype n.s. P = 0.1426. c, Heatmap shows the normalized mean expression (z-score) of neurogenesis-related genes reported by Hochgerner et al. in our dataset. d, UMAP representations of early neuronal marker genes expression. Color represents expression level according to the scale bar on the right. e, f, Scatter plots showing the correlation between AD and exercise effects in Neuroblast I (e) and II (f). Each dot represents a statistically significant DEG in AD (WSvsAS). Dots with black borders represent statistically significant DEGs with exercise in AD mice (ASvsAR). The color gradient illustrates the recovery score (|logFC ASvsAR|). The dot size represents the fraction of non-zero count nuclei in the AR group. g, Dot plots showing Immature Neurons’ recDEGs for all neurogenic cell types. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. Data represent the mean ± s.e.m. (a and b) of biologically independent samples.

Extended Data Fig. 4. Immature Neurons recDEG Atpif1 knock-down disrupts neuronal proliferation and differentiation in vitro.

a-d, Primary cortical neurons were transduced with LV-shRNA for five days. PrestoBlue HS normalized cell viability (a, one-way ANOVA followed by Dunnett’s against shCtrl: shSlc25a4 n.s. P = 0.9967, shAtp6v0c n.s. P = 0.0573, shAtpif1 *P = 0.013), confirmation of the gene knock-down (KD) by qPCR (b, two-way ANOVA, KD ****P < 0.0001, KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl ****P < 0.0001), and gene expression of neuronal markers in response to Atp6v0c (c, two-way ANOVA, KD ****P < 0.0001, KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl: Neurod1 ****P < 0.0001, Dcx **P = 0.0017, Tubb3 n.s. P = 0.8694, Map2 ***P = 0.0003, Dlg4 **P = 0.0071, Syn1 **P = 0.0037, Bax *P = 0.028) and Atpif1 (d, two-way ANOVA, KD ****P < 0.0001, KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl: Neurod1, Dcx, Map2, Dlg4, and Syn1 ****P < 0.0001, Tubb3 **P = 0.0046, Bax **P = 0.0022) knock-down. shAtp6v0c and shSlc25a4 n = 6, shAtpif1 n = 4 (a), shAtp6v0c and shAtpif1 n = 6, shSlc25a4 n = 5 (b-d). e, Representative confocal images of EdU (red) and Nestin (green) staining of embryonic neural stem and progenitor cells transduced with LV-shRNA and maintained in proliferating media for 5 days. Scale bar, 50μm. f-h, Neurospheres were transduced with LV-shRNA and maintained in differentiation media for five days. Confirmation of the gene knock-down by qPCR (f, two-way ANOVA, KD ****P < 0.0001, KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl ****P < 0.0001), and gene expression of neuronal markers in response to Atp6v0c (g, two-way ANOVA, KD n.s. P = 0.0877, KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl: Neurod1 and Tubb3 ****P < 0.0001, Dcx n.s. P = 0.3704, Map2 n.s. P = 0.0826, Dlg4 n.s. P = 0.1893, Syn1 n.s. P = 0.1901) and Atpif1 (h, two-way ANOVA, KD ****P < 0.0001, KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl: Neurod1, Dcx, Tubb3, and Map2 ****P <0.0001, Dlg4 **P = 0.0015, Syn1***P = 0.0004) knock-down. n = 11 per group. i-l, Primary cortical neurons were transduced with LV-shRNA for five days and treated with 20μM recombinant Amyloid-beta 42 for the last 16h (i and j), or Abeta-enriched Tg2576 conditioned-media for the last 3h (k and l). Normalized calcein fluorescent signal indicative of live cells after 16h Amyloid-beta 42 treatment (i, Welch’s ANOVA followed by Dunnett’s T3 against shCtrl ****P < 0.0001), normalized EthD1 fluorescent signal indicative of dead cells after 16h Amyloid-beta 42 (j, Welch’s ANOVA followed by Dunnett’s T3 against shCtrl, shAtp6v0c **P = 0.0055, shAtpif1 ***P = 0.0002), normalized calcein fluorescent signal after 3h Tg2576 conditioned-media (k, Welch’s ANOVA followed by Dunnett’s T3 against shCtrl ****P < 0.0001), and normalized EthD1 fluorescent signal after 3h Tg2576 conditioned-media (l, one-way ANOVA followed by Dunnett’s against shCtrl, shAtp6v0c n.s. P = 0.0726, shAtpif1 n.s. P = 0.0754). n=6 per group. m-o, Adult hippocampus derived neurospheres were transduced with LV-shRNA and maintained in differentiation media for three days. Confirmation of the gene knock-down by qPCR (m, two-tailed unpaired t-test ****P < 0.0001), PrestoBlue HS normalized cell viability (n, two-tailed unpaired t-test **P = 0.0065), and gene expression of neuronal markers in response to Atpif1 knock-down (o, two-way ANOVA, KD ****P < 0.0001, KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl: Neurod1, Dcx, and Map2 ****P < 0.0001, Tubb3 n.s. P = 0.681, Dlg4 *P = 0.0152, Syn1 n.s. P = 0.7519, Bax n.s. P = 0.9609). n=4 (n) and 12 per group (m, o). Data represent the mean ± s.e.m. of biologically independent samples.

Extended Data Fig. 5. Exercise regulates DAM-like microglia in AD mouse models.

a, b, Quantification of Iba-1+ microglia per mm² in the dorsal DG (a, two-way ANOVA, Exercise **P = 0.0021, Genotype ****P < 0.0001, Exercise × genotype **P = 0.0031, followed by Fisher’s LSD, WT-Sed vs Run ns P = 0.8979, APP/PS1 Sed vs Run ***P = 0.0001) and ventral DG (b, two-way ANOVA, Exercise *P = 0.0495, Genotype ****P <0.0001, Exercise × genotype n.s. P = 0.1564) (WT-Sed n = 6, WT-Run n = 6, APP/PS1-Sed n = 5, APP/PS1-Run n = 6). c, UMAP representation of marker genes for perivascular macrophages (Mrc1, Cd163, Cd74), monocytes (S100a4), B cells (Cd79b, Rag1), T cells (Trbc2, Cd3g), and natural killer cells (Nkg7). Color represents expression level according to the scale bar on the right. d, Cell composition in percentage for each subcluster shown in Figure 4g–i. Two-way ANOVA, Exercise n.s. P = 0.9956, Genotype ****P < 0.0001, Exercise × genotype n.s. P = 0.8881. e, Heatmap shows the normalized mean expression (z-score) group of IFN, MHC, and Cyc-M microglia genes reported by Chen et al. in our dataset. f, Dot plots showing microglia subcluster 1 markers. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. g, Violin plots of gene signatures for DAM (Irf8, Trem2, Igf1, Axl, and Csf1) and Homeostatic (P2ry12, Cd33, Tmem119, Csf1r, Cx3cr1) microglia using snRNAseq data of microglia subcluster 1 from AD mice DG. Gene signature = sum of normalized gene expression for all genes of the gene signature per cell (n = 150 and 229 cells for ‘Sed’ and ‘Run’, respectively). Two-tailed Mann Whitney, DAM P = 0.0195 and Homeostatic P = 0.9858. h, Bar plots of gene signatures for DAM (Irf8, Trem2, Igf1, Axl, and Csf1) and Homeostatic microglia (P2ry12, Cd33, Tmem119, Csf1r, Cx3cr1) in isolated CD11b+ cells (microglia) from the cortex and hippocampus of the 5xFAD mouse model using QPCR. Gene signature = sum of normalized gene expression for all genes of the gene signature per animal (n=8 and 7 for ‘Sed’ and ‘Run’, respectively). Two-tailed unpaired t-test, DAM P = 0.0616 and Homeostatic P = 0.9462. Data represented by the mean ± s.e.m. (a, b, d, h) or by the median (middle bold line) and upper and lower quartiles (lighter dotted lines) (g) of biologically independent samples.

Extended Data Fig. 6. Exercise shifts the transcriptional state of astrocytes.

a, b, Quantification of GFAP+ astrocytes per mm² in the dorsal DG (a, two-way ANOVA, Exercise n.s. P = 0.9007, Genotype n.s. P = 0.5594, Exercise × genotype n.s. P = 0.3558) and ventral DG (b, two-way ANOVA, Exercise n.s. P = 0.6633, Genotype n.s. P = 0.9575, Exercise × genotype n.s. P = 0.9111) (WT-Sed n = 6, WT-Run n = 6, APP/PS1-Sed n = 5, APP/PS1-Run n = 6). c, Heatmap shows the normalized mean expression (z-score) of the marker genes for each astrocyte subcluster (subcluster 0 in blue and subcluster 1 in orange). d, UMAP representation of the expression of the high-confidence markers for Radial Glia-like cells in astrocytes. Yellow dots are all nuclei in the astrocyte cluster; grey and green dots represent potentially Radial Glia-like cells based on the expression of listed markers. e, Heatmap shows the normalized mean expression (z-score) per group of previously identified disease-associated astrocytes (DAA) markers in our dataset (subcluster 0 in blue and subcluster 1 in orange). f, Heatmap shows the normalized mean expression (z-score) per group of previously identified reactive astrocytes markers in our dataset (subcluster 0 in blue and subcluster 1 in orange). g, Bar chart of the relevant enriched terms for the astrocyte subcluster 1 marker genes from Enrichr. Enriched terms displayed presented an adjusted p-value <0.05 determined by Fisher exact test with the Benjamini-Hochberg correction for multiple hypotheses. h, CDH4 counts in astrocytes subclusters from human parietal cortex snRNA-seq in Brase et al., 2023. The subclusters presented are the originally described ones from. Linear mixed effect model (covariates, sex and cluster; random effect, sample). i, Violin plots of gene signatures in the astrocyte subcluster 1 (Mfge8, Plxna2, Grin2b, Bmper, Dab1, Pde1c, and Cdh4) using snRNAseq data of astrocytes subcluster 1 from AD mice DG. Gene signature = sum of normalized gene expression for all genes of the gene signature per cell (n = 36 and 60 cells for ‘Sed’ and ‘Run’, respectively). Two-tailed Mann Whitney P = 0.0028. j, Bar plots of gene signatures for astrocyte subcluster 1 (Mfge8, Plxna2, Grin2b, Bmper, Dab1, Pde1c, and Cdh4) in isolated ACSA2+ cells (astrocytes) from the cortex and hippocampus of the 5xFAD mouse model using qPCR. Gene signature = sum of normalized gene expression for all genes of the gene signature per animal (n=7 for ‘Sed’ and ‘Run’). Two-tailed unpaired t-test P = 0.0240. Data represented by the mean ± s.e.m. (a, b, j) or by the median (middle bold line) and upper and lower quartiles (lighter dotted lines) (i) of biologically independent samples.

Extended Data Fig. 7. Astrocytes recDEGs knock-down alters astrocytes states in vitro.

Primary cortical mixed glia cultures were transduced with LV-shRNA for five days in standard growth media (a-d) or treated in the last 24h with a reactive astrocyte activating cocktail of cytokines (e-h) or Amyloid-β 42 peptide (i-l). Confirmation of the gene knock-down by QPCR (a, e, i), and gene expression of reactive astrocyte markers and markers of our subcluster 0 and 1 in response to Nme7 (b, f, j), St7 (c, g, k), and Thra (d, h, l), knock-down. n = 3 for shCtrl in e-l, n = 4 for all other groups. Two-way ANOVA followed by Fisher’s LSD. *p<0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, n.s., not significant. Data represent the mean ± s.e.m. of biologically independent samples. a: two-way ANOVA, KD and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl ****P < 0.0001; b: two-way ANOVA, KD and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl Gfap, Cdh20, Rorb, and Cdh4 ****P < .0001, C3 n.s. P = 0.3171, Serpina3n ***P = 0.0009, Hspb1 **P = 0.004, Csmd1 **P = 0.0068; c: two-way ANOVA, KD n.s. P = 0.2011, KD × gene ****P <0.0001, followed by Fisher’s LSD compared to shCtrl Gfap, C3, Serpina3n, Cdh20, and Cdh4 ****P < 0.0001, Hspb1 *P = 0.0277, Rorb n.s. P = 0.1921, Csmd1 n.s. P = 0.9147; d: two-way ANOVA, KD and KD × gene ****P < .0001, followed by Fisher’s LSD compared to shCtrl Gfap, Serpina3n, Rorb, Cdh4, and Csmd1****P < 0.0001, C3 ***P = 0.0001, Hspb1 n.s. P = 0.283, Cdh20 *P = 0.0174; e: two-way ANOVA, KD and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl ****P < .0001; f: two-way ANOVA, KD and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl Gfap, Serpina3n, Cdh20, and Rorb ****P < 0.0001, C3 n.s. P = 0.7271, Hspb1 n.s. P = 0.107, Cdh4 n.s. P = 0.0856, Csmd1 n.s. P = 0.1875; g: two-way ANOVA, KD and KD × gene ****P < .0001, followed by Fisher’s LSD compared to shCtrl Gfap **P = 0.0092, C3, Cdh20, Cdh4, and Csmd1 ****P < .000, Serpina3n **P = 0.0012, Hspb1 ***P = 0.001, Rorb n.s. P = 0.159; h: two-way ANOVA, KD and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl Gfap, Serpina3n, Cdh20, Rorb, and Cdh4****P < 0.0001, C3 n.s. P = 0.8352, Hspb1 n.s. P = 0.9755, Csmd1 n.s. P = 0.1847; i: two-way ANOVA, KD and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl ****P < 0.0001; j: two-way ANOVA, KD and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl Gfap ***P = 0.0002, C3 n.s. P = 0.2851, Serpina3n **P = 0.003, Hspb1n.s. P = 0.5116, Cdh20, Rorb, and Cdh4 ****P < 0.0001, Csmd1 **P = 0.0093; k: two-way ANOVA, KD n.s. P = 0.7698 and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl Gfap, ***P = 0.0001, C3, Cdh20, and Cdh4 ****P < 0.0001, Serpina3n n.s. P = 0.137, Hspb1 n.s. P = 0.0896, Rorb n.s. P = 0.6144, Csmd1 n.s. P = 0.6805. l: two-way ANOVA, KD n.s. P = 0.4382 and KD × gene ****P < 0.0001, followed by Fisher’s LSD compared to shCtrl Gfap, Serpina3n, Hspb1, Cdh20, Rorb, and Cdh4 ****P < 0.0001, C3 n.s. P = 0.9321, Csmd1 * P = 0.0236.

Extended Data Fig. 8. Exercise remodels AD-dysregulated pathways in mGCs.

a, Average size of amyloid-beta plaques (3D6 staining) in ventral DG sections (APP/PS1-Sed n = 5, APP/PS1-Run n = 6, three section per animal; Two-tailed unpaired t-test P = 0.6679). Data is represented by the median (middle bold line) and upper and lower quartiles (lighter dotted lines). b, Schematic representation of Amyloid precursor protein (APP) processing and amyloid beta degradation pathways. Adapted from the KEGG pathway database. Created with BioRender.com. c, Expression of chimeric mouse/human APPswe and the human PS1-dE9 transgene characteristic of the APP/PS1 mice, and the fraction of cells expressing the gene. d, Expression of alpha-secretase (Adam10), beta-secretase (Bace1), and gamma-secretase (Psenen, Ncstn, Aph1a) coding- genes in all neuronal clusters and the fraction of cells expressing the gene. e, Expression of the Aβ-degrading enzyme coding-gene Ide and Mme in all cell clusters and the fraction of cells expressing the gene. f, Immediate early gene expression in the different neuronal cell types by group. Data represented by biologically independent samples.

Extended Data Fig. 9. Exercise and AD responses in interneurons and vascular cells.

a, Scd2 average expression of all groups in Oligodendrocyte progenitor cells (OPCs) and oligodendrocytes. b, Scd2 expression in oligodendrocytes in different groups. c, Cell compositional analysis for the OPCs subclusters. Two-way ANOVA, Exercise n.s. P = 0.1966, Genotype n.s. P = 0.3081, Exercise × genotype n.s. P = 0.2483. d, Comparison of our exercise effects in DG oligodendrocytes and OPCs with reported exercise effects in the subventricular zone in Buckley et al., 2023. Genes common to both projects, and those we found differentially expressed in exercise vs. sedentary conditions with an FDR-adjusted p value < 0.05 are displayed. Genes with > 0.5 Log2FC change in both projects are labeled. e-g, Interneuron subcluster UMAP representation (e), mean proportion (f), and marker genes for each subcluster (g). h, Scatter plot showing the correlation between AD and exercise effects in Interneurons. Each dot represents a statistically significant DEG in AD (WSvsAS). Dots with black borders represent statistically significant DEGs with exercise in AD mice (ASvsAR). The color gradient illustrates the recovery score (|logFC ASvsAR|). The dot size represents the fraction of non-zero count nuclei in the AR group. i, Dot plots showing recDEGs in Interneurons. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. j, Cell compositional analysis for the Vascular cell subclusters. Two-way ANOVA, subcluster 0: Exercise *P = 0.0266, Genotype n.s. P = 0.9986, Exercise × genotype n.s. P = 0.2262, subcluster 1: Exercise *P = 0.0113, Genotype n.s. P = 0.378, Exercise × genotype n.s. P = 0.5948, subcluster 2: Exercise n.s. P = 0.9024, Genotype n.s. P = 0.492, Exercise × genotype n.s. P = 0.4417.k, Sema3c was a significant recDEGs shared by different cell types. l, Body weights at the start and end of the experiment. Two-way repeated measures ANOVA, Group n.s. P = 0.0603, Time *P = 0.0232, Group × time **P = 0.0018. WT-Sed n = 5, WT-Run n = 5, APP/PS1-Sed n = 4, APP/PS1-Run n = 5 (c and j). WT-Sed n = 12, WT-Run n = 12, APP/PS1-Sed n = 9, APP/PS1-Run n = 9 (l). Data represent the mean ± s.e.m. of biologically independent samples (c, j, l).

Fig. 1.. Neurogenic niche response to exercise and AD at the single-nuclei level.

a, Overview of the experimental design. Seven-month-old male APP/PS1 and WT mice were run for 60 days. Mice were injected with BrdU during the first week. After 60 days of running, behavioral tests were performed at nine months. Afterwards, the dentate gyrus was collected for single-nuclei RNAseq and histological analysis. Created with BioRender.com. b, MWM latency to reach the target platform in reversal (WT-Sed n = 12, WT-Run n = 12, APP/PS1-Sed n = 9, APP/PS1-Run n = 9; three-way ANOVA, Exercise **P = 0.0072, Genotype *P = 0.0434, Exercise × genotype n.s. P = 0.2975). Data represent the mean ± s.e.m. of biologically independent samples. c, UMAP plots of 106,655 annotated nuclei of dentate gyrus from all four groups (WT-Sed or WS, WT-Run or WR, APP/PS1-Run or AR n= 5, APP/PS1-Sed or AS n=4). The 11 distinct cell clusters are shown in different colors. Each dot represents an individual nucleus. d, Canonical marker genes for each cell cluster with the normalized expression and fraction of cells expressing the genes. e, Number of differential expressed genes (DEGs) across the different cell-types. Left bars (darker color), corresponds to WS>WR and right bars (lighter color) corresponds to AS>AR. f, Proportion of DEGs across cell types. Purple, unique DEGs in WS>WR; yellow, unique DEGs in AS>AR; darker grey, DEGs shared between both comparisons; lighter grey, DEGs with opposite direction. g, Scatter plot representations of significant regulator genes in the Immature Neurons via GeneWalk analysis of exercise in WT and APP/PS1 mice. Each dot represents a regulator gene. h, Number of significant cell-cell interactions for all comparisons by CellChat analysis. i, Number of significant ligand-receptor pairs. For each cell type, upregulated connections are displayed in red bars, and downregulated interactions in blue bars. Cells expressing the ligand are designated as “sender cells”. Cells expressing the corresponding receptor are designated as “receiver cells”.

Fig. 2.. Remodeling of adult hippocampal neurogenesis in exercise and AD.

a, Quantification of DG’s BrdU+NeuN+ adult-born neurons and representative confocal images of the DG stained with anti-BrdU (green) and NeuN (red). The white arrows indicate double-positive cells. Scale bar, 100 μm. n = 6 per group. Two-way ANOVA, Exercise ***P = 0.0006, Genotype *P = 0.0446, Exercise × genotype n.s. P = 0.101.b, Quantification of DCX+ cells and representative confocal images of the DG stained with anti-DCX (red) and DAPI (blue). Scale bar, 50 μm. n = 6 per group. Two-way ANOVA, Exercise **P = 0.0035, Genotype ****P <0.0001, Exercise × genotype n.s. P = 0.1222. Data (a, b) represent the mean ± s.e.m of biologically independent samples. c, Schematic of adult hippocampal neurogenesis (adapted from). Molecular layer (ML), granular cell layer (GCL), and subgranular zone (SGZ). Created with BioRender.com. d, Heatmap shows the normalized mean expression (z-score) of markers for each neurogenesis stage. e, Heatmap shows the normalized mean expression (z-score) per group of the top 50 marker genes for Neuroblast I, Neuroblast II, Immature Neurons, and Mature Granule Cell clusters.

Fig. 3.. snRNA-seq of the Immature Neurons reveals specific changes in exercise and AD.

a, Scatter plot showing the correlation between AD and exercise effects in Immature Neurons. Each dot represents a statistically significant DEG in AD (WSvsAS). Dots with black borders represent statistically significant DEGs with exercise in AD mice (ASvsAR). The color gradient illustrates the recovery score (|logFC ASvsAR|). The dot size represents the fraction of non-zero count nuclei in the AR group. b, c, Dot plots showing recDEGs (b) and ‘rare’ recDEGs (c) in Immature Neurons. In each, the hue and size of the dot represent mean expression and fraction of non-zero count nuclei, respectively. d, Embryonic neural stem and progenitor cells were transduced with LV-shRNA and maintained in proliferating media for 5 days. Cell proliferation rate was measured using EdU+ cells normalized to DAPI+ cells. n = 4 per group. One-way ANOVA followed by Dunnett’s against shCtrl, shAtp6v0c ***P = 0.0002, shAtpif1 ***P = 0.0001. e-h, Primary cortical neurons were transduced with LV-shRNA for five days and treated with 20μM recombinant Amyloid-beta 42 for the last 16h (e and f), or Abeta-enriched Tg2576 conditioned-media for the last 3h (g and h). PrestoBlue HS normalized cell viability (e and g), and ratio between dead and live cells (f and h). n = 6 per group. One-way ANOVA followed by Dunnett’s against shCtrl, ****P < 0.0001. i-j, Primary adult hippocampal neural stem and progenitor cells were transduced with LV-shRNA and maintained in proliferating media for 5 days. Representative confocal images of IF staining for EdU (red) and nestin (green) with scale bar = 100μm (i) and quantification of cell proliferation rate using EdU+ cells normalized to DAPI+ cells (j), n = 4 per group. Unpaired two-tailed t-test, ***P = 0.0006. k-m, LV-shRNA was delivered by unilateral stereotaxic injection into the dentate gyrus of WT mice. Representative confocal images of the DG stained with anti-DCX (red) and DAPI (blue) (k), quantification of DCX+ neurons in the DG (l, n = 5 animals per group, unpaired two-tailed t-test, **P = 0.004), Atpif1 gene expression in the hippocampus by QPCR (m, n = 10 animals per group, one-way ANOVA followed by Dunnett’s against shCtrl, ****P < 0.0001). Scale bar, 200μm. Data represent the mean ± s.e.m. of biologically independent samples.

Fig. 4.. Exercise regulates DAM-like microglia in AD mouse models.

a, Representative confocal microscopy images of IBA1+ microglia (red), GFAP+ astrocytes (green), Aβ plaques (3D6 antibody, magenta), DAPI (blue), and composite images from WT and APP/PS1, sedentary or running, mice. Scale bar, 100 μm. b, Quantification of IBA1+ microglia per mm² in the DG (WT-Sed n = 6, WT-Run n = 6, APP/PS1-Sed n = 5, APP/PS1-Run n = 6; Two-way ANOVA, Exercise ***P = 0.0007, Genotype ****P < 0.0001, Exercise × genotype **P = 0.0026, followed by Fisher’s LSD, WT-Sed vs Run ns P = 0.6816, APP/PS1 Sed vs Run ****P < 0.0001). Data represent the mean ± s.e.m. of biologically independent samples. c, Dot plot of Aif1 mean non-zero expression in the microglia cluster. d, Scatter plot showing the correlation between AD and exercise effects in Microglia. Each dot represents a statistically significant DEG in AD (WSvsAS). Dots with black borders represent statistically significant DEGs with exercise in AD mice (ASvsAR). The color gradient illustrates the recovery score (|logFC ASvsAR|). The dot size represents the fraction of non-zero count nuclei in the AR group. e, Dot plots showing recDEGs in Microglia. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. f, Overlap between the mouse AD-dysregulated genes (mAPP/PS1) and their change with running (mAPP/PS1-Run) from the present data set with snRNA-seq data from human parietal cortex microglia of sporadic AD (hAD) or AD autosomal dominant (APP and PSEN1, hADAD) in Brase et al., 2023. LogFC, calculated in comparison to healthy age-matched controls (hAD and hADAD), sedentary WT mice (mAPP/PS1), or sedentary AD mice (mAPP/PS1-Run), are color-coded and listed for each gene and comparison. Bold logFC indicates statistically significant adjusted p values. g, UMAP projection of the microglia subclusters (subcluster 0 in blue and subcluster 1 in orange). h, Cell composition in percentage for each subcluster shown in g (WS, WR, AR n = 5/group; AS n = 4). i, Heatmap shows the normalized mean expression (z-score) of the marker genes for each microglia subcluster (subclusters color-coded as in g). The homeostatic and DAM genes that appeared in our dataset are labeled. j, Heatmap shows the normalized mean expression (z-score) per group of homeostatic and DAM microglia genes reported by Chen et al. in our dataset (subclusters color-coded as in g).

Fig. 5.. Exercise shifts the transcriptional state of astrocytes.

a, Quantification of GFAP+ astrocytes per mm² in the DG (WT-Sed n = 6, WT-Run n = 6, APP/PS1-Sed n = 5, APP/PS1-Run n = 6; Two-way ANOVA, Exercise n.s. P = 0.7639, Genotype n.s. P = 0.7652, Exercise × genotype n.s. P = 0.6449). b, Gfap mean non-zero expression in the astrocyte cluster across different experimental groups. c, Scatter plot showing the correlation between AD and exercise effects in Astrocytes. Each dot represents a statistically significant DEG in AD (WSvsAS). Dots with black borders represent statistically significant DEGs with exercise in AD mice (ASvsAR). The color gradient illustrates the recovery score (|logFC ASvsAR|). The dot size represents the fraction of non-zero count nuclei in the AR group. d, e, Dot plots showing recDEGs (d) and ‘rare’ recDEGs (e) in Astrocytes. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. f, Overlap between the mouse AD-dysregulated genes (mAPP/PS1) and their change with running (mAPP/PS1-Run) from the present data set with snRNA-seq data from human parietal cortex astrocytes of sporadic AD (hAD) or AD autosomal dominant (APP and PSEN1, hADAD) in Brase et al., 2023. LogFC, calculated in comparison to healthy age-matched controls (hAD and hADAD), sedentary WT mice (mAPP/PS1), or sedentary AD mice (mAPP/PS1-Run), are color-coded and listed for each gene and comparison. Bold logFC indicates statistically significant adjusted p values. g, UMAP projection of the astrocytes subclusters (subcluster 0 in blue and subcluster 1 in orange). h, Cell composition in percentage for each subcluster shown in g (WS, WR, AR n = 5/group; AS n = 4). i-l, Validation of the presence of cadherin-4^high astrocytes in the DG, by IF. Representative confocal images of cadherin-4 (CDH4, magenta), GFAP (orange), and vascular endothelium (ICAM2+ cells, green). White arrows mark CDH4^high astrocytes. Scale bar, 100μm. On the far right, examples of CDH4^−/low and CDH4^high astrocytes in the DG. Scale bar, 20μm (i). Quantification of CDH4^high astrocytes in APP/PS1 (j), CDH4^high astrocytic domain size (l, CDH4^high = 61 and CDH4^−/low = 73 cells; two-tailed Mann-Whitney test ****P < 0.0001), and CDH4^high astrocytes distance to blood vessel (k, CDH4^high=81 and CDH4^−/low=82 cells; two-tailed Mann-Whitney test ***P < 0.001). n = 5 animals, 3 dorsal sections per animal. m, n, Heatmap shows the normalized mean expression (z-score) per group of the marker genes for astrocyte subclusters 0 (m) and 1 (n) for each subcluster (color-coded as in g). Data represent the mean ± s.e.m (a and j), violin plots show the median and quartiles (k-l) of biologically independent samples.

Fig. 6.. Exercise remodels AD-dysregulated pathways in mGCs.

a, Average size of amyloid-beta plaques (3D6 staining) in dorsal DG sections (APP/PS1-Sed n = 5, APP/PS1-Run n = 6, 3 section per animal; Two-tailed unpaired t-test, *P = 0.0188). b, Scatter plot showing the correlation between AD and exercise effects in mGCs. Each dot represents a statistically significant DEG in AD (WSvsAS). Dots with black borders represent statistically significant DEGs with exercise in AD mice (ASvsAR). The color gradient illustrates the recovery score (|logFC ASvsAR|). The dot size represents the fraction of non-zero count nuclei in the AR group. c, d, Dot plots showing recDEGs (c) and ‘rare’ recDEGs (d) in mGCs. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. e, f, Representative RNA images of DG showing Chgb puncta (magenta, e) and Scg2 puncta (red, f) in mGCs in WT-Sed, APP/PS1-Sed and APP/PS1-Run. Scale bar, 20μm. g, h, Quantification of Chgb punta (g) and Scg2 punta (h) in the different groups. n = 2 biological independent animals. Kruskal-Wallis followed by the Dunn’s test, Chgb: WSvsAS ****P < 0.0001, ASvsAR ***P = 0.0002, WSvsAR ****P < 0.0001, Scg2: WSvsAS ****P < 0.0001, ASvsAR ****P < 0.0001, WSvsAR ****P < 0.0001. i, j, Overlap between the mouse AD-dysregulated genes (mAPP/PS1) and their change with running (mAPP/PS1-Run) from the present data set with snRNA-seq data from human parietal cortex excitatory neurons of sporadic AD (hAD) or AD autosomal dominant (APP and PSEN1, hADAD) in Brase et al., 2023. LogFC, calculated in comparison to healthy age-matched controls (hAD and hADAD), sedentary WT mice (mAPP/PS1), or sedentary AD mice (mAPP/PS1-Run), are color-coded and listed for each gene and comparison. Bold logFC indicates statistically significant adjusted p values. k, Immediate early genes mean expression and the fraction of cells expressing the genes in mGCs. Violin plots show the median (middle bold line) and upper and lower quartiles (lighter dotted lines) of biologically independent samples.

Fig. 7.. OPCs and oligodendrocytes are highly plastic in exercise and AD.

a, Scatter plot showing the correlation between AD and exercise effects in OPCs. Each dot represents a statistically significant DEG in AD (WSvsAS). Dots with black borders represent statistically significant DEGs with exercise in AD mice (ASvsAR). The color gradient illustrates the recovery score (|logFC ASvsAR|). The dot size represents the fraction of non-zero count nuclei in the AR group. b, Dot plots showing recDEGs in OPCs. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. c, Number of significant ligand-receptor pairs from CellChat analysis. For each cell type, upregulated connections are displayed in red bars, and downregulated interactions in blue bars. Cells expressing the ligand are designated as “sender cells”. Cells expressing the corresponding receptor are designated as “receiver cells”. d-f, OPCs subcluster UMAP representation (d), mean proportion (e), and marker genes for each subcluster (f). g, Scatter plot showing the correlation between AD and exercise effects in oligodendrocytes. The legend details are the same as in a. h, i, Dot plots showing recDEGs (h) and ‘rare’ recDEGs (i) in oligodendrocytes. In each, the hue and size of the dot represent the mean expression and fraction of non-zero count nuclei, respectively. j, Overlap between the mouse AD-dysregulated genes (mAPP/PS1) and their change with running (mAPP/PS1-Run) from the present data set with snRNA-seq data from human parietal cortex oligodendrocytes of sporadic AD (hAD) or AD autosomal dominant (APP and PSEN1, hADAD) in Brase et al., 2023. LogFC, calculated in comparison to healthy age-matched controls (hAD and hADAD), sedentary WT mice (mAPP/PS1), or sedentary AD mice (mAPP/PS1-Run), are color-coded and listed for each gene and comparison. Bold logFC indicates statistically significant adjusted p values. k-m, Vascular cells subcluster UMAP representation (k), marker genes for each subcluster (l), and mean proportion (m).