Cognitive dysfunction and impairment of neuronal synaptic plasticity were presented in diabetic mice
To validate the cognitive dysfunction of diabetes, we assessed metabolic signs, behavioral alterations, and neuronal morphology & function in diabetic and non-diabetic mice (Fig. 1A). Random blood glucose level, food & water intake, and HOMA-IR value were higher in diabetic mice compared to non-diabetic mice. Body weight was clearly lower in the diabetic group compared to the non-diabetic group at baseline, but this observation was reversed by exercise (Fig. 1B-C and Fig. S1A-D). Furthermore, we performed Y-maze and Morris water maze (MWM) behavioral tests to observe learning memory defects in diabetic mice. The alternation triplet (%) was decreased in diabetic mice, whereas exercise improved the alternation triplet (%) in STZ + Run mice (Fig. 1D). Meanwhile, there were no statistically significant differences in Y-maze total arm entries or total distance between each group (Fig. S1E-F). In the MWM test, escape latency was delayed in diabetic mice, while platform crossover and target quadrant retention time (%) were decreased in the STZ group. All of these deficiencies were rescued by exercise in the STZ + Run group (Fig. 1E-G). There were no significant differences in swimming velocity between each of the groups (Fig. S1G). Together, these results indicate that exercise alleviated abnormal metabolic signs and improved hippocampal-dependent learning memory in diabetic mice.
To observe the neuronal morphology of diabetic mice, we evaluated neuronal complexity and classified the 3D structure of dendritic spines. Both the total dendritic length and neuronal complexity were lower in diabetic compared to non-diabetic mice, which were preserved after exercise in STZ + Run mice (Fig. S2A-C). The reconstruction of dendritic spines demonstrated that total spine density was obviously reduced in diabetic mice, especially in stubby and mushroom spines but with exception for filopodia and long thin spines. Exercise (STZ + Run group) was found to increase total spine density in stubby and mushroom spines (Fig. 1H-K and Fig. S2D-E). Meanwhile, we also examined how diabetes impacted the expression of synaptic proteins. The expression levels of excitatory postsynaptic density marker PSD-95 and presynaptic density marker SYP were decreased in diabetic compared to non-diabetic mice, and this deficit was rescued by exercise in STZ + Run mice (Fig. S1H-I). Consistent with the alteration of neuronal morphology and dendritic spine density, the frequency of mEPSCs were abnormally elevated in diabetic compared to non-diabetic mice, which was an effect obviously attenuated by exercise (Fig. 1L-N). However, there was no difference in the amplitude of mEPSCs between the groups (Fig. 1O-P). The ultrastructure observed via electron microscopy showed the representative diagrams of synapse counts per 27.68µm2 and PSD length & thickness in the hippocampus (Fig. S2F). The synapse count per 27.68µm2 was lower in diabetic mice (four) compared to vehicle mice (eight), where this deficit was again rescued by exercise (Fig. S2G). Exercise restored the reductions of PSD length and thickness in diabetic mice (Fig. S2H-I). More specifically, PSD length ranged from 200 nm to 400 nm in vehicle mice, but it decreased to between 100 nm and 250 nm in diabetic mice. PSD length was increased to normal levels in STZ + Run mice (200–350 nm) (Fig. S2J). PSD thickness ranged from 40 nm to 120 nm in vehicle mice, decreasing to between 20 nm and 60 nm in diabetic mice. Once again, PSD thickness was increased to relatively normal levels (40–80 nm) in STZ + Run mice (Fig. S2K). Above all, our findings illustrate that exercise ameliorated many of the impairments in synaptic plasticity and cognition in diabetic mice.
Transcriptomic and proteomic analysis revealed the potential role of astrocytic NDRG2 and complement cascades in regulating neuronal synaptic plasticity at diabetic mice
To illustrate the integrative regulatory mechanism of neurons, astrocytes, and microglia, we analyzed hippocampal tissue from the vehicle, STZ, and STZ + Run groups for quantitative transcriptomic sequencing. The mRNA transcripts were processed using String Tie and ballgown (http://www.bioconductor.org/packages/release/bioc/html/ballgown.html) to arrive at 55,450 total genes. We identified 2,664 differentially-expressed genes by filtering with fold change > 2 or < 0.5 and p-value < 0.05 in DESeq2 analysis (http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html). PCA analysis of the 15 samples indicated that the distributions of samples among the three groups were discrete and obviously different from each other, whereas the distribution of samples within each group was aggregative and showed great consistency (Fig. S3A). Next, we performed weighted gene co-expression network analysis (WGCNA) of the 55,450 total genes from the 15 independent samples. A soft threshold of 0.8 and merged cut height of 0.4 were used to ensure that the gene correlation network satisfied scale-free-topological distribution. From WGCNA, we obtained a cluster dendrogram with 18 modules (Fig. 2A). PCA analysis of the 18 modules confirmed that the distribution of the modules was dispersed, representing different groups of genes based on biological function (Fig. S3B). PC1 values reflected weighted gene expression patterns in each module, which was called the module eigengene value. We identified hub genes and biological functions according to the weighted value of each gene and Gene Ontology (GO) enrichment analysis of each module . Then, we clarified the cellular nature of each module by checking whether it consisted of particular cell-type biomarkers .
We observed enrichment of neuronal synaptic plasticity genes – Syt1, Manf, Syp and Bdnf – in module 1 (Fig. 2C). The module eigengene for module 1 was clearly decreased in diabetic mice, and it was moderately improved after exercise (Fig. S3C). We also saw enrichment of astrocytic and oligodendrocytic genes – Lcn2, Ndrg2 and Mobp – in module 2 (Fig. 2D). Furthermore, we found the relationships between each module and identified a highly negative relationship between module 1 and module 2, which reflected that astrocytic and oligodendrocytic genes play a negative role in the neuronal synaptic plasticity of diabetic mice (Fig. 2B). Moreover, endoplasmic reticulum and immune response genes – Cxcr1 and Cxcl11 – were enriched in module 4 (Fig. S3D), which had a negative correlation with module 1 and positive correlation with module 2, thus forming a trilaterally logic-reasonable relationship. Characterized by having the most abundant genes, astrocytic inflammatory response genes were enriched in module 17 (Fig. S3E). Hub genes related to complement activation of microglia – Klf17, C1s2, and F3 – were enriched in module 8, whereas genes associated with complement and coagulation cascades, including Fga, F9, Cxcr2, and Fgg, were enriched in module 11 (Fig. S3F-G). We identified positive relationships between module 17, module 8, and module 11 (Fig. 2B), which reflected the biological mechanism that complement activation of microglia, astrocyte inflammatory response, and complement & coagulation cascades are intimately linked with each other. Specifically, microglia-derived C1q, TNFα, and Il-1α work together to induce A1 reactive astrocytes, then upregulate the expression of astrocytic C3 to initiate complement & coagulation cascades, which causes astrocytes to lose their protective function of promoting synaptic formation and acquire strong neurotoxicity, thus injuring neuronal synaptic plasticity.
KEGG enrichment analysis of the 2,664 differentially expressed genes included the PI3K − Akt signaling pathway, MAPK signaling pathway, Glutamatergic synapse, and Neuroactive ligand − receptor interaction terms (Fig. S3H). However, to ignore the influence of thresholds on differentially expressed genes, we used gene set enrichment analysis (GSEA) to screen out the most biologically significant gene sets. In diabetic mice, the MAPK signaling pathway concentrated at bottom of the ranked list (P = 0.0003, STZ vs. vehicle) and the normalized enrichment score (NES) was found to be -1.69. Conversely, this pathway concentrated at top of the list (P = 0.0014, STZ + Run vs. STZ, NES = 1.53) in STZ + Run mice. These results were closely correlated with the injury of neuronal synaptic plasticity in the WGCNA analysis of diabetic mice, which reversed after exercise (Fig. S4A). The neuroactive ligand receptor interaction pathways enriched at the top of the ranked list in diabetic mice compared to vehicle mice (P = 0.0251, STZ vs. vehicle, NES = 1.41), and concentrated at bottom of the ranked list in STZ + Run mice (P = 0.0086, STZ + Run vs. STZ, NES=-1.50). This finding reflects the upregulation of neuroactive ligand receptor interaction in diabetic mice and subsequent downregulation following exercise (Fig. S4B).
Proteins are the ultimate executors in life activities and animal behaviors . To validate the correlation network of RNA-sequencing, we analyzed the results of TMT-labeled proteomic sequencing of hippocampus tissue in vehicle, STZ, and STZ + Run mice, with three independent biological duplicates in each group. We selected the Uniprot_MusMusculus_17027_20200226 database for qualitative analysis and acquired 295,274 total spectra with 114,660 peptide spectra matches, screening out 53,640 unique peptides and 6,557 proteins. PCA analysis illustrated that the distribution of samples was discrete among groups and concentrated within each group (Fig. S5A). The threshold used for differentially expressed proteins were fold changes > 1.2 or < 0.83 accompanied by a p-value < 0.05. We detected 151 upregulated proteins and 29 downregulated proteins in diabetic mice, whereas a volcano plot showed six upregulated proteins and 78 downregulated proteins in STZ + Run mice (Fig. S5B-D). The results of proteomic GSEA were consistent with correlation network analysis of RNA-sequencing, which involved complement coagulation cascades and long-term potentiation pathways. The complement and coagulation cascades enriched at the top of the ranked list in STZ mice compared to vehicle mice (P = 0.0002, STZ vs. vehicle, NES = 2.22). Furthermore, these cascades concentrated at the bottom of the list in STZ + Run mice (P = 0.0035, STZ + Run vs. STZ, NES=-1.82). Overall, these findings reflect the upregulation of complement and coagulation cascades in diabetic mice and subsequent downregulation following exercise (Fig. 3A). The long-term potentiation pathways were enriched at the bottom of the ranked list in diabetic mice (P = 0.0445, STZ vs. vehicle, NES=-1.40), and at the top of the list in the STZ + Run group (P = 0.0041, STZ + Run vs. STZ, NES = 1.69). These results are consistent with WGCNA analysis, which had shown that injury in neuronal synaptic plasticity occurred in diabetic mice and was reversed after exercise (Fig. 3B). The expression levels of C3, FGA, FGG, and FGB proteins involved in complement and coagulation cascades, as well as proteins involved in long-term potentiation pathways, are illustrated in the heat map (Fig. 3C-D).
Although omics provided details for an integrated mechanism from a macro perspective, it was also critical to verify the reliability of our results at a micro level. Accordingly, the expressions of complement C3 and astrocytic NDRG2 were examined by western blotting and immunostaining. As a key node in complement cascades, we found that C3 expression levels were increased in diabetic mice and decreased following exercise (Fig. 3E-F), suggesting that astrocytes were activated into their neurotoxic A1 form in diabetic mice and attenuated after exercise. Then, we sought to assess the expression of astrocytic marker NDRG2. We found that NDRG2 levels were decreased in the STZ group compared to vehicle mice, but exercise led to a significant increase in the expression of NDRG2 (Fig. 3E-G). Meanwhile, immunofluorescence staining was performed by confocal microscopy to observe the intensity of C3 and NDRG2. We observed that NDRG2 mainly colocalized with astrocytic maker GFAP, but not with neurons, in the hippocampus. Consistent with western blotting results, the confocal intensities of C3 in the STZ groups were increased than that of the vehicle groups, and the intensities of NDRG2 in the STZ groups were weaker than that of the vehicle groups, which were all rescued by exercise (Fig. S5E and Fig. S6).
Overexpression of NDRG2 alleviated damages to synaptic structure and cognitive dysfunction by inhibiting p-NF-κB/C3 signaling
Although we confirmed that complement C3 levels were increased in diabetic mice and more seriously elevated after astrocytic NDRG2 loss-of-function, we sought to find a better intervention to reduce the production of C3 at its source. To accomplish this, we investigated the effects of NDRG2 overexpression, which included a remodeling of the pathological structure of astrocytes to inhibit the production of C3. We then clarified the effectiveness of rAAV-GfaABC1D-NDRG2-2A-mCherry-WPRE-pA, AAV2/9 (AAV-NDRG2), finding that the expression of NDRG2 was elevated in the vehicle + AAV-NDRG2 and STZ + AAV-NDRG2 groups (Fig. 5A-C). Consistent with our previous results, the levels of NDRG2 were decreased in the STZ + AAV-Ctrl group compared to the vehicle + AAV-Ctrl group. The overexpression of NDRG2 downregulated heightened levels of p-NF-κB and C3 in diabetic mice (Fig. 5D-F), indicating that a gain-of-function in NDRG2 could inhibit complement signaling and neurotoxic characteristic.
Working memory and spatial memory were assessed using Y-maze and MWM testing. Although there were no significant differences in Y-maze total arm entries and total distance among the four groups, the alternation triplet (%) was decreased in the STZ + AAV-Ctrl group compared to the vehicle + AAV-Ctrl group, and was increased with overexpression of NDRG2 in the STZ + AAV-NDRG2 group (Fig. 5G and Fig. S7A-B). In the MWM tests performed at three to five days, escape latency was delayed in the STZ + AAV-Ctrl group compared to the vehicle + AAV-Ctrl group, whereas NDRG2 overexpression clearly shortened escape latency in the STZ + AAV-NDRG2 group (Fig. 5H). There were no significant differences in swimming velocity between the groups (Fig. S7C). Platform crossover and target quadrant retention time (%) were decreased in the STZ + AAV-Ctrl group compared to the vehicle + AAV-Ctrl group, and this deficit was rescued in the STZ + AAV-NDRG2 group (Fig. 5I-J). These results indicate that NDRG2 overexpression improves hippocampal-dependent memory in diabetic mice.
The expression levels of PSD95 and SYP were decreased in the STZ + AAV-Ctrl group compared to the vehicle + AAV-Ctrl group, but NDRG2 overexpression improved protein levels in the STZ + AAV-NDRG2 group (Fig. 5K-L). The automatic classification of dendritic spine density demonstrated that overexpression of NDRG2 reversed the reductions in stubby, mushroom, and total spine densities in diabetic mice, but did not significantly impact long thin or filopodia spine density (Fig. 6A-F). Representative ultrastructure images describing PSD length, thickness, and synapse counts in the hippocampus were obtained via electron microscopy and shown in 10K and 25K (Fig. S7D). The synapse count per 27.68µm2 was reduced from 7.25 in the vehicle + AAV-Ctrl group to 3.3 in the STZ + AAV-Ctrl group, and it was increased to normal in the STZ + AAV-NDRG2 group (Fig. S7E). PSD length and thickness were reduced in the STZ + AAV-Ctrl group compared to the vehicle + AAV-Ctrl group, whereas NDRG2 overexpression restored both PSD length and thickness (Fig. S7F-G). More specifically, PSD length ranged from 150 nm to 400 nm in the vehicle + AAV-Ctrl group, decreased to 50–300 nm in the STZ + AAV-Ctrl group, and was restored to 150–400 nm in the STZ + AAV-NDRG2 group (Fig. S7H). PSD thickness ranged from 40 nm to 80 nm in the vehicle + AAV-Ctrl group, decreased to between 20 nm and 40 nm in the STZ + AAV-Ctrl group, and was increased to 20–60 nm in the STZ + AAV-NDRG2 group (Fig. S7I). Above all, these findings indicate that NDRG2 overexpression protected diabetic mice from cognitive defects and neuronal synaptic injury.