Transcriptome analysis in calorie-restricted rats implicates epigenetic and post-translational mechanisms in neuroprotection and aging

Abstract

Background: Caloric restriction (CR) can increase longevity in rodents and improve memory function in humans. α-Lipoic acid (LA) has been shown to improve memory function in rats, but not longevity. While studies have looked at survival in rodents after switching from one diet to another, the underlying mechanisms of the beneficial effects of CR and LA supplementation are unknown. Here, we use RNA-seq in cerebral cortex from rats subjected to CR and LA-supplemented rats to understand how changes in diet can affect aging, neurodegeneration and longevity.

Results: Gene expression changes during aging in ad libitum-fed rats are largely prevented by CR, and neuroprotective genes are overexpressed in response to both CR and LA diets with a strong overlap of differentially expressed genes between the two diets. Moreover, a number of genes are differentially expressed specifically in rat cohorts exhibiting diet-induced life extension. Finally, we observe that LA supplementation inhibits histone deacetylase (HDAC) protein activity in vitro in rat astrocytes. We find a single microRNA, miR-98-3p, that is overexpressed during CR feeding and LA dietary supplementation; this microRNA alters HDAC and histone acetyltransferase (HAT) activity, which suggests a role for HAT/HDAC homeostasis in neuroprotection.

Conclusions: This study presents extensive data on the effects of diet and aging on the cerebral cortex transcriptome, and also emphasises the importance of epigenetics and post-translational modifications in longevity and neuroprotection.

Fig. 1

Age-related differential gene expression in rats fed AL and subjected to CR. a Two simplified networks of related, statistically significant enriched Gene Ontology (GO) terms using the Cytoscape add-on ClueGO [30, 31]. The network comparing 6 months AL (6 m AL) and 28 months AL (28 m AL) is shown on the left and 12 months AL (12 m AL) and 28 months AL (28 m AL) is shown on the right. DE genes were used to generate a GO term network. The filled coloured circles (nodes) represent each statistically significant parent GO term. The lines (edges) between the nodes show that there are overlapping genes between terms. The Cytoscape add-on ClueGO allows enrichment analysis and the collapsing of GO terms into parent categories for each comparison. Each of the terms is statistically significant (Benjamini-Hochberg correction <0.05). Colours represent shared GO terms. The different sizes of the nodes relate to how many genes fall into the terms. b Heatmap of the DE genes with age across all AL and CR datasets (Table 2); created in R using the heatmap3 package. The y-axis represents all the DE genes. Red = up-regulated, blue = down-regulated, log2 fold change reported for statistically significant DE genes. White means no statistically significant change in expression. 6 m = 6 months of age, 12 m = 12 months of age, 28 m = 28 months of age

Fig. 2

CR and LA supplementation elicit a similar neuroprotective gene expression profile. a Simplified networks of related statistically significant enriched GO terms using the Cytoscape add-on ClueGO [30, 31]. The network was made from the overlapping differentially expressed genes between all diets with CR and LA supplementation (a.k.a neuroprotective profile; see text for details). The Cytoscape add-on ClueGO allows enrichment analysis and the collapsing of GO terms into parent categories for each comparison. Each of the terms is statistically significant (Benjamini-Hochberg correction <0.05). The filled coloured circles (nodes) represent each statistically significant enriched parent GO term. The lines (edges) between the nodes show that there are overlapping genes within terms. The different sizes of the nodes relate to how many genes fall into the terms. b Heatmap of the differentially expressed genes (Table 1) across all diets with CR and LA supplementation when compared with AL; created in R using the heatmap3 package. Red = up-regulated, blue = down-regulated, log2 FC reported for statistically significant DE genes. White means no statistically significant change in expression. c Quantitative PCR confirmation of RNA-seq results, showing glutamate receptor differential expression (GRIN2c, GRM1 and GRID2) when comparing CR (left) and AL supplemented with LA (right) with AL alone. Orange bars are RNA-seq FC and blue bars are qPCR FC. The horizontal dotted line represents the one FC cutoff. The vertical dotted line separates the CR and LA supplemented comparisons. Error bars represent the standard error calculated for log2 FC as follows: (Standard error/Mean) × log2e. Hprt1, B2m and Ywhaz were used as reference genes. N = 6 per group, unpooled RNA from the RNA-seq experiment

Fig. 3

Each diet group and age elicits its own miRNA expression profile but miR-98-3p is overexpressed in all CR and LA diets and may affect HAT and HDAC activity. a Heatmap of the 175 DE miRNAs across all diets and ages; created in R using the heatmap3 package. mir-98-3p is highlighted by a green box. Red = up-regulated, and blue = down-regulated, log2 fold change reported for statistically significant DE genes. White means no statistically significant change in expression. b qPCR confirmation of RNA-seq results, showing miR98-3p differential expression in all CR and LA supplemented diets. Orange bars are RNA-seq log2 FC and blue bars are qPCR FC (corrected to provide the true FC in the sample by Ratio – 1 if the sample was up-regulated and by (–1/Ratio) + 1 for down-regulated samples). The horizontal dotted line represents the 1 FC cutoff. Errors bars represent the standard error calculated for log2 FC as follows: (Standard error/Mean) × log2e. Snord96a, Snord95 and Snord68 were used as reference genes. c The difference in HDAC and HAT activity was calculated as a ratio of the transfection (TF) control and then transformed into true FC (by Ratio – 1 if the sample was up-regulated and by (–1/Ratio) + 1 for down-regulated samples). A cutoff of 1 FC was used, represented by the dotted line. The figure shows that the miR-98-3p inhibitor significantly (p < 0.001, two-tailed t-test) decreases HAT activity when compared with HDAC activity. A miR-98-3p mimic appears not to affect HDAC/HAT balance (no statistically significant difference). LA reduces HDAC activity (p < 0.001) and the untreated cells show a slight increase in both HAT and HDAC activity, indicating that transfection might affect HDAC and HAT activity but not HDAC/HAT balance (no statistically significant difference). The error bars represent the standard deviation. Experiments were performed independently three times with n = 3 wells per treatment

Fig. 4

Dietary memory effect approach and networks. a The red boxes represent diet groups that have no effect on longevity (baseline longevity) and the green boxes represent diets that extend longevity. This shows how we compared the diet groups to identify the key genes involved in the dietary memory effect. First the DE gene list for AL versus CR > AL + LA (“>” = switched), which results in extended longevity, was compared with the DE gene lists for two conditions that do not result in extended longevity but do have a diet switch and LA supplementation, thereby identifying the genes for extended longevity but excluding the confounding effect of LA supplementation and diet switching. Next, DE genes from AL versus AL + LA > CR (baseline longevity) were compared with those from the extended longevity AL versus AL > CR condition, identifying genes LA induces that are important in disrupting the longevity effect of switching to CR. Finally, the effect of the order of LA supplementation (before or after diet switch) was identified. The three resulting lists were then compared to identify key genes involved in the dietary memory effect of LA. b Simplified networks of related statistically significant enriched GO terms using the Cytoscape add-on ClueGO [30, 31]. The network shows enrichment of gene function for the dietary memory effect genes (Additional file 6). The Cytoscape add-on ClueGO allows enrichment analysis and the collapsing of GO terms into parent categories for each comparison. The filled coloured circles (nodes) represent each statistically significant parent GO term. The lines (edges) between the nodes show that there are overlapping genes within each term. Each of the terms is statistically significant (Benjamini-Hochberg correction <0.05). The different sizes of the nodes relates to how many genes fall into that term

Fig. 5

Overlap of longevity genes identified in the rat cerebral cortex with the longevity associated networks in mouse, Drosophila and C. elegans. The 28 longevity genes identified in rats in this study were found to be longevity-associated genes (LAGs) or partners of LAGs in the longevity networks for mouse, Drosophila and C. elegans (overlap significantly higher than expected by chance; p = 8.37E − 03, Fisher’s exact test). LAGs are genes which result in noticeable changes in the ageing phenotype and/or lifespan. These are identified experimentally through knockout, mutation, overexpression or RNA interference [76]. Longevity networks (previously described in detail [79]) are protein–protein interaction networks, include a core of LAGs, depicted in green in the figure, and their first order protein-interacting partners, shown in light green. The size of the network for each species is thus dependent on the protein–protein interaction data, but also on the number of LAGs available from the literature. All overlaps are higher than expected by chance (though this is not always statistically significant), with the overlap being significantly higher for worm LAGs, network partners of fly LAGs and network partners of mouse LAGs. Fisher’s exact test (one-tailed) p values for each of the overlaps are as follows: worm LAG p value = 6.11E − 05, fly LAG p value = 0.55, mouse LAG p value = 0.35, p value for LAG partners in worm longevity network = 0.09, p value for LAG partners in fly longevity network 0.027, p value for LAG partners in mouse longevity network = 0.034