Mouse brain metabolome coverage and data quality assessment
To depict the brain metabolome atlas across distinct regions and investigate the impact of the feeding regimen on global metabolic remodeling, we performed GC-MS-based metabolomics and LC-MS-based lipidomics on seven brain regions of C57BL/6 female mice subjected to different periods of fasting (3, 6, 12, and 24 h) or fed ad libitum (Fig. 1a). The seven anatomically defined regions included the olfactory bulb (OB), frontal cortex (COR), hypothalamus (HYT), hippocampus (HIP), cerebellum (CBL), brainstem (BST) and spinal cord (SC). Sagittal views of these brain regions are shown in Fig. 1a.6 For GC-MS analysis, using retention time, retention index, and mass spectral information from the NIST11 library and in-house database, the chromatographic peaks of 144 ion features were manually integrated, and 126 of them were annotated with definite structures (Supplementary Data 1). These metabolites were defined into 14 chemical classes according to the HMDB classification scheme. The number and percentage of metabolites in each chemical category are shown in Fig. 1b, c. These metabolites were mapped into 54 metabolic pathways using MetaboAnalyst 5.0 (https://www.metaboanalyst.ca, Fig. 1d).13 For lipidomics analysis, 801 lipid ion features in positive ionization mode and 431 lipid ion features in negative ionization mode were identified using MS-DIAL (ver. 4.93) software.14 After removing the lipids detected repeatedly by two ionization modes and those with low signal-to-noise ratio and high coefficient of variation, 671 unique lipid species of 22 lipid classes were retained (Supplementary Data 2, Fig. 1e, f). These data illustrate the adequate coverage of pathway modules in this study for in-depth profiling of complex brain regions.
Overview of the mouse brain metabolome dataset. a Graphic illustration of the experimental design to acquire mouse brain metabolome dataset. b Number of the detected metabolites in each chemical class using HMDB classification scheme (GC-MS). c Percentage of each chemical class in the metabolites detected by GC-MS. d Top-20 mapped metabolic pathways of the annotated metabolites in GC-MS. e Number of the detected lipids in each chemical class. f Percentage of each lipid class
Quality control (QC) samples were constructed by pipetting equal aliquots of brain extracts from all tissues and analyzed along with the samples to monitor signal fluctuation and reproducibility during the analytical process. To visualize the variation in the dataset, principal component analysis (PCA) in two-dimensional and three-dimensional score plots were performed. As shown in Supplementary Fig. 1a–e, the QC samples clustered tightly, demonstrating minimal residual technical errors. Using univariate analysis of the detected metabolites in QC samples, 92.1% and 89.7% of the annotated metabolites and lipids, respectively, showed excellent biological reproducibility with relative standard deviations of <30% in GC-MS and lipidomic analyses, suggesting that the data we acquired were robust and dependable (Supplementary Fig. 1c, f).
Metabolic differences across brain regions
Since regional specialization is a hallmark of the architecture and function of the brain, we first investigated the differences in the metabolome across brain regions. To begin, we performed PCA on the metabolites identified across all brain regions and found segregation primarily according to the region type (Fig. 2a and Supplementary Fig. 2a). The metabolic profiles of the seven brain regions were integrated into five clusters. One cluster comprised BST and SC, which contain more axons than neuronal cell bodies and dendrites on the cellular component.15 One cluster comprised HIP and COR, which are enriched with neuronal cell bodies and dendrites and depleted in axons.15 As judged by PCA, the remaining three regions (OB, CBL, and HYT) exhibited three distinct metabolic profiles. PCA of samples from mice fed ad libitum also resulted in similar sample clustering (Fig. 2b and Supplementary Fig. 2b).
Metabolic profiles across various brain regions. a Three-dimensional diagram of PCA of brain metabolome from mice fed ad libitum or subjected to STF. R2X = 0.614, Q2 = 0.420 (GC-MS data). b Three-dimensional diagram of PCA of brain metabolome from mice fed ad libitum. R2X = 0.869, Q2 = 0.599 (GC-MS data). c Heat map of the differentially expressed metabolites (detected by GC-MS) among seven brain regions in mice fed ad libitum. d Pathway analysis of the highly expressed and lowly expressed metabolites in different brain regions. Pathway enrichment analysis was performed utilizing the Pathway Analysis module from online MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/)
To identify the differentially expressed metabolites (DEMs) among the different brain regions, we performed a one-way analysis of variance (ANOVA) and identified 538 DEMs across the seven brain regions. Consistent with the PCA results, the variation in these DEMs was clustered into five patterns (Fig. 2c and Supplementary Fig. 2–5). Notably, we did not find any differences in the levels of glucose, fructose, or lactate across the brain regions. To better visualize the differences in metabolic pathways, we performed pathway enrichment analysis using metabolites with high or low expression in each brain region (Fig. 2d). For instance, the most abundant metabolites in the OB engaged in taurine and hypotaurine metabolism and glutathione metabolism, whereas low-abundance metabolites were involved in glycine, serine and threonine metabolism and nicotinate and nicotinamide metabolism. However, in HYT, the most abundant metabolites were implicated in the pathways of beta-alanine metabolism and lysine degradation, whereas metabolites in the pathways of taurine and hypotaurine metabolism, and glutathione metabolism were expressed at low levels. In addition, different brain regions exhibited distinct lipid compositions. Hierarchical clustering showed that phosphatidylethanolamine (PE) O- and hexosyl ceramide (HexCer) were enriched in SC and BST. Conversely, the COR and HIP brain regions showed diametrically opposing lipid compositions. High levels of ceramide (Cer) and sphingomyelin (SM) and low levels of N-acylethanolamine (NAE) were observed in HYT (Supplementary Fig. 3, 5).
Effects of STF on the metabolome of different brain regions
Next, we focused on the spatiotemporal effects of STF on brain metabolome. To achieve this, we performed partial least squares discriminant analysis (PLS-DA, Fig. 3a, Supplementary Fig. 6) and Student’s t test to identify significantly changed metabolites after different periods of fasting and found that the metabolic response to STF in the brain was region-dependent. By calculating the number of DEMs induced by STF in each brain region, we found that the metabolite components in the SC, OB, and HIP were the most affected, followed by CBL, COR, HYT, and BST (Fig. 3b). Our data also showed that the effect of STF on the brain metabolome was enhanced with the duration of fasting in the SC, OB, CBL, HYT, and BST. Conversely, the most profound effects on the metabolome of COR and HIP were observed after 6 h of fasting (Fig. 3c). We previously mentioned that similar metabolic phenotypes were identified between SC and BST, as well as between COR and HIP, and our results further indicated that the impact of STF duration was similar between SC and BST, as well as between COR and HIP. However, more profound metabolic alterations were observed in the SC and HIP than in the BST and COR during STF (Fig. 3d, e, Supplementary Fig. 7). Except for DEMs that were identified only in SC (or COR) or BST (or HIP), similar response patterns were found for DEMs identified in both SC and BST (or COR and HIP) during STF (Fig. 3f, g, Supplementary Data 3). Detailed information on the differentially expressed lipids and metabolites after each fasting period is provided in Supplementary Data 4, 5. The results showed that STF elicited broad metabolic remodeling of lipid and amino acid metabolism in the brain.
Spatiotemporal effects of STF on the metabolome in multiple brain regions. a Three-dimensional diagrams of PLS-DA of metabolic responses to different periods of STF in SC, R2X = 0.478, R2Y = 0.585, Q2 = 0.160; OB, R2X = 0.508, R2Y = 0.543, Q2 = 0.203; HIP, R2X = 0.497, R2Y = 0.603, Q2 = 0.0753; CBL, R2X = 0.426, R2Y = 0.562, Q2 = −0.0154; COR, R2X = 0.472, R2Y = 0.539, Q2 = 0.129; HYT, R2X = 0.522, R2Y = 0.497, Q2 = −0.00197 and BST, R2X = 0.472, R2Y = 0.522, Q2 = 0.0736. b Number of DEMs upon STF in seven brain regions. c Number of DEMs after various periods of STF in seven brain regions. Venn diagram of the number of DEMs (GC-MS data only) after STF in (d) BST and SC, (e) COR and HIP. Venn diagrams were generated using online tools (https://jvenn.toulouse.inrae.fr/app/example.html). More metabolites were found to be changed after STF in SC and HIP, compared with BST and COR, respectively. Heat maps of the DEMs (GC-MS data only) upon STF in (f) BST and SC, (g) COR and HIP
Remodeling of lipid metabolism
STF induced marked alterations in the glycerolipid levels in all brain regions. The abundance of diacylglycerol (DG), monoacylglycerol (MG), and fatty acid (FA) increased, whereas total triacylglycerol (TG) levels declined after STF (Fig. 4a). Each lipid species in the MG, FA, and TG followed the same pattern, except for DG. DGs having a long chain containing a total number of carbon atoms greater than 50 decreased, while those containing total carbon atoms less than 50 increased after STF (Supplementary Data 4). STF also affected the levels of sphingolipids (Cer and SM) and phospholipids in the brain. However, the alteration of each lipid species in these lipid classes followed different patterns across brain regions. In addition, elevated levels of NAEs, a class of neuroprotective lipids, were noted after STF. These findings indicate remodeling of lipid composition in the brain during fasting.
Visualization and pathway analysis of the differentially expressed lipids and metabolites that responded to fasting in the brain. a Change of total abundance of each lipid class after STF in seven brain regions. *0.01 < p < 0.05, **0.001 < p < 0.01, ***p < 0.001. CAR, acylcarnitine; Cer, ceramide; DG, diacylglycerol; FA, fatty acid; PC, phosphatidylcholine; MG, monoacylglycerol; NAE, N-acylethanolamine; PG, phosphatidylglycerol; PS, phosphatidylserine; CL, cardiolipin; LPC, lysophosphatidylcholine; LPE, lysophosphatidylethanolamine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; SM, sphingomyelin; ST, sterol lipids; TG, triacylglycerol. b Summary of the categories of differential metabolites detected by GC-MS in seven brain regions. c Heat map of DEMs (GC-MS data) in OB after STF. d Percentage of DEMs (GC-MS data) in different metabolic pathways in OB. e Top-25 metabolic pathways (GC-MS data) that were associated with STF in OB. The rich factor indicates the ratio of the number of DEMs in the indicated metabolic pathway to the number of all metabolites annotated to this pathway, the higher the value, the higher the enrichment degree. The “Number” in circle size represents the number of DEMs that fall into the indicated pathway. Pathway enrichment analysis was performed utilizing the Pathway Analysis module from online MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/)
Apart from lipid metabolism, a global alteration in amino acids was also observed during STF (Fig. 4b–e, Supplementary Fig. 8–13). Pathway enrichment analysis indicated that the biosynthesis of valine, leucine, and isoleucine (branched-chain amino acids, BCAAs) and phenylalanine, tyrosine, and tryptophan (aromatic amino acids, ArAAs), together with arginine, alanine, aspartate, and glutamate metabolism, were the most representative pathways related to STF (Fig. 4e, Supplementary Fig. 8–13).
Remodeling of amino acid metabolism
First, we noted a global elevation in BCAAs levels across seven brain regions. In contrast, decreased level of tyrosine (SC, COR, HIP, OB, and CBL) and increased level of phenylalanine (SC, COR), both of which are ArAAs, were observed during STF (Fig. 5a). Moreover, the ratio of BCAAs/ArAAs markedly increased in the SC, BST, COR, HIP, and OB during fasting (Fig. 5b, Supplementary Data 6). Considering that fasting for 3 h had a slight effect on the levels of these metabolites, we divided the samples into two phases (phase I: Ctrl and F-3 h; phase II, F-6 h, F-12 h, and F-24 h) to investigate the correlation between metabolites in different fasting states. In the COR, a significant inverse correlation was found between the tyrosine and isoleucine levels after a longer period of fasting (Fig. 5c, d). However, positive phenylalanine/aspartate (Asp)-isoleucine/leucine correlations were disrupted after a longer fasting duration (Fig. 5d).
Changes of BCAAs and ArAAs in the brain during STF. a Heat map of BCAA and ArAA levels after different periods of fasting in seven brain regions. b Changes of BCAAs/ArAAs ratios after STF in seven brain regions. One-way ANOVA was used to identify differences between different time points in each tissue. c Correlation between metabolites in phase I. d Correlation between metabolites in phase II. Isoleu, isoleucine; Val, valine; Leu, leucine; Tyr, tyrosine; Phe, phenylalanine; Asp, aspartic acid. *0.01 < p < 0.05, **0.001 < p < 0.01, ***p < 0.001
BCAAs can provide nitrogen for the synthesis of glutamic acid (Glu) and GABA, both of which increased during STF (Fig. 6a). Excess of Glu can be neurotoxic and can be consumed through alanine aminotransferase by converting pyruvate to alanine. As a result, declined pyruvate and elevated alanine levels were noted in most brain regions during STF (Supplementary Data 5). To better understand the potential mechanism underlying the metabolic changes, we reanalyzed previously published RNA-sequencing data on transcriptomic changes across different brain regions during STF.4 Consistently, a significant increase in the mRNA expression of Gpt2, which encodes alanine aminotransferase, was noted in CBL and BST (Supplementary Data 7). However, levels of N-acetylaspartylglutamic acid (NAAG, COR, and OB) and N-acetyl-l-aspartic acid (NAA, SC, COR, and OB) decreased after fasting (Fig. 6a). In addition, the NAA/Asp ratio decreased in the SC, indicating a possible reduction in enzyme activity that catalyzes the synthesis of NAA from Asp during fasting (Fig. 6b). The NAAG/NAA and NAAG/Glu ratios were also reduced in the OB (Fig. 6c, d). To support this, the mRNA expression of Rimkla and Rimklb, which encode enzymes that convert NAA to NAAG, was significantly decreased in the OB after STF (Supplementary Data 7), which might indicate suppressed synthesis of NAAG during fasting. In contrast, the NAAG/Glu ratio was elevated in HYT after fasting for 6 h, which may have contributed to the elevation of NAA in HYT (Fig. 6e). In addition, our results showed that a longer fasting period could induce a positive NAAG-Asp correlation but an inverse NAA-Asp correlation in the brain (Fig. 6f, g).
Changes of amino acid neurotransmitters in the brain during STF. a Heat map of the differentially expressed amino acid neurotransmitters upon STF in seven brain regions. b Change of NAA/Asp ratio after STF in SC. c, d Changes of NAAG/NAA and NAAG/Glu ratios after STF in OB. e Change of NAAG/Glu ratio after STF in HYT. f Correlation between metabolites in phase I. g Correlation between metabolites in phase II. Gly, glycine; Ala, alanine; Glu, glutamate; GABA, gamma-aminobutyric acid; Gln, glutamine; Asp, aspartic acid; NAA, N-acetyl-L-aspartic acid; NAAG, N-acetylaspartylglutamic acid. *0.01 < p < 0.05, **0.001 < p < 0.01, ***p < 0.001
In addition to the results mentioned above, elevated levels of citrulline (OB, SC and BST) and ornithine (OB, CBL, COR, HIP, SC and BST) and reduced levels of O-succinylhomoserine (OB), homocysteine (Hcy, OB, HIP, HYT and BST)) and methionine (seven brain regions) were also observed in the brain during STF (Supplementary Data 5). Due to the lack of ornithine carbamoilotransferase in the mouse brain, citrulline cannot be produced from ornithine normally.16 Alternatively, citrulline can be recycled via citrulline–nitric oxide (NO) cycle through argininosuccinate synthetase, argininosuccinate lyase and nitric oxide synthase.16 We found that the mRNA expression levels of Ass1 (OB), Asl (CBL) and Nos1 (OB and CBL) increased after fasting, which might indicate an activation of citrulline–NO cycle during STF (Supplementary Data 7).
Distinct effects of STF on the brain and liver metabolome
Given that the crosstalk between the liver and brain is crucial for maintaining glucose homeostasis during fasting, we investigated and compared the metabolic effects of STF in the brain and liver to identify potential network connections between these two organs. In total, 446 STF-related metabolites, including 376 lipids, were identified in the liver. As expected, increased acylglycerol and FA levels and decreased cholesteryl ester and phospholipid levels were determined during fasting (Supplementary Fig. 14a). Besides, elevated 3-hydroxybutyric acid and reduced levels of monosaccharides, disaccharides and trisaccharides were observed after 6 h of fasting (Supplementary Fig. 14b). The level of 3-hydroxybutyric acid positively correlated with the levels of FAs (Supplementary Fig. 15). In addition, decreased levels of malic acid and fumaric acid, two major metabolites of the tricarboxylic acid (TCA) cycle, were observed after 3 h of fasting. These results are consistent with those of previous studies demonstrating enhanced FA oxidation and reduced glucose production and utilization in the liver in the fasting state. However, we did not find any global alterations in amino acid levels in the liver after STF. In contrast to the brain, the level of Hcy markedly increased in the liver after STF. The remodeling of metabolic pathways during STF in the brain and liver is shown in Fig. 7.
Metabolic remodeling in the brain and liver following STF. STF induced a global remodeling of lipid and amino acid metabolism, especially with respect to BCAA, ArAA, neurotransmitter, methionine and the citrulline–NO cycle for the adaptive responses of the brain under limited nutrition conditions. In contrast to the brain, STF induced significant remodeling of energy metabolism in the liver. Enhanced FA oxidation and declined glycolysis and TCA cycle, were found in liver in the fasted state. The metabolites and genes were linked to pathways using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (https://www.kegg.jp/kegg/pathway.html)
To ascertain whether STF induces adaptive metabolic responses in the brain and liver, we investigated the lipidome of mice after 24 h of fasting and 24 h of refeeding (Ref group). The OPLS-DA score plot demonstrated that the Ref group remained distinguishable from the Ctrl group, indicating that the metabolic effects of fasting were not entirely reversed even after refeeding (Supplementary Fig. 6). The impact of refeeding on lipid species appeared to be multifaceted, with some lipid levels returning to the pre-fasted state, whereas others were further elevated or decreased (Supplementary Data 4, 8, Supplementary Fig. 16).