Imaging brain glucose metabolism in vivo reveals propionate as a major anaplerotic substrate in pyruvate dehydrogenase deficiency

Abstract

A vexing problem in mitochondrial medicine is our limited capacity to evaluate the extent of brain disease in vivo. This limitation has hindered our understanding of the mechanisms that underlie the imaging phenotype in the brain of patients with mitochondrial diseases and our capacity to identify new biomarkers and therapeutic targets. Using comprehensive imaging, we analyzed the metabolic network that drives the brain structural and metabolic features of a mouse model of pyruvate dehydrogenase deficiency (PDHD). As the disease progressed in this animal, in vivo brain glucose uptake and glycolysis increased. Propionate served as a major anaplerotic substrate, predominantly metabolized by glial cells. A combination of propionate and a ketogenic diet extended lifespan, improved neuropathology, and ameliorated motor deficits in these animals. Together, intermediary metabolism is quite distinct in the PDHD brain, it plays a key role in the imaging phenotype, and it may uncover new treatments for this condition.

Graphical Abstract

In brief

In this study, Marin-Valencia et al. show that elevated glucose uptake and glycolysis in the PDH-deficient (PDHD) brain in vivo are indicators of disease progression. Propionate is a key anaplerotic substrate in this condition, and its supplementation with a ketogenic diet, prolongs lifespan and improves neurological outcomes in PDHD mice.

Introduction

In mitochondrial medicine, we still have a limited capacity to evaluate the extent of brain disease in vivo. Markers of neurological involvement in patients with mitochondrial diseases rely mostly on brain structural changes. These include stroke-like lesions, hypoplasia, atrophy, calcifications, and necrosis that variably involve the cortex, white matter, basal ganglia, and cerebellum¹. Biochemical markers detected by magnetic resonance spectroscopy (MRS) are used to support the diagnosis. The typical pattern is elevated lactate and decreased choline and N-acetyl-aspartate (NAA)^2,3. However, little is known about the metabolic mechanisms that drive such structural and molecular changes in brain imaging, which limits our capacity to identify new biomarkers and therapeutic targets in these conditions.

Pyruvate dehydrogenase (PDH) deficiency (PDHD) is one of the prototypical mitochondrial diseases in humans^4,5. It encompasses a broad clinical spectrum that ranges from moderate neurological deficits, such as cerebellar ataxia, to severe brain malformations at birth. It is one of the most common causes of primary lactic acidosis in children⁶, presenting typically with elevated pyruvate and lactate in plasma and urine with normal lactate/pyruvate ratio (≤ 20)⁷. Imaging modalities used in the clinical setting, such as 1.5T to 3T MRS, possess limited detection capabilities, potentially leading to the oversight of disease-relevant biomarkers. Further, the lack of standard imaging protocols to evaluate the extent of brain disease in these patients, including the inconsistent use of MRS at the time of diagnosis nor in the context of acute neurological deficits, hinders further characterization of the disease. As a result, it is still unclear how the in vivo metabolic profile in the brain changes throughout the course of the disease and the mechanisms that drive such changes^8,9.

Here, leveraging a well-defined murine model of PDHD¹⁰, we combined multi-modality imaging and isotope tracing to better characterize the brain metabolic state in this condition. Given that the cellular and molecular features suggested a higher utilization of glucose in the PDHD brain than in control, we elucidated the role of glucose metabolism in these animals. We found that glucose uptake in vivo was consistently higher in the PDHD brain than in control, and such difference increased with disease progression. Glucose oxidation was interrupted at the PDH level, suggesting that alternative substrates bypassing PDH were utilized to sustain the Krebs cycle. We previously identified acetate as an alternative source of acetyl-CoA in the PDHD brain¹⁰. Acetate is avidly oxidized through the Krebs cycle, improving energy production and the electrophysiological dysfunction in the PDHD brain. Here, we found that succinyl-CoA is another entry point of carbons to the Krebs cycle in these animals and identified propionate as an essential precursor. Propionate is a 3-carbon fatty acid that, in contrast to acetate, provides net carbons to the Krebs cycle for biosynthetic purposes, primarily in peripheral organs. In the brain, propionate has been considered to play a limited role in intermediary metabolism based on the lack of labeled byproducts after the administration of its ¹³C form¹¹. In this study, we resolved, for the first time, propionate metabolism in the brain in vivo. Guided by the metabolic profile on MRS and ¹³C tracing analysis, we found that propionate contributes to the in vivo imaging phenotype in the PDHD brain and identified that its conversion to succinyl-CoA occurs primarily in PDH-deficient glia. Its oral administration in conjunction with a ketogenic diet extended the lifespan of PDHD mice, attenuated neuronal loss and reactive astrocytosis, and mildly improved the motor phenotype. Taken together, the metabolic reprogramming in the PDHD brain is quite distinctive. It plays an essential role in the imaging phenotype in vivo and opens a new avenue for potential therapeutic interventions in this condition.

Results

PDHD mice recapitulate neuropathological features of the human condition

Mice were generated by crossing FVB-Tg(GFAP-cre)25Mes/J^+/− males (Jackson Laboratory) with B6.129P2-Pdha1tm1Ptl/J^fl/fl females (Jackson Laboratory). We chose the hGFAP-Cre promoter because it is expressed throughout the central nervous system, including neurons and glial cells, and because, as we previously showed, it replicates the severe form of the human disease^10,12. All experiments were performed in male mice (Pdha1^hGFAP animals and controls) in a mixed background. The rationale for using males was to ensure phenotypic consistency. The breeding of heterozygous females for Pdha1 and Cre to produce females homozygous for Pdha1 and heterozygous for Cre (Pdha1^hGFAP hereafter) could result in progenitors exhibiting neurological deficits of varying degrees due to the random inactivation of the × chromosome. Such deficits could, in turn, affect the phenotypic outcomes in both control and Pdha1^hGFAP progeny. First, we evaluated the expression of the hGFAP-Cre promoter in the brain by crossing FVB-Tg(GFAP-cre)25Mes/J ^+/− males with Gt(ROSA)26Sor-EGFP females (Fig. 1A). The promoter was highly expressed in regions with high cell density, such as the cortex, hippocampus, and cerebellum. As previously described in Pdha1^hGFAP animals on a C57BL/6 background¹⁰, the condition was lethal in our Pdha1^hGFAP mice, with an average postnatal lifespan of 30 ± 6.4 days (mean and standard error of the mean will be used hereafter). These mice manifested a similar neuropathological phenotype to patients, including microencephaly, cortical atrophy, hypoplasia of the corpus callosum, and atrophy of the hippocampus and cerebellum (Fig. 1B)¹⁰. The expression of PDHA1 was significantly reduced in the cortex, cerebellum, and hippocampus by 35%, 47.6%, and 89% compared to controls, respectively, which resemble the pattern of decreased PDH activity in these animals relative to controls (Fig. 1C). The Pdha1 gene was targeted in both neuronal and glial populations by the hGFAP-Cre promoter, as evidenced by the reduction in PDHA1 protein expression in glia and granule cells in the Pdha1^hGFAP cerebellum relative to the control (Fig. 1C).

Figure 1. Structural and molecular characterization of the Pdha1^hGFAP brain.

(A) Immunofluorescence staining to visualize Cre-recombinase expression indicated by GFP in a P21 hGFAP-cre;Rosa26^GFP reporter mouse brain. Scale bar: 2 mm. (B) Cresyl violet staining on sagittal sections of control and Pdha1^hGFAP mouse brain. Scale bar: 2 mm. (C) Western blot and quantification of Pdha1 expression and total PDH activity in control and Pdha1^hGFAP cortex, hippocampus, and cerebellum (N=3/group). Western blot of Pdha1 in granule cell neurons and glial cells from P7 cerebella of Pdha1^hGFAP and control mice. Molecular weight markers are displayed in Kd. (D) Tunel staining and quantification of Tunel positive area normalized to the Dapi area in cortex, hippocampus, and cerebellum (N=3/group). (E) Immunofluorescence staining and quantification of neuronal density, as reported by NeuN, and density of reactive astrocytes, as reported by cell size of GFAP⁺ cells, in cortex (scale bar: 200 μm), hippocampus (scale bar: 200 μm), and cerebellum (scale bar: 50 μm) (N=5/group). GCL: granule cell layer. Unpaired Student t-test was used to carry out the statistical analysis hereafter unless specified otherwise. Values are expressed in mean ± S.E.M in all figures unless indicated otherwise. *: p< 0.05, **: p< 0.01, ***: p< 0.001.

The Pdha1^hGFAP brain had ongoing cell death, as illustrated in the Tunel staining (Fig. 1D). Cell death occurred in the neuronal population based on the decreased density of NeuN⁺ cells in the cortex, hippocampus, and cerebellum (Fig. 1E). This contrasts with the increased density and size of astrocytes in those brain regions, a feature commonly seen in the context of cell death and neurodegeneration¹³. The analysis of reactive astrocyte transcripts revealed a mixed population of A1, A2, and PAN-positive astrocytes (Fig. S1A–B)^14,15. Following the current consensus, we used the term reactive astrocytes to define these cells, given their morphological and molecular changes in the Pdha1^hGFAP brain¹⁴. Neuronal loss and reactive astrocytosis increased with age in the Pdha1^hGFAP animal, such that no differences in density of NeuN⁺ cells, presence of reactive astrocytes, and reactive astrocyte transcripts were observed between animal groups at P15, in contrast to P23 mice (Fig. S1B, Fig. S2A). In turn, transcripts of pro-inflammatory cytokines were overall elevated in the Pdha^hGFAP brain, particularly in the hippocampus, relative to that of the control at P23 (Fig. S1C). Together, neuronal loss, reactive astrocytosis, and inflammation mirrored those observed in the brain of patients with PDHD⁵, indicating that neurodegeneration is taking place¹⁶.

In vivo metabolic profile of the Pdha1^hGFAP brain is consistent with neuronal loss and defective glucose metabolism

To characterize the structural phenotype and the static metabolic profile in vivo in the Pdha1^hGFAP mouse brain, we carried out MRI and ¹H-MRS imaging. As seen in patients with this condition¹⁷, Pdha1^hGFAP mice had symmetric T₂-weighted hyperintensity lesions in the gray matter, including the cortex, hippocampus, and to a lesser extent the cerebellum (Fig. 2A). This is consistent with the ongoing cell death and inflammation described earlier in histological specimens. The in vivo metabolic pattern was well differentiated between control and Pdha1^hGFAP animals, as shown in the PCA model scores (Fig. 2B). When normalizing proton spectral data to the creatine peak, the decreased NAA and glutamate/glutamine (glx) peaks were statistically different between groups across all brain regions, which are consistent with neuronal loss (Fig. 2C). Increased lactate/lipid reached statistical significance in the cerebellum (Fig. 2C). Ex-vivo lactate analysis in the Pdha1^hGFAP and control brain substantiated this finding, supporting the notion that glucose metabolism through PDH is impaired in these animals (Fig. S3). Other detected metabolites by ¹H-MRS, such as taurine, choline, myo-inositol, and creatine/phosphocreatine had lower levels in the Pdha1^hGFAP hippocampus (Fig. S4A), except branched-chain amino acids, which were similar in both groups. Unexpectedly, aspartate was elevated in the Pdha1^hGFAP cerebellum compared to its counterpart in control mice (Fig. 2C), and it was equivalent in the cortex and hippocampus between both groups. These data indicate that aspartate metabolism differs from other Krebs cycle-derived metabolites, such as glutamate and glutamine, which were low in the Pdha1^hGFAP brain.

Figure 2. Brain MRI, ¹H-MRS, and ¹³C-Hyperpolarized MRI.

(A) Sagittal and coronal T₂-weighted MRI of control and Pdha1^hGFAP brain. Yellow squares depict where voxels were placed to perform ¹H-MRS analysis. Yellow and orange arrows illustrate T_2-weighted hyperintensities in the cortex and hippocampus in the Pdha1^hGFAP mouse brain, respectively. (B) PCA analysis of animal groups and individual metabolites. (C) Representative ¹H-MRS of cortex, hippocampus, and cerebellum normalized to creatinine and quantification of peak intensity areas for each metabolite (N=8/group). A Welch T-test with Benjamini-HochBerg correction for multiple comparison analysis was performed, and p-adjusted values were reported. (D) Schematic of ¹³C-hyperpolarized (HP) pyruvate metabolism in the cell and quantification of bicarbonate:lactate ratio are illustrated (Control: N=5, Pdha1^hGFAP: N=4).

To visualize intermediary metabolism as it happens in the intact brain, we used hyperpolarized [1-¹³C]pyruvate imaging to assess the downstream glycolytic switch where carbons are shunted into the Krebs cycle through PDH. Hyperpolarized (HP) magnetic resonance imaging (MRI) amplifies the MRI signal ~10,000 fold, such that it allows the detection of labeled metabolic products at micromolar concentrations and the analysis of metabolic dynamics in real-time¹⁸. Recent work has translated this approach to humans for imaging brain tumors and the normal brain¹⁹. Dynamic magnetic resonance spectroscopic imaging was acquired over the whole brain, separating HP pyruvate and its metabolic products, primarily lactate and bicarbonate (Fig. S4B). To quantify pyruvate metabolism through PDH in the brain in vivo, we calculated the bicarbonate:lactate ratio, an indicator of pyruvate metabolism through PDH vs. lactate dehydrogenase, respectively. The ratio was statistically higher in the control brain relative to the Pdha1^hGFAP brain (Fig. 2D), supporting the notion that glucose oxidation through PDH and downstream pathways is impaired in the brain of the Pdha1^hGFAP animal in vivo.

Molecular and histological findings suggest the metabolic role of reactive astrocytes in the Pdha1^hGFAP brain

Given the abundance of reactive astrocytes in the context of defective energy metabolism, we hypothesized that a potential role of these cells is to increase the incorporation of nutrients into the Pdha1^hGFAP brain²⁰. In particular, blood vessels and astrocytes regulate glucose uptake into the brain by modulating the expression of the primary glucose transporter in the blood-brain barrier: glucose transporter type 1 (GLUT1) (Fig. 3A)²¹. In support of this hypothesis, reactive astrocytes were abundantly present around blood vessels in the Pdha1^hGFAP brain at P23. The overlap between astrocyte GFAP signal and the wall of blood vessels was significantly higher in Pdha1^hGFAP mice compared to controls (Fig. 3A). The diameter of blood vessels was overall increased in the brain of Pdha1^hGFAP animals, except in the hippocampus (Fig. 3B). An analysis of blood vessel diameter by ranges showed that enlarged vessels (> 6 μm)²² were more abundant in the cortex, hippocampus, and cerebellum of Pdha1^hGFAP mice, suggesting an increased blood supply to these brain regions (Fig. 3B). Based on these results, we then quantified the expression of the primary glucose transporters in the brain: GLUT1 and GLUT3. The two isoforms of GLUT1, 55 kDa (vessel isoform) and the 45 kDa (astrocyte isoform), were highly expressed in the cortex, hippocampus, and cerebellum of Pdha1^hGFAP mice compared to controls (Fig. 3C). In contrast, GLUT3 (54 kDa), which is primarily expressed in neurons, was consistently decreased in Pdha1^hGFAP animals in those regions. These findings progressed with age (Fig. S2). Neuronal density and expression of GLUT3 decreased in Pdha1^hGFAP mice compared to controls from P15 to P23 (Fig. S2A–B), and, on the contrary, reactive astrocytosis, expression of GLUT1, and blood vessel diameter increased (Fig. S2A–D). These data suggest that glucose uptake into the brain is upregulated as the disease progresses in Pdha1^hGFAP mice, and it might be driven by reactive astrocytes.

Figure 3. Gliovascular unit and expression of glucose transporters.

(A) Schematic of the gliovascular unit and immunostaining of blood vessels and astrocytes in the hippocampus using GLUT1 and GFAP markers, respectively (3 sections/animal, 3 animals/group). Scale bars: 20 μm. (B) Immunofluorescence staining of blood vessels using GLUT1 marker (Bar = 200 μm), Imaris 3D reconstruction of blood vessels (Bar = 20 μm), and quantification of average and total blood vessel diameter in cortex, hippocampus, and cerebellum (N = 5/group). (C) Quantitative Western blot analysis of GLUT1 and GLUT3 in cortex, hippocampus, and cerebellum (N = 5/group).

Glucose uptake in vivo increases with age in the Pdha1^hGFAP brain

To analyze the potential upregulation of glucose uptake in the Pdha1^hGFAP brain in vivo, we performed positron emission tomography (PET) imaging with [¹⁸F]fluorodeoxyglucose (FDG). This analogue of glucose serves as an indicator of the amount of glucose uptake in the brain over time. Once phosphorylated, FDG predominantly accumulates in the cytoplasm of brain cells as FDG-6-phosphate, given its lack of the 2-hydroxyl group needed for further glycolysis (Fig. 4A). In contrast to P15 in which the FDG signal was similar between animal groups (Fig. S5A–B), the FDG uptake in the brain at P23 was significantly higher in Pdha1^hGFAP mice compared to controls and positively correlated with the expression of GLUT1 (Fig. 4B, Fig. S5C). The estimated FDG uptake in the cortex, hippocampus, and cerebellum was also increased in these animals at that age. To evaluate the distribution of glucose uptake with a higher anatomical resolution, we performed ex-vivo autoradiography imaging with ¹⁴C-2-deoxyglucose (¹⁴C-2DG) at P23. This tracer accumulates in the brain by the same chemical principle as FDG, indicating the amount of glucose uptake into the brain. Results were consistent with FDG-PET data, such that the ¹⁴C-2DG signal was overall increased in the cortex, hippocampus, and cerebellum of Pdha1^hGFAP mice relative to controls (Fig. 4C). These results support the idea that in vivo glucose uptake increases with disease progression in the Pdha1^hGFAP animal.

Figure 4. FDG-PET/CT and ¹⁴C-2DG autoradiography analyses from control and Pdha1^hGFAP brains.

(A) Schematic showing metabolism of FDG and ¹⁴C-2DG in neurons and astrocytes. (B) Representative coronal and sagittal FDG-PET/CT brain images of control and Pdha1^hGFAP mice and brain atlas–guided PET quantification in standardized uptake value (SUV) (Control: N=8, Pdha1^hGFAP: N=6). (C) Autoradiography and cresyl violet staining of coronal sections from both groups and quantification of ¹⁴C-2DG signal intensity (N=4/group). Cereb: cerebellum, Hip: hippocampus, DCN: deep cerebellar nuclei.

While glycolysis is enhanced, alternative substrates to glucose are used to support Krebs cycle activity in the Pdha1^hGFAP brain

To analyze intermediary metabolism downstream of glucose uptake, we performed a non-steady state ¹³C isotopologue analysis in flash-frozen brains after the administration of [U-¹³C]-glucose at P15 and P23. [U-¹³C]-glucose was administered intraperitoneally (2 mg/g of body weight) to awake animals after 5 hours of fasting, and brain and blood samples were obtained 30 min later (Fig. 5A). We did not reach isotopic steady state, which is achieved approximately 1 hour after a 2 mg/g of glucose injection¹⁰, because we aimed to evaluate how fast glucose is incorporated and metabolized in the brain. The concentration and ¹³C enrichment of glucose in blood right before tissue extractions were similar in both groups at P15 and P23 (Table S1).

Figure 5. Analysis of [U-¹³C]-glucose metabolism in the brain.

(A) Design of the experiment. (B) Analysis of metabolites pool size based on mass spectrometry signal intensity in blood, cortex, hippocampus, and cerebellum and Pdha1^hGFAP:control ratio of signal intensities (N=7/group). (C) Heatmaps of Pdha1^hGFAP:control fold change of isotopologue enrichments of metabolites identified in blood, brain, cortex, hippocampus, and cerebellum. Scale restricted from 1 to −1 for illustration purposes. (D) Average of Pdha1^hGFAP:control fold change of total ¹³C enrichment of metabolites in the brain. Cluster dendrogram demarcating metabolite groups based on labeled/total pools. (E) Isotopologue analysis of brain metabolites in the Pdha1^hGFAP brain relative to the control brain expressed as fold changes. (F) Schematic of ¹³C enrichments of glycolytic metabolites and the first turn of the Krebs cycle metabolites in the Pdha1^hGFAP brain relative to the control expressed as fold changes. 3-PG: 3-phosphoglycerate, OAA: oxaloacetate, α-KG: α-ketoglutarate.

Given the severity of the disease and the significant increase in glucose uptake at advanced disease stages, we started our analysis at P23. We first evaluated the total pool sizes of metabolites upstream and downstream of PDH as revealed by mass spectrometry intensity signal (Fig. 5B). While the pool sizes of glucose and lactate in blood were similar between both groups (Pdha1^hGFAP:control ratio of 0.87 and 0.91, respectively), the pools of glucose, 3-phosphoglycerate (3-PG), and lactate were 4.9, 2.3, and 1.5 times higher in the Pdha1^hGFAP brain compared to controls, respectively. These results suggest that glucose was incorporated faster and/or that glycolytic intermediates accumulated in the Pdha1^hGFAP brain compared to the control. Downstream of PDH, the Pdha1^hGFAP:control ratio was <1 for most Krebs cycle intermediates and metabolites that derive directly from the Krebs cycle, following a U-shape distribution (Fig. 5B). The Pdha1^hGFAP:control ratio went down until succinate (citrate: 0.79, glutamate: 0.47, and succinate: 0.52), and then increased in fumarate (0.97), malate (0.96), and aspartate (1.46). This finding suggests that the incorporation of substrates to the Krebs cycle is different between animal groups, such that glucose-derived metabolites support the Krebs cycle through PDH in the brain of control animals, whereas non-glucose-derived substrates are potentially incorporated downstream of PDH in the Pdha1^hGFAP mouse brain.

We further investigated this finding by performing a ¹³C isotopologue analysis in blood and brain. While the enrichment of glucose M6 (six ¹³C carbons within the molecule) in blood was not statistically different between animal groups at P23 (Fig 5C, Table S1), in the brain the enrichment was higher in the Pdha1^hGFAP brain compared to the control. Furthermore, the enrichment of the glycolytic intermediates 3-PG M3 (three ¹³C carbons) and lactate M3 was also higher in the Pdha1^hGFAP brain (Fig 5C, Table S2). These data indicate that, at P23, the incorporation of [U-¹³C]-glucose to the brain and its metabolism through glycolysis was faster in Pdha1^hGFAP mice than controls. These results contrast to those at P15, in which brain glucose uptake by FDG-PET and the ¹³C enrichment of lactate and alanine in the brain were not different between control and Pdha1^hGFAP mice (Fig. S5A–B, Table S3). Downstream of PDH, the enrichment of labeled ions (≥M1) in Krebs cycle intermediates and intermediate-derived metabolites went from similar values between animal groups at P15 (except M0 to M2 in fumarate, malate, and aspartate) to globally decreased values at P23 in the Pdha1^hGFAP animal compared to the control (Fig. 5C, Table S2, Table S3), suggesting that there is significant incorporation of unlabeled substrates to the Krebs cycle in the Pdha1^hGFAP brain, primarily at advanced disease stages.

To identify the potential entry points of unlabeled substrates to the Krebs cycle in the Pdha1^hGFAP brain at P23, we computed the fold change, namely (Pdha1^hGFAP-Control)/Control, of the labeled pool vs. the total pool of each metabolite in Pdha1^hGFAP animals relative to controls. Figure 5D shows the averaged data from the three brain regions in each animal group (data from individual regions is shown in Fig. S6A). Whereas the labeled pool vs. the total pool fold change was >0 in glucose, 3-PG, and lactate, the fold change was <0 in metabolites downstream of PDH, particularly after succinate. Such distribution was well demarcated by cluster analysis, where upstream metabolites and intermediates located before and after succinate were well separated (Fig. 5D). On this basis, we next performed a carbon-by-carbon isotopologue analysis of Krebs cycle intermediates and derivates (Fig. 5E). The enrichment of M0 was approximately 25–50% higher in metabolites located before succinate in the Krebs cycle in the Pdha1^hGFAP brain compared to the control, whereas intermediates located at and after succinate, the enrichment of M0 was consistently >60% in the PDH brain. This suggests that there is a potential entry point of unlabeled substrates at or before succinate at P23. The analysis of M2 of Krebs cycle intermediates provided further insights into this feature. Starting from lactate, in which the averaged M3 enrichment was 23% higher in the Pdha1^hGFAP brain than the control brain (Table S2), the enrichment of M2 of citrate dropped to approximately 5%, and then it became higher in the control in subsequent reactions of the Krebs cycle: 25–40% higher in glutamate, glutamine, and GABA and approximately 50% in succinate, fumarate, malate, and aspartate (Fig. 5E). These data indicate that there are two potential entry points of unlabeled carbons to the Krebs cycle in the Pdha1^hGFAP brain (Fig. 5F): between pyruvate and citrate and between α-ketoglutarate and succinate.

Propionate serves as an anaplerotic substrate in the Pdha1^hGFAP brain

We previously identified acetate as an alternative energy substrate to glucose in the Pdha1^hGFAP brain (Fig. 5F)¹⁰. Acetate is metabolized to acetyl-CoA, which enters the Krebs cycle for further oxidation. It contributes to the drop of the ¹³C labeling between lactate M3 and citrate M2, which here it was 40.1% in the Pdha1^hGFAP brain and 25.4 % in the control brain. As described earlier, we found another potential entry point of unlabeled carbons to the Krebs cycle in Pdha1^hGFAP brain: between α-ketoglutarate and succinate. It is known that the main source of biosynthetic precursors to the Krebs cycle between α-ketoglutarate and succinate is succinyl-CoA, and, in turn, one of the main substrates that enters the cycle through succinyl-CoA is propionate. Propionate crosses the blood-brain barrier through monocarboxylate transporters. Once in the brain, it can be carboxylated and converted to succinyl-CoA through epimerase and mutase-mediated reactions. Based on these findings, we analyzed the incorporation of ¹³C-propionate to the Krebs cycle in the brain of Pdha1^hGFAP and control mice.

We used the same experimental approach as in [U-¹³C]glucose injections. We administered [U-¹³C]propionate intraperitoneally (0.5 mg/g of body weight) to control and Pdha1^hGFAP mice at P23 after a 5-hour fasting (Fig. 6A). At 30 min of the injection, several regions of the brain were collected for a multipurpose analysis (Fig. 6A): 1) cortex, hippocampus, and cerebellum of one hemisphere were extracted for polar metabolite analysis using gas-chromatography mass-spectrometry (GC-MS); and 2) the frontal 2/3 of the contralateral hemisphere was used for metabolic compartmentation analysis by ¹³C nuclear magnetic resonance (¹³C-NMR) spectroscopy and the 1/3 left for CoA enrichment analysis by liquid-chromatography mass-spectrometry (LC-MS). In these animals, we measured the concentration and ¹³C enrichment of glucose in blood because propionate can serve as a gluconeogenic substrate²³. There were no differences in the content or ¹³C enrichment of glucose in blood between animal groups (Table S1). The labeling pattern of Krebs cycle intermediates and derivates was the opposite of that observed in [U-¹³C]-glucose injections (Fig. 6B; Fig. S6B; Table S4). The Pdha1^hGFAP:control fold change of all intermediates and most intermediate-derived metabolites labeled in M3 (generated from [U-¹³C]propionate) was >0, primarily from succinate to malate in which the average fold change between animals was 3.3 (Fig. 6B–D). In addition, the ¹³C enrichment of M3 propionyl-CoA and methylmalonyl-CoA were significantly higher in the Pdha1^hGFAP brain compared to that of the control (Fig. 6G; Table S5). These results indicate that propionate accesses the brain and is metabolized to and through the Krebs cycle to a greater extent in the Pdha1^hGFAP brain than in the control brain. The enhanced expression of monocarboxylate transporter type 1 (MCT1) in the cortex, hippocampus, and cerebellum of the Pdha1^hGFAP brain, as opposed to the control, supports this observation (Fig. S7A–B). In contrast to [U-¹³C]-glucose injections in which the enrichment of aspartate M2 (derived primarily from PDH metabolism of [U-¹³C]glucose) and aspartate M3 (derived from pyruvate carboxylase and also from PDH after several turns through the Krebs cycle) was decreased in Pdha1^hGFAP mice, the enrichment of aspartate M3 derived from [U-¹³C]propionate was not different between groups, suggesting that propionate may contribute to aspartate synthesis in the Pdha1^hGFAP brain.

Figure 6. Analysis of [U-¹³C]-propionate metabolism in the brain.

(A) Design of the [U-¹³C]propionate experiment. (B) Heatmap of Pdha1^hGFAP:control fold change of isotopologue enrichment of metabolites identified in cortex, hippocampus, and cerebellum (Control: N=7, Pdha1^hGFAP: N=6). (C) Pdha1^hGFAP : control fold change of M3 enrichment of metabolites in the brain. (D) Isotopologue analysis of brain metabolites in the Pdha1^hGFAP brain relative to the control brain expressed in fold change. (E) Representative ¹³C-NMR spectrum from control and Pdha1^hGFAP mice. Demarcated regions in each spectrum (star and parenthesis) are expanded in panel F. (F) ¹³C-NMR spectrum and isotopomer analysis of glutamate, glutamine, and aspartate C3. (G) Isotopologue fractional enrichment of CoAs and ratios of M3 enrichments between Pdha1^hGFAP and control brains (H) Schematic of the proposed compartmentalized propionate metabolism in astrocytes and neurons. Glut: glutamate, Gln: glutamine, OAA: oxaloacetate, α-KG: α-ketoglutarate, Asp: aspartate, Ala: alanine, Tau: taurine, NAA: N-acetyl-aspartate. C#: carbon labeled in position #. Sx: singlet, Dxx: doublet, Q: quartet.

We further corroborated the potential anaplerotic role of propionate in the Pdha1^hGFAP brain by co-injecting [U-¹³C]glucose with unlabeled propionate (Fig. S8A). If propionate incorporates to the Krebs cycle despite the oxidation of ¹³C-glucose, it is expected to further drop the ¹³C labeling of Krebs cycle intermediates in the Pdha1^hGFAP brain compared to when only [U-¹³C]glucose is injected. We administered an equivalent amount of [U-¹³C]glucose and unlabeled sodium propionate intraperitoneally (2 mg/g of body weight total) following a 5-hour fasting. After 30 minutes, we collected the cortex, hippocampus, and cerebellum and extracted polar metabolites for isotopologue analysis by GC-MS. Glucose enrichment in plasma was similar between both groups, and the concentration was approximately two-fold higher in Pdha1^hGFAP mice (Table S1). The Pdha1^hGFAP:control fold change of ¹³C enrichment of Krebs cycle intermediates and derivates dropped an average of 21.9% from single [U-¹³C]glucose injections to [U-¹³C]glucose and unlabeled propionate injections, in which aspartate showed the highest drop (30.5%) (Fig. S8B–D; Fig. S9; Table S6). This indicates that propionate contributes to Krebs cycle anaplerosis in the Pdha1^hGFAP brain despite glucose metabolism. A similar pattern was observed in metabolites upstream of PDH. The fold change of labeled lactate dropped 45.5% when propionate was administered with [U-¹³C]glucose compared to single [U-¹³C]glucose injections. These data suggest that propionate can exert a glucose-sparing effect as observed with other monocarboxylates, such as acetate and even-carbon ketone bodies^10,24,25.

To identify the cell metabolic compartment where propionate is metabolized in the brain in vivo, we performed a ¹³C-NMR isotope distribution analysis (isotopomer analysis hereafter) on the extracted contralateral hemisphere. In contrast to mass spectrometry, NMR allows the analysis of neuronal and astrocyte metabolism in vivo by tracking the labeling change of compartment-specific metabolites over time²⁶. Neurons obtain most of their energy from oxidation of glucose and are where glutamate and GABA are produced, whereas metabolism of acetate and glucose and net synthesis of glutamine occur in astrocytes^21,26. In tissues like the heart and liver^27,28, it has been shown that [U-¹³C]propionate labels succinyl-CoA in three carbons (M3, carbons 1,2,3 or 2,3,4 given its symmetry), but only as it enters the Krebs cycle for the first time. As succinyl-CoA M3 is metabolized through the cycle, it will label succinate, fumarate, malate, aspartate, and citrate in three carbons, and glutamate, glutamine, and GABA in three or two carbons depending on whether carbon 1 of isocitrate gets labeled, which is removed via decarboxylation to make α-ketoglutarate. At the end of the first turn, one or two ¹³C carbons derived from [U-¹³C]propionate are lost, such that succinyl-CoA will be labeled in one or two carbons, but never in three, assuming that [U-¹³C]propionate is the only source of ¹³C to the cycle (Fig. S10). We used these principles to analyze and interpret the ¹³C-NMR data from the brain tissue.

Figure 6E shows a representative ¹³C-NMR spectrum from control and Pdha1^hGFAP brains. To our knowledge, this is the first time that ¹³C multiples (the NMR signals resulting from ¹³C-¹³C spin-spin coupling) have been resolved in the brain after the administration of ¹³C-propionate¹¹. There is an increased signal of lactate and alanine in the Pdha1^hGFAP animal compared to the control, and a decreased signal of metabolites located downstream of PDH, including glutamate and glutamine. Given that glutamate and glutamine will be primarily labeled in carbons 2 and 3, and 1, 2, 3 from [U-¹³C]propionate (Fig. S11; Table S7), we focused the isotopomer analysis on carbons 2 and 3. Figure 6F shows that the labeling pattern of glutamate and glutamine C3 was different between animals. Since there is only one transamination reaction between glutamate and glutamine, the labeling pattern of both metabolites is expected to be similar (Fig. S11). Here, the labeling of these metabolites was different, and such difference was more pronounced in the Pdha1^hGFAP animal. This suggests that these metabolites do not completely exchange during the 30 min of [U-¹³C]propionate exposure, namely there is no full transfer of the ¹³C labeling pattern from one metabolite to the other. Because the doublet 3 (D3) of glutamine was higher than that of glutamate in both Pdha1^hGFAP and control animals (Fig. 6F), it indicates that propionate is primarily metabolized in glial cells and to a higher extent in the Pdha1^hGFAP brain compared to the control. A similar pattern was observed in carbon 2 of both metabolites (Fig. S11). The doublet D23, which derives primarily from [U-¹³C]propionate, was higher in the Pdha1^hGFAP brain, whereas the doublet D12, which derives primarily from ¹³C-glucose (from [1,2-¹³C] to [U-¹³C] generated from [U-¹³C]propionate), was higher in the control, such that the D23:D12 ratio of glutamate and glutamine C2 was significantly increased in the Pdha1^hGFAP animal compared to the control. To corroborate these findings, we conducted a comparative analysis of gene expression levels for enzymes implicated in propionate metabolism. This analysis was performed on nuclei isolated from astrocytes and neurons from Pdha1^hGFAP and control brains. As illustrated in Figure S7C, we observed a marked elevation in the expression of genes encoding propionyl-CoA carboxylase alpha subunit (Pcca), propionyl-CoA carboxylase beta subunit (Pccb), and methylmalonyl-CoA mutase (Mmut) within astrocytes when compared to neurons in the brains of Pdha1^hGFAP mice, as well as in comparison with controls. This increase corroborates the heightened incorporation and metabolism of propionate in the glial compartment, as evidenced by our ¹³C-NMR data. Collectively, these findings, reinforced by the expression pattern of MCT1, support the notion that propionate oxidation is enhanced in the brain of Pdha1^hGFAP mice compared to controls, and that glial cells are the cell type where propionate metabolism takes place primarily (Fig. 6G–H).

The labeling pattern of aspartate also contains important information about how propionate is metabolized in the brain (Fig. 6E–F, Fig. S10). Figure S10 simulates the labeling pattern of aspartate C3 under different percentages of ¹³C enrichment of propionate and glucose-derived pyruvate. When [U-¹³C]propionate is 70% enriched, similar to our experimental conditions in Pdha1^hGFAP mice (Table S5), the doublet D23 of aspartate C3 is higher than the doublet D34, and such difference increases as propionate’s contribution to Krebs cycle anaplerosis goes up. In contrast, when glucose-derived [U-¹³C]pyruvate is enriched 50%, and propionate is not labeled, the doublet D34 of aspartate C3 is higher than the doublet D23, and such difference increases as propionate contribution to Krebs cycle anaplerosis rises. Overall, these indicate that propionate contributes mostly to the D23 enrichment of aspartate C3, whereas glucose contributes primarily to D34 labeling. In our experimental data, the doublet D23 of aspartate C3 was significantly higher in the Pdha1^hGFAP animal, whereas the doublet D34 was higher in the control (Fig. 6F). These results, along with the mass spectrometry data discussed earlier, indicate that propionate labels aspartate to a higher extent in the Pdha1^hGFAP brain than the control, and that it could contribute to its net synthesis and the prominent ¹H-MRS signal in the brain of Pdha1^hGFAP mice.

Propionate, along with a ketogenic diet, improves survival and the brain structural and motor phenotypes of Pdha1^hGFAP mice

We next examined the impact of dietary propionate on the survival, clinical, and neuropathological manifestations of Pdha1^hGFAP mice. We first determined the optimal dose of sodium propionate, administered via drinking water, in four-week-old wild-type mice maintained on a standard chow diet. Adhering to established guidelines²⁹, we monitored the animals’ weight gain at three-day intervals for a fortnight. Starting with a 500 mM sodium propionate, as documented previously³⁰, we subsequently tested concentrations of 300 mM, 150 mM, and included a control group without propionate (N=4/group). Tolerance to propionate was evaluated based on the trajectory of the animals’ weight, considering a variance within ±10% of the control group as indicative of acceptable tolerance. Conversely, a deviation in weight >10% was deemed to reflect poor tolerance. Our cutoff was more stringent than the 15% recommended by available guidelines for testing new dietary supplements or drugs in mice²⁹. This decision was based on two considerations: firstly, our animals are of developmental age and therefore more sensitive to weight changes, and secondly, any weight reduction in Pdha1^hGFAP mice could potentially precipitate additional metabolic stress. As shown in Figure S12A, the cohort subjected to 500 mM sodium propionate experienced a significant weight reduction, approximating 120% to that of the control group. Mice receiving 300 mM had a weight decrease of 14.8%, whereas those on 150 mM exhibited a marginal weight increase of 1.5% over control animals. Based on these results, we proceeded with 150 mM sodium propionate for further testing.

Given that the ketogenic diet (KD) is the cornerstone treatment for individuals with PDHD and based on the avid incorporation of acetyl-CoA sources alternative to glucose by the Pdha1^hGFAP brain^10,31, we further tested the tolerance of propionate in conjunction with the KD. Initially, we developed a KD formulation suitable for gestation. The conventional dietary fat-to-carbohydrate ratio prescribed for PDHD patients is 4:1, yet evidence from murine studies suggests potential fetal harm associated with this ratio^32,33. By moderating the ratio to approximately 2.5:1, comprising 70% fat, 15% carbohydrates, and 15% proteins, we achieved a state of ketosis in adult animals (β-hydroxybutyrate levels in plasma 0.83 ± 0.23 mM), while avoiding adverse effects such as low birth weight, miscarriages, or reduced offspring size in wild-type mice (Fig. S12A–B). This tailored KD was used throughout this study in both pregnant and non-pregnant mice. We co-administered 150 mM sodium propionate solution via drinking water with our KD to first evaluate tolerance of this combination in four-week-old, wild-type, non-pregnant mice. Over a period of two weeks, these mice exhibited a weight gain comparable to those on a regular diet and showed approximately 9% more weight gain than those solely on KD, as depicted in Figure S12A. Next, we evaluated the effects of sodium propionate intake during pregnancy and early postnatal periods. We monitored the litter size, birth weight, and weight progression of the progeny for the initial 40 postnatal days. One group of breeding pairs was provided with a chow diet and 150 mM sodium propionate, while another group received the same concentration of sodium propionate in conjunction with KD (N=3 breeding cages/group). Offspring numbers in the chow diet group supplemented with 150 mM sodium propionate were consistent with those on regular chow diet (Fig. S12B). However, the KD group with 150 mM sodium propionate manifested adverse effects: two pregnant females had to be euthanized due to dystocia at parturition, and a third female’s litter of five succumbed three days post-birth, leading to the termination of this experimental branch. In response to these outcomes, we tested a lower concentration of sodium propionate in combination with a KD during pregnancy. We found that 50 mM of sodium propionate with a KD resulted in litter sizes that were comparable to those observed in either the chow diet or KD-alone groups (Fig. S12B).

We next evaluated the postnatal growth of mice administered sodium propionate at 150 mM in a chow diet and at 50 mM in a KD, comparing these groups to a standard chow diet and KD cohorts (N = 5–6/group). The growth trajectories of the propionate groups largely mirrored those of the non-propionate group, as delineated in Figure S12C. However, a notable deviation was observed at P20 and P30, where mice receiving 150 mM sodium propionate with a chow diet exhibited a reduction in weight relative to their chow diet counterparts. Mice on KD presented with reduced body weights at P20 when compared to those on a regular diet. Contrastingly, no significant difference in weight progression was identified between the group on 50 mM sodium propionate with KD and the chow diet group (Fig. S12C).

Based on these data, we used 150 mM sodium propionate with chow diet, 50 mM sodium propionate with KD (KDP hereafter), and KD without propionate as experimental diets on Pdha1^hGFAP mice (Fig. 7A). We quantified plasma levels of propionate, glucose, and β-hydroxybutyrate at P15 to assess propionate intake by the animals during lactation, and the impact of the various diets on ketone body and glucose levels at that age. As delineated in Figure S3, plasma propionate was most concentrated in the 150 mM group (28.8 ± 2.9 μM), with the KDP group following (16.8 ± 0.97 μM) (Fig. S3A). The KD group did not show differences in plasma propionate levels compared to the chow diet group. No significant differences in glucose concentrations were observed across groups (Fig. S3B). The concentration of β-hydroxybutyrate was only different in the 150 mM propionate group, which was lower (0.08 ± 0.006 mM) than the other three groups.

Figure 7. Analysis of dietary propionate on survival, weight gain, pathological findings, and motor dysfunction of Pdha1^hGFAP mice.

(A) Schematic of the three dietary regimens administered during gestation and the postnatal period, and the metabolic pathways targeted with each diet. (B) Survival curves of Pdha1^hGFAP mice on chow diet (N=7), ketogenic diet (KD) (N=8), KD with sodium propionate 50 mM on the drinking water (KDP) (N=8), and chow diet with 150 mM of sodium propionate (Prop) (N=7). Log-rank statistical test was used to compare the survival distributions of animals among experimental diets. (C) Analysis of weight gain during the first 30 postnatal days of Pdha1^hGFAP mice subjected to each dietary condition (Chow, N=5, KD, N=9, KDP, N=8, Prop, N=7). A pairwise t-test analysis followed by Benjamini-Hochberg p-value correction was used hereafter in this figure to compare the chow diet group with other experimental dietary groups individually. (D) Analysis of neuronal density as defined by NeuN⁺ cells in each region (normalized to the wildtype group of each experimental condition for comparison purposes), and reactive astrocytosis, denoted by GFAP⁺ cells of sizes established in Pdha1^hGFAP mice on a regular diet, within the cortex, hippocampus, and cerebellum (Chow, N=7, KD, N=5, KDP, N=4, Prop, N=3). Scale bars in cortex (100 μm), hippocampus (200 μm), and cerebellum (50 μm). (E) Analysis of lateral body sway (pixels) of both control and Pdha1^hGFAP mice while walking on an open field platform (Chow, N=9, KD, N=9, KDP, N=8, Prop, N=9). GCL: granule cell layer. NS: not significant.

The survival of Pdha1^hGFAP mice improved when exposed to propionate 50 mM with KD and with KD alone compared to those on a chow diet (Fig. 7B). The maximum survival of Pdha1^hGFAP mice on a chow diet was 31 days (this is a different cohort than the one describing the lifespan of these animals in the neuropathological findings section). Conversely, approximately 62% of mice on the KDP and 25% on the KD surpassed the 40-day mark, with the longest survivors reaching 50 days on the KDP and 100 days on the KD (p = 0.01 for both RD:KDP and RD:KD pairwise comparison using the log-rank test). No significant survival differences were detected between Pdha1^hGFAP mice on a chow diet and those administered a 150 mM propionate chow diet (Fig. 7B). Of note, two animals on 150 mM propionate survived for 41 and 45 days, respectively. Subsequent analysis focused on the weight trajectories of Pdha1^hGFAP mice during the initial 30 days postnatally since most animals died before that age. Mice on the KDP and KD exhibited an increased weight over the control diet group at P15 and P20 (Fig. 7C). This initial gain in weight was not sustained over time, with average weights of both groups reducing to levels comparable to the chow diet group from P25 onward. Mice receiving 150 mM propionate mirrored the weight trajectory of the control diet group. Notably, at day 30, the two surviving mice on the 150 mM propionate regimen showed weights of 7 grams and 4.8 grams, contributing to the observed rise in average weight at this time point.

Next, we analyzed the effects of these dietary interventions on the neuropathological findings of Pdha1^hGFAP mice. We focused the analysis on P21-P23 mice as neuropathological manifestations are florid at this age, as depicted in Figure 1E. Neuronal density in the cortex was found to be 15–20% increased in the KD and 150 mM propionate groups relative to the Pdha1^hGFAP mice on a chow diet. No significant differences in neuronal density were observed in the hippocampus across groups (Fig. 7D). As for the cerebellum, the thickness of the granular cell layer was found to be 10–13% greater in all experimental groups compared to the control group. As for reactive astrocytosis, the KDP, KD, and 150 mM propionate groups exhibited a decrease in reactive astrocyte density—up to 80%—in comparison to the group on a chow diet, specifically within the cortex and hippocampus. In the cerebellum, a statistically significant reduction in reactive astrocytosis was observed exclusively in the KD group.

We then evaluated the impact of experimental diets on Pdha1^hGFAP mouse motor function. For this purpose, we used the open field test. This assay involved recording the mice motor behavior from above and below an acrylic cage (see methods section for further details) (Fig. 7E). At P20, Pdha1^hGFAP mice on a chow diet exhibited abnormal motor behaviors such as prominent body’s lateral sway during ambulation, a typical manifestation of poor balance and coordination in mice^64,65 (Videos S1–2). These mice also displayed a narrowed step width and reduced stride length relative to wild-type mice and exhibited a preference for the periphery of the cage over the center, indicative of increased anxiety-like behavior (Fig. S13). To quantify the body’s sway, we analyzed the lateral body movements during locomotion (see methods section). Figure 7E and supplementary videos 2 to 5 show that only the KDP group exhibited some reduction in the amplitude of body sway during locomotion, with an approximate decrease of 20% compared to both the standard chow diet group and the group receiving 150 mM propionate. No significant differences were noted in the time spent in either the periphery or center of the cage among Pdha1^hGFAP mice on experimental diets (Fig. S13). Regarding gait metrics such as step width and stride length, they generally remained consistent with those of Pdha1^hGFAP mice on a standard chow diet, with a few exceptions. Those fed a KD demonstrated stride lengths and step widths comparable to wild-type mice. No differences were noted in locomotor speed among the groups (Fig. S13).

Discussion

In this study, we have mapped the metabolic network that underlies the structural and in vivo metabolic phenotypes of the Pdha1^hGFAP brain. As occurs in children with the severe form of this disease, Pdha1^hGFAP mice had brain inflammatory lesions reflected by the T₂-weighted hyperintensities on MRI in the context of ongoing cell death and reactive astrocytosis. The neuronal loss was reflected by the low NAA and glx peaks on ¹H-MRS, and the defective metabolism through PDH by the accumulation of lactate and the decreased bicarbonate:lactate ratio on ¹H-MRS and HP-MRI, respectively. Brain glucose uptake in vivo has not been previously evaluated in patients with PDHD because no clinical indication has been identified so far. The reactive astrocytosis, increased expression of GLUT-1, and vasodilation suggested a potential upregulation of brain glucose uptake in Pdha1^hGFAP animals.

The increased FDG-PET signal in the Pdha1^hGFAP brain compared to the control at P23 is consistent with that observed in acute inflammatory lesions of patients with other mitochondrial diseases. Ikawa et al. used double PET imaging with FDG and ⁶²Cu-ATSM to quantify glucose metabolism, cerebral blood flow, and oxidative stress at different stages of stroke-like lesions in mitochondrial myopathy, encephalopathy, lactic acidosis, and stroke-like episodes (MELAS)³⁴. Acute lesions showed increased glucose metabolism, cerebral blood flow, and oxidative stress as depicted by the enhanced FDG signal, the increased ⁶²Cu-ATSM signal at the early uptake phase, and the increased ⁶²Cu-ATSM signal at the delayed uptake phase, respectively. Authors hypothesized that the increased energy demand of cells with respiratory chain defects results in vasodilation, hyperemia, oxygen overload, and increased glucose uptake. In subacute stages (~1 month after the initial lesion), oxidative stress was markedly enhanced based on the increased ⁶²Cu-ATSM signal in the delayed phase, which authors attributed to the accumulation of reactive oxygen species. Cerebral blood flow and glucose metabolism were, however, reduced. In chronic stages (~1 year after the initial lesion), blood flow, glucose uptake, and oxidative stress were all significantly reduced, probably because of the cell loss and resolved inflammation. In the context of chronic brain damage, low FDG-PET signal has also been documented in patients with other mitochondrial diseases, such as Leigh syndrome or mitochondrial neurogastrointestinal encephalomyopathy (MNGIE)^35–37. It is challenging to perform FDG-PET imaging in the context of acute patient care because of the limited availability, logistic efforts, the requirement of specific facilities and technical support, and financial cost (Heckmann et al., 2019). Yet, in recent years, FDG-PET is becoming a key player in the acute management of patients with systemic inflammatory diseases, such as vasculitis, tuberculosis, and atherosclerotic plaque inflammation^38,39. In these conditions, inflammatory cells contribute to the increased glucose uptake in the involved tissue as much as 60% of the total signal in vivo, showing a high correlation between the degree of inflammation and the FDG-PET signal⁴⁰. Thus, FDG-PET is a feasible method that may contribute to stage brain disease in children with PDHD and adjust clinical management accordingly.

Other causes of neuroinflammation, such as intracranial bleeding or CNS infections, have also been associated with upregulation of glucose uptake in the brain^38,40,41. In mice subjected to intracerebral hemorrhage, the FDG-PET signal correlated with the GLUT1 expression. At early stages (day 1–3 post-injury), the FDG uptake and GLUT1 content decreased in the injured hemisphere, whereas 7 to 14 days post-injury, both parameters increased in parallel to reactive astrocytosis and proliferation of other inflammatory cells³⁸. In Pdha1^hGFAP mice, a similar process occurred. The increased expression of GLUT1 positively correlated with the rise of glucose uptake in the Pdha1^hGFAP brain compared to the control from P15 to P23 (Fig. S5C). Based on the distribution of ¹⁴C-2DG, the regions with the highest glucose uptake in the Pdha1^hGFAP brain were the cortex, hippocampus, and cerebellum, likely to compensate for the oxidative energy deficit in these regions (Fig. 4C). The increased glucose uptake was associated with augmented glycolytic metabolism. After the administration of [U-¹³C]glucose, the ¹³C enrichment and mass spectrometry signal of glucose, the glycolytic intermediate 3-PG, and the glycolytic end-product lactate were higher in the Pdha1^hGFAP brain than the control, indicating that glycolysis was faster in the brain of these animals. In contrast, downstream of PDH, the ¹³C enrichment was lower in the Pdha1^hGFAP brain compared to the control. We found that such a drop of ¹³C labeling was not uniform across Krebs cycle intermediates. There were two spots in the Krebs cycle where the enrichment dropped the most: 1) between pyruvate and citrate; and 2) between α-ketoglutarate and succinate. This finding indicates that unlabeled substrates are incorporated at these entry points probably to sustain Krebs cycle activity in the context of PDHD.

We previously identified the first entry point in the cortex of Pdha1^hGFAP mice¹⁰. The reduced enrichment of ¹³C-acetyl-CoA derived from [U-¹³C]glucose in the Pdha1^hGFAP cortex suggested that alternative, non-labelled substrates are metabolized independently of PDH activity at that level. After the co-administration of [1,2-¹³C]acetate, an immediate precursor of acetyl-CoA that bypasses PDH, and [1,6-¹³C]glucose, we found that the acetate-to-glucose oxidative ratio was four times higher in the Pdha1^hGFAP cortex than that of the control, supporting the notion that acetate is oxidized through the Krebs cycle in the context of impaired glucose metabolism. In addition to its impact on intermediary metabolism, acetate improved brain electrophysiological dysfunction in these animals. It enhanced cortical activation, potentiated local synaptic transmission as assayed by in vivo intracortical stimulation and recording, and improved the number and duration of electographic and clinical epileptiform events in these animals. As a result, acetate has become a potential new therapy for this condition. In this study, we found another entry point of unlabeled substrates to the Krebs cycle in the Pdha1^hGFAP brain.

The aspartate signal on ¹H-MRS gave us the first hint that alternative substrates to glucose may be entering the Krebs cycle in the brain of these animals. In contrast to glutamate and glutamine, aspartate levels in vivo were higher in the Pdha1^hGFAP cerebellum and preserved in the cortex and hippocampus compared to controls. At first, we thought this could result from a mass effect phenomenon due to the accumulation of pyruvate and its subsequent conversion to oxaloacetate and aspartate through pyruvate carboxylase (PC), as seen in cultured cells with defective PDH or pyruvate transport^42,43. With due caution for not being at isotopic steady state, the isotopologue analysis of [U-¹³C]glucose metabolism indicated that PC was not the main source of aspartate synthesis in the Pdha1^hGFAP brain. Most aspartate in these animals was unlabeled (M0 = 70–80%), and the enrichment of M3 (the labeling pattern derived from PC metabolism of [U-¹³C]glucose-derived pyruvate) only represented 7–10% of the total aspartate pool (Table S2). In controls, in contrast, most aspartate was ¹³C-labeled (50–60%) and the enrichment of M3 was 15–18% of the total pool (p <0.001). These data suggested that unlabeled substrates are a major source of aspartate synthesis in the Pdha1^hGFAP brain. Given the difference of total ¹³C enrichment of metabolites located before and after succinate synthesis in the Krebs cycle (Fig. 5D), we hypothesized that an entry point of unlabeled carbons in the Pdha1^hGFAP brain locates between α-ketoglutarate and succinate. At this level, the main entry point of substrates to the Krebs cycle is succinyl-CoA²¹.

There are several precursors of succinyl-CoA, including branched-chain amino acids, methionine, odd-chain fatty acids, and propionate. These metabolites undergo a series of reactions culminating in propionyl-CoA carboxylase and methylmalonyl-CoA mutase to generate succinyl-CoA⁴⁴. The difference between precursors of succinyl-CoA and acetyl-CoA is that succinyl-CoA precursors provide net carbons to replace Krebs cycle intermediates that engage in other reactions away from the cycle, like cataplerotic or biosynthetic pathways. The process of replenishing Krebs cycle intermediates is called anaplerosis. In contrast, the two carbons of acetyl-CoA that enter the cycle are lost in the form of CO₂ during the first turn, such that it does not provide net carbons to replenish Krebs cycle intermediates. Anaplerosis is essential to sustain biosynthesis, which in turn is necessary for the proper formation of the nervous system²¹. Developmental processes such as cell proliferation, migration, and myelination rely on a constant supply of biosynthetic precursors derived from the Krebs cycle, for example, citrate-derived acetyl-CoA for fatty acid and cholesterol synthesis, α-ketoglutarate-derived glutamate and glutamine for protein formation, and oxaloacetate-derived aspartate for DNA/RNA synthesis⁴⁵. This is probably the reason precursors of acetyl-CoA, such as acetate or even-chain ketone bodies, which have no anaplerotic properties, have minimal impact on the developmental structural defects in the brain of children with PDHD³¹.

Among succinyl-CoA precursors, we tested propionate as a potential anaplerotic substrate in Pdha1^hGFAP mice. Propionate crosses the blood-brain barrier in a dose-dependent manner^11,24. Its primary source is the bacterial anaerobic fermentation of dietary fibers and resistant starch, serving as a key player in the microbiota-gut-brain interactions⁴⁶. Propionate can modulate GABA neurotransmission by increasing GABA levels and inducing its release to the extracellular space²⁴, a property that has been recently used to treat seizures in animal models of epilepsy⁴⁷. Along with other short-chain fatty acids, propionate can reinforce blood-brain barrier integrity, influence levels of neurotrophic factors, and exert anti-inflammatory effects in the context of an autoimmune disorder or infection of the nervous system^30,46,50,51. The mechanisms by which propionate carries out such effects are still unknown. So far, its role in intermediary metabolism in the brain has been considered limited due to the lack of labeling of byproducts after the systemic administration of its ¹³C form¹¹. The dose administered, the field strength, and the innate properties of propionate metabolism in the brain may play a role in the limited sensitivity of the ¹³C-NMR signal obtained so far.

Here, we resolved, for the first time, propionate’s metabolism in the brain in vivo. Propionate is incorporated into the Pdha1^hGFAP brain to a greater extent than the control brain. The M3 enrichment of immediate products of propionate metabolism, such as succinate, fumarate, and malate, was approximately three-fold higher in the Pdha1^hGFAP brain than in the control. In turn, we found that propionate is primarily metabolized in glial cells in the Pdha1^hGFAP brain in vivo. Nguyen et al. found that [3-¹⁴C]propionate is a glial metabolic substrate based on the higher radioactivity of its byproducts in cultured astrocytes and tissue extracts from control mice¹¹. Given the fact that the net synthesis of glutamine occurs in glial cells⁴⁸, the higher enrichment of glutamine M3 over glutamate M3 (approximately two-fold in both controls and Pdha1^hGFAP mice) supported the notion that glia is where propionate metabolism takes place in vivo. An NMR isotopomer analysis provided further insights into this finding. The labeling pattern of glutamate and glutamine was different within and between animal groups. On the one hand, the higher doublet D3 of glutamine and glutamate in the Pdha1^hGFAP brain compared to the control indicates that propionate metabolism was increased in the brain of Pdha1^hGFAP mice. On the other hand, the higher doublet D3 of glutamine vs. glutamate in Pdha1^hGFAP and control mice supports the notion that glial cells drive the uptake and metabolism of propionate in the brain in vivo, primarily in PDH-deficient glia. The upregulation of genes encoding enzymes involved in propionate metabolism further corroborated these findings. In the Pdha1^hGFAP brain, EEAT1-isolated nuclei (astrocyte specific) exhibited increased transcript levels of Pcca, Pccb, and Mmut relative to NeuN-isolated nuclei (neuron specific) (Fig. S7C). In light of these results, propionate could represent a new potential therapeutic intervention for this condition.

The KD is the standard treatment for patients with PDHD, starting commonly after birth. Its application during gestation has been linked with anomalies in organ development, including neurodevelopmental deficits, in murine and rat models^32,49. Protracted gestation has been observed in humans on low carbohydrate diets⁶⁶. The standard therapeutic KD typically adheres to a 4:1 fat to carbohydrate ratio, with fats accounting for 80–90% of the caloric intake. To minimize the potential side effects during pregnancy, we administered a 2.5:1 ratio KD starting at preconception, where fats comprised approximately 70% of the caloric breakdown, and proteins and carbohydrates each contributed 15%. This composition was sufficient to induce ketosis in adult animals. The progeny count, birth weight, and postnatal growth of wild-type mice were commensurate with those on a regular diet, the reason we used this diet alone or in combination with propionate to test its therapeutic effects. In the early postnatal period, Pdha1^hGFAP mice subjected to KDP and KD exhibited an increase in weight gain compared to those on a regular diet, becoming statistically significant at postnatal days 15 and 20. Neuropathological assessments also showed improvements. Neuronal density in the cortex increased in the KD and 150 mM propionate groups, and all groups showed a thicker cerebellar granular cell layer relative to the regular diet group. Additionally, a reduction in reactive astrocyte density was observed across the brain regions examined in all experimental diets, except the KDP in the cerebellum. These findings suggest that both KD and sodium propionate may independently contribute neuroprotective and immunomodulatory benefits in the context of PDHD. While the experimental diets had an overall modest effect on motor deficits, a 20% reduction in the body’s sway of Pdha1^hGFAP mice was observed with the KDP diet relative to those on a chow diet. Those animals on a KD showed improvement on stride lengths and step widths, which were comparable to wild-type mice. No notable motor improvements were discerned in mice given 150 mM of sodium propionate with a regular diet. Together, these data indicate that propionate, when used in conjunction with the KD from pre-conception through pregnancy and continuing post-delivery, offers potential therapeutic advantages in the Pdha1^hGFAP mouse model.

Limitations of the study

Future investigations are necessary to establish the precise timing for the administration of propionate that would be most beneficial during gestation and in the postnatal period, along with determining the maximum dose that is tolerable. Additionally, it is crucial to ascertain both systemic and neurological potential side effects resulting from long-term exposure to propionate, whether in conjunction with a ketogenic diet or not, in Pdha1^hGFAP and control mice. Furthermore, the range of doses that could induce adverse effects must be established. Our findings indicate that propionate metabolism predominantly occurs in glial cells and it mitigates reactive astrocytosis. It is necessary to investigate the mechanisms through which propionate modulates inflammation in this specific context and in other neuroinflammatory disorders, apart from its effects when used alongside the ketogenic diet. An essential objective is to bridge the gap between these preclinical findings and clinical applications for patients.

STAR METHODS

Resource availability

Lead contact

Requests for resources and reagents should be directed to the lead contact, Isaac Marin-Valencia (isaac.marin-valencia@mssm.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

This paper reports original code to analyze mouse lateral sway, which is available at the following

GitHub repository: https://github.com/Neurometabolomics/Mouse-body-sway-analysis/blob/main/Mouse_behavioral_analysis_top_video_titubation.ipynb

Uncropped high-resolution scans of all the blots, data that have been used to create all graphs in the article, as well as supplementary tables and videos are provided in Data S1.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental model and study participants details

Mice

All animal experiments were approved by Icahn School of Medicine at Mount Sinai, Memorial

Sloan Kettering Cancer Center (MSKCC), and The Rockefeller University Institutional Animal Care and Use Committees and Institutional Review Boards. Mice used in all experiments were from P15 to P23. Brain-specific Pdha1^hGFAP mice were generated by breeding C57BL6/J Pdha1^flox/flox females with FVB/N hGFAP-Cre males. Genotyping of Pdha1 (forward primer 5’- AGCAGCCAGCACGGACTACT-3′ and reverse primer 5′-GCAGCCAAACAGATTACACC- 3′) and the Cre transgene (forward primer 5’-ATTTGCCTGCATTACCGGTC-3’ and reverse primer 5’- ATCAACGTTTTCTTTTCGG-3’) was carried out by PCR¹⁰.

Method details

Pyruvate dehydrogenase (PDH) activity

Total PDH complex activity was assayed using the PDH Activity Assay Kit from Biovision (cat# K679), with modifications. Briefly, cortex, hippocampus, and cerebellum (10–50 mg) from three control and Pdha1^hGFAP mice were quickly extracted and homogenized (10 strokes) in 100 μl ice cold PDH assay buffer with protease and phosphatase inhibitors (¼ minitablet of both complete protease inhibitor and phosphatase inhibitor in 1 ml of PDH assay buffer). Samples were left on ice for 10 min and then centrifuged at 10,000 g for 5 min at 4 °C. For colorimetric detection of NADH, we added 20 μL of each sample (1 μg/μL dilution) to a well of a 96-well plate and then added a PDH assay buffer to reach 50 μL. We then added 50 μL of the reaction mix to each well, incubated the plate at 37 °C for 2–3 min, and then read the absorbance at 450 nm in a BioTek Synergy NEO2 plate reader.

¹³C tracing analysis

[U-¹³C]glucose injections were carried out as previously described with some modifications¹⁰. Briefly, P23 (N=7/group) and P15 mice (N=6/group) fasted for 5 hours prior to tracer administration and remained awake during the experiment. [U-¹³C]glucose (Cambridge Isotope Laboratories, cat# CLM-1396) was injected (2 mg/g of weight) intraperitoneally to control and Pdha1^hGFAP mice. Cortex, hippocampus, cerebellum, and blood from each mouse were collected 30 min after the injection. [U-¹³C]propionate experiments were completed as previously reported¹¹. In this case, P23 animals fasted for 5 hours, and then injected 0.5 mg/g of the weight of [U-¹³C] sodium propionate (Cambridge Isotope Laboratories, cat# CLM-1865) intraperitoneally (Control: N=8, Pdha1^hGFAP: N=6). Cortex, hippocampus, cerebellum, and blood from each mouse were collected 30 min after each injection. [U-¹³C]glucose and unlabeled propionate were administered at 2 mg/g of weight total (~7 μmol/g of weight each), after 5 hours of fasting (Control: N=6, Pdha1^hGFAP: N=8). For each experiment, cortex, hippocampus, cerebellum, contralateral hemisphere, and blood from each mouse were collected 30 min after the injection. For all these experiments, mice were sedated right before decapitation with isoflurane (~20 sec) until they became unresponsive to painful stimuli. The brain was quickly extracted, and anatomical regions of interest were dissected in ~30–40 sec total, frozen in liquid nitrogen, and stored at −80°C. Blood was simultaneously collected from the cervical stump, and glucose concentration was measured using a clinical glucometer. For mass spectrometry analysis, ice-cold 80% methanol was added to each tissue at a volume of 10:1 (μL/mg), homogenized on ice, incubated for 4 hours at −80°C, and centrifuged at 14,000 g for 20 min at 4°C. The supernatant was transferred to a clean 1.5 ml tube, lyophilized on a SpeedVac with no heat, and then stored at −80°C until further processing for mass spectrometry analysis.

For NMR analysis, we performed the sample processing as we previously described⁵². Briefly, frozen brain or blood samples were finely grounded in a mortar under liquid nitrogen. Perchloric acid (4%; 1:4, w/v) was added to each sample, followed by centrifugation at 47,800 g for 15 min. The supernatant was transferred to a new tube where chloroform/tri-n-octylamine (78%/22%; v/v) was added in a 1:2 volumetric ratio to a pH of 6. The samples were centrifuged at 3300g for 15 min, and the aqueous phase was removed and transferred to a microfuge tube, and then lyophilized. 50 μL of deuterium oxide (99.96%, Cambridge Isotope Laboratories) was added to each sample, and the pH was adjusted to 7.0 with 2–3 μL of 1 M sodium deuteroxide (99.5%, Cambridge Isotope Laboratories). The pH-neutral samples were then centrifuged at 18,400g for 1 min, and the supernatant was removed and placed into a 1.7-mm NMR tube for subsequent NMR analysis.

Polar Metabolomics Profiling

Tracing analysis was performed using a gas chromatography-mass spectrometry in an Agilent 5977B GC/MSD or a Q Exactive Orbitrap mass spectrometer (Thermo Scientific) coupled to a Vanquish UPLC system (Thermo Scientific). For GC/MSD analysis, 20mg of brain or 20ul of plasma were extracted using 1 ml of water/methanol/acetonitrile mixture (1:2:2). Solvents were transferred into a GC vial and then dried with N₂ gas. The extracts were reconstituted with 20ul of MTBSTFA +1 % TBDMCS (N-metil-N-(tert-butildimethylsylil) trifluoroacetamide) and 40ul of acetonitrile for TCA cycle intermediates and 20ul MSTFA +1% TMCS (N-methyl-N-trimethylsylil) trifluoroacetamide and 40ul of acetonitrile for glucose. Extracts were heated for 45min at 75C to generate TBDMS or TMS derivatives respectively and detected in EI mode (70eV). Quantitation was performed based on MS signal intensities with Trycarballylic acid added as a surrogate. For LC-MS, the Q Exactive operated in polarity-switching mode. A Sequant ZIC-HILIC column (2.1 mm i.d. × 150 mm, Merck) was used for separation. The flow rate was set at 150 μL/min. Buffers consisted of 100% acetonitrile for mobile B, and 0.1% NH₄OH/20 mM CH₃COONH₄ in water for mobile A. Gradient ran from 85% to 30% B in 20 min followed by a wash with 30% B and re-equilibration at 85% B. For acyl-CoA analysis, the Q Exactive operated in positive ion mode. A Cadenza CD-C18 3 μm packing column (Imtakt, 2.1 mm id × 150 mm) was used for separation. The flow rate was set at 150 μL/min. Buffers consisted of 100% methanol for mobile B, and 5 mM CH₃COONH₄ in water for mobile A. Gradient ran from 2% to 15% B in 3 min, 15% to 95% B in 2.5 min, followed by a wash with 95% B and re-equilibration at 2% B. Compounds were identified on the basis of exact mass within 5 ppm and standard retention times. Quantitation was performed based on MS signal intensities. Natural isotope abundances were corrected using R statistics Isocorrector package⁵³.

Immunohistochemistry

Mice (P23: N=5/group on regular diet (Fig 1), N=2/group added for Fig 7D to control for new experimental conditions, P21-P23: N=5/group on KD, N=4/group on KDP, N=3 on 150 mM propionate, P15: N=5 controls, N=3 Pdha1^hGFAP mice) were anesthetized with Nembutal and perfused transcardially with cold PBS for approximately 3 min (until the liver became pale), followed by 4% paraformaldehyde (PFA) in 0.1 M PBS (pH 7.4). Brains were dissected and additionally fixed overnight in 4% PFA at 4°C. Brain tissues were then sequentially processed in 20% and 30% sucrose in PBS at 4°C, embedded in optimal cutting temperature compound (O.C.T., Sakura, cat#4583), and sliced into 50 μm sections in a cryostat. The staining was conducted in free-floating sections on a 24-well plate. Sections were first washed on PBS for 5 min and then permeabilized and blocked at 4°C overnight with PBS/0.3% Triton X-100/10% normal donkey serum (blocking solution). The following day, sections were incubated overnight at 4°C in the following primary antibodies: mouse anti-NeuN (1:200, Millipore-Sigma, cat#:MAB377), goat anti-GFP (1:500, Rockland, cat#: 600–101-215), chicken anti-GFAP (1:2000, Encor Biotechnology, cat#: CPCA-GFAP), rabbit anti-Glut1 (1:100, Millipore, cat#: 07–1401). On the third day, sections were washed at room temperature and then incubated at 4°C overnight in their respective secondary antibodies: Alexa flour 488-conjugated goat anti-chicken IgY, Alexa flour 488-conjugated donkey anti-goat IgG, Alexa flour 555-conjugated donkey anti-mouse IgG, Alexa flour 555-conjugated donkey anti-rabbit IgG (Thermo Scientific). The following day, sections were washed in blocking solution twice at room temperature, DAPI (1:5000, Thermo Scientific) was added for 10 min and then washed in PBS for 20 min before mounting with Prolong Gold antifade mountant (ThermoFisher, cat#: P10144). Images were obtained on a Zeiss inverted LSM 780 laser scanning confocal microscope and analyzed using ImageJ (NIH) and Imaris softwares.

To differentiate reactive astrocytes from non-reactive astrocytes, we used the cell size as a differentiating factor. Using ImageJ, we quantified the size of these cells (in μm²) in the cortex, hippocampus, and cerebellum of Pdha1^hGFAP mice on a regular chow diet and compared them to their counterparts in control brains (N= 5 mice/group; 5–25 cells/animal/anatomical region). Astrocytes from control brains were labeled as non-reactive. We used the OptimalCutpoints R package to calculate the cutoff size point that differentiates the two cell populations, such that an astrocyte with an equal or bigger size than the cutoff point was considered a reactive astrocyte, and vice versa. The defined sizes of reactive astrocytes in the cortex, hippocampus, and cerebellum of PDHA1^hGFAP mice on a standard chow diet served as the benchmark for assessing reactive astrocytosis in animals subjected to the experimental diets.

qPCR

Total RNA was extracted from brain tissue (N=3/group/age) using the RNeasy micro kit (Qiagen, cat.#: 74004), and cDNA synthesis was performed using the SuperScript IV First-Strand Synthesis System (ThermoFisher, cat.#: 18091050) according to manufacturer’s protocol. qPCR reactions were run in 96-well format as duplicates in a total volume of 10 μl, which consisted of LightCycler 480 SYBR Green I Master mix (Roche, cat.#: 04707516001), 300 nM each of gene-specific primers and 10 ng cDNA. The amplification conditions were as follows: pre-incubation for DNA polymerase activation at 95°C for 5 min, followed by 45 amplification cycles at 95°C for 10 sec and 60°C for 10 sec, and 72°C for 10 sec. At the end of the amplification cycles, a melting curve analysis was performed to verify specific amplification. One reference gene Actb, was used to normalize transcript expression in both animal groups. Data processing and analysis were completed using R. Reactive astrocyte primers were used as described previously¹⁵.

Cresyl violet staining

Mice (N=1/group) were anesthetized with Nembutal. The whole brain was carefully extracted and immersed in a small beaker with ice-cold isopentane (Fisher, cat#A451–1) for 15 min and then stored at −80 °C until embedding. Frozen brains were placed in a pre-chilled block sitting on dry ice and embedded in O.C.T. prior to cryostat sectioning. We obtained 20 μm sections and mounted them onto microscope glass slides. Sections were stained with 0.1% cresyl violet in sodium acetate to reveal cell nuclei. Images were obtained on an Olympus IX71 inverted microscope.

Tunel staining

Mice (N=3/group) were anesthetized with Nembutal. The brain was immersed in a small beaker with ice-cold isopentane (Fisher, cat#A451–1) for 15 min and then stored at −80 °C until further processing. Frozen brains were placed in a pre-chilled block sitting on dry ice and embedded in O.C.T. prior to sectioning. We obtained 8 μm sections and mounted them onto microscope glass slides. Sections were stained with In Situ Cell Death Detection Kit (Sigma, Cat# 11684795910), with some modifications from the company’s protocol. Briefly, sections were fixed with 4% PFA for 20 min, washed with PBS for 30 min, and then permeabilized with 0.3% Triton in PBS for 30 min. Sections were then washed on PBS twice for 3 min each, then incubated for one hour at 37 °C in the dark on a Tunel reaction mixture. Sections were then washed once in PBS for 3 min, incubated in DAPI (1:5000) for 5 min, washed again in PBS for 3 min, and then mounted in Prolong Gold Antifade mountant. Images were obtained on a RS-G4 resonant scanning confocal microscope, and the analysis was performed in ImageJ.

Western Blot

Cortex, hippocampus, and cerebellum (N=5/group at P23, N=3/group at P15) were dissected, immersed in liquid nitrogen, and kept at −80°C. Cytoplasmic proteins were extracted using a home-made lysis buffer containing protease inhibitor cocktail (Roche) and homogenized under ice with an ultrasonic processor. Membrane proteins were extracted using the solubilization buffer from the Mem-PER Plus Membrane Extraction Kit (Pierce, cat#: 89842) containing protease inhibitor cocktail and homogenized in a Dounce tissue grinder (6–10 strokes). Afterwards, samples were incubated for 30 min at 4°C with constant mixing. Samples were then centrifuged at 16,000 g for 15 min at 4°C. Protein content was quantified by BCA assay (Bio-Rad). Approximately 15 μg of total protein per sample was separated in a home-made 10% gradient SDS-PAGE gels and transferred to PDVF membranes. The membranes were blocked with 5% milk in Tris-buffered saline, 0.1% Tween-20 (TBST) for 1 h and then incubated overnight with primary antibody to PDHA1 (rabbit monoclonal, 1:3000, Abcam), GLUT1 (rabbit polyclonal, 1:10000, Millipore), GLUT3 (rabbit polyclonal, 1:1000, Thermo), MCT1 (rabbit polyclonal, 1:100, Millipore). They were then washed three times with TBST, followed by incubation of goat rabbit HRP-conjugated secondary antibody (1:5000, Santa Cruz Biotechnology) for 1 h. Membranes were washed again three times in TBST and treated with Pierce ECL Western Blotting Substrate (Thermo Scientific) prior to film development in a dark room. Blots were stripped using OneMinutePlus Western Blot Stripping Buffer (GM Biosciences) and reprobed with β-actin (1:5000, Sigma-Aldrich) as a loading control. Protein content was determined by band densitometry and normalized to β-actin using ImageJ.

FDG-PET

Six male PDHD and 8 control mice aged P23–25 were fasted 3–5 hours prior to PET imaging. After injection of ~100 μL (~300 μCi) of [¹⁸F]fluordeoxyglucose (FDG) via the lateral tail vein, mice were immediately anesthetized using 0.5 to 1% isoflurane at a flow rate of 1–1.5 L/min of medical air as the carrier gas. PET/CT images were acquired using a small-animal Inveon microPET/microCT (Siemens Healthcare GmbH) at MSKCC. Five and 10-min static CT and PET images, respectively, were acquired one hour after the FDG injection. The counting rates in the reconstructed images were converted to activity concentrations (percentage injected dose per gram of tissue, %ID/g) and then to standardized uptake value (SUV) using a system calibration factor and animals’ weight for normalization. Quantification of brain FDG-PET signal intensity was carried out using VivoQuant software (version 3.0; InviCRO), as previously described⁵⁴. In addition, eight Pdha1^hGFAP and control mice were imaged at P15 and then at P22-P23 to evaluate changes in glucose uptake into the brain with disease progression. The FDG was injected intraperitoneally to facilitate its administration at the younger age. The dose, timing, anesthesia, and imaging protocol were the same as the one used for P23 mice.

Autoradiography

Ex-vivo ¹⁴C-2DG brain autoradiography was performed as described previously⁵⁵. In brief, a dose of 0.4 μCi/g of body weight was injected intraperitoneally into each animal (controls = 4, Pdha1^hGFAP = 4), following a 3–5 hour fasting period. One hour after the injection, mice were sedated via inhalation of 0.5 to 1% isoflurane and then intracardially perfused with 0.9% of saline to wash out the blood prior to brain extraction. The brain was dissected out, embedded in O.C.T., and frozen in dry ice. Coronal brain sections of 20 μm thickness were collected and affixed to positively charged slides and air-dried for 30 min. Sections were exposed to digital autoradiography films for 3 weeks at −20°C, and each film was scanned using a Typhoon phosphor imager (Amersham). Sections were then stained with cresyl violet following the protocol outlined earlier to match the radioactivity signal with the anatomical location. ImageJ was used to visualize and analyze images.

MRI and ¹H-MRS

Mice (N=8/group) were scanned on a 400-MHz Bruker 9.4 T Biospec MRI scanner (Bruker Biospin) equipped with a Bruker 530 mT/m 114-mm (ID) gradient at the Small Animal Imaging Core at MSKCC. Mice were anesthetized with 0.5–1 % isoflurane in medical air. Sedated animals were monitored for temperature, and respiratory rate during scanning (SA Instruments), and anesthesia was adjusted accordingly. For anatomical imaging, axial T₂-weighted brain images were acquired using the fast spin–echo RARE sequence (rapid acquisition with relaxation enhancement) with TR 2.5 s, TE 11 ms, RARE factor 8, slice thickness 0.5 mm, field of view 25 × 20 mm, in-plane resolution 98 × 78 μm, and 4 averages. For ¹H MRS, the acquisition was obtained using a voxel size of 1.5 × 1.5 × 1.5 mm³ for cortex, hippocampus, and cerebellum at TR: 2 s, 256 averages, resolution enhancement with shifted Gauss ltering, TE: 16.5 ms, shift: 7%, broadening: 5 Hz. Using 10 Hz lb, the baseline spectra from each region of interest were corrected and normalized to the creatine peak (all performed in MestreNova). All spectra were binned from 4.49 – 0 ppm, bin size was set to 0.0005 ppm, and the integrals from binned data were taken using MATLAB (The Mathworks Inc., Natick, MA). A Welch t-test with Benjamini-HochBerg correction on these integrals was performed, along with a principal component analysis (PCA) using univariate-scaled data, with a R² of 92.7% and 10 components. R packages FactoExtra, ggplot2, ggpubr, and rstatix were used for PCA and Welch tests analysis and to perform all MRS data visualization.

NMR spectroscopy acquisition and analysis

Proton decoupled ¹³C spectra were acquired on an 800 MHz Bruker magnet using a 5 mm TXO cryoprobe optimized for ¹³C detection (Bruker Corporation, Billerica, MA). Proton decoupling was performed at 2.3 kHz using a Waltz-16 sequence. ¹³C NMR spectroscopy parameters included a 30° flip angle per transient, a relaxation delay of 1.5 s, an acquisition time of 1.5 s, and a spectral width of 36.7 kHz at 25 °C. A heteronuclear 2H lock was used to compensate for magnet drift during data acquisition. To achieve adequate signal-to-noise in brain spectra, the number of scans was typically 10,000–12,000.

NMR spectral analyses were performed with ACD/Labs Spectrus Processor 2021 software (Advanced Chemistry Development, Inc., Toronto, ON, Canada) as previously described⁴. Time-series free induction decays were zero-filled and windowed with an exponential weighting function prior to Fourier transformation to analyze spectral contents. Metabolite peaks were then identified based on the chemical shift position referenced to the glutamate C4 singlet at 34.2 ppm. Each peak was fitted with a Gauss–Lorentz function, and the area measurements for each fitted resonance peak and their multiplets were estimated. For each isotopomer, multiplet areas were defined as a fraction of the total atomic resonance area, here called multiplet fractional amount.

¹³C-labeling simulations

To predict the ¹³C labeling pattern of glutamate C2, C3, and aspartate C3 NMR spectra, as well as their corresponding isotopologue analyses in the context of different enrichments of ¹³C propionate, glucose, and anaplerotic rates through succinyl-CoA with constant rates at PDH and PC enzymes, several simulation studies were carried out using the open-source software tcaSIM⁵⁶. The simulations were performed in a single-cell compartment as previously described⁵².

¹³C Hyperpolarized MRSI

Spectroscopic imaging was performed on a small animal imaging system (Bruker Biospec 3T, Bruker Corporation, Billerica, MA) on a dual-tuned ¹H-¹³C coil. Anatomic ¹H imaging was performed at high-resolution (0.1 × 0.1 × 1 mm) over the whole mouse brain. Dynamic nuclear polarization (DNP) was performed by a SpinLab Hyperpolarizer (GE Inc) on a 100ul sample of 14.2 M [1-¹³C] pyruvic acid containing 15 mM trityl radical (General Electric), as we have done previously for mouse brain imaging⁵⁷. Following hyperpolarization the sample was rapidly dissolved into the buffer (40 mM Tris at pH 7.4, 1 mM EDTA), and using NaOH (10 N), the final pH was adjusted to 7.0. Subsequently, approximately 150 mM of HP [1-¹³C]-pyruvate in approximately 250 uL of volume was injected through the tail vein under anesthesia (0.5–1 % isoflurane).

Time-resolved, slice-selective, 2D EPSI ¹³C imaging was modified from the standard Bruker library sequence by allowing for 8 phase encodes; modifications in timing delays to allow completely phaseable absorptive spectra were not made (Miloushev et al., 2017). The following parameters were set: coronal in-plane resolution 4 × 4 mm, with direct (read) dimension-oriented AP/SI (FOV 64 mm) and indirect (phase encode) dimension-oriented RL (FOV 32 mm); center out alternating phase encode ordering (0,−1,1,−2,2,−3,3,−4); coronal slice thickness 7 mm; EPSI spectral bandwidth 3125 Hz and acquisition time 80 ms; time resolution (TR) 800ms and dynamic acquisition length 1 min (75 reps); flip angle 5 degrees; ¹³C carrier frequency 170 ppm.

Data processing was performed using standard MATLAB (Mathworks, Inc.) functions. Alternating lines in EPSI dimension k-space were time-reversed and subsequently averaged, folding the spectral dimension and averaging artifacts. The EPSI spectral dimension was zero-padded prior to Fourier transformation, and all dimensions were then zero-padded to twice the size. Exponential apodization (line-broadening = 2.5 Hz) was used for the EPSI spectral dimension. Baseline offset correction was performed using a histogram method (the baseline offset was equated to the most frequent value). Spectral referencing and first-order phasing were performed on the pyruvate resonance. Metabolite maps were made from 1 ppm spectral integrals of each resonance (abs intensity mode).

The matrix pencil (MP) method (10 components) was used to fit mean spectra from the brain regions of interest. This method demonstrated robust performance in fitting our spectra, which resulted in inverted lactate and dispersive phase bicarbonate resonances because of acquisition delays. While the fitting easily quantified high amplitude signals (i.e. pyruvate, lactate), a different strategy was used for quantifying the smaller bicarbonate resonance (essentially absent from the Pdha1 animals) and quantifying the bicarbonate:lactate ratio. First, two sets of residual pseudo-spectra were made from the MP fits, in one case excluding the region of the lactate resonance (2 ppm) and the other excluding the region of the bicarbonate resonance (this was done to accurately account for the dominant effect of the pyruvate signal and to lesser extent lactate signals on the dispersive phase small amplitude bicarbonate resonance which cannot be accurately integrated from real spectra phased on pyruvate where the integral is null or unfiltered imaginary spectra where the integral is dominated by the dispersive tail from the pyruvate resonance). Second, 2 ppm wide integrals over the bicarbonate and lactate regions were taken from the respective residual pseudo-spectra (the bicarbonate integral was performed on the imaginary spectra since the resonance was dispersive phase relative to pyruvate and the lactate integral was performed on the negative real spectrum). The ratio was then taken of the respective integrals (physically unrealistic opposite sign integrals for bicarbonate were thresholded to zero). The t-test was used for statistical significance testing between control and Pdha1^hGFAP groups (Control: N=5, Pdha1^hGFAP: N=4). Statistical analysis was performed using R. Figures were prepared with MATLAB, R, and Adobe Illustrator (Adobe Inc.).

Mouse behavior analysis

Motor function was quantitatively assessed using an open field platform conducted within a transparent acrylic cage measuring 10×10×10 inches. These assessments were captured using video cameras strategically positioned to record both ventral and dorsal perspectives. Concurrent recordings, each lasting three minutes, were obtained for individual mice. The camera for the top view was a HERO8 Black (GoPro Inc.) camera configured with the following settings: a resolution of 1080p, frame rate of 60 frames per second (FPS), a linear field of view, and Protune disabled. For bottom recordings, a HERO7 Black (GoPro Inc.) was used, with settings adjusted to a resolution of 1440p, standard field of view, a frame rate of 60 FPS, a narrow lens, and standard Protune settings. These configurations were meticulously chosen to encompass the entire open field space without distorting the captured images.

Tracking of specific body parts was performed using the DeepLabCut software (version 2.3.5) as described by Mathis et al.⁵⁸. For the analysis of the ventral recordings, 540 frames sourced from 18 videos of wild-type mice at age P15 and P20 were manually annotated, with 95% of these frames used for the training phase. A neural network based on ResNet-50 architecture, as proposed by Insafutdinov et al.⁵⁹ and He et al.⁶⁰, was employed with the standard parameters over 500,000 iterations for training. Validation was executed with a single shuffle, revealing a test error of 3.47 pixels and a training error of 2.34 pixels, with the image size normalized to a resolution of 1148×1080 pixels. For subsequent analyses, the X and Y coordinates were conditioned on a probability cutoff of 0.95.

Regarding the analysis of the top videos, a total of 390 frames extracted from 13 videos of wild-type mice aged P15 and P20 underwent labeling, with 95% designated for network training. Employing the same ResNet-50-based framework and default parameters, the network underwent a half-million training iterations. Post-analysis of four novel videos, outliers were identified, corrected manually, and the network was retrained for an additional 350,000 iterations. Validation involving two shuffles yielded a test error of 2.98 pixels and a training error of 2.59 pixels, after image downscaling by a factor of 1.5. Analogous to the dorsal analysis, a probability cutoff of 0.95 was applied to refine the X and Y coordinates data for further examination.

With this setting, we recorded the spontaneous behavior of control and Pdha1^hGFAP mice when on the various experimental diets, and these recordings were subsequently analyzed using the trained neural network. Specifically, on the standard chow diet, the behavior of nine control and nine Pdha1^hGFAP mice was monitored. Similarly, while on the KD, we recorded another set of nine control and nine Pdha1^hGFAP mice. For the KDP group, recordings were made for ten control mice and eight Pdha1^hGFAP mice. Lastly, on the diet with propionate supplementation, the behavior of nine control and nine Pdha1^hGFAP mice was captured.

The analysis of the bottom videos was carried out using the R package DLCAnalyzeR (https://github.com/ETHZ-INS/DLCAnalyzer/tree/master). Raw cvs files generated from DeepLabCut were processed using the DLCAnalyzeR package to quantify mouse body area, longitudinal length, stride length, step width, speed, distance between paws, trajectories, and density paths. We also quantified the animals’ body lateral sway during walking from the top videos using a Python code (https://github.com/Neurometabolomics/Mouse-behavior-analysis). DeepLabCut data and corresponding video footage from the top camera were analyzed using this computational sequence:

• Long axis formulation: first, the longitudinal axis of the rodent was ascertained by computing the vector difference between the coordinates of the tail base and the neck, mathematically represented as:

  $$\text{long}\_\text{axis}=\text{tailbase}‒\text{neck}.$$

• Perpendicular vector calculation: subsequently, a vector perpendicular to the long axis was derived. For a vector delineated as (a, b), the orthogonal vector was defined as (−b, a).

• Normalization of the perpendicular vector: The perpendicular vector was then normalized to unit length to ensure uniformity in scale. This was accomplished by dividing the perpendicular vector by its own norm. The norm of a vector is a measure of its magnitude in multidimensional space, calculated as the square root of the sum of the squares of its components (see the code in Github for further details).

• Neck to center of mass (CoM) vector: The center of mass (CoM) represents the point at which the animal’s mass is concentrated. We used the neck, trunk, and tailbase body points to localize and calculate the CoM. The CoM along the x-axis was determined by averaging the x-coordinates of the specified body parts, and similarly, the CoM along the y-axis was calculated by averaging the y-coordinates as described in the code. We then used the CoM as a reference point to calculate the lateral body displacement. The vector indicative of the CoM lateral movement relative to the neck point was computed by:

  $$\text{neck}\_\text{to}\_\text{CoM}=\text{CoM}-\text{neck}.$$

• Projection of lateral movements: The degree of lateral sway of the CoM with respect to the animal’s long axis was quantified by projecting the neck-to-CoM vector onto the normalized perpendicular vector. The magnitude of lateral sway was ascertained by the dot product of these vectors, indicating the lateral deviation at a specific frame.

• Quantification of lateral sway: The projection’s magnitude was calculated for each video frame, yielding a scalar value in pixels that encapsulated the extent of lateral movement (see the code for further details).

Propionate titration assessment

We conducted a dose-titration experiment with sodium propionate using four-week-old, C57BL/6, wild-type mice (N=4/condition). The primary objective was to observe survival rates and patterns in weight gain at three-day intervals for a duration of two weeks. All animals were maintained on a standard chow diet. We started with a sodium propionate concentration of 500 mM, as reported by Sumbria et al.³⁰, and subsequently administered 300 mM and 150 mM dosages to ascertain the maximum tolerable concentration.

Upon determining that the 150 mM concentration of sodium propionate was the most suitable for administration with a chow diet, we proceeded to evaluate its impact on survival and weight gain in a separate group of mice placed on a KD. The ketogenic formulation was defined by the following caloric composition: 15% from protein, 15% from carbohydrates, and a 70% from fat. Comparative analysis was drawn against a control group fed a regular chow diet (RD).

An additional phase of the analysis was to assess the influence of 150 mM sodium propionate in combination with a chow diet or KD on offspring development during pregnancy and the postnatal period. For this purpose, five breeding cages were set up. We quantified the size, weight, and viability of the progeny from birth to P40, at three-day intervals. Throughout the experimental timeline, both special diets and the sodium propionate-enriched water were renewed on a weekly basis. Given the detrimental effects of the combination of 150 mM propionate with the KD on pregnant females and their progeny, an additional group of 50 mM sodium propionate and KD was set up with five breeding cages.

Propionate, b-hydroxybutyrate, glucose measurements in plasma

Blood samples were collected from a cohort of six P15 mice, each subjected to one of four dietary regimens: a standard chow diet, a ketogenic diet, a ketogenic diet supplemented with 50 mM of sodium propionate on the drinking water, or a chow diet with 150 mM of sodium propionate on the water. The procedure involved extracting blood from the distal end of the tail. This sampling method was specifically designed to measure glucose levels efficiently using the Contour Next EZ meter (Ascensia Diabetes Care Inc., NJ, USA). Simultaneously, β-hydroxybutyrate concentrations were determined using the Precision Xtra meter (Abbott, OH, USA). For the determination of propionate levels, a further volume of blood was drawn into heparin-treated tubes. The concentrations were quantified via gas chromatography-mass spectrometry, using an Agilent 5977B GC/MSD system. The process entailed mixing 40 μl of plasma with 600 μl of water, acidification with 80 μl of 5N HCl, followed by an extraction phase with 600 μl of diethyl ether. To the extracted solution, 50 μl of 1% pentafluorobenzylbromide (PFBBr) and 50 μl of 1% diisopropylethylamine were added, and the mixture was incubated for 30 minutes at ambient temperature to synthesize short-chain fatty acid-PFB (SCFA-PFB) derivatives.

Post-derivation, the solvents were evaporated using a Savant^™ SpeedVac^™ Vacuum Concentrator SPD140DDA-115 (Thermo Fisher Scientific Inc.), and the dry extracts were then re-dissolved in 60 μl of hexane. A sample volume of 1 μl was injected into the GCMS for analysis. The SCFA-PFB derivatives were identified using Electron Capture Negative Ionization, with methane as the moderating gas. SCFA concentrations were deduced from internal standard calibration curves, which were established using d7-isobutyric acid as an internal standard surrogate in the samples.

Lactate concentration in brain

The brains of P21 control and Pdha1^hGFAP mice (N=7/group) were harvested and snap-freezed as described above for isotope tracing experiments. The quantification of lactate levels was performed using an Agilent 5977B GC/MSD system. In brief, a 20 mg sample of whole brain tissue was homogenized in a 1 ml solvent mixture of water, methanol, and acetonitrile in a ratio of 1:2:2. The resulting solution was then transferred to a GC vial and dried using a Savant^™ SpeedVac^™ Vacuum Concentrator SPD140DDA-115 (Thermo Fisher Scientific Inc.). The dried extracts were reconstituted with 20 μL of N-methyl-N-(tert-butyldimethylsilyl) trifluoroacetamide (MTBSTFA) with 1% tert-butyldimethylchlorosilane (TBDMCS) and 40 μl of acetonitrile. The samples were subsequently heated for 45 minutes at 75°C to facilitate the formation of tert-butyldimethylsilyl (TBDMS) derivatives, which were detected in Electron Ionization (EI) mode at 70 eV. Quantitative analysis was executed using MassHunter software (Agilent Technologies, Inc.), referencing the mass spectrometric signal intensities against an eight-point calibration curve, which exhibited a coefficient of determination (R²) exceeding 0.998. Tricarballylic acid was incorporated as an internal standard surrogate.

Nuclei isolation, nuclei labeling and sorting, RNA purification, and cDNA amplification

The protocol for nucleus isolation was refined from the method established by Mätlik et al.⁶¹ and Pressl et al⁶², incorporating several modifications. Briefly, brains from P21 control and Pdha1^hGFAP mice were harvested, snap-frozen, and stored at −80°C for later processing. The tissue was disrupted using a homogenization buffer composed of 0.25 M sucrose, 150 mM KCl, 5 mM MgCl2, 20 mM Tricine-KOH (pH 7.8), 0.15 mM spermine, 0.5 mM spermidine, and supplemented with an EDTA-free protease inhibitor cocktailand 1 mM DTT) This homogenization was executed through 10 strokes with a loose (A) pestle, followed by an additional 10 passes with a tight (B) pestle in a glass Dounce homogenizer, all performed on ice. The homogenate was then combined with 92% of an iodixanol solution (50% iodixanol/optiprep, 150 mM KCl, 5 mM MgCl2, 20 mM Tricine at pH 7.8, 0.15 mM spermine, 0.5 mM spermidine, EDTA-free protease inhibitor cocktail and 1 mM DTT) layered on top of a 27% iodixanol cushion and centrifuged at 17,200 g for 30 minutes at 4°C to pellet the nuclei.

The resulting nuclear pellet was re-suspended in homogenization buffer, filtered through a cell strainer, and then fixed with 3% glyoxal solution at room temperature for 5 min. The fixed nuclei were then centrifuged at 1000 g for 5 min at 4°C, washed sequentially with homogenization buffer and then with a wash buffer containing 50 ng/m RNase-free BSA (Thermo Scientific, cat#: AM2616), 0.05% TritonX-100, 1 mM DTT, 10 U/ml Superase-In (Thermo Fisher cat#: AM2696) and 20 U/ml Rnasin (Promega, cat#: N2515) RNase Inhibitors in PBS. For immunolabeling, the nuclei were blocked on ice in a blocking buffer containing 100 ng/ml RNase-free BSA 0.05% Triton X-100 in PBS for 30 mins. They were then incubated on ice for 1 hour with primary antibodies against NeuN (1:400 dilution, Millipore cat#: ABN91) and Eaat1 (1:100 dilution, Santa Cruz Biotechnology, cat#: sc-515839). Following primary antibody incubation, nuclei were washed and incubated with Alexa fluor 488 (1:250 dilution) and Alexa fluor 647 (1:400 dilution) conjugated secondary antibodies, then subjected to additional 3 washes with wash buffer containing 0.05% Triton X-100.

For flow cytometry-based nuclei sorting, the labeled nuclei were suspended in a Blocking buffer without Triton X-100 and stained with DAPI at 2 μg/ml for 15 minutes on ice, protected from light, and sorted at Mount Sinai’s Flow Cytometry CoRE facility. Sorting was performed with a BD FACSAria^™ II cell sorter, employing 100 μm nozzle. The gating strategy first identified singlet nuclei that were positive for DAPI, followed by the sorting of the NeuN⁺ (labeled with an Alexa Fluor 647 dye) and Eaat1⁺ (labeled with Alexa Fluor 488 dye) population in two different Eppendorf tubes containing RLT lysis buffer supplemented with beta mercaptoethanol (Qiagen). Post-sorting, the lysed nuclei in the RLT buffer were stored at −80°C for later RNA extraction and qPCR analyses.

Granule cells and glia isolation from cerebellum

Cerebellar granule cells (GC) and glia were isolated using the Hatten protocol⁶³. In brief, the cerebella from one control and one Pdha1^hGFAP mouse were dissected within a chilled, calcium- and magnesium-free phosphate-buffered saline (CMF-PBS) solution. The tissues were then subjected to enzymatic dissociation using trypsin and DNase I for a duration of 5 minutes at 37°C. This was followed by a centrifugation step at 700 g for 5 minutes at a temperature of 4°C. After the removal of the trypsin-DNase I solution, the tissues underwent a gentle trituration process in DNase supplemented CMF-PBS. This involved repeated pipetting: 10 times trituration using a transfer pipette, followed by 10 passages through a finely tapered fire-polished glass Pasteur pipette, and a final series of 10 pipetting with an extra-fine glass Pasteur pipette. The triturated homogenate was then centrifuged once more at 700 g for 5 minutes at 4°C. The resulting cell pellet was reconstituted in DNase-CMF-PBS solution.

For the enrichment of GC and glia, a Percoll gradient sedimentation was performed. Glial cells were collected from the top of the gradient, whereas GC were collected from the interphase between the 35% and 60% Percoll solutions. Both cell suspensions underwent a preplating protocol: initially for a brief period of 15 to 30 minutes on a standard Petri dish, and subsequently for 1 to 2 hours on a tissue culture dish. After preplating, the respective media containing GC and glia were collected, centrifuged at 700 g for 5 min at 4°C to obtain the cell pellet, and then further processed for protein extraction as indicated in the western blot section.

Quantification and statistical analysis

R programming language was used for statistical analyses, including independent Student’s t-test assuming or not equal variances based on the F-test. For multi-comparison analysis against a single group, for example regular diet group with other dietary groups, the pairwise t-test, or Mann-Whitney U test in case the data was not normally distributed, was used followed by Benjamini-HochBerg testing. For the analysis of three or more groups with normally distributed data, a One-way ANOVA test was carried out with Tukey post-hoc correction. For the analysis of three or more groups with not normally distributed variables, a Kruskal-Wallis statistical test was used along with Dunn post-hoc testing. To compare the survival distributions of Pdha1^hGFAP mice between diets, we used the log-rank test followed by the Benjamini-HochBerg post hoc test. For correlation analysis between numeric variables, the Pearson correlation test was used. The value of alpha (significance level) was set at 0.05, and all tests of significance were two-sided. For hierarchical clustering, all values were scaled, the distance matrix was built using the Euclidean method based on the continuous numerical values of all variables, and clustering was performed using the average linkage method. All data are expressed as means ± standard error of the mean (SEM).

Supplementary Material

1

Data S1: Additional Supplemental Tables, related to Figures 5, 6, 7, and S10.

Table S1. Concentration (mg/dl) and ¹³C enrichment of glucose in the blood from [U-¹³C]glucose experiments at P15 and P23, [U-¹³C]glucose and unlabeled propionate at P23, and [U-¹³C]propionate at P23, related to Figures 5 and 6.

Table S2. Complete isotopologue analysis from [U-¹³C]glucose injections at P23 of metabolites detected in blood, cortex, hippocampus, and cerebellum in control and Pdha1^hGFAP mice, related to Figure 5.

Table S3. Complete isotopologue analysis from [U-¹³C]glucose injections at P15 of metabolites detected in blood, cortex, hippocampus, and cerebellum in control and Pdha1^hGFAP mice, related to Figure 5.

Table S4. Complete isotopologue analysis from [U-¹³C]propionate injections of metabolites detected in blood and cortex, hippocampus and cerebellum in control and Pdha1^hGFAP mice, related to Figure 6.

Table S5. Analysis of most abundant CoA isotopologues derived from [U-¹³C]propionate metabolism in the brain in control and Pdha1^hGFAP mice at P23, related to Figure 6.

Table S6. Complete isotopologue analysis from [U-¹³C]glucose and unlabeled propionate injections of metabolites detected in blood and cortex, hippocampus, and cerebellum in control and Pdha1^hGFAP mice at P23, related to Figure S10.

Table S7. Complete isotopomer analysis from [U-¹³C]propionate of metabolites detected in the brain in control and Pdha1^hGFAP mice at P23, related to Figure 6.

3

Video S1. Analysis of the lateral sway of the center of mass during walking in a P20, wild-type mouse within an open field platform, related to Figure 7.

4

Video S2. Analysis of the lateral sway of the center of mass in a P20, Pdha1^hGFAP mouse on a chow diet, related to Figure 7.

5

Video S3. Analysis of the lateral sway of the center of mass in a P20, Pdha1^hGFAP mouse on a ketogenic diet, related to Figure 7.

6

Video S4. Analysis of the lateral sway of the center of mass in a P20, Pdha1^hGFAP mouse on a ketogenic diet and 50 mM of sodium propionate, related to Figure 7.

7

Video S5. Analysis of the lateral sway of the center of mass in a P20, Pdha1^hGFAP mouse on 150 mM of sodium propionate with a chow diet, related to Figure 7.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
PDHA1	Abcam	Cat# ab168379, not in RRID
GFAP	Encor Biotechnology Inc	Cat# CPCA-GFAP, not in RRID
GLUT1	Millipore	Cat# 07–1401 RRID:AB_1587074
GLUT3	Thermo Fisher	Cat# OSG00012W, RRID:AB_10979336
MCT1	Millipore	Cat# AB3538P, RRID:AB_11210679
NeuN (IF)	Millipore	Cat# MAB377, RRID:AB_2298772
NeuN (single nuclei isolation)	Millipore	Cat#ABN91, RRID:AB_11205760
EAAT1	Santa Cruz Biotechnology	Cat# SC-515839, not in RRID
GFP	Rockland	Cat# 600–101-215, RRID:AB_218182
β-Actin	Sigma Aldrich	Cat# A5441–100UL, not in RRID
Donkey anti-Chicken IgY (H+L) Alexa Fluor 647	Thermo Fisher	Cat# A78952, RRID:AB_2921074
Donkey anti-Mouse IgG (H+L) Alexa Fluor 488	Thermo Fisher	Cat# A21202, RRID:AB_141607
Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 488	Thermo Fisher	Cat# A21206, RRID:AB_2535792
Donkey anti-Mouse IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 555	Thermo Fisher	Cat# A31570, RRID:AB_2536180
Alexa Fluor 488 AffiniPure Donkey Anti-Chicken IgY (IgG) (H+L)	Jackson ImmunoResearch Inc	Cat# 703–545-155, not in RRID
Sheep Anti-Mouse IgG (H+L)-HRP conjugate	Jackson ImmunoResearch Inc	Cat# 515–035-062, not in RRID
Goat Anti-Rabbit IgG (H + L)-HRP Conjugate	Bio-Rad Laboratories	Cat# 1706515, not in RRID
Chemicals, peptides, and recombinant proteins
Amino Acid Standard H	Thermo Fisher	Cat# PI20088
Acetone	Fisher Scientific	Cat# A18S-4
DL-B-Hydroxybutiric acid sodium salt	Sigma Aldrich	Cat# H6501–5G
Hydrochloric Acid	Fisher Scientific	Cat# A144S-500
Hexanes	Fisher Scientific	Cat# H303SK-4
Acetonitrile	Fisher Scientific	Cat# A996–4
Diethyl ether	Fisher Scientific	Cat# AC615080040
Water, Optima	Fisher Scientific	Cat# W64
D-Glucose (U-13C, 99%)	Cambridge Isotope Laboratories	Cat# CLM-1396-PK
Heparin sodium salt	Sigma Aldrich	Cat# H3149–25KU
Isobutyl chloroformate	Spectrum Chemical	Cat# I2018–25GM
Tricarballylic acid	Acros Organics	Cat# 139360500
Sodium Propionate	Sigma-Aldrich	Cat# P1880
Propionic acid	EMD Millipore	Cat# 800605
2,3,4,5,6-Pentafluorobenzylbromide	EMD Millipore	Cat# 841643
MSTFA + 1% TMCS	Thermo Fisher	Cat# TS-48915
Pentafluorobenzyl Bromide	TCI Chemicals	Cat# P0809–1G
Maleic acid	Sigma-Aldrich	Cat# M0375–500G
Sodium pyruvate	Sigma-Aldrich	Cat# P2256–5G
Ammonium Hydroxide	Fisher Scientific	Cat# A669S-500
Amino Acids Mix Solution	Sigma-Aldrich	Cat# 79248–5X2ML
Butyric Acid-d7	Cayman Chemical Company	Cat# 29408
D-(−)-3-Phosphoglyceric acid disodium salt	Sigma-Aldrich	Cat# P8877–10MG
Hexanoic Acid-d11	Cayman Chemical Company	Cat# 28082
(±)-β-Hydroxybutyrate-d4	Cayman Chemical Company	Cat# 14158
N,N-Diisopropylethylamine	TCI Chemicals	Cat# D1599–25ML
Short-chain Fatty Acid Mixture 3	Cayman Chemical Company	Cat# 28682
(±)-β-Hydroxybutyrate	BioVision	Cat# B2829–100
Tricarballylic acid	Sigma-Aldrich	Cat# T53503–25G
L-(+)-Lactic acid	Sigma-Aldrich	Cat# L6402–1G
Sodium citrate	Sigma-Aldrich	Cat# 1613859–1G
Succinic Acid	Sigma-Aldrich	Cat# PHR1418–1G
α-Ketoglutaric acid	Sigma-Aldrich	Cat# 61234–10MG
D-(+)-Glucose	Sigma-Aldrich	Cat# 47249
γ-Aminobutyric acid	Sigma-Aldrich	Cat# 03835–250MG
4-Aminobutyric acid-2,2-d2	Sigma-Aldrich	Cat# 617458–250MG
Glutaric acid	Sigma-Aldrich	Cat# 89147–100MG
MTBSTFA (with 1% t-BDMCS)	Sigma-Aldrich	Cat# M-108–5X1ML
Phenol red solution	Sigma Aldrich	Cat# P0290–100ML
Basal Medium Eagle	Sigma Aldrich	Cat# B1522–500ml
DMEM	Thermo Fisher	Cat# A1443001
Bovine Serum Albumin	Sigma Aldrich	Cat# A9418–50G
Perchloric acid	Sigma Aldrich	Cat# 311421
MOD TEST DIET 5TGE W/15% PROTEIN 15% CARBOHYDRATES 70% FAT	Test diet	Cat# 1815470–186
SUPERaseIn RNase Inhibitor	Thermo Fisher	Cat# AM2694
Invitrogen™ ProLong™ Gold Antifade Mountant	Fisher Scientific	Cat# P36934
Fisherbrand superfrost Plus Stain Slides	Fisher Scientific	Cat# 22–034979
Xylenes	Sigma Aldrich	Cat# CAS 1330–20-7
SUCROSE CRYSTAL CERT	Fisher Scientific	Cat# S5–3
Boric Acid	Fisher Scientific	Cat# A73–500
16% Paraformaldehyde	Electron Microscopy Sciences	Cat# 15710
Cresyl violet acetate	Sigma Aldrich	Cat# C5042
Hoechst	Thermo Fisher	Cat# H3570
Eukitt quick-hardening mounting medium	Sigma Aldrich	Cat# 03989–100ML
Triton X-100 (100 ml)	Sigma Aldrich	Cat# T8787–100ML
TWEEN 20	Sigma Aldrich	Cat# P1379–500ML
Tris-HCl	Thermo Fisher	Cat# 15568025
Bovine Serum Albumin	Sigma Aldrich	Cat# 001–000-161
EDTA (0.5 M)	Fisher Scientific	Cat# AM9260G
Dextran from Leuconostoc spp	Sigma Aldrich	Cat# 31390–25G
Dynabeads MyOne Stretavidin T1	Thermo Fisher	Cat# 65601
Protein Solubilizer 40	AG Scientific	Cat# P-1505–10X5ML
Acetic Acid, Glacial	Fisher Scientific	Cat# A38S-212
Ammonium Hydroxide	Fisher Scientific	Cat# A669s
Tri-n-octylamine	TCI	Cat# I0362
Ethanol	Fisher	Cat# AC611050040
Stainless Steel Beads 0.9 – 2.0 mm blend	Next Advance	Cat# SSB14B
Methanol	Fisher Scientific	Cat# A412SK4
Potassium hydroxide	Sigma Aldrich	Cat# 221473–500G
Diethyl ether, ACS reagent, anhydrous	Fisher Scientific	Cat# AC615080040
Chloroform	Fisher Scientific	Cat# C297–4
Dichloromethane	Thermo Fisher	Cat# AC61005004
2-Propanol	Fisher Scientific	Cat# A416P-4
Heparin sodium salt	Sigma Aldrich	Cat# H3149–25KU
Sodium propionate (13C₃ 99%)	Cambridge Isotope Laboratories	Cat# CLM-1865-PK
1 Kb Plus DNA Ladder	Thermo Fisher	Cat# 10787018
10X Blue Juice Gel loading buffer	Thermo Fisher	Cat# 10816015
GelRed Nucleic Acid Gel Stain	Biotium	Cat# 41003
MyTaq HS Red DNA Polymerase	Meridian Bioscience	Cat# BIO-21115
1 kb Plus DNA Ladder	NEB	Cat# N3200S
Platinum Taq DNA Polymerase	Thermo Fisher	Cat# 15966005
Ethylenediaminetetraacetic acid disodium (EDTA Salt)	Fisher Scientific	Cat# AA3331236
Proteinase K Solution	Bioline	Cat# BIO-37085
Agarose	Fisher Scientific	Cat# BP160–100
Amersham ECL Western Blotting Detection Reagents	Cytiva	Cat# 45–000-885
Immobilon-P PVDF Membrane	Millipore	Cat# IPVH00010
Invitrogen Ambion PBS Phosphate Buffered Saline	Thermo Fisher	Cat# AM9624
Acetone Optima	Fisher Scientific	Cat# A929SK-4
Non Fat Dry Milk	Fisher Scientific	Cat# 50–751-7666
Pepsatin	Sigma Aldrich	Cat# 10253286001
PhosSTOP EASY pack	Roche	Cat# 4906837001
Protein Assay Dye Reagent Concentrate	Biorad	Cat# 5000006
SDS	Fisher Scientific	Cat# BP166–500
Sodium azide	Sigma Aldrich	Cat# S8032–100G
TEMED Sigma	Sigma Aldrich	Cat# T7024–50ML
Tris-Base	Fisher Scientific	Cat# BP1521
Acrylamide	Fisher Scientific	Cat# BP1410–1
Potassium Ferrocyanide Trihydrate	Fisher Scientific	Cat# P232500
2-Mercaptoethanol	Sigma Aldrich	Cat# M3148–100ML
Pierce RIPA Buffer	Thermo Fisher	Cat# 89900
Deoxyribunoclease I	Worthington	Cat# NC9131722
Glycine	Fisher Scientific	Cat# BP381–1
IGEPAL	Sigma Aldrich	Cat# I8896–50ML
UltraPure Glycerol	Fisher Scientific	Cat# 15514011
TaqMan Fast Advanced Master Mix	Applied Biosystems	Cat# 4444556
TaqMan Gene Expression Assays	Thermo Fisher	Cat# 4331182
RNase free PCR Tubes, 0.5mL	Thermo Fisher	Cat# AM12275
RNase AWAY	Thermo Fisher	Cat# 7003PK
RNaseZap™ RNase Decontamination Solution	Thermo Fisher	Cat# AM9780
Deuterium oxide	Cambridge Isotope Laboratories	Cat# DLM-4–1L
Sodium Deuteroxide	Cambridge Isotope Laboratories	Cat# 14014–06-3
SuperScript IV First-Strand Synthesis System	Thermo Fisher	Cat# 18091050
Complete mini protease inhibitors	Roche	Cat# 11836153001
OneMinutePlus Western Blot Stripping Buffer	GM Bioscences	Cat# GM6011
14C-2DG	American Radiolabeled Chemicals	Cat# ARC 0112D-50 μCi
[1–13C]pyruvate	Cambridge Isotopes Laboratories	Cat# CLM-1082–0.25
Trityl radical	General electric	Cat# AH-111501
RNasin Ribonuclease Inhibitor (40 U/ul)	Promega	Cat# N2515
RLT Buffer	Qiagen	Cat# 79216
Trypsin	Worthington Biochemical	Cat# LS003707
DNAase	Worthington Biochemical	Cat# LS002139
BME	Gibco	Cat# 41100–025
Percoll	Sigma Aldrich	Cat# P1644–100ML
Sodium hydroxyde	Fisher Scientific	Cat# S318–500
Poly-D-lysine	EMD Millipore	Cat# A-003-E
L-Glutamine	Gibco	Cat# 25030–016
Pen/Strep	Gibco	Cat# 15140–015
Pierce ECL Western Blotting Substrate	Thermo Fisher	Cat# 32106
Amersham ECL Western Blotting Detection Reagent	Fisher Scientific	Cat# 45–000-885
SODIUM DODECYL SULFATE	Fisher Scientific	Cat# BP166–500
Ammonium persulfate	Sigma Aldrich	Cat# A3678–100G
FDG	Sofie Radiopharmacy	Cat# NDC # 49609–101-01
OptiPrep Density Gradient Medium	Sigma Aldrich	Cat# D1556–250ML
Glyoxal	Sigma Aldrich	Cat# 50649–25ML
Spermidine trihydrochloride	Sigma Aldrich	Cat# 85578–1G
Spermine tetrahydrochloride	Sigma Aldrich	Cat# S1141–1G
Ammonium chloride	Sigma Aldrich	Cat# A9434–500G
DL-Dithiothreitol	Sigma Aldrich	Cat# D9779–1G
PBS 10x	Thermo Fisher	Cat# AM9625
2M KCl	Thermo Fisher	Cat# AM9640G
1 M MgCl2	Thermo Fisher	Cat# AM9530G
Nuclease-free water	Thermo Fisher	Cat# AM9930
UltraPure BSA	Thermo Fisher	Cat# AM2616
HEPES	Thermo Fisher	Cat# 15630080
UltraPure DNase/RNase-Free Distilled Water	Thermo Fisher	Cat# 10977015
Diethylpyrocarbonate	Gold Bio	Cat# D-340–25
Critical commercial assays
Pyruvate Dehydrogenase Activity Kit	Biovision	Cat# K679–100
Mitochondria Isolation Kit	Biovision	Cat# K288–50
In Situ Cell Death Detection Kit	Roche	Cat# 11684795910
Mem-PER Plus Membrane Protein Extraction Kit	Thermo Fisher	Cat# 89842
RNeasy Micro Kit	Qiagen	Cat# 74004
NEBNext Single Cell/Low Input cDNA Synthesis & Amplification Module	NEB	Cat# E6421S
Invitrogen superscript IV First Strand Synthesis System	Thermo Fisher	Cat# 18091050
LightCycler 480 SYBR Green I Master	Roche	Cat# 4707516001
LightCycler 480 Multiwell Plate 96	Roche	Cat# 4729692001
Experimental models: Organisms/strains
C57BL/6J	The Jackson Laboratory	Strain #:000664, RRID:IMSR_JAX:000664
B6.129P2-Pdha1tm1Ptl/J	The Jackson Laboratory	Strain #:017443, RRID:IMSR_JAX:017443
Tg(Gt(ROSA)26Sor-EGFP)I1Able/J	The Jackson Laboratory	Strain #:007896, RRID:IMSR_JAX:007896
FVB/NJ	The Jackson Laboratory	Strain #:001800 RRID:IMSR_JAX:001800
FVB-Tg(GFAP-cre)25Mes/J	The Jackson Laboratory	Strain #:004600, RRID:IMSR_JAX:004600
Oligonucleotides
Genotyping and qPCR primers	Table S8	N/A
Software and algorithms
ImageJ	NIH	NA
AdobeIllustratorCS5	AdobeSystems	https://www.adobe.com/de/creativecloud/membership.html
R version 4.1.1	Posit	https://posit.co/
VSCode	Microsoft	https://code.visualstudio.com/
DeepLabCut	Mackenzie Mathis Lab	https://www.mackenziemathislab.org/deeplabcut#:~:text=DeepLabCut%E2%84%A2%20is%20an%20efficient,typically%2050%2D200%20frames).
Matlab	MathWorks	https://www.mathworks.com/products/matlab.html
ACD/Labs Spectruss Processor 2021	AdvancedChemistryDevelopment	https://www.acdlabs.com/products/adh/spectrusprocessor/
TCASim	Jeffry R. Alger et al.	PMID: 31745452
MSDChemStation	AgilentTechnologies	E02.01.1177;RRID:SCR_015742
Biorender	Biorender	https://www.biorender.com
Mouse lateral sway analysis code	This paper	https://github.com/Neurometabolomics/Mouse-body-sway-analysis/blob/main/Mouse_behavioral_analysis_top_video_titubation.ipynb

Highlights.

• Glucose uptake and metabolism is increased in the PDHD brain in vivo.

• Two entry points of carbons sustain Krebs cycle activity in the PDHD brain.

• Propionate is a major anaplerotic substrate to the PDHD brain.

• Propionate in conjunction with a ketogenic diet improves neurological outcomes in PDHD.