Abstract

   Ketogenic diet (KD) alleviates refractory epilepsy and reduces seizures
   in children. However, the metabolic/cell biologic mechanisms by which
   the KD exerts its antiepileptic efficacy remain elusive. Herein, we
   report that KD-produced β-hydroxybutyric acid (BHB) augments brain
   gamma-aminobutyric acid (GABA) and the GABA/glutamate ratio to inhibit
   epilepsy. The KD ameliorated pentetrazol-induced epilepsy in mice.
   Mechanistically, KD-produced BHB, but not other ketone bodies,
   inhibited HDAC1/HDAC2, increased H3K27 acetylation, and
   transcriptionally upregulated SIRT4 and glutamate decarboxylase 1
   (GAD1). BHB-induced SIRT4 de-carbamylated and inactivated glutamate
   dehydrogenase to preserve glutamate for GABA synthesis, and GAD1
   upregulation increased mouse brain GABA/glutamate ratio to inhibit
   neuron excitation. BHB administration in mice inhibited epilepsy
   induced by pentetrazol. BHB-mediated relief of epilepsy required high
   GABA level and GABA/glutamate ratio. These results identified BHB as
   the major antiepileptic metabolite of the KD and suggested that BHB may
   serve as an alternative and less toxic antiepileptic agent than KD.

   Subject terms: Post-translational modifications, Cancer metabolism,
   Mechanisms of disease

Introduction

   The ketogenic diet (KD) switches cell energy metabolism from
   carbohydrates to fatty acid and amino acid oxidation^[53]1, from which
   ketone bodies, such as β-hydroxybutyrate (BHB), acetone, and
   acetoacetate (AcAc), are generated^[54]2. A consensus is yet to be
   reached regarding whether physiologic effects of the KD, including
   weight loss^[55]3, decreasing blood sugar via normalizing beta cell
   function^[56]4, and cholesterol level reduction^[57]5, are due to
   altered energy metabolism or ketone body production^[58]6,[59]7. One of
   the recognized efficacies of the KD is its beneficial neuronal effects.
   The KD has been reported to help treat brain disorders such as
   Parkinson’s disease^[60]8, Alzheimer’s disease^[61]9, multiple
   sclerosis^[62]10,[63]11, sleep disorders^[64]12, autism^[65]13, and
   brain cancer^[66]14. Since its first clinical practice in epilepsy
   treatment in 1921^[67]15, KD has been shown to be effective in treating
   children with refractory epilepsy^[68]16,[69]17, in which onset often
   occurs in childhood and older adulthood^[70]18.

   Epilepsy is a common neurological condition which affects 0.8%–1% of
   the human population, and ~30% of patients with epilepsy have
   drug-resistant epilepsy. Epileptic seizures are mostly due to a
   relative imbalance of excitatory and inhibitory neurotransmitters,
   including glutamate and gamma-aminobutyric acid (GABA)^[71]19, that
   leads to abnormal and excessively synchronized firing of neurons in the
   brain^[72]20. In the central nervous system, GABA is primarily produced
   by glutamate decarboxylase 1 (GAD1), which decarboxylates glutamate
   within GABAergic axon terminals^[73]21. The interictal epileptic
   activity in the human temporal lobe is critically dependent on the
   depolarizing action of GABA^[74]22. Disturbance of central glutamate
   and GABA homeostasis is consistently associated with neurological
   disorders, such as schizophrenia and epilepsy^[75]23,[76]24. The
   proposed antiepileptic mechanisms include inhibiting the presynaptic
   release of glutamate by directly competing with Cl^– for allosteric
   activation of vesicular glutamate transporters and increasing levels of
   GABA in cerebrospinal fluid^[77]25. BHB can be produced in astrocytes
   from oxidation of fatty acids or catabolism of amino acids, and is
   metabolized in the mitochondria of all brain cell types^[78]26. The
   open probability of K[ATP] channels in the hippocampus in vitro is
   enhanced in the presence of BHB^[79]27.

   Glutamate receives amines from most amino acids for deamination and is
   the most abundant excitatory neurotransmitter in the vertebrate nervous
   system^[80]28. Glutamate can be converted to α-ketoglutarate (α-KG) to
   enter the tricarboxylic acid cycle^[81]29 via glutamate dehydrogenase
   (GDH) or to GABA via GAD1. It has been suggested that SIRT4 represses
   enzymatic activity and limits the catabolism of glutamate and glutamine
   by ADP-ribosylation of GDH^[82]30. GDH-produced ammonia is converted to
   carbamoyl phosphate (CP), which spontaneously forms protein lysine
   carbamylation (CP-K) to activate mitochondrial proteins, including
   ornithine transcarbamylase (OTC) of the urea cycle; CP-K effects on OTC
   are reversed via mitochondrial sirtuin SIRT4^[83]31. SIRT4 regulates
   GDH activity and GABA/glutamate ratio that determines neuronal
   excitation, it is possible that CP-K and SIRT4 can regulate glutamate
   and GABA homeostasis, and consequently determine GABA/glutamate ratio
   in epilepsy control.

Results

KD increased the GABA, glutamate, and GABA/glutamate ratio in the brains of
mice

   To investigate the seizure-preventing effects of the KD, 8-week-old
   male C57/BL6 mice were fed with either a KD or normal diet (ND) for 12
   weeks before they were intraperitoneally injected with pentetrazol
   (PTZ), which induces acute epilepsy in mice^[84]32. Female mice were
   not used in this study to exclude the effects of hormone changes during
   puberty and hormone fluctuations over estrous cycle^[85]33. The
   development of both clonic and tonic-clonic seizures began with
   stereotypic behaviors, such as Straub tail, myoclonic jerks and
   hindlimb flexions, which progressed to forelimb clonus and head bobbing
   and ended with relaxed muscle tone. And an isoelectric period^[86]34
   was monitored. The severity of epilepsy was calculated according to the
   Racine scale (Fig. [87]1a)^[88]35.

Fig. 1. KD increased GABA, glutamate, and GABA/glutamate ratio in mouse
brains.

   [89]Fig. 1
   [90]Open in a new tab

   a Schematic diagram of the timeline of dietary exposure and PTZ
   treatment of mice and the Racine scale used to assess the intensity of
   epilepsy in mice. b–e PTZ-induced epilepsy in mice. KD- and ND-fed mice
   were intraperitoneally administered with a shot of PTZ before being
   monitored for epilepsy symptoms that were staged by the Racine scale.
   Latency time (b), the highest level (c), duration of (d) and frequency
   of seizures (e) were monitored. Data are means ± SD, n = 10 mice in
   each treatment. f The KD weakened neuronal activity in mice. The
   representative EEG recordings of WT mice fed the ND or KD with
   epileptiform activity are shown. g Pathway enrichment analysis showing
   metabolic pathways affected by the KD. The rich factors represent the
   ratio of differentially expressed protein in this pathway annotated by
   number. h The KD increased GABA and glutamate in mouse brains. Levels
   of GABA, glutamate, glutamine and aspartate were determined in the
   brains of C57 mice fed with either ND or KD. n = 6 mice. i The KD
   increased the GABA/glutamate ratio in mouse brains. The ratio of
   GABA/glutamate was compared between ND- and KD-fed mice. n = 6 mice.

   The KD-fed mice had a longer seizure latency than the ND-fed mice (Fig.
   [91]1b). Moreover, the KD-fed mice also had less severe seizure
   symptoms. Among ten mice tested in each group, only three KD-fed mice
   reached grade III or above; in contrast, most ND-fed mice reached grade
   IV or V (Fig. [92]1c). Furthermore, while PTZ induced a similar
   epilepsy duration in both KD- and ND-fed mice (Fig. [93]1d), it induced
   lower seizure frequency in KD-fed mice than in ND-fed mice (Fig.
   [94]1e). The epilepsy-relieving effects of the KD were further
   confirmed via electroencephalography (EEG), which showed that
   PTZ-induced less electrical activity in KD-fed mice than in ND-fed mice
   (Fig. [95]1f).

   Proteomic analysis revealed that metabolic pathways, including
   glutamatergic and GABA-like receptor-related pathways, were altered by
   the KD (Fig. [96]1g). These results were consistent with those of the
   targeted metabolomic analyses, which showed that KD-fed mice had higher
   levels of glutamate, GABA, and associated metabolites in the brains
   than those of ND-fed mice (Fig. [97]1h). Notably, KD feeding resulted
   in a higher GABA/glutamate ratio (Fig. [98]1i). These findings
   highlighted the possibility that the KD exerts seizure-relieving
   effects via augmenting GABA and GABA/glutamate ratio, both inhibit the
   excitation of neurons^[99]36.

KD inactivated GDH and activated GAD1

   The KD altered glutamate homeostasis, prompting us to examine whether
   it affected the levels of enzymes involved in the Glu/GABA metabolism
   pathway (Fig. [100]2a) in vivo^[101]37. The KD had negligible impacts
   on other Glu/GABA metabolism pathway enzymes, except for increased mRNA
   and protein levels of GAD1 (Fig. [102]2b, c), consistent with the
   proteomics results (Figs. [103]1g, [104]2d; Supplementary Fig.
   [105]S1a). As the KD caused neither GDH nor glutaminase (GLS) changes
   in mouse brains (Fig. [106]2b, c), it may increase glutamate (see Fig.
   [107]1h) by inactivating GDH, which oxidizes glutamate to α-KG, or by
   inactivating GAD1, which converts glutamate to GABA^[108]38. When GAD1
   was overexpressed in U87MG (a human brain glioblastoma cell line)
   cells, it increased GABA levels; however, it failed to decrease
   glutamate levels (Fig. [109]2e), suggesting that GAD1 activity is more
   important for GABA/glutamate ratio than for glutamate levels.
   Collectively, these results suggest that the KD may decrease
   GDH-specific activity to increase glutamate in mouse brains.

Fig. 2. KD relieved epilepsy by changing GDH, GAD1, and SIRT4.

   [110]Fig. 2
   [111]Open in a new tab

   a Diagram of the enzymes involved in glutamate/GABA metabolism
   pathways. Enzymes studied in the current study (GDH and GAD1) are
   highlighted. b, c The KD upregulated GAD1 expression. Protein (b) and
   mRNA (c) levels of glutamate/GABA metabolism enzymes were determined in
   the brains of mice fed ND or KD. n = 4 mice. d KD induced SIRT4 and
   GAD1 expression. The histogram reflects the protein levels of SIRT4 and
   GAD1. n = 3 mice. e GAD1 had a limited impact on glutamate levels. GAD1
   was overexpressed in U87MG cells; GABA and glutamate levels were
   compared between overexpressed and control U87MG cells. f The KD
   induced SIRT4 expression. SIRT4 levels in mouse brains were compared
   between KD- and ND-fed mice. n = 4 mice. g–k SIRT4 was required for the
   KD to relieve epilepsy in mice. KD- and ND-fed WT and Sirt4^–/– mice
   were each administered with an intraperitoneal PTZ injection before
   being monitored for epilepsy symptoms. Latency time (g), duration (h),
   the highest level (i), and frequency of seizures (j) and EEG (k) were
   monitored. Data were means ± SD, n = 10 mice in each treatment.

SIRT4 was involved in seizure relief via the KD

   Notably, SIRT4, a known GDH regulator^[112]30, was also induced by the
   KD (Fig. [113]2d; Supplementary Fig. [114]S1a). Upregulation of SIRT4
   by the KD was confirmed via western blot (Fig. [115]2f) and real-time
   qPCR (Supplementary Fig. [116]S1b) in the mice brains. The involvement
   of SIRT4 in the KD-mediated control of epilepsy was further
   investigated. 8-week-old wild-type (WT) and Sirt4 knockout male C57
   mice (Sirt4^–/– mice) were fed with ND or KD. The Sirt4^–/– mice grew
   normally, similar to WT mice, upon chow feeding (Supplementary Fig.
   [117]S1c, d). Moreover, indiscernible differences in the brain tissue
   structure were observed using Nissl (Supplementary Fig. [118]S1e) and
   hematoxylin and eosin (HE) staining (Supplementary Fig. [119]S1f) after
   12 weeks of ingesting KD. PTZ-induced pronounced epilepsy symptoms in
   Sirt4^–/– mice, as manifested by faster onset of symptoms such as limb
   tremors and side-to-side swaying (Fig. [120]2g), accompanied by longer
   seizures (Fig. [121]2h). Sirt4^–/– mice exhibited more frequent and
   intense symptoms, including outstretched hind limbs and front paws
   tucked under the chin, which suggest death during an episode of status
   epilepticus (Fig. [122]2i, j) upon PTZ administration. These behavioral
   seizures and epileptiform interictal activity in PTZ-challenged
   Sirt4^–/– mice were confirmed using video/EEG (Fig. [123]2k). Notably,
   the KD failed to relieve the epilepsy symptoms in PTZ-challenged
   Sirt4^–/– mice (Fig. [124]2g, k). Collectively, these results suggest
   that the KD prevents epilepsy, which is dependent on SIRT4 function.

   We injected Cre-adeno-associated virus (AAV) into the hippocampus of
   Sirt4^flox/flox mice to delete Sirt4 from hippocampus (Sirt4-cKO,
   Supplementary Fig. [125]S2a, b). We found that Sirt4-cKO mice
   phenocopied constitutive Sirt4 knockout mice in response to PTZ
   challenge and to KD feeding (Supplementary Fig. [126]S2c–i). These
   results confirmed that Sirt4 in central nervous system is involved in
   seizure relief effects of the KD.

KD-produced BHB-activated SIRT4 and GAD1 transcription

   We investigated how the KD induces SIRT4 and GAD1 expression. The KD
   has a low carbohydrate content. Therefore, we tested whether sugar
   deprivation could result in KD-induced SIRT4 expression. Culturing
   U87MG (Supplementary Fig. [127]S3a) and mouse neuroblastoma (Neuro-2a)
   (Supplementary Fig. [128]S3b) cells in low-glucose medium failed to
   induce changes in SIRT4 and GAD1 protein and mRNA levels, ruling out
   the possibility that the KD drives SIRT4 overexpression solely owing to
   its low carbohydrate content.

   Ketone bodies, such as BHB, AcAc, and acetone, significantly (P < 0.05)
   increased in the serum (Fig. [129]3a) and brain (Fig. [130]3b) of
   KD-fed mice. However, at physiologic relevant higher levels, only
   BHB-induced protein and mRNA expression of SIRT4 and GAD1, when U87MG
   and Neuro-2a cells were treated with each of 5 mM sodium BHB, sodium
   AcAc, or sodium chloride, or with 0.5 mM acetone (Fig. [131]3c, d;
   Supplementary Fig. [132]S3c). Moreover, the BHB-induced protein and
   mRNA expression of SIRT4 and GAD1 was dose-dependent in mouse
   hippocampal HT22 cells, and in U87MG (Fig. [133]3e, f; Supplementary
   Fig. [134]S3d). To further determine the effect of BHB on SIRT4 and
   GAD1 expression, we overexpressed β-hydroxybutyrate dehydrogenase 1
   (BDH1), which catalyzes the first reaction of BHB oxidation to convert
   BHB back into AcAc^[135]39, in HT22, U87MG, and Neuro-2a cells. We
   found BDH1 overexpression, while decreasing cellular BHB levels (Fig.
   [136]3g), decreased the expression of SIRT4 and GAD1 in HT22, U87MG,
   and Neuro-2a cells (Fig. [137]3h, i). Conversely, knockdown BDH1 with
   small interfering RNAs that increased intracellular BHB concentration
   (Fig. [138]3j), increased mRNA and protein levels of SIRT4 and GAD1 in
   U87MG and HT22 cells (Fig. [139]3k, l; Supplementary Fig. [140]S3f).
   These results, together with the fact that SIRT4 and GAD1 degradation
   in protein synthesis-inhibited CHX-treated cell lysates were unaffected
   by BHB (Supplementary Fig. [141]S3g), supported the hypothesis that the
   KD promotes SIRT4 and GAD1 transcription by upregulating BHB levels.

Fig. 3. KD-produced BHB activated SIRT4 and GAD1 transcription.

   [142]Fig. 3
   [143]Open in a new tab

   a, b The KD increased mice ketone body levels. Serum (a) and brain (b)
   levels of the BHB, AcAc, and acetone were detected in ND- and KD-fed
   mice. c, d BHB increased SIRT4 and GAD1 expression levels. U87MG cells
   were treated with ketone bodies and the protein (c) and mRNA (d) levels
   of SIRT4 and GAD1 were measured. e, f BHB induced SIRT4 and GAD1
   expression in a dose-dependent manner. Protein (e) and mRNA (f) levels
   of SIRT4 and GAD1 were determined in HT22 cells that were treated with
   BHB at the indicated concentrations. g BDH1 overexpression decreased
   BHB levels. The BHB levels were measured in HT22, U87MG and Neuro-2a
   cells and in BDH1-overexpressing HT22 cells, U87MG and Neuro-2a cells.
   h, i BDH1 overexpression decreased SIRT4 and GAD1 expression. Protein
   (h) and mRNA (i) levels of SIRT4 and GAD1 were detected in HT22 cells,
   U87MG, and Neuro-2a cells and in BDH1-overexpressing HT22 cells, U87MG
   and Neuro-2a cells. j–l BDH1 knockdown increased BHB levels and
   increased SIRT4 and GAD1. BDH1 was silenced with independent siRNAs in
   HT22 and U87MG cells. BHB levels (j), protein (k), and mRNA(l) levels
   of SIRT4 and GAD1 were compared between siRNA-untreated and -treated
   cells.

BHB inhibited HDAC1/HDAC2, thus activating SIRT4 and GAD1 transcription

   To investigate the mechanism by which BHB promotes SIRT4 and GAD1
   expression in cells, we used actinomycin D to inhibit mRNA synthesis
   and measured the half-lives of SIRT4 and GAD1 mRNA in U87MG cells. No
   significant differences in SIRT4 and GAD1 mRNA half-lives were observed
   between the BHB-treated and control cells (Fig. [144]4a; Supplementary
   Fig. [145]S4a), indicating that BHB promotes SIRT4 and GAD1 mRNA
   synthesis rather than inhibiting mRNA degradation. Given that BHB is an
   established inhibitor of histone deacetylases (HDACs)^[146]40, we
   analyzed the chromatin accessibility of the SIRT4 and GAD1 promoter
   regions and found that BHB treatment markedly enhanced chromatin
   accessibility of the SIRT4 and GAD1 promoters (Fig. [147]4b;
   Supplementary Fig. [148]S4b). This result is consistent with the key
   function of BHB, which enables the transcriptional machinery to inhibit
   histone deacetylation^[149]40.

Fig. 4. BHB inhibited HDAC1/HDAC2 to activate SIRT4 and GAD1 transcription.

   [150]Fig. 4
   [151]Open in a new tab

   a mRNA degradation did not account for the BHB-promoted SIRT4 and GAD1
   mRNA increase. SIRT4 mRNA in U87MG cells was detected at the indicated
   time points with 10 mM actinomycin D. b BHB increased chromatin
   accessibility of the SIRT4 promoter. Quantitative PCR was performed on
   DNase I-pretreated nuclei of U87MG cells that were treated with BHB
   with the indicated concentrations. c BHB specifically elevated H3K27Ac.
   Histone acetylation levels were assayed in U87MG cells and in BHB- or
   trichostatin A (TSA)-treated U87MG cells, employing site-specific and
   pan-anti-acetyl-lysine (Ac) antibodies. d BHB-treatment elevated
   H3K27Ac at the promoter regions of SIRT4 and GAD1. The H3K27Ac antibody
   was used for immunoprecipitation (IP) after BHB treatment, and qRT-PCR
   of SIRT4 and GAD1 promoter output was performed. e, f The
   HDAC1/HDAC2-specific inhibitor FK228 increased SIRT4 and GAD1 protein
   (e) and mRNA (f) levels. The ability of FK228 to upregulate SIRT4 and
   GAD1 expression was tested in U87MG cells. g, h BHB elevated SIRT4 and
   GAD1 expression, dependent on HDAC1 and HDAC2 acetylation. The effects
   of BHB (10 mM) treatment on SIRT4 and GAD1 expression were assayed in
   HeLa cells with HDAC1, HDAC2, or double knockout of both (g, quantified
   4 replicates in right), and in U87MG cells overexpressing HDAC1, HDAC2,
   or both (h). i HDAC1 and HDAC2 knockout enhanced H3K27Ac at the SIRT4
   promoter. ChIP assays were performed for the promoter region of SIRT4,
   using the H3K27Ac antibody for IP and qRT-PCR for the SIRT4 promoter
   output.

   Tracing the acetylation levels at different sites on histone H3 via
   immunoblotting revealed that the acetylation levels at H3K27 increased
   with BHB (Fig. [152]4c). Moreover, chromatin immunoprecipitation (ChIP)
   assays showed that BHB enhanced H3K27Ac levels and histone interactions
   with SIRT4 and GAD1 promoters (Fig. [153]4d), suggesting that
   BHB-induced SIRT4 and GAD1 transcriptional activation may correlate
   with H3K27Ac, which is deacetylated by the HDAC family, including
   HDAC1–HDAC10^[154]41.

   Treatment of U87MG cells with different HDAC inhibitors revealed that
   neither the HDAC3 inhibitor RGFP966 (Supplementary Fig. [155]S4c, d)
   nor the HDAC 4/5/7/9 inhibitor TMP195 (Supplementary Fig. [156]S4e, f)
   affected the expression of SIRT4 and GAD1, and only the HDAC1 and HDAC2
   inhibitor FK228 upregulated protein (Fig. [157]4e) and mRNA (Fig.
   [158]4f) levels of SIRT4 and GAD1.

   To determine whether HDAC1 and HDAC2 are involved in the
   transcriptional regulation of SIRT4 and GAD1 via BHB, we knocked out
   either HDAC1 or HDAC2 to induce SIRT4 and GAD1 transcription and
   protein expression; however, only knockout of both HDAC1 and HDAC2
   caused pronounced increase of SIRT4 and GAD1 expression and prevented
   BHB from activating SIRT4 and GAD1 transcription in HeLa cells (Fig.
   [159]4g). In contrast, the overexpression of HDAC1 and HDAC2 in U87MG
   (Fig. [160]4h) and HeLa cells (Supplementary Fig. [161]S4g) decreased
   SIRT4 and GAD1 expression. These results, consistent with those of ChIP
   assays, showed that HDAC1 and HDAC2 knockouts extensively increased the
   H3K27Ac of histones, which interacted with the SIRT4 and GAD1 promoters
   in HeLa cells (Fig. [162]4i). HDAC1/HDAC2-mediated histone
   deacetylation inhibits the transcription of GAD1^[163]42 and this study
   confirmed that BHB activates SIRT4 and GAD1 transcription by inhibiting
   HDAC1- and HDAC2-induced H3K27 deacetylation.

Lysine carbamylation-activated GDH

   Although GAD1 activity was unresponsive to SIRT4 variation
   (Supplementary Fig. [164]S5a, b) in SIRT4-silenced and
   SIRT4-overexpressing HEK293T cells, GDH was activated (Fig. [165]5a)
   and inactivated (Fig. [166]5b), respectively, suggesting that GDH is
   inactivated by GDH mono-ADP-ribosylation^[167]43. However, the activity
   of the ADP-ribosylation-null mutant GDH^C119G ^[168]44 was still
   responsive to SIRT4 variation (Fig. [169]5a, b), suggesting that a
   mechanism other than ADP-ribosylation regulates GDH enzyme activity to
   convert glutamate into α-KG.

Fig. 5. Lysine carbamylation activated GDH.

   [170]Fig. 5
   [171]Open in a new tab

   a, b The activity of the ADP-ribosylation-null GDH^C119G mutant was
   under regulation by SIRT4. Flag-tagged GDH or GDH^C119G was ectopically
   expressed in HEK293T cells or SIRT4-silenced HEK293T cells (a).
   Flag-tagged GDH or GDH^C119G was ectopically expressed in HEK293T cells
   or SIRT4-overexpressing HEK293T cells (b). The specific activities of
   GDH and the GDH^C119G mutant were determined and compared. c GDH
   interacted with CPS1. Flag-tagged CPS1 or GDH was ectopically expressed
   in HEK293T cells together with individual HA-tagged GDH or CPS1 as
   indicated. Flag bead- or HA bead-purified proteins were subject to
   western blotting analysis. d Carbamylation of endogenous GDH relied on
   CPS1 to generate CP. Endogenous GDH of HeLa cells was purified via IP
   with Protein A beads conjugated with the GDH antibody, and
   carbamylation of the purified GDH was detected. e KD decreased mouse
   brain mitochondrial CP-K levels. Mitochondrial CP-K levels were
   detected in brains of mice fed with either ND or KD. f Carbamylation
   activated GDH. Flag-tagged GDH was ectopically expressed in HEK293T
   cells treated with CP (5 mM) and purified via IP with Flag beads. The
   specific activity of GDH was determined. g GDH K162 was carbamylated.
   MS/MS spectra from tryptic peptide libraries of HEK293T cells matching
   MS/MS spectra from synthetic peptide (composed of ^12C, ^1H, ^14N, and
   ^16O) identified GDH CP-K162. h Screening of the major regulatory
   carbamylation sites in GDH. Detected carbamylated residues were each
   mutated to arginine (R), and the specific activity of each mutant GDH
   was compared with that of WT GDH. i In vitro formation of CP-K162. The
   synthetic GDH K162-containing peptide was treated with 10 mM CP,
   followed by mass spectrometry analysis.

   Glutamate oxidation via GDH produces α-KG and ammonia, which are
   converted to CP that spontaneously forms CP-K^[172]45,[173]46, which is
   reversed by SIRT4^[174]31. GDH interacted with CPS1, an enzyme that
   spontaneously produces CP to carbamylate lysine (Fig. [175]5c)^[176]45.
   Moreover, CPS1 knockout in HeLa cells reduced the level of GDH CP-K
   (Fig. [177]5d). Furthermore, in the mitochondria of mouse brains, the
   KD-induced lower CP-K levels (Fig. [178]5e). Thus, we investigated the
   possibility that the KD induces SIRT4, which inactivates GDH by
   decarbamylating it. Moreover, the incubation of recombinant GDH with
   5 mM CP increased GDH CP-K levels and activated GDH (Fig. [179]5f),
   suggesting that carbamylation activates GDH.

   Using a Thermo Q-Exactive HF (QE) Quadrupole-Obitrap mass spectrometer,
   we detected multiple CP-modified lysine sites, including K84, K162,
   K191, K200, K352, and K503 (Fig. [180]5g; Supplementary Fig. [181]S6)
   in GDH purified from U87MG cells. Eliminating potential carbamylation
   sites in GDH by switching lysine to arginine (K/R) revealed that
   GDH^K162R, GDH^K352R, and GDH^K503R inactivated GDH and diminished the
   ability of CP to activate GDH (Fig. [182]5h), suggesting that these
   were the key carbamylation sites that activated GDH. To confirm that CP
   spontaneously carbamylates GDH, we synthesized a GDH peptide containing
   K162, which is evolutionarily conserved from Trachemys scripta elegans
   to Homo sapiens (Supplementary Fig. [183]S7a), and close to the active
   site of GDH, K183 (Supplementary Fig. [184]S7b). Treatment with 5 mM CP
   induced carbamylation in the K162-containing peptide but not in the
   peptide in which K162 was replaced by arginine (K162/R) (Fig. [185]5i).
   Collectively, these results suggest that lysine carbamylation enhances
   GDH activity without affecting its expression.

SIRT4 decarbamylated and inactivated GDH to increase both glutamate and GABA
but GABA/glutamate ratio

   GDH and SIRT4 were co-immunoprecipitated when they were simultaneously
   expressed in HEK293T cells (Supplementary Fig. [186]S8a). Moreover, the
   decarbamylase activity of SIRT4 was confirmed by detecting
   decarbamylated peptide formation from the K162R-containing GDH peptide
   using reverse-phase high-performance liquid chromatography (Fig.
   [187]6a). This suggests that SIRT4 is a decarbamylase of GDH, which was
   confirmed by the CP-K level and specific activity of ectopically
   expressed GDH in HEK293T cells (Fig. [188]6b). Endogenous GDH in U87MG
   cells (Supplementary Fig. [189]S8b) cultured in CP-supplemented media
   was decreased by SIRT4 overexpression, and not by that of SIRT3 or
   SIRT5. Conversely, silencing SIRT4 in HEK293T cells increased CP-K
   levels and the specific activity of GDH (Fig. [190]6c). In contrast,
   SIRT4 had no effect on the specific activity of GDH^K162R, which was
   diminished when it was co-expressed with GDH^K162R (Fig. [191]6c, d).
   Affinity-purified GDH from WT mice brains were carbamylated less and
   were less active than that from Sirt4^–/– mouse brains (Fig. [192]6e,
   f). Moreover, the KD, which upregulated SIRT4 in the mouse brains (Fig.
   [193]2d–f), decreased CP-K levels and inactivated endogenous GDH (Fig.
   [194]6e, f). By using ^13C-glutamine as a tracer (Supplementary Fig.
   [195]S8c), we detected the relative cytosolic and mitochondrial
   ^13C-GABA and ^13C-α-KG levels in both U87MG cells and SIRT4-silenced
   U87MG cells. SIRT4 knockdown decreased the GABA and glutamate levels,
   but increased α-KG levels (Fig. [196]6g). Moreover, the GABA and
   glutamate levels were decreased in Sirt4^–/– mouse brains (Fig.
   [197]6h), and the GABA/glutamate ratio was not altered by SIRT4 removal
   (Fig. [198]6i). These results suggest that GAD1 is not the downstream
   of SIRT4, and collectively supports the idea that SIRT4 decarbamylates
   and inactivates GDH.

Fig. 6. SIRT4 decarbamylated and inactivated GDH to increase glutamate and
GABA and not the GABA/glutamate ratio.

   [199]Fig. 6
   [200]Open in a new tab

   a SIRT4 decarbamylated GDH CP-K162. RP-HPLC analyses were used to
   detect decarbamylation of the synthetic GDH CP-K162 peptide catalyzed
   by SIRT4. b SIRT4 decarbamylated GDH. HA-tagged mitochondrial SIRT3,
   SIRT4, and SIRT5 were each individually expressed in untreated and CP
   (5 mM)-treated HEK293T cells. The GDH ectopically expressed in these
   cells were purified and subject to CP-K and specific activity analysis.
   c, d K162 was the major regulatory carbamylation site in GDH.
   Ectopically expressed GDH and GDH^K162R mutant from HEK293T cells and
   SIRT4 knockdown (c) or SIRT4 overexpression (d) HEK293T cells assayed
   for CP-K and specific activities. e GDH in Sirt4^–/– mouse brain was
   more carbamylated. The GDH protein was purified using Protein A beads
   conjugated with the GDH antibody from the brains of WT and Sirt4^–/–
   mice that were fed either ND or KD. GDH carbamylation was detected for
   purified GDH. f GDH was inactivated by the KD only in WT mouse brains.
   The WT and Sirt4^–/– mice were each fed the ND and KD from week 8. The
   mouse brain GDH purified at week 20 was assayed for specific activity.
   g SIRT4 knockdown decreased GABA and glutamate and increased α-KG.
   Culture media was supplemented with 2 mM ^13C-glutamine. The M + 5
   glutamate and α-KG (carbons fully labeled) and M + 4 GABA (carbons
   fully labeled) were determined in the mitochondria and cytoplasm of WT
   and SIRT4 knockdown U87MG cells after 6 h of chasing. h Sirt4 ablation
   caused more pronounced GABA elevation than glutamate elevation. The
   levels of GABA and glutamate were compared between WT and Sirt4^–/–
   mouse brains. i Sirt4 ablation did not alter GABA/glutamate ratio in
   mouse brains. GABA/glutamate ratios were detected in WT and Sirt4^–/–
   mouse brains.

BHB effectively relieved epilepsy

   To determine whether BHB alone may have seizure-relieving effects, the
   ND was supplemented with a BHB ester via oral gavage, which increased
   serum BHB levels > 4-fold (Supplementary Fig. [201]S9a); however, it
   did not affect body weight (Supplementary Fig. [202]S9b) and food
   intake (Supplementary Fig. [203]S9c).

   BHB supplementation induced higher GABA/glutamate ratio in the brains
   (Fig. [204]7a), longer latency of seizures (Fig. [205]7b), less severe
   seizures (Fig. [206]7c), and lower frequency of them (Fig. [207]7d).
   However, it had no impact on seizure duration (Fig. [208]7e) in
   PTZ-induced C57 mice, reinstating the antiepileptic efficacy of the KD
   (see Fig. [209]1b–e). Notably, no seizure-relieving efficacy was
   observed in PTZ-treated Sirt4^–/– mice (Fig. [210]7b, e). The lower
   potency of BHB in inhibiting epilepsy in Sirt4^–/– mice than in WT mice
   was further confirmed by video/EEG (Supplementary Fig. [211]S9d). The
   carbamylation (Supplementary Fig. [212]S9e) and activity (Supplementary
   Fig. [213]S9f) of mice GDH were decreased by BHB in WT but not in
   Sirt4^–/– mouse brain. Moreover, H3K27Ac (Supplementary Fig. [214]S9g)
   and SIRT4 and GAD1 protein levels (Supplementary Fig. [215]S9e) were
   increased by BHB. These are all consistent with the findings that BHB
   elevates H3K27Ac to overexpress SIRT4 and GAD1 and shift GABA/glutamate
   ratio in the brains.

Fig. 7. BHB relieved epilepsy in mice.

   [216]Fig. 7
   [217]Open in a new tab

   a BHB increased the GABA/glutamate ratio in WT and Sirt4^–/– mouse
   brains. Data were represented as individual values. b–e BHB alone
   SIRT4-dependently relieved seizure phenotypes. The ND- and BHB-fed WT
   and Sirt4^–/– mice were subject to measurements after they were
   administered with PTZ. Latency time (b), the highest level (c), the
   frequency (d) and the duration of seizures (e) were monitored. Data
   were means ± SD, n = 10 mice in each treatment. f–k GDH inhibition
   potentiated BHB to relieve seizures. BHB or the GDH inhibitor EGCG
   alone, or BHB together with EGCG were employed to treat Sirt4^–/– mice.
   The seizure-relieving effects of these treatments were monitored after
   mice were administered with PTZ. Brain GABA and glutamate (f) and
   GABA/glutamate ratio (g), latency time (h), the highest level of
   seizures (i), the frequency of seizures (j) and the duration of
   seizures (k) were monitored. Data were means ± SD, n = 10 mice in each
   treatment. l Schematic illustration showing KD upregulates brain GABA
   levels and GABA/glutamate ratio. Compared to ND, KD promotes BHB, which
   inhibits HDAC1/HDAC2 to upregulate SIRT4 and GAD1 transcription,
   results in inactivated GDH and activated GAD1, and consequently higher
   GABA and GABA/glutamate ratio.

   Inactivation of GDH through SIRT4-catalyzed decarbamylation may account
   for the inability to relieve PTZ-induced epilepsy in Sirt4^–/– mice. We
   used epigallocatechin-3-gallate (EGCG), an inhibitor of GDH^[218]47, to
   reverse Sirt4 knockout effects, i.e., preserving and directing
   glutamate to GABA. Oral gavage of EGCG (100 mg/kg) and BHB (500 mg/kg)
   for 4 weeks^[219]48, which had no impact on body weight (Supplementary
   Fig. [220]S10a) and food intake (Supplementary Fig. [221]S10b) of
   Sirt4^–/– mice, helped BHB to increase GABA and glutamate levels (Fig.
   [222]7f) and the GABA/glutamate ratio (Fig. [223]7g). It also led to
   remission of epileptic symptoms, which included a slower onset of the
   first seizure after PTZ administration (Fig. [224]7h), reduced
   intensity (Fig. [225]7i) and frequency of seizures (Fig. [226]7j), and
   shorter duration of seizures (Fig. [227]7k). These were consistent with
   the findings of the video/EEG for PTZ-induced Sirt4^–/– mice
   (Supplementary Fig. [228]S10c). Lastly, the seizure-relieving efficacy
   of BHB was not observed in PTZ-treated Sirt4-cKO mice (Supplementary
   Fig. [229]S10d–j). These findings collectively support the notion that
   high GABA/glutamate ratio and GABA levels in central nerve systems
   induced by BHB are required to relieve epilepsy.

Discussion

   Our results revealed that the KD-produced BHB accounts for
   antiepileptic efficacy of the KD. By inhibiting HDAC1/HDAC2, BHB
   increases H3K27Ac levels and facilitates the transcription of SIRT4 and
   GAD1, which function synergistically to inhibit neuronal activation. By
   upregulating SIRT4, the decarbamylase of GDH, BHB inactivates GDH and
   accumulates glutamate for GABA production. Moreover, upregulation of
   GAD1 by BHB favors the production of the inhibitory neurotransmitter
   GABA from glutamate and increases the GABA/glutamate ratio to inhibit
   epilepsy (Fig. [230]7l)^[231]49. Notably, the activation of both SIRT4
   and GAD1 seems to be essential for BHB to exert antiepileptic effects,
   as BHB exhibits limited antiepileptic potency in Sirt4^–/– mice
   although BHB increased brain GABA in both WT and Sirt4^–/– mice brains.
   Inhibiting GDH alone to proportionally increase glutamate and GABA also
   has limited antiepileptic potency. However, the GDH inhibitor EGCG
   potentiates BHB to inhibit epilepsy in Sirt4^–/– mice. This design
   suggests that a sufficiently high level of GABA is required in addition
   to high GABA/glutamate ratio to accomplish neuronal inhibition, given
   the array of other excitatory neurotransmitters besides glutamate that
   activate neurons^[232]50,[233]51. It is worth to note that this
   mechanism may be related to mammal physiology. BHB and/or ketone bodies
   may be fat-sufficient signals, which inhibit neuronal excitation,
   required for hunting in mammals^[234]52, to keep energy expenditure in
   check and therefore exert antiepileptic efficacy. Therefore, it makes
   physiologic sense that KD and BHB inhibit epilepsy. KD induces cardiac
   fibrosis^[235]53 as high fat diet do^[236]54, supports that KD may
   generate fat-sufficient signals.

   BHB is an endogenous and specific inhibitor of class I HDACs^[237]40.
   Moreover, GAD1 is down-regulated by HDAC1/HDAC2-mediated histone
   de-acetylation^[238]42. These results supported the upregulation of
   GAD1 by BHB or KD. The identification of BHB that upregulates SIRT4 in
   the mouse brains adds important clues to how BHB, whose production is
   upregulated by the KD, exerts antiepileptic effects. Therefore, in
   addition to its ADP-ribosyltransferase^[239]30, lipoamidase^[240]55 and
   deacylase^[241]56 activities, the newly identified de-carbamylase
   activity of SIRT4^[242]31 expands its roles in glutamate and GABA
   metabolism and neuronal activity regulation.

   In addition to many mechanisms that had been proposed to explain the
   use of KD to constrain epilepsy, the identification of BHB as a major
   antiepileptic metabolite in KD is clinically noteworthy. As KD has been
   used as a dietary treatment for refractory epilepsy since the 1920s,
   side effects include poor taste and low compliance^[243]57,[244]58, and
   prolonged KD administration tends to cause gastrointestinal
   disturbances and vomiting^[245]59, malnutrition^[246]60, cardiovascular
   diseases^[247]61, kidney stones^[248]62, and poor growth^[249]63.
   However, these limitations may be overcome by employing BHB or its
   efficacy-optimized molecules to replace the KD, which have fewer
   limitations in terms of toxicity and high doses.

   Overall, our study highlights that BHB may serve as fat-rich signals to
   inhibit neuronal excitation. Moreover, the identification of BHB as an
   antiepileptic molecule has translational potential for the development
   of low toxic and high tolerable antiepileptic drugs. We were unable to
   obtain samples from human patients, mechanistic studies were performed
   on cultured cells, and in vivo studies were performed on mice model.
   Given that we study the GABA and glutamate homeostasis on epilepsy, we
   used non-competitive GABA(A) receptor inhibitor PTZ to generate mouse
   epilepsy model, which simulates the generalized symptoms of clinical
   human myoclonic epilepsy and is used to screen anti-epilepsy drugs.
   However, PTZ model fails to find the anti-non-spasmodic seizure effect
   of lamotrigine and tiagabine, and other epilepsy models, especially
   genetic animal epilepsy models are necessary to be employed to confirm
   whether KD and/or BHB will exert general anti-epilepsy efficacies.

Materials and methods

Cell culture

   HEK293T (human embryonic kidney), HeLa (human adenocarcinoma), U87MG
   (human glioblastoma), and Neuro-2a (mouse neuroblastoma) cells were
   cultured in normal DMEM (Gibco) supplemented with 10% fetal bovine
   serum (FBS) (Gibco), 100 U/mL penicillin (Invitrogen) and 100 μg/mL
   streptomycin (Invitrogen). All cells were cultured an incubator at
   37 °C and 5% CO[2]. Cell treatments are carried out by adding chemicals
   to the culture media to the levels specified in legends. BHB treatment
   takes 12 h.

Animal experiments

   Four-week-old C57BL/6 Sirt4^–/– mice (constructed using CRISPR-mediated
   deletion of exons 3 and 4 of the mouse Sirt4 gene) were purchased from
   GemPharmatech Co., Ltd. (Shanghai, China). Genotypes were determined
   via PCR using genomic DNA obtained from tails and two primer pairs
   (sequences are provided in Supplementary Table [250]S1). All mice were
   housed in a specific pathogen-free facility at 20–22 °C on a 12 h
   light/dark cycle with ad libitum access to food and water. All
   experiments were started when mice were 6 weeks old.

   Sirt4-cKO mouse was generated based on C57BL/6 Sirt4^flox/flox mice.
   2–3-month-old C57BL/6 Sirt4^flox/flox mice, purchased from Shanghai
   Research Center of the Southern model organisms, were anesthetized with
   a mixture of ketamine (100 mg/kg) and xylazine (10 mg/kg). Thirty
   three-gauge syringe needles (Hamilton, Reno, Nevada) were used to
   bilaterally infuse 1.0 μL of AAV virus into mouse dorsal hippocampus
   (AP = –2.0, ML = ±1.5, and DV = –2.0 from Bregma (mm)) or 0.5 μL of AAV
   virus into nucleus accumbens (NAc) (AP = +1.5, ML = + 1.6, and
   DV = –4.4 from Bregma (mm) at 10°) at a rate of 0.2 μL (0.1 μL for NAc)
   per min and the needle remained in place for 5 min before removal to
   prevent backflow. Intra-hippocampal microinjection of AAV vectors
   containing the gene for Sirt4 or GFP alone were prepared by Genechem,
   Shanghai, China. AAV-injected mice were allowed to recover for 21 days
   before behavioral testing.

   Carbon dioxide asphyxiation was adapted to euthanize mice. The animals
   were brought from their home cages to the CO[2] euthanasia apparatus in
   the procedure room. Remove the filter top from the cage, and cover the
   cage with the stainless-steel euthanasia lid. Turn on the flow of CO[2]
   to the cage by using the flow meter attached to the pressure regulator.
   Expose the animals to CO[2] until complete cessation of breathing for a
   minimum of 2 min after 5 to 10 min CO[2] exposing. Continue the
   following experiments after visualizing all mice in the cage were
   euthanized.

   Adult male mice were euthanized, and serum and brain tissues were
   collected according to standard procedure. Briefly, brains were
   perfused with saline to remove blood cells before isolation, and the
   tissues were quickly removed and snap frozen in liquid nitrogen for
   further analyses. All animal procedures were conducted in accordance
   with the animal care committee at Fudan University, Shanghai, China.

Mouse KD or BHB treatment

   Eight-week-old WT and Sirt4^–/– mice received an ND (Bio-Serv, F0761)
   or KD (Bio-Serv, F3666) with free access to drinking water for 12
   weeks. BHB ester (HVMN, 200 µL by oral gavage daily, 80 mg per mouse)
   treatments to mice last 4 weeks before analysis.

PTZ-induced acute seizure mouse model

   Twenty-week-old male mice received a single dose of intraperitoneal
   injection of PTZ (Sigma), with the dosage ranging between 40 and
   45 mg/kg of body weight after being fed ND or KD. Mice were observed
   for 10 min after each injection.

   The scores of seizure were staged as follows: 0 grade: no response; I
   grade: facial movements, including blinking, twitching, rhythmic
   chewing, etc; II grade: I grade plus rhythmic nodding; III grade: II
   grade plus forelimb clonic; IV grade: III grade plus standing on hind
   legs; V grade: IV grade plus falling.

Mouse EEG

   Under Ketamine/Xylazine anesthesia, chronic EEG electrodes were
   surgically implanted using aseptic technique in mice. Micro-screws
   anchored into the cranium served as epidural electrodes and were
   connected to a microplug adhered to the cranium with dental acrylic.
   Recording electrodes were positioned bilaterally over each hemisphere
   (anterior-posterior, –2.0 mm from bregma; lateral, ±1.75 mm); a
   reference electrode was placed anterior to bregma (anterior-posterior,
   +0.5 mm) and a ground electrode was placed over the cerebellum. Mice
   were allowed to recover for 5–7 days before recording. The Telefactor
   Video/EEG Monitoring System was used to record EEGs and to videotape
   behavior simultaneously from un-anesthetized free-moving mice. The
   low-frequency filter setting was 0.3 Hz, and the high-frequency filter
   was set at 70 Hz.

CRISPR/Cas9-mediated deletion of HDAC1 and HDAC2

   HDAC1-knockout and HDAC2-knockout HeLa were generated by a CRISPR/Cas9
   plasmid, respectively targeting HDAC1 and HDAC2. The plasmids were
   constructed by cloning the annealed sgRNA into px459 vector. The sgRNAs
   were designed by CRISPR Design website ([251]http://crispr.mit.edu) and
   the sequences are listed in Supplementary Table [252]S1.

Plasmid construction

   The cDNA encoding full-length human SIRT3, SIRT4, SIRT5, GDH, GAD1,
   BDH1, CPS1, HDAC1, and HDAC2 was cloned into Flag, Myc, HA-tagged
   vectors (pcDNA3.1-Flag; pcDNA3.1-Myc; pcDNA3.1-HA). Point mutations of
   GDH were generated by Mut Express II Fast Mutagenesis Kit V2 (Vazyme).

Metabolomics analysis

Metabolite extraction

   Frozen tissues were homogenized in 80% methanol (methanol:water = 4:1)
   and centrifuged at 14,000× g for 15 min (4 °C). The supernatant was
   transferred to a new tube and lyophilized into a pellet. The sample was
   resuspended in 1500 μL of extract solution
   (acetonitrile:methanol:water = 2:2:1, with isotopically labelled
   internal standard mixture). After 30 s of vortexing, samples were
   sonicated for 10 min in an ice-water bath. Next, the samples were
   incubated for 1 h at –40 °C and centrifuged at 12,000× g for 15 min at
   4 °C. The resulting supernatant was transferred to a fresh glass vial
   for analysis.

LC-MS/MS analysis

   LC-MS/MS analyses were performed using a UHPLC system (Vanquish,
   ThermoFisher Scientific) with an UPLC BEH Amide column
   (2.1 mm × 100 mm, 1.7 μM) coupled to a QExactive HFX mass spectrometer
   (Orbitrap MS, ThermoFisher Scientific). The mobile phase consisted of
   25 mmol/L ammonium acetate and 25 mmol/L ammonia hydroxide in water
   (pH = 9.75) (A) and acetonitrile (B). The analysis was carried out with
   an elution gradient as follows: 0–0.5 min, 95% B; 0.5–7.0 min, 95%–65%
   B; 7.0–8.0 min, 65%–40% B; 8.0–9.0 min, 40% B; 9.0–9.1 min, 40%–95% B;
   9.1–12.0 min, 95% B. The column temperature was 30 °C. The temperature
   of the autosampler was 4 °C, and the injection volume was 2 μL.

   The QE HFX mass spectrometer was used due to its ability to acquire
   MS/MS spectra in information-dependent acquisition mode under the
   control of acquisition software (Xcalibur, ThermoFisher Scientific). In
   this mode, the acquisition software continuously evaluated the
   full-scan MS spectrum. The ESI source conditions were set as follows:
   sheath gas flow rate of 30 Arb, Aux gas flow rate of 25 Arb, capillary
   temperature of 350 °C, full MS resolution of 60,000, MS/MS resolution
   of 7500, collision energy of 10/30/60 in NCE mode, and spray voltage of
   3.6 kV (positive) or –3.2 kV (negative).

Data analysis

   Raw data were converted to the mzXML format using ProteoWizard and
   processed using an in-house programme, which was developed using R and
   based on XCMS, for peak detection, extraction, alignment, and
   integration. An in-house MS2 database (BiotreeDB) was applied for
   metabolite annotation. The cut-off for annotation was set at 0.3.

^13C labelling of metabolites

   Cells were plated in 10-cm dishes for 24 h before labelling. The next
   day, the medium was changed for unlabelled medium without glutamine 1 h
   before adding the ^13C[5]-glutamine medium. Cells were incubated in
   labelling media containing 2 mM ^13C[5]-glutamine (Sigma) for 6 h
   before collection. At the time of collection, cells were washed with
   prechilled saline, lyzed with (40%) methanol:(40%) acetonitrile:(20%)
   water with 0.1 M formic acid, and processed as described above
   (Metabolomics). Metabolites were reconstituted in 100 μL of 0.1% formic
   acid in LC/MS-grade water, vortex-mixed, and centrifuged to remove
   debris. The supernatants were analyzed using LC-MS/MS.

Proteomics analysis

Protein digestion

   The procedures for protein digestion were performed according to the
   filter-aided sample preparation (FASP) technique^[253]64 with slight
   modifications. Briefly, FASP digestion was adapted for the following
   procedures in Microcon PL-10 filters. Following a three-time buffer
   displacement with 4 M guanidine hydrochloride and 100 mM Tris-HCl, pH
   8.0, proteins were reduced using 10 mM DTT at 37 °C for 30 min,
   followed by alkylation with 30 mM iodoacetamide at 25 °C for 45 min in
   the dark. Digestion was carried out with trypsin (enzyme:protein =
   1:50) at 37 °C overnight after washing with 20% acetone nitrile and
   digestion buffer (30 mM Tris-HCl, pH 8.0). After digestion, the
   solution was filtrated out, the filter was washed with 15% ACN, and all
   filtrates were pooled, vacuum-dried, and kept at –80 °C until use.

Carbamyl peptide enrichment

   Carbamyl peptide enrichment was performed using pan-anti-carbamyl
   lysine antibody (Abcam, Cat# ab175132) conjugated with protein A beads
   (Sigma, Cat# 16-156). Peptides were mixed with antibody-conjugated
   beads at 4 °C for 6–8 h under rotary shaking. The mixture was briefly
   centrifuged at 1000× g (< 10 s), and the supernatant was carefully
   removed. Antibody-conjugated beads were subsequently washed thrice with
   phosphate buffered saline (PBS) buffer. Finally, the bound carbamyl
   peptides were eluted with 30% ACN/0.1% TFA and concentrated in a
   SpeedVac. Before MS analysis, all carbamyl peptides were desalted.

LC-MS/MS analysis

   LC-MS analysis was performed using a nanoflow EASYnLC 1200 system
   (ThermoFisher Scientific, Odense, Denmark) coupled to an Orbitrap
   Fusion Lumos mass spectrometer (ThermoFisher Scientific, Bremen,
   Germany). A one-column system was adopted for all analyses. Samples
   were analyzed on a C18 analytical column (75 µm i.d. ×25 cm,
   ReproSil-Pur 120 C18-AQ, 1.9 µm (Dr. Maisch GmbH, Germany)) constructed
   in-house. The mobile phases consisted of Solution A (0.1% formic acid)
   and Solution B (0.1% formic acid in 80% ACN). The derivative peptides
   were eluted using the following gradients: 2%–5% B in 2 min, 5%–35% B
   in 100 min, 35%–44% B in 6 min, 44%–100% B in 3 min, 100% B for 10 min,
   at a flow rate of 200 nL/min. Data-dependent analysis was employed in
   MS analysis: the time between master scan was 3 s, and fragmented in
   HCD mode, normalized collision energy was 30.

Data processing and analysis

   Raw MS files were searched using Protein Discoverer (version 2.4,
   ThermoFisher Scientific) with Mascot (version 2.7.0, Matrix Science).
   Data were processed using the UniProt mouse protein database (55,462
   entries). The mass tolerances were 10 ppm for precursor and fragment
   mass tolerance 0.05 Da. Up to two missed cleavages were allowed. The
   search engine set protein N-acetylation, pyroglutamate on peptide N
   termini, oxidation on methionine, and carbamylation on Lys as variable
   modifications and cysteine carbamidomethylation as a fixed
   modification.

Recombinant SIRT4 expression and purification

   Recombinant SIRT4 expression was performed as previously described with
   slight modifications^[254]65. The NAD^+-dependent deacetylase domain
   encompassing residues 27–314 of SIRT4 was subcloned into a pSUMOH10
   vector (modified based on pET28b) containing an N-terminal 10× HIS-SUMO
   tag. SIRT4^27–314 were expressed in Escherichia coli BL21 (DE3) strain
   (Transgene, China). After induction with 0.2 mM isopropyl
   β-D-thiogalactoside at 16 °C in LB medium supplemented with 0.1 mM
   ZnCl[2] overnight, cells were harvested by centrifugation at 4000× g
   for 20 min at 4 °C. The pellets were resuspended in buffer A containing
   20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 5% glycerol, and 20 mM imidazole
   and were then disrupted using an EmulsiFlex-C3 homogenizer (Avestin) at
   4 °C. The lysate was further cleared by centrifugation at 14,000× g for
   60 min at 4 °C, and the supernatant was loaded onto a His-trap affinity
   column (GE Healthcare). After washing with buffer A, bound proteins
   were subjected to on-column cleavage by ULP1 SUMO protease. The
   tag-free SIRT4^27–314 protein was collected as flow-through,
   centrifuge-concentrated, and then purified on an anion-exchange QHP
   column followed by size exclusion chromatography on a Superdex 75
   column (GE Healthcare). Purified peak fractions were pooled,
   concentrated, aliquoted, and stored at −80 °C for future use.

MS-based in vitro decarbamylation assays

   The ability of SIRT4 to hydrolyze carbamyl lysine was measured using
   MALDI TOF/TOF. Carbamyl peptide (200 µM) was incubated with 5 µM SIRT3,
   SIRT4, or SIRT5 in reaction buffer containing 50 mM Tris-HCl, pH 7.8,
   1 mM DTT, and 1 mM MgCl[2] without or with 2 mM NAD^+ for 2 h at 37 °C.
   Before MS analysis, all carbamyl peptides were desalted and eluted
   using C18 tips. Samples were analyzed using a 4800 Plus MALDI TOF/TOF
   Analyzer (Applied Biosystems/MDSSCIEX) and operated in Reflector
   Positive mode using 4000 Series Explorer Software. The laser intensity
   was set at 3500 V.

SDS-PAGE and western blotting assay

   SDS-PAGE and western blotting assay were performed following standard
   protocols. Briefly, cells were harvested with loading buffer containing
   50 mM Tris-HCl pH 6.8, 10% glycerol (v/v), 2% SDS (w/v), 4%
   β-mercaptoethanol (v/v), and 0.0012% bromophenol blue (w/v). For
   western blotting analysis, each sample was subjected to SDS-PAGE and
   transferred to nitrocellulose membranes (GE Healthcare Life Science).
   The membranes were blocked in 5% (w/v) skim milk in Tris-buffered
   saline with 0.1% (v/v) Tween-20 (TBST) for 1 h at room temperature and
   were then probed with primary antibodies in antibody dilution buffer
   (QuickBlock^TM, Beyotime) at 4 °C overnight. After incubation with
   horseradish peroxidase-conjugated secondary antibodies in TBST
   (containing 5% skim milk), membranes were developed using ECL-Plus
   (Thermo Fisher Scientific) and visualized using Typhoon (GE Healthcare
   Life Science).

Protein half-life assays

   Cells were treated under the indicated conditions. For half-life
   studies, cycloheximide (100 mg/mL, Sigma) was added to the medium.
   Cells were collected at the indicated time points and protein
   concentrations were measured, then analyzed by immunoblot analysis.

ChIP assays

   ChIP assays were conducted using the ChIP Assay Kit (Beyotime). First,
   U87MG and HeLa cells treated with or without BHB were crosslinked by 1%
   formaldehyde for 10 min. DNA was then sonicated into fragments with a
   mean length of 200–500 bp, and sheared chromatin was immunoprecipitated
   with antibodies against H3K27Ac (Abcam) or non-specific rabbit IgG
   (Santa Cruz) overnight at 4 °C. The precipitated DNA fragments were
   then identified by PCR and quantified by quantitative real-time PCR
   (qRT-PCR).

RNA isolation and qRT-PCR analyses

   RNA was extracted from cells using TRIzol (Transgene, China) according
   to the manufacturer’s instructions, and the concentration/purity were
   determined using NanoDrop Spectrophotometer (ThermoFisher Scientific).
   cDNA was synthesized using the HiScript III 1st Strand cDNA Synthesis
   Kit (Vazyme) according to the manufacturer’s instructions. qRT-PCR was
   performed with SYBR qPCR Master Mix (Vazyme) on a Real-Time
   Thermocycler (BioRad). qPCR analyses in cells or tissues are relative
   to β-actin. The primer sequences are listed in Supplementary Table
   [255]S1.

Mitochondrial isolation

   Cells were washed with prechilled PBS, harvested with isolation buffer
   (20 mM HEPES, pH 7.4, 140 mM KCl, 10 mM EDTA, 5 mM MgCl[2] with a
   protease and phosphatase inhibitor cocktail), and homogenized using a
   chilled glass-Teflon homogenizer with 20 strokes. The homogenate was
   centrifuged twice at 700× g for 10 min at 4 °C. The supernatant was
   transferred to a new tube and centrifuged at 9000× g for 30 min at 4 °C
   to obtain the mitochondrial pellet. The pellet was collected and washed
   thrice with washing buffer (20 mM HEPES, pH 7.4, 800 mM KCl, 10 mM
   EDTA, and 5 mM MgCl[2]).

Histone extraction

   Cells were washed twice with PBS and resuspended in 200 μL of buffer A
   (10 mM HEPES (pH 7.9), 10 mM KCl, 1.5 mM MgCl[2], 0.34 M sucrose, 10%
   glycerol, 1 mM DTT, 0.1% Triton X-100, and protease inhibitor mixture
   (Roche Molecular Biochemicals). After incubation for 5 min on ice, the
   nuclei were collected in the pellet by low-speed centrifugation
   (1500× g, 4 min, 4 °C). The nuclei were washed once with buffer A
   without 0.1% Triton X-100 and then lysed in 200 μL of buffer B (3 mM
   EDTA, 0.2 mM EGTA, 1 mM DTT, and protease inhibitor mixture). After
   10 min incubation on ice, soluble histones were separated from
   chromatin by centrifugation (2000× g, 4 min).

RNA interference

   Small RNA interference and stable knockdown were used in this study. In
   RNA interference, double-stranded siRNAs targeting GDH and BDH1,
   respectively, were purchased from GenePharma and transfected into cells
   via RNAiMax (Invitrogen) according to the manufacturer’s instructions.
   siRNA transfections were performed with Lipo2000 (Life technology)
   according to the manufacturer’s instruction. Cells were cultured in
   six-well plate (ThermoFisher Scientific) and siRNAs were used at a
   concentration of 20 nM. Cells were harvested at 48 h post-transfection.

   Stable shRNA knockdown cells were generated by co-transfecting cells
   with pCMV-VSV-G, pCMV-Gag-Pol, and shRNA plasmids by the calcium
   phosphate method. DMEM containing 10% FBS was used to culture the cells
   for 6 h after the transfection. After 24 h of transfection, supernatant
   of the cultured medium was collected and used as retrovirus preparation
   to infect cells at a density of 10% confluence in 90 mm dishes. Cells
   were re-infected 48 h after the initial infection using 5 μg/mL
   puromycin (Amresco) for selecting cells. All the sequences of
   siRNA/shRNA are listed in Supplementary Table [256]S1. The knock-down
   efficiency was verified by qPCR or western blotting assay.

Measurement of GDH enzyme activity

   Fresh tissues were weighed and transferred to solubilization buffer (1%
   Triton X-100, 150 mM NaCl, 10 mM KCl, 5 mM EDTA, 50 mM Tris-HCl, pH
   8.0) containing protease and phosphatase inhibitors in 5 µL of
   extraction buffer per mg tissue and homogenized using ceramic beads.
   The samples were centrifuged at 14,000× g for 15 min (4 °C). The
   supernatant was transferred to a new tube for immunoprecipitation and
   enzymatic activity measurements, and protein concentrations were
   determined using a BCA protein assay kit (Beyotime).

   The enzyme activity of GDH were determined by using a Glutamate
   Dehydrogenase Activity Assay Kit (Sigma). In short, the GDH protein was
   mixed with the assay buffer and incubated for 2 min at 37 °C in
   darkness. The change in absorbance at 450 nm was measured every 30 s
   for 10 min at 37 °C in a SpectraMax M5 Microplate Reader (Molecular
   Devices). Each sample was analyzed in triplicate, and the enzyme
   activity of GDH was calculated following the manufacturer’s
   instruction.

In vitro GAD1 enzymatic assay

   Pre-warmed (37 °C) assay buffer (50 mM KPO[4] pH 8.0, 0.2 mM PLP, 1 mM
   AET, 100 μg/mL BSA) containing 20 mM sodium glutamate was added to test
   tube containing the different cell lysates in ratio 1:1. After 120 min
   of incubation at 37 °C, samples were immediately placed in ice to stop
   the reaction and GABA was immediately quantified. Each determination
   was performed in triplicates and the whole experiments were repeated
   twice.

BHB treatment and measurement

   Mice were administered with BHB esters (HVMN), 200 µL by oral gavage
   daily, 80 mg per mouse.

   Levels of BHB in serum and brain were determined by using a BHB Assay
   Kit (Sigma) according to the manufacturer’s instruction. In short, the
   samples were mixed with the assay buffer and incubated the reaction for
   30 min at room temperature in darkness. Absorbance was measured at
   450 nm (A450) in a SpectraMax M5 Microplate Reader (Molecular Devices).
   Each sample was analyzed in triplicate.

Measurements of the concentrations of AcAc and acetone

   Serum and brain AcAc levels were measured using the Ketone Body Assay
   Kit (Abnova, KA1630) according to the manufacturer’s instructions. The
   acetone concentrations were measured using the Goat Acetone Elisa Kit
   (MyBioSource, MBS9310393) according to the manufacturer’s instructions.

Immunofluorescence analysis

   Animals were deeply anesthetized with vaporized sevoflurane and
   perfused with 50 mL saline, followed by 50 mL 4% PFA. The brains were
   extracted and soaked in 4% PFA at 4 °C overnight, cryoprotected by
   transferring to 30% sucrose solution until the brains were saturated.
   Coronal brain sections (40 µm) were cut using a freezing microtome
   (CM1950, Leica). The slices undergoing immunohistochemical staining
   were washed in PBS three times (5 min each time) and pre-incubated in
   PBS containing 5% normal goat serum and 0.3% Triton X-100 for 1 h at
   room temperature, before incubating with the following primary
   antibodies at 4 °C overnight: rabbit anti-Cre recombinase (1:1000,
   15036, Cell Signaling, USA). Then, the sections were washed in PBS
   three times (5 min each time) and incubated with a secondary antibody
   at 37 °C for 1 h. The slices were washed in PBS thrice (5 min each
   time). Finally, all of the slices were mounted onto glass slides,
   dried, and covered with Antifade Mounting Medium (Solarbio, S2100,
   China).

Quantification and statistical analysis

   Statistical tests were performed using GraphPad Prism 8 software and
   two-tailed unpaired Student’s t-test was used to compare two groups.
   All data are presented as the means ± standard deviation (SD). P values
   were calculated assuming a normal distribution and were corrected for
   multiple hypotheses using the Benjamini–Hochberg procedure. P values
   are denoted in figures as: ns, P > 0.05; *P < 0.05; **P < 0.01;
   ***P < 0.001; ****P < 0.0001.

Supplementary information

   [257]Supplementary information^ (1.9MB, pdf)

Acknowledgements