Ketogenic diet dampens excitatory neurotransmission by shrinking synaptic vesicle pools

TL;DR

A ketogenic diet changes brain activity by altering gene expression and reducing excitatory signals in the hippocampus, which may explain its benefits in epilepsy and neurodegenerative diseases.

Contribution

The study reveals that KD causes transcriptional reprogramming in the hippocampus via histone modifications, leading to synaptic adaptations.

Findings

KD induces extensive transcriptional reprogramming, including altered expression of synaptic genes.

KD reduces excitatory synaptic gain and short-term plasticity at CA3-CA1 synapses.

Functional changes are driven by a reduction in the readily releasable vesicle pool at excitatory synapses.

Abstract

Ketogenic diet (KD) is used for the treatment of drug-resistant childhood epilepsy and has been proposed to improve outcomes in neurodegenerative diseases. However, the mechanisms by which KD alters brain circuitry remain unclear. Here, we investigated the impact of KD on hippocampal function through integrative analysis of gene expression and neurotransmission. We found that KD induces extensive transcriptional reprogramming, including altered expression of numerous synaptic genes. Proteomic and genomic profiling revealed significant changes in histone modifications, particularly at promoters of KD-regulated genes. Electrophysiological recordings showed that KD reduces excitatory synaptic gain and short-term plasticity at CA3-CA1 synapses, dampening the summation of excitatory inputs and enhancing the summation of inhibitory inputs. These functional changes were driven, in part, by a reduction in the readily releasable vesicle pool at excitatory synapses under KD. Together, our findings demonstrate that KD drives transcriptional remodeling of hippocampal circuits, leading to synaptic adaptations that may underlie its anti-epileptic and neuroprotective effects.

Stunault et al. show that the ketogenic diet drives extensive transcriptional reprogramming of the hippocampus through histone post-translational modifications, leading to altered short-term plasticity and reduced synaptic integration at the circuit level.

Figures

D](#F1)), a similar fraction (58.7%) was downregulated (Figure S2B), consistent with the pattern observed in adult mice ([Figure 1D](#F1)). The downregulated genes in young KD mice include *Npas4* and *Kcnj2*, which encode an activity-dependent transcription factor^[24](#R24)^ and an inward rectifying potassium channel, respectively^[25](#R25)^ (Figure S2C; Table S1). Pathway analysis of downregulated DEGs in young mice revealed that the KD affects neurogenesis, fatty acid metabolism, and synaptic transmission (Figure S2D; Table S1). To assess the degree of conservation in KD-induced transcrip

D](#F2)). Interestingly, H2A.Z preferentially regulates the expression of synaptic genes, and its acetylation is implicated in hippocampal memory formation.^[32](#R32)^ Here, we showed that KD significantly increased the enrichment of H3K4me1 at promoters of both downregulated and upregulated DEGs, consistent with previous reports.^[38](#R38),[39](#R39)^ This finding strongly implicates K4 mono-methylation in KD-induced transcriptional regulation. Altogether, our results suggest that KD engages complex epigenetic changes across many genomic loci^[30](#R30)^ whose specific regulatory roles rema

E](#F3)) despite similar baseline ([Figure 3F](#F3)). Analysis of cumulative release ([Figure 3G](#F3)) revealed a significantly smaller functional RRP size at excitatory CA3-CA1 synapses in KD mice ([Figure 3H](#F3)), consistent with our hypothesis. Structurally, the RRP consists of synaptic vesicles docked at the active zone (AZ).^[44](#R44)^ To confirm KD-induced reduction in RRP size, we used electron microscopy and examined synaptic vesicle populations in hippocampal slices from chow-fed and KD mice ([Figures 3I](#F3) and [3J](#F3)). Quantification of vesicle populations in excitatory CA3

A](#F4)). The activation of CA1 pyramidal neurons also provides excitatory drive to the local interneurons that produce feedback inhibition ([Figure 4A](#F4)). KD-induced dampening of excitatory gain in CA1 synapses and the resulting changes in excitability of CA1 neurons may therefore alter both the excitatory and inhibitory components of the circuit and change the excitation/inhibition circuit balance in a complex way. To examine the combined circuit-level effects of KD on CA1 pyramidal cell excitability, we examined the summation of excitatory post-synaptic potentials (EPSPs) in CA1 pyramid

Full text

INTRODUCTION

Ketogenic diets (KDs) are high-fat, low-carbohydrate, and moderate-protein dietary regimens that have been used for nearly a century to treat intractable childhood epilepsies, which account for approximately 30% of all pediatric epilepsy cases.^1^ More recently, KDs have been shown to improve disease outcomes in various neurological disorders.^2-4^ KDs induce a metabolic state known as ketosis^5,6^ in which the liver releases ketones, mainly β-hydroxybutyrate (BHB) and acetoacetate, into the bloodstream.^7^ In patients placed on KD, the gradual increase in circulating ketones coincides with seizure control, which starts within days of the diet,^8^ reaching full effects within a few months.^9^ The clinical effectiveness of KDs is limited by poor adherence, highlighting the need for alternative therapies that capture their neuroprotective mechanisms. However, the cell and molecular mechanisms underlying the effects of KD remain poorly understood.

The metabolic switch to ketosis exerts both acute and chronic effects on the brain, altering metabolism and neuronal activity over different time scales.^10,11^ Ketones acutely suppress excessive firing by opening the K_ATP_ channels to trigger membrane hyperpolarization^12^ and by inhibiting glutamate loading into synaptic vesicles,^13^ thus dampening excitatory neurotransmission. Unlike these acute effects, the long-term effects of KDs on the brain are not fully understood. A handful of studies suggest that KDs or ketone ingestion induces the expression of brain-derived neurotrophic factor,^14^ and alters steady-state levels of brain neurotransmitters, including GABA and glutamate.^15,16^ Ketones also function as signaling molecules,^17^ and KD profoundly alters gene^18^ and protein^10^ expression across the brain, but the physiological significance of these changes for neurotransmission and circuit function remains largely unknown.

Here, we examined the effects of KD on the hippocampus, an area of the brain that is frequently involved in epileptic seizures^19^ and is highly responsive to KDs.^20,21^ We found significant downregulation of synaptic gene expression and major changes in histone modifications in the hippocampus, resulting in profound structural remodeling of hippocampal synapses, impacting short-term plasticity and summation of excitatory and inhibitory synaptic inputs, which likely contribute to the anti-epileptic properties of the diet.

RESULTS

Short-term KD alters the expression of synaptic genes in mouse hippocampus

Metabolic adaptations to nutrient availability often involve changes in gene expression patterns;^22^ therefore, we investigated transcriptional rewiring of the hippocampus in response to the KD. Although the effects of KD on gene expression are often studied after several months of the diet, we hypothesized that transcriptional rewiring would emerge early during ketosis. To determine the onset of ketosis, we monitored blood BHB levels in 10-week-old C57BL/6NJ male mice placed on a KD (see STAR Methods for diet composition) (Figure 1A). The blood BHB level increased from 0.4 to 1.0 mM within the first week of the KD and stabilized after 3–4 weeks of the diet, indicating the establishment of a state of ketosis (Figure 1B). However, BHB levels did not change in mice placed on a chow diet. Consistent with previous reports,^23^ the body weight of KD mice remained lower than that of chow-diet mice over the duration of the diet (Figure S1A).

We then performed RNA sequencing on the hippocampi of 12-week-old (adult) male mice placed on a KD or chow diet for 4 weeks. The postmortem analysis of serum metabolites revealed similar glucose and fatty acid levels in both dietary groups, while BHB and triglyceride levels were significantly higher in KD mice (Figures S1B-S1E). Analysis of differentially expressed genes (DEGs) revealed a transcriptionally distinct group of KD mice relative to chow-fed controls, demonstrating diet-driven alterations in gene expression (Figure 1C). We detected 503 DEGs between chow- and KD-fed mice (false discovery rate threshold [FDR] < 0.05), with 308 (61.2%) downregulated and 195 (38.8%) upregulated (Figures 1D and 1E). Pathway analysis uncovered significant enrichment of Gene Ontology (GO) cellular components related to neuronal synapses, specifically, excitatory synapses, dendritic spines, and other synaptic signaling components, as well as histone modifying enzymes (Figures 1F and S1L; Table S1). Downregulated DEGs in KD mice were predominantly associated with synapses, post-synaptic density, and spines, as well as extracellular matrix and astrocyte projections (Figures 1G and S1M; Table S1).

Given that the KD is used for the treatment of pediatric epilepsies, we also examined hippocampal gene expression in young (4-week-old) mice placed on KD. As with adult mice, young KD mice consistently had lower body weights and higher BHB content in blood compared to chow-fed mice (Figures S1F and S1G). Analysis of serum metabolites after 3 weeks on the diet confirmed higher BHB and lower glucose levels in young KD mice as compared to chow-diet mice (Figures S1H-S1K). Although fewer DEGs were identified in young KD mice (179) (Figures S2A-S2C) compared to adults (503) (Figure 1D), a similar fraction (58.7%) was downregulated (Figure S2B), consistent with the pattern observed in adult mice (Figure 1D). The downregulated genes in young KD mice include Npas4 and Kcnj2, which encode an activity-dependent transcription factor^24^ and an inward rectifying potassium channel, respectively^25^ (Figure S2C; Table S1). Pathway analysis of downregulated DEGs in young mice revealed that the KD affects neurogenesis, fatty acid metabolism, and synaptic transmission (Figure S2D; Table S1). To assess the degree of conservation in KD-induced transcriptional responses across age groups, we constructed a dot plot illustrating the common downregulated GO biological process pathways in adult and young mice (Figure S2E). Both age groups showed changes in the expression of genes associated with neurogenesis, learning and memory, and extracellular matrix organization. We validated some of the RNA sequencing findings with the qPCR of transcript levels in young mice, confirming the reduced expression of Gria1,^26^ which encodes the GluA1 subunit of the AMPA glutamate receptor (Figure S2F), as well as increased Gabrg2 expression (Figure S2G), which encodes the γ-subunit of the GABA_A_ receptor, and is implicated in epilepsy.^27^ Similarly, qPCR analysis confirmed reduced Kcnj2 (an inward rectifying potassium channel) expression in adult mice (Figure S2H). These findings suggest that KD may alter the excitation and inhibition balance in the hippocampal circuits. Taken together, these results suggest that KD triggers broad remodeling of synaptic transcriptional programs in the hippocampus that likely impacts neurotransmission.

KD regulates hPTMs in the hippocampus

Recent studies indicate that ketone bodies can serve as substrates or cofactors for histone post-translational modifications (hPTMs),^28^ linking metabolic state to gene expression.^29^ To assess how KD impacts hippocampal histone modifications (hereafter referred to as epigenetic regulation), we performed liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) to analyze the relative abundance of hippocampal hPTMs. Four-week-old male mice were placed on chow or KD diet for 3 weeks, and their hippocampi were collected for histone extraction (Figure 2A). Agglomerative hierarchical clustering of histone peptides revealed a clear separation of hPTM enrichment profiles in the hippocampi of chow versus KD mice (Figure 2B). The average relative abundance of 22 histone peptides (7 down and 15 up) was significantly altered in KD mice as compared to chow-diet mice (Figure 2C). These included hPTMs (both acetylation and methylation) of canonical histone H3, which have been tightly linked to gene regulation^30^ as well as the overall levels and PTMs of several histone variants (Figure 2D). Specifically, we observed increased acetylation of the H2A.Z variant (Figure 2D), which has been linked to hippocampal gene expression and long-term memory formation.^31,32^

Importantly, we found that KD altered the levels of both activating and repressive histone marks in the hippocampus. For example, the average relative abundance of H3K4ac, a hPTM enriched at active gene promoters,^34^ was significantly reduced in KD mice (Figures 2D and S3A), which could contribute to the downregulation of specific genes by KD (Figures 1 and S2). Similarly, the significant upregulation of repressive marks H3K79me1^35^ and H3K9me3^36^ (Figure 2D) could contribute to gene repression in KD. On the other hand, increased expression of genes might be linked to loss of repressive marks, such as H3K9me1K14ac (Figure 2D), which has been implicated in the silencing of active genomic regions by recruiting the repressive K9 methyltransferase SETDB1.^37^ We also observed a trend for increased H3K4me1 in KD (Figure S3B), which has been linked to both active transcription^38^ and gene repression.^39^

While our findings outlined global changes in hippocampal histone modifications linked to KD, the expression of specific genes is more closely linked to local epigenetic landscapes at gene promoters and transcription start sites.^30^ Thus, to assess how KD impacts histone modifications specifically at DEGs, we then performed chromatin immunoprecipitation coupled with high-throughput sequencing (ChIP-seq). We targeted H3K4me1 given its well-known role in transcriptional regulation.^30,38,39^ We found that the enrichment of H3K4me1 was significantly increased at the promoters of both upregulated (Figure 2E) and downregulated (Figure 2F) DEGs in KD mice, suggesting that H3K4 mono-methylation might be an important epigenetic mechanism underlying KD-induced transcriptional outcomes. Emphasizing specificity, H3K4me1 was not affected by KD at a similar set of genes that were not differentially expressed (Figure 2G).

Lysine β-hydroxybutyrylation (K-BHB) is another histone modification that has emerged in metabolic regulation of gene expression.^40^ We examined K-BHB modification of histones by immunoblotting histone extracts from chow- and KD-fed hippocampi with a pan-K-BHB antibody, which detected a band corresponding to histone H3. We found that K-BHB modification of histone H3 was significantly increased in KD mice when normalized to total histone H3 levels (Figures S3C and S3D), suggesting that this modification may regulate hippocampal gene expression in response to KD. In summary, our results show that KD has a marked effect on the hippocampal epigenome, consistent with the global transcriptional effect of KD in the hippocampus.

KD dampens synaptic gain and reduces the vesicle pools in excitatory hippocampal synapses

Changes in synaptic gene expression under KD suggest effects on hippocampal synaptic transmission. We performed electrophysiological recordings in acute slices prepared from chow-diet or KD mice to investigate excitatory neurotransmission in the hippocampal CA3-CA1 synapses (Figure 3A). We particularly focused on the rapid activity-dependent dynamics of neurotransmission, known as short-term plasticity (STP), which plays an essential role in information processing.^41,42^ Whole-cell patch-clamp recordings from CA1 pyramidal neurons were performed to assess synaptic dynamics during high-frequency (40 Hz) trains at near-physiological temperatures (33° C–34° C) and in the presence of gabazine to block inhibitory responses (Figure 3B). Excitatory post-synaptic currents (EPSCs) during stimulus trains were normalized to an average of five low-frequency (0.2 Hz) controls preceding each train, thus representing relative changes in synaptic strength, i.e., synaptic gain. We observed that synaptic gain was significantly reduced during 40 Hz trains (Figures 3B and 3D), which was not due to differences in stimulation intensity since the baseline EPSC amplitudes were comparable between groups (Figure 3C).

Rapid changes in synaptic gain during stimulus trains are driven by the interplay of several counteracting STP processes, namely facilitation and short-term depression. We found no significant changes in facilitation, as measured with the paired-pulse protocol in KD mice (Figures S4A-S4C). The attenuation of synaptic gain that we observed in KD mice could thus arise from increased short-term depression, a fundamental mechanism of STP, which is determined predominantly by the availability of synaptic vesicles for release. Therefore, we hypothesized that a decrease in the readily releasable pool (RRP) of vesicles docked at the plasma membrane and immediately available for release could be responsible for the diminished STP in KD mice. To obtain a functional estimate for the RRP size, we used a well-established approach based on the linear back-extrapolation of the cumulative steady-state EPSC amplitude at the end of a prolonged high-frequency train, which has been shown to be proportional to the RRP size.^43^ We thus applied a train of 150 stimuli at 100 Hz to evoke steady-state depression and observed stronger depression in KD compared to chow-diet mice (Figure 3E) despite similar baseline (Figure 3F). Analysis of cumulative release (Figure 3G) revealed a significantly smaller functional RRP size at excitatory CA3-CA1 synapses in KD mice (Figure 3H), consistent with our hypothesis. Structurally, the RRP consists of synaptic vesicles docked at the active zone (AZ).^44^ To confirm KD-induced reduction in RRP size, we used electron microscopy and examined synaptic vesicle populations in hippocampal slices from chow-fed and KD mice (Figures 3I and 3J). Quantification of vesicle populations in excitatory CA3-CA1 synapses within stratum radiatum revealed a significant reduction in the number of docked vesicles (defined as those localized within 50 nm of AZ plasma membrane^45^) as well as total vesicle pool in KD mice (Figures 3K and 3L). In contrast, the number of docked and total vesicles in inhibitory synapses was not significantly affected (Figures S4D and S4E). The AZ size and the number of synapses also remained unchanged for both excitatory and inhibitory hippocampal synapses in KD (Figures 3M, 3N, S4F, and S4G). Altogether, our results suggest that KD causes a selective reduction of vesicle pools in excitatory synapses, which leads to diminished synaptic gain and STP of excitatory neurotransmission.

KD alters the summation of synaptic inputs in the hippocampal circuit by remodeling excitatory and inhibitory pathways

In addition to directly targeting CA1 pyramidal neurons, Schaffer collateral inputs from the CA3 project onto CA1 inhibitory interneurons, which in turn provide feedforward inhibition onto CA1 pyramidal cells (Figure 4A). The activation of CA1 pyramidal neurons also provides excitatory drive to the local interneurons that produce feedback inhibition (Figure 4A). KD-induced dampening of excitatory gain in CA1 synapses and the resulting changes in excitability of CA1 neurons may therefore alter both the excitatory and inhibitory components of the circuit and change the excitation/inhibition circuit balance in a complex way. To examine the combined circuit-level effects of KD on CA1 pyramidal cell excitability, we examined the summation of excitatory post-synaptic potentials (EPSPs) in CA1 pyramidal cells evoked by trains of Schaffer collateral stimulation. This temporal summation is a critical determinant of neuronal excitability that integrates all of the direct excitatory and indirect inhibitory inputs in the local circuit, including both feedforward and feedback inhibition evoked by Schaffer collateral stimulation. A train of five stimuli at 40 Hz was used to examine EPSP summation in the absence of any glutamate or GABA receptor blockers to keep both excitatory and inhibitory pathways intact. We observed that KD significantly dampened the EPSP summation at 40 Hz (Figures 4B-4D). This was not due to the differences in the stimulation intensity, because the first EPSP amplitude was comparable in both conditions (Figure 4C). We further examined EPSP summation in response to 20 and 100 Hz stimulation. Interestingly, KD had no measurable effect on altering EPSP summation of these lower or higher frequency inputs (Figure 4E). These results suggest that the KD causes an overall reduction in excitatory-inhibitory balance in the CA3-CA1 circuit at a specific frequency range.

The net effect of KD on EPSP summation at different stimulation frequencies is determined by the complex interplay between the timing and dynamics of excitatory and inhibitory synaptic inputs in the local circuits. To better understand how this dynamic interplay contributes to the circuit effects of KD, we recorded the underlying excitatory and inhibitory post-synaptic currents (EPSC and IPSC) evoked in CA1 pyramidal cells by Schaffer collateral stimulation. EPSCs were recorded at −65 mV and IPSCs at 0 mV in the same cell and with the same stimulation intensity in both measurements. First, we examined baseline transmission at 0.2 Hz and found that KD had no effects on the kinetics of baseline EPSCs (Figures 4F-4K), while it significantly increased the rise time, peak time, and decay time of baseline IPSCs (Figures 4N-4P). These changes were not due to different stimulation intensities or locations, because the amplitudes and latencies were comparable between the groups for both EPSCs and IPSCs (Figures 4G, 4H, 4L, and 4M). Notably, IPSC latency was significantly longer than EPSC latency (Figures 4H and 4M), suggesting that IPSCs represent feedforward and feedback inhibition driven by SC stimulation (multi-synapse delay) rather than direct stimulation of inhibitory inputs. Second, we analyzed the dynamics of EPSCs and IPSCs during high-frequency trains. Interestingly, we found that while EPSC summation was significantly reduced in KD during 40 Hz stimulation, IPSC summation was concomitantly increased (Figures 4Q-4U). These opposite effects of KD on excitatory and inhibitory input summation combine in an additive manner to result in net attenuation of EPSP summation during 40 Hz stimulation (Figures 4B-4E). However, the combined dynamics of EPSC and IPSC summation at lower or higher frequencies resulted in a different net outcome. Specifically, we found that both EPSC and IPSC summations were slightly enhanced at 20 Hz in KD (Figures S4H, S4J, and S4L) (although not statistically significant for IPSCs), effectively canceling each other, while at 100 Hz stimulation, both EPSCs and IPSCs exhibit similarly strong summation that was largely unaffected by KD (Figures S4M-S4Q). Consequently, at both 20 and 100 Hz stimulation, the interplay of excitatory and inhibitory input dynamics results in no net effect of KD on EPSP summation (Figure 4E).

Taken together, our results suggest that, at the circuit level, KD alters the dynamics and interplay of the excitatory synaptic inputs and their dependent inhibitory pathways, resulting in dampened temporal summation in CA1 pyramidal cells at a specific frequency range.

DISCUSSION

The KD exerts protective effects against epilepsy and neurodegeneration, yet the mechanisms underlying these benefits are not fully understood due to the pleiotropic nature of metabolic reprogramming. Our study demonstrates that KD profoundly alters hippocampal synaptic gene expression, highlighting synapses as particularly sensitive targets of the diet. Although bulk RNA sequencing may obscure cell-type-specific effects, the strong enrichment of synaptic signatures in our dataset, together with recent cell-type-resolved proteomic studies,^10^ underscores a robust neuronal transcriptional response to KD. Given the high energetic demands of synapses, this preferential remodeling likely reflects adaptive responses to altered metabolic conditions. Many of the DEGs in KD hippocampi encode post-synaptic components, which may reflect homeostatic adaptations for maintaining network stability, and further studies are needed to determine how the KD impacts post-synaptic signaling and dendritic function.

Consistent with these transcriptional changes, we found broad modifications in histone post-translational marks within the hippocampus. Ketone bodies are implicated in diverse epigenetic mechanisms, including direct deposition on histones, such as K-BHB,^46^ as well as by regulating the enrichment of canonical histone marks.^47^ We found that K-BHB modification of histone H3 was modestly increased in the hippocampi of KD mice (Figures S3A and S3B), but it is not known how this modification impacts hippocampal transcription. Although we did not detect a global increase in histone acetylation, KD selectively enhanced acetylation of the histone variant H2A.Z (Figure 2D). Interestingly, H2A.Z preferentially regulates the expression of synaptic genes, and its acetylation is implicated in hippocampal memory formation.^32^ Here, we showed that KD significantly increased the enrichment of H3K4me1 at promoters of both downregulated and upregulated DEGs, consistent with previous reports.^38,39^ This finding strongly implicates K4 mono-methylation in KD-induced transcriptional regulation. Altogether, our results suggest that KD engages complex epigenetic changes across many genomic loci^30^ whose specific regulatory roles remain to be defined.

How do broad changes in synaptic gene expression ultimately shape hippocampal function? We examined excitatory neurotransmission in hippocampal CA3-CA1 synapses and found a decrease in synaptic gain during high-frequency stimulation along with a shift in STP toward increased short-term depression in KD animals. In the hippocampus, bursts of high-frequency firing generated by place cells are believed to carry spatial information about the environment^48,49^ and may also represent a response to various non-spatial sensory inputs.^50^ STP is generally considered to play a central role in these computations^41,51^ by modulating information processing and transfer.^41,42,52,53^ STP arises from the complex interplay among multiple, interdependent synaptic mechanisms, including synaptic facilitation driven by the accumulation of residual presynaptic calcium and opposing short-term depression driven primarily by the depletion of vesicles available for release.^54^ Our results suggest that reduction of synaptic gain and the shift in STP are caused primarily by smaller RRP size in KD mice rather than changes in facilitation, ultimately reducing glutamate release during high-frequency activity and dampening the responsiveness of CA1 neurons. At the circuit level, this results not only in altered summation of excitatory inputs but also concomitant changes in inhibitory input dynamics, thereby shifting the circuit balance between excitation and inhibition. Notably, the increased summation of IPSCs we observed in KD conditions would not arise trivially from the reduced synaptic drive from CA1 pyramidal cells onto local interneurons, suggesting additional mechanisms. Indeed, we observed slower kinetics of IPSCs in KD conditions, which may arise from a shift in the balance between feedforward versus feedback inhibition or from circuit adaptation engaging more distal inhibitory inputs. Given the complexity of local inhibitory networks and the immense cellular and functional diversity of inhibitory interneurons within the hippocampal circuit, additional studies are needed to investigate the remodeling of inhibitory pathways in KD. Overall, we observed that temporal summation in CA1 neurons, which combines both excitatory and inhibitory synaptic inputs, is dampened in KD mice, particularly in a frequency range relevant to the naturally occurring firing of place cells.^48,49^ This form of integration is a critical determinant of neuronal excitability, and its dampening in KD mice may thus represent a major mechanism contributing to the anti-epileptic effects of the KD. These results align with an earlier report describing a KD-induced shift toward a more inhibitory neural state.^15^ KD has also been suggested to affect long-term potentiation, although these studies yielded conflicting outcomes thus far.^55,56^ In contrast, our results identify STP and its presynaptic regulation as robust targets of KD.

Our morphological analyses revealed sustained reductions in vesicle pool size at excitatory synapses following prolonged ketosis in vivo. Although the molecular pathways underlying this remodeling remain unknown, our data suggest that KD induces lasting presynaptic adaptations that reshape hippocampal circuit function. Future studies are needed to define the molecular mechanisms linking metabolic state to presynaptic architecture and to determine whether similar adaptations occur across brain regions and epilepsy models.

Limitations of the study

While our study defines epigenetic, transcriptional, and functional adaptations of excitatory hippocampal synapses to the KD, it does not resolve how KD affects distinct neuronal subtypes, particularly diverse inhibitory interneurons, or how individual transcriptional changes causally drive synaptic and circuit-level remodeling. Future work using targeted epigenetic editing^57^ will be needed to link specific histone modifications to gene expression and synaptic function under KD.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animals and diets

All animal procedures were in compliance with the US National Institutes of Health Guide for the Care and Use of Laboratory Animals and conformed to Washington University Animal Studies Committee guidelines. Wild-type male C57BL/6NJ mice (Jackson laboratory, Strain #005304) of 4 and 12 weeks of age were either fed a chow diet (control; PicoLab Rodent Diet 20) or a high fat (75%), low carbohydrate (3%) Ketogenic Diet (KD, Bio-serv Cat# S3666) for 3–4 weeks ad libitum. Diet compositions are as follows: chow (11.3% fat, 53.4% carbohydrates, and 21% proteins), KD (75% fat, 3% carbohydrates, and 8.6% protein).Cage bedding was replaced in both groups by aspen chips to avoid any additional nutritional value, and a chewing bone was added in each cage to prevent tooth overgrowth. Weight was monitored once a week for the duration of the diet. Adult mice were fasted for 4 h prior to hippocampal collection for RNA sequencing. This fasting step was omitted for young mice because preliminary testing indicated that it induced excessive stress.

METHOD DETAILS

Blood collection and serum preparation

Glucose and BHB blood levels were monitored weekly and always at the same time of day for the duration of the diet to avoid circadian variability. One drop of blood (~20 μL) was collected from the lateral saphenous vein using a sterile 20G needle following mouse immobilization. Glucose and BHB levels were then immediately analyzed using the KetoMojo GK + Blood Glucose and β-ketone Meter and associated strips. For serum analysis, blood was collected by terminal submandibular bleeding on the day of euthanasia with a 21G x 1 ½ Needle. Mice were then immediately euthanized by CO2 overdose and cervical dislocation, and blood was left to clot for two hours at room temperature, and overnight at 4° C. Blood was then centrifuged for 10 min at 4° C at 2000xg and supernatant was collected for metabolite analysis.

Serum metabolite analysis

Total non-esterified fatty acid (NEFA), triglycerides (TG), and glucose levels were analyzed in the mouse serum by the Washington University NORC, Nutrition Obesity Research Center. Serum β-hydroxybutyrate levels were measured using the β-hydroxybutyrate colorimetric assay (Cayman chemicals, #700190) according to the manufacturer’s instructions. Briefly, sera and standards were incubated in the dark for 30min at room temperature with the corresponding reagent. Sample concentrations were determined from the corresponding standard curve and corrected for blanks.

Tissue harvesting

Following euthanasia, animal death was ensured by toe pinching. If no reaction was observed, the hippocampal region of the brain was harvested on ice. Hippocampi were then snap frozen or stored in RNAprotect Tissue Reagent (QIAGEN, #1017980) at −80° C until further utilization.

RNA extraction and cDNA synthesis

Frozen hippocampi were thawed on ice and 25-30mg of tissue was used for RNA extraction. Hippocampi were dissociated mechanically using a motor tissue grinder (Fisherbrand, #12-1413-61) and 1.5mL pestles- RNase-free (KIMBLE, #749521-1590) in lysis buffer (QIAGEN, #1015762) and β-mercaptoethanol. RNA extraction was performed using the RNeasy mini-kit (QIAGEN, #74104) according to manufacturer’s instructions. RNA libraries were prepared according to library kit manufacturer’s protocol, indexed, pooled, and sequenced on an Illumina NovaSeq 6000.

RT-qPCR assay and analysis

cDNA synthesis for qPCR was performed using ABM-All-In-One 5× RT Master Mix (ABM, #G592) according to manufacturer’s instructions. Real-time qPCR was performed on cDNA using TaqMan Fast Advanced Master Mix (Applied Biosystems, #4444965) and TaqMan probes for genes of interest. qPCRs were performed on the StepOne device (Applied Biosystem). All conditions were performed in triplicates. Results were normalized to GapdH for Kcnj2 and Gabrg2, and Hprt1 for Gria1, gene expression according to their respective amplicon length. Fold change in gene expression were normalized to chow diet controls.

Histone preparation

Fresh frozen hippocampus was used to extract the histones as previously described.^58^ Briefly, tissue was homogenized in 200 μL of Nuclear Isolation Buffer (NIB) (15 mM Tris-HCl, 15 mM sodium chloride, 60 mM potassium chloride, 5 mM magnesium chloride, 1 mM calcium chloride, and 250 mM sucrose at pH 7.5; with 0.5 mM 4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride (AEBSF), 10 mM sodium butyrate, 5 mM microcystin, and 1 mM dithiothreitol) and 10% NP-40. Homogenates were centrifuged at 600 rcf for 5 min at 4° C, and the supernatant was discarded. Nuclear pellets were resuspended in 200 μL NIB and centrifuged again under the same conditions. The resulting pellets were resuspended in 200 μL of 0.4 N sulfuric acid and incubated with rotation at 4° C for 4 h. Samples were centrifuged at 3,400 rcf for 5 min at 4° C, and the supernatant containing solubilized histones was collected. Histones were precipitated by adding 50 μL of 100% trichloroacetic acid (TCA) and incubating the samples on ice overnight at 4° C. Precipitates were washed sequentially with 0.1% HCl in acetone and then with 100% acetone (4,000 rcf, 2 min, 4° C). Pellets were air-dried for 20 min and resuspended in 30 μL of nuclease-free water. Protein concentrations of histone preparations were determined using the Qubit Protein Assay Kit (Life Technologies, Cat. No. Q33211).

Western blot

Protein samples were prepared in 1× sample buffer with reducing agent, denatured at 95° C for 10 min. Proteins were separated on 4–15% Mini-PROTEAN precast in Tris–Glycine–SDS buffer, then transferred onto 0.2 μm nitrocellulose membranes in chilled Tris–Glycine–methanol buffer. Membranes were washed in TBST, blocked in 5% milk/TBST for 1 h at room temperature, and incubated overnight at 4° C with anti–β-hydroxybutyryl-lysine antibody (1:2000; Cat. No. PTM-1201RM). Membranes were incubated with HRP-conjugated secondary antibody (1:6000; Invitrogen, Cat. No. A27036) for 1 h, washed, and developed using chemiluminescent substrate (Thermo Scientific, Cat. No. 34577) on an iBright FL1500 system. Membranes were stripped (0.5 M Tris-HCl, 10% SDS, 2-mercaptoethanol, deionized water) at 50° C for 20 min) and washed before reprobing with anti–histone H3 (Invitrogen, Cat. No. AHO1432) for loading control detection, followed by secondary incubation and imaging as above.

Band intensities were quantified in Fiji. A consistent region of interest was applied across lanes, and background was estimated from three random areas per image. Background-subtracted, inverted pixel densities were calculated, and β-hydroxybutyrylation signal for each sample was normalized to its corresponding histone H3 loading control.

Chromatin immunoprecipitation and sequencing

Chromatin immunoprecipitation followed by high throughput sequencing (ChIP-seq) was carried out using established protocols.^59^ Briefly, fresh frozen hippocampal tissue (both hemispheres) from each mouse (n = 4 for each chow and keto diet) was finely minced on cold plate, crosslinked in 1% formaldehyde in PBS for 10 min at RT, and the reaction was quenched using 125 mM glycine for 5 min at RT. Crosslinked tissues were washed with ice-cold PBS. The pelleted tissue was suspended in Sucrose/TKM buffer (0.25 M sucrose, 50 mM Tris-HCl (pH 7.5), 25 mM potassium chloride, 5 mM magnesium chloride, 1× Protease Inhibitors, 10 mM sodium butyrate) and subjected to dounce homogenization to isolate nuclei. Samples were centrifuged for 5 min at 5,000 rcf at 4° C and the pelleted nuclei were resuspended in nuclei lysis buffer (10 mM Tris-HCl, pH 8, 100 mM NaCl, 1mM EDTA, 0.5 mM EGTA, 0.1% sodium deoxycholate, 0.5% N-lauroylsarcosine). Chromatin was sheared to approximately 250 bp using a Covaris E220 sonicator (150 Watts peak incident power, 13% duty factor, 200 cycles/burst, 20 min/sample (1200 s/sample)). 0.1× volume of 10% Triton X- was added to the sonicated lysate to stabilize proteins. Equivalent quantities of sheared chromatin were incubated with Dynabeads protein G to generate bead-antibody complex overnight at 4° C with gentle rotation using H3K4me1 antibody (Abcam, Cat. No. 8895). Next day, the immunoprecipitated samples were washed with low salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.0, 150 mM sodium chloride), high salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.0, 500 mM sodium chloride), lithium chloride wash buffer (0.25 M lithium chloride, 1% CA-630, 1% sodium deoxycholate, 1 mM EDTA, 10 mM Tris-HCl pH 8.0) and 1× TE buffer (10 mM Tris-HCl pH 8.0, 1 mM EDTA). DNA was eluted from beads, followed by reverse crosslinking and DNA purification (QIAquick PCR Purification Kit Qiagen, Cat. No. 28106) together with input samples. The DNA was quantified using Qubit dsDNA high sensitivity assay kit (Cat. No. Q32851) and size distribution of ChIP DNA was assessed using an Agilent Bioanalyzer. Following quantification, the DNA samples were submitted to Genome Technology Access Center, McDonnell Genome Institute at Washington University School of Medicine for library preparation and sequencing. Briefly, 10ng of DNA was used for the library preparation. DNA fragments were end-repaired to generate blunt ends, followed by the addition of a single adenosine (A) residue to the 3′ ends. Illumina sequencing adapters were then ligated to the DNA fragments. Size selection was performed using AMPure XP beads to enrich for fragments between 150 and 700 bp. Adapter-ligated fragments were amplified by PCR for 15 cycles using primers containing unique dual index sequences. The resulting libraries were sequenced on an Illumina NovaSeq X Plus platform using paired-end 150 bp reads.

Mass spectrometry

Histone extraction and sample preparation for mass spectrometry were performed as previously described.^58^ Following the acetone wash (see under histone preparation), air-dried pellets were resuspended in 30 μL of 50 nM ammonium bicarbonate (pH 8.0). To derivatize primary and secondary amines, the histone sample was mixed with 15 μL derivatization mix (propionic anhydride and acetonitrile in a 1:3 ratio (v/v)), immediately followed by addition of 7.5 μL ammonium hydroxide to maintain pH 8.0. The sample was incubated for 15 min at room temperature, and the derivatization procedure was repeated once. Samples were then dried and resuspended in 50 mM ammonium bicarbonate and incubated with sequencing-grade modified trypsin (Promega, Cat. No. V5113) in a 1:20 enzyme:sample ratio overnight at room temperature. Following trypsin digestion, derivatization was performed twice to derivatize the N-termini of the peptides. Samples were desalted with C18 stage tips (made using Fisher Scientific, Cat. No. 13-110-019). An LC-MS/MS system consisting of a Vanquish Neo UHPLC coupled to an Orbitrap Q Exactive or Ascend (Thermo Scientific) was used for peptide analysis. Histone peptide samples were maintained at 7° C on a sample tray in LC. Separation of peptides was carried out on an Easy-Spray PepMap Neo nano-column (2 μm, C18, 75 μm × 150 mm) at room temperature with a mobile phase. The chromatography conditions consisted of a linear gradient from 2 to 32% solvent B (0.1% formic acid in 100% acetonitrile) in solvent A (0.1% formic acid in water) over 48 min and then 42–98% solvent B over 12 min at a flow rate of 300 nL/min. The mass spectrometer was programmed for data-independent acquisition (DIA). One acquisition cycle consisted of a full MS scan and 35 DIA MS/MS scans of 24 m/z isolation width starting from 295 m/z to 1100 m/z. Full MS scans were typically acquired in the Orbitrap mass analyzer across 290–1200 m/z at a resolution of 70,000 or 120,000 in positive profile mode with an injection time of 50 ms and an automatic gain control (AGC) target of 1.0E-06 or 200%. MS/MS data from higher energy collisional dissociation (HCD) fragmentation was collected in the Orbitrap. These scans typically used a nominal collision energy (NCE) of 30 or 25, AGC target of 1000%, and maximum injection time of 60 ms. EpiProfile 3^60^ was used to analyze histone MS data and calculate the ratio at which each modification was present for each peptide (relative abundance).

Electrophysiology

C57bl6/NJ male mice (4-week-old) were fed a chow diet or a KD for 3 weeks. Hippocampal slices were then prepared as previously described.^61^ Briefly, following deep CO_2_ anesthesia, mice were decapitated and brains were dissected out in ice-cold saline containing the following (in mM): 130 NaCl, 24 NaHCO_3_, 3.5 KCl, 1.25 NaH_2_PO_4_, 0.5 CaCl_2_, 5.0 MgCl_2_, and 10 glucose (replacing 5 mM glucose with equimolar BHB in the keto diet animals), pH 7.4 (saturated with 95% O_2_ and 5% CO_2_). Horizontal hippocampal slices (350 μm) were cut using a vibrating microtome (Leica VT1100S).^61^ Slices were initially incubated in the above solution at 35° C for 1 h for recovery and then kept at room temperature (~23° C) until use.

Whole-cell patch-clamp recordings using a MultiClamp 700B amplifier (Molecular Devices) in voltage-clamp or current-clamp mode were made from CA1 pyramidal neurons visually identified with infrared video microscopy (Olympus BX50WI; Dage-MTI) and differential interference contrast optics. All the recordings were conducted at near-physiological temperature (33° C–34°C). The recording electrodes were filled with the following (in mM): 130 K-gluconate, 0.1 EGTA, 2 MgCl_2_, 5 NaCl, 2 ATP-Na_2_, 0.4 GTP-Na, and 10 HEPES, pH 7.3. The extracellular solution contained the following (in mM): 130 NaCl, 24 NaHCO_3_, 3.5 KCl, 1.25 NaH_2_PO_4_, 2 CaCl_2_, 1 MgCl_2_, and 10 glucose (replacing 5 mM glucose with equimolar BHB in the keto diet animals), pH 7.4 (saturated with 95% O_2_ and 5% CO_2_). In all experiments, NMDA receptors were blocked with AP-5 (50 μm) to prevent long-term effects, unless stated otherwise. Excitatory synaptic gain in Figures 3B-3H only was recorded in the presence of 5 μM Gabazine to isolate excitatory responses. EPSCs (V-clamp, held at −65 mV), IPSCs (V-clamp, held at 0 mV) or EPSPs (I-clamp) were recorded from CA1 pyramidal neurons by stimulating Schaffer collaterals with a monopolar electrode positioned at least 300 μm away from the recorded cell. For recordings of EPSCs and IPSCs from the same CA1 cells, the stimulation intensity and location was kept the same for both measurements. All data were filtered at 2 kHz, digitized at 20 kHz, acquired using custom software written in LabView or Clampex 11.3, and analyzed using programs written in MATLAB. Sample size of recordings was based on our previous recordings, consistent with the literature.^61-63^

Imaging hippocampal slices with scanning electron microscopy

For sample analysis via Large-Area Scanning Electron Microscopy (LaSEM), mouse brain samples were fixed in a solution containing 2.5% glutaraldehyde and 2% paraformaldehyde in 0.15 M cacodylate buffer with 2 mM CaCl2 pH 7.4. 100 μm coronal vibratome sections of the brains were then stained according to the methods described by Deerinck et al.^64^ In brief, sections were rinsed in cacodylate buffer 3 times for 10 min each and subjected to a secondary fixation for one hour in 2% osmium tetroxide/1.5% potassium ferrocyanide in cacodylate buffer for one hour, rinsed in ultrapure water 3 times for 10 min each, and stained in an aqueous solution of 1% thiocarbohydrazide for one hour. After this, the tissues were once again stained in aqueous 2% osmium tetroxide for one hour, rinsed in ultrapure water 3 times for 10 min each, and stained overnight in 1% uranyl acetate at 4°C. The samples were then washed in ultrapure water 3 times for 10 min each and en bloc stained for 30 min with 20 mM lead aspartate at 60°C. After staining was complete, coverslips were briefly washed in ultrapure water (3 × 5 min), dehydrated in a graded acetone series (50%, 70%, 90%, 100% x2) for 10 min in each step, and infiltrated with microwave assistance (Pelco BioWave Pro, Redding, CA) into Durcupan resin. Samples were flat embedded between Teflon coated slides and cured in an oven at 60°C for 48 h.

Post resin curing, sample regions of interest (hippocampus) were selected, removed from the slide with a razor blade, and mounted on a blank block with cyanoacrylate. Thin sections 90 nm in thickness were cut and placed onto 10 mm square silicon wafer chips (Ted Pella, Redding, CA). These chips were then adhered to SEM pins with silver paint and large areas (~330 × 330 μm) were then imaged at high resolution in an FE-SEM (Zeiss Merlin, Oberkochen, Germany) using the ATLAS (Fibics, Ottowa, Canada) scan engine to tile large regions of interest. High-resolution tiles were captured at 16,384 × 16,384 pixels at 5 nm/pixel with a 5 μs dwell time and line average of 2. The SEM was operated at 8 KeV and 900 pA using the solid-state backscatter detector. Tiles were aligned and exported using ATLAS 5.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical tests used to measure significance, the corresponding significance level (p-value), and the values of n are provided for each panel in Table S2. Data are reported as mean ± SEM, and unless otherwise noted, p < 0.05 was considered statistically significant.

Analysis of electrophysiological data

EPCSs during the stimulus trains were normalized to an average of five low-frequency (0.2 Hz) control stimuli preceding each train, to give relative changes in synaptic strength. Each stimulus train was presented four to six times in each cell, and each presentation was separated by ~2 min to allow complete recovery to the baseline. To correct for the overlap of EPSCs at short interspike intervals (ISIs), a normalized template of EPSC waveform was created by averaging EPSCs from the same cell with ISI >100 ms. Every EPSC in the train was then approximated by a template waveform scaled to the peak of the current EPSC, and its contribution to the next EPSC was digitally subtracted. For determination of paired-pulse ratio (PPR), two EPSCs were recorded at an inter-pulse interval of 25 ms or 10 ms. PPR is the ratio of EPSC2 to EPSC1 amplitude. For measurement of EPSP temporal summation, EPSPs were recorded in response to a train of 5 stimuli at 20, 40 or 100 Hz. EPSP summation is expressed as the percentage increase in EPSP amplitude of 2^nd^ to 5^th^ EPSP relative to the 1^st^ EPSP, i.e., calculated as 100×[(EPSPn – EPSP1)/EPSP1], where EPSP1 is the amplitude of the 1^st^ EPSP and EPSPn is the amplitude of the n-^th^ EPSP in the train. For measuring baseline EPSC and IPSC kinetics, the stimulation frequency was 0.2 Hz. EPSC or IPSC latency was determined as the duration from stimulation start to the point of 10% peak amplitude. Rise time was determined as the duration from 10% to 90% peak amplitude. Peak time was determined as the duration from stimulation start to peak point. Decay time was determined as the duration from the peak point to the point of 37% peak amplitude. For EPSC and IPSC dynamics during train stimulation, we tested 20, 40 and 100 Hz and used the same quantification methods as those used for EPSP summation.

Analysis of CA3-CA1 synapses in electron micrographs

Individual synapses were selected from the stratum radiatum of the CA1 region of the hippocampus, as neurons in this layer are known to receive input from the Schaffer collaterals from the CA3 region. Both excitatory and inhibitory synapses with clearly visible active zone (AZ), were selected from slices prepared from three chow-diet (control) and three KD-fed mice for randomization and further analysis by experimenters blinded to sample type. Synapses showing small post-synaptic area (spine), and large post-synaptic density (PSD) (>30 μm) were identified as excitatory synapses; Inhibitory synapses were characterized by the absence of PSD (<5 μm) and large post-synaptic area (cell body or dendrite). Number of synapses, active zone length, number of “docked vesicles” within 50nm of the active zone, and total number of synaptic vesicles were determined using ImageJ.

RNA sequencing and analysis

Basecalls and demultiplexing were performed with Illumina’s bcl2fastq software and a custom python demultiplexing program with a maximum of one mismatch in the indexing read. RNA-seq reads were then aligned to the Ensembl release 101 primary assembly with STAR version 2.7.9a.^65^ Gene counts were derived from the number of uniquely aligned unambiguous reads by Subread:featureCount version 2.0.3.^66^ Isoform expression of known Ensembl transcripts were quantified with Salmon version 1.5.2.^67^ Sequencing performance was assessed for the total number of aligned reads, total number of uniquely aligned reads, and features detected. The ribosomal fraction, known junction saturation, and read distribution over known gene models were quantified with RSeQC version 4.0.^68^ All gene counts were then imported into the R/Bioconductor package EdgeR^69^ and TMM normalization size factors were calculated to adjust for samples for differences in library size. Ribosomal genes and genes not expressed in the smallest group size minus one sample greater than one count-per-million were excluded from further analysis.

Gene-level raw count matrices were provided separately for young and adult cohorts (10 samples per cohort; 5 control and 5 ketogenic diet). Duplicate Ensembl gene IDs were aggregated by summing counts across duplicates. The two cohorts were aligned to the identical set of 55,487 genes prior to filtering, and low-abundance genes were filtered prior to statistical testing: a gene was retained if it had ≥10 raw counts in ≥3 samples in either age group, yielding 22,934 tested genes used as the background “universe” for enrichment analyses.

Differential expression testing was performed within each age using DESeq2 (v1.42.1) comparing ketogenic diet (keto) versus control. Size-factor normalization and dispersion estimation followed DESeq2 defaults; Wald tests were used for inference. p values were adjusted for multiple testing using Benjamini–Hochberg; genes with FDR (padj) < 0.05 were considered differentially expressed.

Over-representation analysis of Gene Ontology (GO) terms was performed using clusterProfiler (v4.10.1, ‘enrichGO’) with Mus musculus annotations from org.Mm.e.g.,.db. Prior to enrichment, significant genes (padj <0.05) were split by direction (up vs. down) and by age. Gene symbols were mapped to Entrez IDs via org.Mm.e.g.,.db. The tested gene set after filtering (22,934 genes) served as the universe for all GO analyses, with enrichment parameters pAdjustMethod = “BH”, pvalueCutoff = 0.05, and qvalueCutoff = 0.1.

ChIP sequencing analysis

Bioinformatic analysis of ChIP-seq data was performed following the ENCODE consortium pipeline. Briefly, paired-end sequencing reads were aligned to the Mus musculus reference genome (mm10) using Bowtie2 (version 2.3.4.3) with a maximum fragment length of 2000 bp. PCR duplicates were identified and marked using Picard (version 2.20.7). MACS2 (version 2.2.9) was used for peak calling with default parameters. For visualization, bigWig files representing fold enrichment of histone mark signals over their corresponding input controls were generated. Promoter regions (−1000 bp to +100 bp relative to the transcription start site) and gene body regions of differentially expressed genes (DEGs) were defined according to the latest Mus musculus NCBI RefSeq annotation (mm10). ChIP-seq signal intensities over the defined genomic regions were quantified using deepTools multiBigwigSummary (version 3.5.5). Statistical comparisons between groups were performed using the Wilcoxon rank-sum test.

Supplementary Material

1

2

3

Supplemental information can be found online at 10.1016/j.celrep.2026.116945.