Ketogenic diet sensitizes glucose control of hippocampal excitability1

A high-fat low-carbohydrate ketogenic diet (KD) is an effective treatment for refractory epilepsy, yet myriad metabolic effects in vivo have not been reconciled clearly with neuronal effects. A KD limits blood glucose and produces ketone bodies from β-oxidation of lipids. Studies have explored changes in ketone bodies and/or glucose in the effects of the KD, and glucose is increasingly implicated in neurological conditions. To examine the interaction between altered glucose and the neural effects of a KD, we fed rats and mice a KD and restricted glucose in vitro while examining the seizure-prone CA3 region of acute hippocampal slices. Slices from KD-fed animals were sensitive to small physiological changes in glucose, and showed reduced excitability and seizure propensity. Similar to clinical observations, reduced excitability depended on maintaining reduced glucose. Enhanced glucose sensitivity and reduced excitability were absent in slices obtained from KD-fed mice lacking adenosine A1 receptors (A1Rs); in slices from normal animals effects of the KD could be reversed with blockers of pannexin-1 channels, A1Rs, or KATP channels. Overall, these studies reveal that a KD sensitizes glucose-based regulation of excitability via purinergic mechanisms in the hippocampus and thus link key metabolic and direct neural effects of the KD.

(Series 1000, Vibratome). Slices were incubated in aCSF saturated with 95% O 2 plus 5% CO 2 for 30-40 min at 37°C, then kept at room temperature for 1-5 h until recording. A slice was placed on a nylon net in the recording chamber under nylon mesh attached to a U-shaped platinum frame and submerged in and continuously perfused with aCSF at a fl ow rate of 2 ml/min at 32-34°C. Only one manipulation was tested in each slice. Slices in all treatment conditions (CD versus KD feeding, 3 versus 11 mM glucose incubation) remained recordable out to the longest postslicing recovery incubation tested (5 h).
For extracellular recordings, medium wall (1.5 mm) capillary fi lament glass was pulled on a Sutter P-97 micropipette puller (Novato, CA) using a 4-cycle program, giving electrode resistances of 8-12 M ⍀ . The recording electrode fi lled with 3 M NaCl was placed in the stratum pyramidale of the CA3 region for recording population spikes (PSs) or, in some recordings, in the stratum lucidum of the CA3 region for extracellular fi eld excitatory postsynaptic potentials (fEPSPs). A twisted bipolar insulated tungsten electrode was placed as stimulation electrode in the hilus of the dentate gyrus; stimuli were delivered at 30 s intervals. Pulse duration was 100 s and the intensity was adjusted such that the amplitude of evoked PS responses was between 0.6 and 1.4 mV. All electrophysiological responses were recorded via an AC amplifi er (World Precision Instruments) and fi ltered at 1 kHz. Data were digitized (16-channel A/D board, National Instruments) at a rate of 4 kHz and analyzed on-line using custom NeuroAcquisition software (Galtware, Denver, CO). All time courses of PS are moving averages of fi ve data points (graphs in the fi gures show sparse markers every 3 points).
The pannexin-1 mimetic blocking peptide 10 panx (WRQAAF-VDSY, with C-terminal amidation) and its scrambled counterpart were synthesized by Biomatik. Other drugs and chemicals were obtained from Sigma. All drugs were dissolved in aCSF at 100 times the desired fi nal concentration and applied via syringe pump upstream in the perfusion line to reach fi nal concentration before reaching the slice chamber ( 28,32 ). In all fi gures, the point indicated as the onset of drug or altered glucose application is the calculated time when the solution fi rst begins to mix into the volume of the slice chamber. Bicuculline was applied for 20 min before subsequent treatments. Other pharmacological agents were applied for at least 15 min before subsequent treatments.
Recorded extracellular fi eld potentials were analyzed off-line with NeuroAnalysis software (Galtware) and Igor Pro 5 (Wave-Metrics, Lake Oswego, OR). All data are expressed as mean ± standard error. The area of the PS was measured at 20 min after bicuculline application ( Fig. 1B ), 15 min after other drug applications [8-cyclopentyl-1,3-dipropylxanthine (DPCPX), 10 panx, and tolbutamide; Fig. 3 ], or 20 min after increased extracellular glucose concentration ( Fig. 1C ). The amplitude of the PS was also measured and all results of the amplitude data were the same as the results of the area (data not shown). Differences of evoked potentials with 11 mM glucose were compared with the nonparametric Mann-Whitney U test for normalized values. Evoked potential areas between CD and KD or between before and after drug treatment were compared with one-way ANOVA. P < 0.05 was considered signifi cant.

RESULTS
We fed a CD or KD to rats or mice for 13-18 days and prepared acute hippocampal slices for extracellular fi eld potential recordings in CA3. Analysis of rat blood plasma indicated signifi cant elevation of the ketone body ␤hydroxybutyrate at time of euthanization (0.97 ± 0.14 mM proposed that anticonvulsant effects are mediated directly by increased levels of ketone bodies (acetone, acetoacetate, ␤ -hydroxybutyrate) produced through ␤ -oxidation of lipids in liver mitochondria; however, blood ketones correlate poorly with seizure control in most animal and clinical studies (16)(17)(18) and do not translate clearly into changes in neuronal activity. Nevertheless, ketone esters are being developed as a means to elevate ketone levels without a drastic change in diet ( 19,20 ). Other established metabolic effects are increased brain ATP and, as noted above, a decreased and stable glucose level (21)(22)(23). Improved seizure control has also been observed with a low glycemic index diet ( 24 ) and a modifi ed Atkins diet ( 25 ), further suggesting the importance of reduced glucose. We proposed a glucose-related mechanism such that KDs increase adenosine, a purine nucleoside with antiseizure effects at the inhibitory adenosine A 1 receptor (A 1 R ) ( 12,26,27 ). Extracellular ATP is dephosphorylated rapidly into adenosine ( 28 ), thus placing adenosine squarely between KD-induced changes in metabolism and neuronal activity. Consistent with this, we found that a KD decreases spontaneous seizures due to adenosine defi ciency in mice with A 1 Rs, but is ineffective in mice lacking A 1 Rs ( 29 ).
Here, we demonstrate that KD feeding decreases in vitro seizure susceptibility and sensitizes glucose-based control of excitability in the CA3 region of the hippocampus. KD feeding neither reduced excitability nor induced glucose sensitivity in A 1 R knockout mouse slices, and blocking pannexin-1 channels, A 1 Rs or K ATP channels abolished these effects in slices from normal animals. The present methods may represent a useful tool for the in vitro study of KDs. Taken together, the present experiments, initiated in vivo and evaluated in vitro, link key metabolic and direct neural effects of the KD.

MATERIALS AND METHODS
All experiments were performed in conformity with Public Health Service Policy as defi ned in the Institute for Laboratory Animal Research Guide for the Care and Use of Laboratory Animals, and were approved by the Trinity College Animal Care and Use Committee. All measures were taken to minimize animal discomfort. Sprague-Dawley rats and C57Bl/6 mice [wild-type or lacking A 1 R ( 30 )] of either sex were fed standard rodent chow [control diet (CD); LabDiet 5001] or a KD (BioServ F3666) ad libitum for 13-18 days before slice preparation at age 5-8 weeks. F3666 has a fat:(protein+carbohydrate) ratio of 6.6:1 and a protein:carbohydrate ratio of 2.6:1 ( 31 ).
Standard slice preparation and recording conditions were used, similar to our previous publications ( 32,33 ). Briefl y, rats were anesthetized with isofl urane and decapitated; trunk blood was collected and centrifuged to isolate plasma; plasma was later tested for ␤ -hydroxybutyrate (StanBio, Boerne, TX). Four to fi ve coronal hippocampal slices of 400 m thickness were made in ice-cold artifi cial cerebrospinal fl uid (aCSF) with low (3 mM) or high (11 mM) glucose concentrations containing the following: 126 mM NaCl, 3 mM KCl, 1.5 mM MgCl 2 , 2.4 mM CaCl 2 , 1.2 mM NaH 2 PO 4 , 3 or 11 mM glucose, 8 or 0 mM sucrose (to balance osmolarity with the two concentrations of glucose), and 26 mM NaHCO 3 (osmolarity 320 mOsm, pH 7.4 when saturated with 95% O 2 plus 5% CO 2 ) with a vibrating slice cutter glucose, seizure-like activity induced by blocking ␥aminobutyric acid-mediated (GABAergic) inhibition (bicuculline, 10 M) was diminished in slices obtained from KD-fed rats compared with those from CD-fed rats ( Fig. 1B ). Reduced excitability promoted by the KD was masked by 11 mM glucose incubation: compared with slices from CDfed rats, prior KD feeding had minimal effects on the input/output relationship ( Fig. 1D ) and no signifi cant effects on the area of the epileptiform discharge evoked by bicuculline ( Fig. 1E ) in 11 mM glucose. These results argue strongly that the effect of the KD depends on maintaining reduced glucose (3 mM). A comparison of the input/output curves of slices from CD-fed rats incubated in 3 and 11 mM glucose showed minor differences that were only signifi cant at the lowest intensity (analysis not shown), generally consistent with the reported hippocampal PS stability during extended perfusion with 4 or 10 mM glucose ( 36 ). The dynamic infl uence of glucose on KD vs. 0.05 ± 0.02 mM CD, P < 0.001). Similar and consistent changes in blood chemistry were found in mice (data not shown). Stimulation intensity was not signifi cantly different in slices from KD-fed and CD-fed rats (0.72 ± 0.09 mA KD vs. 0.51 ± 0.13 mA CD; P > 0.05); also, the average adjusted PS amplitude before the application of bicuculline was not signifi cantly different between CD and KD groups (1.00 ± 0.05 mV KD vs. 1.18 ± 0.12 mV CD; P > 0.05).
To maintain in vitro conditions like those in vivo during KD feeding (stable, low blood glucose), some hippocampal slices were incubated and recorded in aCSF with glucose at a low concentration (3 mM) ( 34,35 ); other slices were incubated in high-glucose aCSF (11 mM; typical for acute slices). KD feeding reduced excitability as quantifi ed by PS current/voltage input/output curves, particularly at higher stimulation intensities in 3 mM glucose-incubated slices ( Fig. 1A ). Furthermore, after incubation in 3 mM Fig. 1. KD feeding in vivo and reduced glucose in vitro limit excitability and control seizure-like activity in rat hippocampus. A-C: Data from hippocampal slices incubated in reduced (3 mM) glucose. D-F: Data from hippocampal slices incubated in standard (11 mM) glucose. A: PS input-output curves demonstrate that hippocampal CA3 in KD-fed rats is less excitable across a range of stimulation intensities, and the maximum response amplitude was signifi cantly lower. CD (n = 5), KD (n = 20); && P < 0.01 compared between CD and KD. B: After matching for initial response amplitude, block of GABAergic inhibition (bicuculline, 10 M) induced seizure-like activity in all slices (quantifi ed as area under evoked response). The response area was reduced signifi cantly in slices from KD-fed rats. CD (n = 5), KD (n = 20); *NS, not signifi cantly different; * P < 0.05 between CD and KD; $$ P < 0.01 between baseline and bicuculline. C: Acutely increasing glucose (from 3 mM to 11 mM) augments bicuculline-induced seizure-like activity signifi cantly in the CA3 region of slices from KD-fed rats, but has no effect in slices from CD-fed rats. For comparability, seizure-like activity prior to acute glucose [which differed between CD and KD treatment; see (B)] is set to 100% to form new baselines for better comparison of acute glucose effects. n = 4-5; #NS, not signifi cantly different; ## P < 0.01 compared with 100% (Mann-Whitney U test); ** P < 0.01 between CD and KD. D, E: Slices from KD-fed rats incubated and recorded in 11 mM glucose showed minor electrophysiological changes in hippocampal pyramidal neurons, even during block of GABAergic inhibition . CD (n = 14), KD (n = 27); & P < 0.05 between CD and KD; *NS, not signifi cantly different between CD and KD; $$ P < 0.01 between baseline and bicuculline. F: When glucose was reduced acutely (from 11 mM to 3 mM), there was a reduction in bicucullineinduced excitability only in slices from KD-fed rats. CD (n = 13), KD (n = 7); #NS, not signifi cantly different; ## P < 0.01 compared with 100% (Mann-Whitney U test); * P < 0.05 between CD and KD. hippocampal excitability was selective to slices obtained from KD-fed rats. Increasing glucose to 11 mM after 3 mM glucose slice incubation increased excitability in slices from KD-fed but not CD-fed rats ( Fig. 1C ), similar to breakthrough seizures in KD-fed epileptic patients after carbohydrate ingestion ( 21 ). Note that in Fig. 1C , baseline levels are set to 100%, but areas were higher in slices from CDfed rats. Conversely, reducing glucose to 3 mM after 11 mM glucose incubation decreased the response area significantly solely in slices obtained from KD-fed rats ( Fig. 1F ). Thus, glucose dynamically controls CA3 excitability after in vivo KD feeding.
Regarding underlying mechanisms, we revealed an essential role for adenosine, an endogenous neuromodulator that links metabolism to decreased neuronal activity via A 1 Rs ( 37 ). In 11 mM glucose-incubated slices from KDfed rats, reduced excitability upon exposure to 3 mM glucose was blocked completely by the A 1 R antagonist DPCPX ( Fig. 2A ). Additionally, KD-related reduced excitability was evident in slices obtained from wild-type mice but absent in those from A 1 R knockout mice ( Fig. 2A ). These data suggest that KD feeding activates A 1 R in the hippocampus. Activation of A 1 R is known to cause presynaptic reduction of glutamate input and postsynaptic increase of K + conductance in CA3 pyramidal neurons ( 38 ). Interestingly, in recordings in the stratum lucidum, the amplitude of fEPSPs and paired-pulse ratios did not change with reduced glucose in 11 mM glucose-incubated slices from KD-fed rats, suggesting that these A 1 R effects are mainly postsynaptic (data not shown). Previous in vitro work has shown that, during reduced glucose, adenosine can be produced from ATP released via pannexin-1 channels and consequently reduce excitability via A 1 Rs linked to postsynaptic K ATP channels ( 39 ); other studies have also implicated K ATP channels in the effects of a KD ( 40,41 ). Here, blockade of pannexin channels with a pannexin-selective dose of carbenoxolone ( 42 ) or a specifi c peptide antagonist, 10 panx, eliminated effects of reduced glucose, similar to the A 1 R antagonist ( Fig. 2B ). Reduced excitability also depended on K ATP channels: all effects were blocked by increasing extracellular K + (from 3 mM to 5 mM) or antagonizing K ATP channels selectively with tolbutamide ( Fig. 2C ).
To further explore this phenomenon, we determined the involvement of these targets on the increased excitability in CA3 produced by switching slices from KD-fed rats from 3 mM to 11 mM aCSF during recording. We found that blocking A 1 Rs, pannexin-1 channels, or K ATP channels all enhanced seizure-like activity, and this enhanced activity occluded the excitatory effects of 11 mM glucose ( Fig. 3 ).
in vitro; effects of the KD were reversed or masked by 11 mM glucose (a standard for most brain slice physiology). Reduced excitability and heightened glucose sensitivity were absent in slices obtained from mice with a genetic deletion of A 1 Rs and abolished in slices during a pharmacological blockade of A 1 Rs, pannexin-1 channels, or K ATP channels. Because A 1 Rs couple to K ATP channels to reduce postsynaptic excitability, these experiments identify lowered glucose and elevated A 1 R activity as key links to specifi c neuronal mechanisms of the KD, and suggest a new experimental preparation, in vivo KD feeding followed by reduced glucose in vitro, for further study of the KD.
Even though we lowered extracellular glucose, we observed no signs that slices incubated in 3 mM glucose were signifi cantly hypoglycemic. Hypoglycemia is well-known to release adenosine ( 43 ), which would have driven the input/output curve downward compared with slices incubated in 11 mM glucose; such a change in the curve did not occur. Slices incubated in 3 mM glucose (and 11 mM) remained similarly recordable and thus apparently healthy out to our maximum slice recovery time of 5 h. In a previous study using identical slicing and recording conditions in the identical hippocampal substructure (in tissue from CD-fed animals), we presented data inferring that adenosine tone at A 1 Rs was similar in 3 mM and 11 mM glucose. The A 1 R antagonist DPCPX alone reduced tonic outward (K + ) current mildly in 11 mM glucose; when applied after ‫ف‬ 20 min of 3 mM glucose, DPCPX reduced outward current to a virtually identical extent [compare Fig. 3A "pretreatment" DPCPX vs. "reversal" DPCPX in ( 39 )]. Adenosine tone is thus similar in both conditions; therefore, significant hypoglycemia is unlikely in our particular experimental parameters.
It is well-established that the brain regulates glucose metabolism ( 1 ) and that glucose can infl uence seizures ( 21 ). Our fi ndings are consistent with research observations that blood glucose level can correlate directly with seizure frequency ( 44,45 ), and observations that anticonvulsant effects of the KD in vivo reverse quickly upon glucose injection or by ingesting carbohydrate-rich food in both animal models of epilepsy ( 29,46,47 ) and epileptic patients ( 21 ). Thus, KD feeding sensitizes hippocampal circuitry to changes in glucose whereby: 1 ) maintaining reduced glucose (3 mM) in acute hippocampal slices in vitro sustains the reduced excitability promoted by the KD in vivo; and 2 ) elevating glucose (11 mM) models the breakthrough seizures in patients on a KD who ingest carbohydrates. Based on these fi ndings, limited consequences of KD feeding in prior experiments with acute in vitro slices might be due to incubation and superfusion with 11 mM aCSF glucose: we found minimal changes in the input/output relationship when slices from KD-fed animals were recorded in standard aCSF.
In a prior study, we modeled a KD in vitro by acutely lowering extracellular glucose and maintaining or elevating intracellular ATP in CA3. Under these metabolic conditions, designed to mimic a KD, we also demonstrated inhibitory effects in pyramidal neurons mediated by A 1 Rs linked to K ATP channels, an effect that was not present in astrocytes ( 39 ). However in these previous experiments we did not use a dietary treatment: their focus was on establishing metabolic endpoints of the diet. Accordingly, our approach was similar to other studies using in vitro electrophysiology to increase understanding of neural mechanisms underlying the effects of ketone-based metabolism, for example, by applying ketones in vitro ( 40,(48)(49)(50). Whereas in vitro manipulations can offer exact control over experimental variables and elucidate mechanisms, overall they lack a connection to the diverse metabolic changes that occur in vivo with a dietary treatment.
Here, after KD feeding for several weeks, the cohort of mechanisms described with our acute in vitro model of the KD was recapitulated. The A 1 R-based control of excitation observed here is consistent with the KD's effects quantifi ed in vivo in transgenic mice with electrographic seizures due to adenosine defi ciency ( 29 ). Elevation of adenosine and heightened activation of A 1 Rs could explain the KD's anticonvulsant success against a wide range of seizure disorders ( 26 ), because A 1 R activation is effective in virtually every tested animal model of seizures ( 51 ) including Fig. 3. Acute elevation in glucose blocks the KD's effect on hippocampal excitability via an A 1 R-pannexin-K + channel pathway. All slices were incubated in 3 mM glucose aCSF and extracellular glucose concentration was acutely increased to 11 mM glucose for 25 min. Bicuculline was applied for 20 min before other drugs. A: DPCPX application (1 M) augmented bicuculline-induced seizure-like activity in slices from KD-fed rats and blocked 11 mM glucose-induced increase in this activity (n = 4). B: Blocking A 1 Rs, pannexin-1 channels, or K ATP channels (DPCPX, 1 M; 10 panx, 100 M; tolbutamide, 500 M, respectively) increased epileptiform activity similarly in slices from KD-fed rats. The excitatory effect of acutely increased glucose was prevented by all three antagonists. n = 4-5; %% P < 0.01 compared pre-and postdrug application (Mann-Whitney U test); *NS, not signifi cantly different between baseline and 11 mM glucose; ** P < 0.01 between baseline and 11 mM glucose. pharmacoresistant seizures ( 52 ). To date, an established model of the KD in vitro has never been established; a recent paper examining CSF from mice fed a KD helps address this knowledge gap ( 53 ), and we suggest that the match among the present experiments, previous in vivo experiments ( 29 ), and our metabolic mimic in vitro ( 39 ) suggest that reduced glucose and suffi cient ATP are critical in mobilizing adenosine-based anticonvulsant effects.
Interest has intensifi ed recently toward understanding key mechanisms underlying the KD's anticonvulsant effects and, to that end, in establishing an effective protocol to assess in vitro the effects of KD feeding. This interest is due to increasingly widespread and international use of the KD for epilepsy and, in parallel, a burgeoning interest in metabolic approaches as a platform for new therapies for diverse neurological disorders. Overall, the present experiments represent the fi rst study delineating processes mobilized in vivo by KD feeding that: 1 ) link to and depend on known metabolic effects (limited glucose); 2 ) identify specifi c anticonvulsant neuronal mechanisms, i.e., reducing excitability in a seizure-prone area via pannexin-1 channels, adenosine A 1 Rs, and ultimately K ATP channels; and 3 ) as observed clinically, reverse with increased glucose.