Effects of ketogenic diet and ketone monoester supplement on acute alcohol withdrawal symptoms in male mice

Abstract

Rationale

After alcohol ingestion, the brain partly switches from consumption of glucose to consumption of the alcohol metabolite acetate. In heavy drinkers, the switch persists after abrupt abstinence, leading to the hypothesis that the resting brain may be “starved” when acetate levels suddenly drop during abstinence, despite normal blood glucose, contributing to withdrawal symptoms. We hypothesized that ketone bodies, like acetate, could act as alternative fuels in the brain and alleviate withdrawal symptoms.

Objectives

We previously reported that a ketogenic diet during alcohol exposure reduced acute withdrawal symptoms in rats. Here, our goals were to test whether 1) we could reproduce our findings, in mice and with longer alcohol exposure, 2) ketone bodies alone are sufficient to reduce withdrawal symptoms (clarifying mechanism), 3) introduction of ketogenic diets at abstinence (a clinically more practical implementation) would also be effective.

Methods

Male C57BL/6NTac mice had intermittent alcohol exposure for three weeks using liquid diet. Somatic alcohol withdrawal symptoms were measured as handling-induced convulsions, anxiety-like behavior was measured using the light-dark transition test. We tested a ketogenic diet, and a ketone monoester supplement with a regular carbohydrate-containing diet.

Results

The regular diet with ketone monoester was sufficient to reduce handling-induced convulsions and anxiety-like behaviors in early withdrawal. Only the ketone monoester reduced handling-induced convulsions when given during abstinence, consistent with faster elevation of blood ketones, relative to ketogenic diet.

Conclusions

These findings support the potential utility of therapeutic ketosis as an adjunctive treatment in early detoxification in alcohol-dependent patients seeking to become abstinent.

Introduction

Worldwide, harmful use of alcohol kills more than 3 million people annually, more than diabetes, hypertension, or road injuries ((WHO) 2018). Alcohol use disorder (AUD) exacts a tremendous personal and economic toll worldwide. A challenge of treatment is the withdrawal symptoms that occur upon cessation of prolonged heavy drinking. Acutely, severe alcohol withdrawal can result in life-threatening seizures and brain damage, classically attributed to increased neuronal excitability (Becker and Mulholland 2014; Finn and Crabbe 1997). Withdrawal symptoms are alleviated by alcohol consumption and thus contribute to relapse to alcohol use (Brower 2003; Engel et al. 2016; Heilig et al. 2010). During a life-time, as many as 80% of abstinent alcohol users relapse (Barrick and Connors 2002; Connor et al. 2016; Jin et al. 1998). Benzodiazepines, the most common treatment, are effective against withdrawal symptoms but have side effects including addictive properties. Indeed, patients with AUD have a high prevalence of non-medical benzodiazepine use and benzodiazepine dependence, with estimates ranging from 14% current to 78% lifetime (Johansson 2003; Kan et al. 2001; Morel et al. 2016; Votaw et al. 2019). While few studies directly compared benzodiazepine dependence in AUD vs. general population, one study showed 15% and 1%, respectively (Johansson 2003). The disease remains severely undertreated: a recent US survey reported that only 4.2% of patients diagnosed with AUD received specialty treatment ((SAMHSA) 2019). There exists a clear need for new treatment options based on a better understanding of the effects that heavy drinking and withdrawal have on the brain.

Brain imaging studies show that acute alcohol administration lowers glucose metabolism in the human brain (de Wit et al. 1990; Volkow et al. 1990). Because the decrease also happens at low alcohol doses that have minimal behavioral effects (Volkow et al. 2006), it was hypothesized that the reduction in glucose metabolism reflects utilization of an alternate energy substrate by the brain, e.g., the alcohol metabolite acetate (Pawlosky et al. 2010; Volkow et al. 1993). Acetate is readily taken up into the brain via the monocarboxylic acid transporter 1 and is used as an energy substrate by astrocytes (Cruz et al. 2005; Lebon et al. 2002; Minchin and Beart 1975; Waniewski and Martin 1998). Brain acetate metabolism is increased while glucose metabolism is decreased during alcohol intoxication, an effect more pronounced in heavy drinkers than in control subjects (Jiang et al. 2013; Volkow et al. 2013; Volkow et al. 2015). The findings led to the hypothesis that the brain may be in a state of reduced energy turnover when acetate availability drops upon abrupt cessation of sustained alcohol drinking, despite normal plasma glucose, contributing to the neurotoxicity and symptoms observed in acute withdrawal. Providing the brain with a non-glucose energy substrate like acetate may then alleviate withdrawal symptoms.

The ketone bodies acetoacetate, acetone, and β-hydroxybutyrate (BHB) also rise during ethanol consumption (Baraona and Lieber 1979; Lefèvre et al. 1970; Lukivaskaya and Buko 1993). Ketone bodies are readily consumed by the brain, whether in starvation (Hasselbalch et al. 1995; Hawkins et al. 1986) or by exogenous administration (Gjedde 1983; Jiang et al. 2011b; Pan et al. 2002). Therefore, alcohol consumption has may promote consumption of both acetate and ketone bodies in the brain. Ketone bodies are also produced in the liver when glucose is not available, such as during fasting or adherence to a ketogenic diet (KD; very high fat, moderate protein, very low carbohydrate). Similar to acetate, ketone bodies are taken up into the brain via monocarboxylic acid transporters, and can serve as energy substrate for all brain cell types, with oxidation distributed in neurons vs. astrocytes in proportions similar to glucose utilization (Achanta and Rae 2017; Jiang et al. 2011a; Pan et al. 2002).

Thus, we hypothesized that ketone bodies could replace ethanol-derived acetate and ketones as energy substrates during alcohol withdrawal. Indeed, we previously reported that rats maintained on a KD during alcohol exposure had reduced withdrawal symptoms when alcohol was discontinued, relative to control diet (Dencker et al. 2018). In practice, adherence to a strict KD would be challenging for alcohol-dependent patients before initiation of sobriety, because the KD is difficult to maintain, requires medical supervision and adequate micronutrient supplementation, and its effects on long-term health are debated. However, ketosis (defined as an increase in blood ketone concentration >0.5 mM) can also be achieved by intake of exogenous ketones, typically as ketone salts or esters.

Here, we replicated our findings from rats, in mice with longer alcohol exposure, and we extended the studies by asking two main questions to clarify both mechanism of action and clinical potential. First, we asked whether ketone bodies alone are sufficient to reduce withdrawal symptoms. We tested this hypothesis using a regular diet supplemented with a ketone monoester (KME) that produces acute elevations of blood BHB (Clarke et al. 2012a). Second, we asked whether diet-induced ketosis could reduce alcohol withdrawal symptoms when introduced at the start of abstinence rather than during alcohol exposure, i.e., in a manner more suitable for clinical application.

Material and methods

Subjects

Male C57BL/6NTac (Taconic, Lille Skensved, Denmark) were used because they are not prone to spontaneous convulsions and to facilitate comparisons with planned studies involving alcohol drinking. Mice were acquired at 5–6 weeks of age and acclimated at least one week to the facilities before testing. Mice were group-housed four per cage (not necessarily litter mates) in type III cages (42.5 × 26.6 × 15.5 cm) with aspen bedding (Tapvei) and with hiding devices, nesting material, and wooden chewing blocks as enrichment, in a room maintained under a reversed light-dark cycle (light on 19:00 – 07:00), temperature maintained at 22 ± 2°C and relative humidity at 55 ± 10%. A total of 118 mice were used. Three mice were euthanized during the studies and no data were included: two due to fighting during the habituation period, and one due to sudden weight loss (control diet alcohol group). Procedures were approved by the Animal Experiments Inspectorate under the Danish Ministry of Food, Agriculture, and Fisheries in accordance with the EU directive 2010/63/EU (approval number 2017-15-0201-01334). Method of euthanasia was cervical dislocation. Tap water was available freely throughout. Standard rodent chow (Altromin 1310, Brogaarden, Denmark) was available freely at arrival, specialty diets were provided during experiments as described below.

Experiments and diets

Alcohol dependence and withdrawal were induced using a liquid diet in three cycles of five days on 6.7% v/v alcohol, two days forced abstinence, a common liquid-diet approach in the literature and a concentration used in commercially available alcohol liquid diets [Figure 1A; (Overstreet et al. 2002; Snell et al. 1996)]. After acclimation, mice were gradually switched to liquid diets over 2 days (liquid diet with 3% alcohol + 3g chow/mouse; then liquid diet with 4.5% alcohol + 1.5g chow/mouse). For feasibility, mice were tested in two consecutive cohorts in each experiment, each cohort comprising all experimental groups, i.e., each experiment followed a randomized block design. In each block, cages were allocated randomly to experimental group. The experimenter was blinded to experimental group for all tests.

Fig. 1. Experimental setup and diets.

(a) Timeline of alcohol exposure (grey) and forced abstinence (open), with test times as hours since alcohol removal shown for the last cycle. Periods with test diets are indicated by diagonal hatching. Arrows indicate blood alcohol sampling (pointing down) or light-dark transition test (pointing up). (b) Composition of diets as calories provided by each nutrient, per mL diet. Reg.: regular diet. Alco: alcohol-containing. All diets also contained vitamins, minerals, and 5.5–5.7% fiber.

In Experiment 1 “Keto Throughout”, in service of rigor/reproducibility of our previous report (Dencker et al. 2018), we tested the effects of diet-induced ketosis started during alcohol exposure: mice were fed alcohol-containing regular, KD, or KME diet, plus a regular-control (no alcohol) group. On the test days, food was replaced by the corresponding (regular, KD, KME) no-alcohol diet at 13:00 to start forced abstinence. Due to a technical error, no data could be collected from the last cycle in Experiment 1 first cohort, including light-dark transition test and blood measures, resulting in a lower sample size.

In Experiment 2 “Keto After”, we tested the effects of ketosis induced after alcohol exposure. Mice were fed alcohol-regular diet, then no-alcohol regular, KD, or KME diet during abstinence. To ensure fast and uniform onset of ketosis in the KME group, a bolus dose of KME (3g/kg by gavage, 7.8 mL/kg) was administered once at abstinence start, in addition to the available KME diet.

To mix alcohol into a KD without evaporation (e.g., provided in closed bottles), and be able to compare KD, KME, and regular diets as directly as possible, we developed our own diets. The diets were isocaloric, closely based on commercial formulations to meet both nutritional and scientific needs, and were prepared fresh daily. The regular-control and regular-alcohol diets were based on the commonly used open source formulations of diet AIN-76 liquid, the alcohol diets providing 35.6% of calories from alcohol. The KD had a caloric distribution (before addition of alcohol) comparable to the commercial diet used in our previous study (Dencker et al. 2018). The KME diet was based on the regular diet but provided 30% of calories from the ketone ester (D-BHB ester; HVMN ketone ester, HVMN Inc., San Francisco, CA), a dose previously shown to be well tolerated and produce ketosis in rodents (Clarke et al. 2012b; Murray et al. 2016; Srivastava et al. 2012). The KME is converted to BHB and acetoacetate (Clarke et al. 2012c; Desrochers et al. 1995; Tate et al. 1971). Figure 1B shows the nutritional composition of each diet. Unshelled peanuts were offered weekly (never during abstinence) in all groups as enrichment and to limit weight loss (Anji and Kumari 2008).

Apparatus

Light-dark transition test was conducted in open field activity chambers fitted with beam-break movement detection systems (OFA-510, Med Associates, St Albans, VT, USA). A partition of dark red plastic (not transparent for mice but allowing beams through) was used to create two compartments each measuring 27 × 13.5 cm, 30 cm tall. Walls and lid were clear in one compartment, opaque black in the other. The light side had low illumination (~40 lux; Fisherbrand Traceable dual-range light meter) not anxiogenic alone (Hascoet et al. 2001), allowing detection of anxiogenic-like effects of alcohol withdrawal (McGinnis et al. 2020; Vranjkovic et al. 2018). The partition had a 4 × 4 cm opening allowing free movement between compartments. Operant procedures used mouse operant-conditioning chambers (Med Associates ENV-307A) each containing two nose-poke holes fitted with a photocell and a yellow cue light, and a steel dish into which reinforcers were delivered from a syringe pump (Thomsen and Caine 2005; Thomsen et al. 2017). All chambers were individually enclosed in sound-attenuating cubicles equipped with a light and a ventilation fan.

Alcohol withdrawal testing: somatic symptoms

Handling-induced convulsions (HIC) were assessed after 1, 2, 3, 4, 6, and 8 hours of abstinence in the “Keto Throughout” experiment, and after 1, 2, 3, 5, 7, 9, 11, 20, and 28 hours in the “Keto After” experiment to allow more time for ketosis to develop from the KD. Alcohol diets were again provided at 08:00 the next day, i.e., after 43h of abstinence, starting the next 5-day alcohol access cycle. HIC scoring was adapted from Crabbe et al. (Crabbe et al. 1991) with possible scores 0–7, and is described in detail in Supplemental methods. HIC scores were averaged over test cycles, then, area under the curve (AUC) was calculated in each mouse for the first three hours (3h-AUC), i.e., the maximum time predicted to produce ketosis after KME administration, and for the entire observation period.

Alcohol withdrawal testing: anxiety-like behavior

Mice were tested in the light-dark transition test 6h after abstinence onset (19:00), in the last cycle only to avoid habituation. Mice were placed in the lit side of the apparatus and activity was recorded for 10 min. Primary measures were light-dark crossings and voluntary time on lit side (Hascoet et al. 2001; Kliethermes 2005). Data from no-alcohol regular diet controls tested in different experiments did not differ significantly (p=0.10 and p=0.38) and were pooled for data presentation and analysis.

Blood measures

Blood BHB and glucose levels were assessed from needle sticks to the tail using strips and FreeStyle Precision Neo analyzer (Abbott). In the “Keto Throughout” experiment, in which test diet exposure was continuous, blood levels were measured 24h before HIC testing in each cycle, to minimize stress on test days. In the “Keto After” experiment, BHB levels were assessed in cycle 2 and 3, at 24h before and after HIC testing at 1, 2, 7.5, and 24h after withdrawal start in the KME group, and in regular and KD groups, at −24 and 2h. Blood glucose was tested at −24, 2, and 24h. Blood alcohol levels were determined using a GL6 analyzer, Analox Instruments (Stokesley, UK). In the “Keto Throughout” experiment, blood samples were taken 24h before the last cycle of HIC testing by retro-orbital bleeding under brief sevoflurane gas anesthesia. In the “Keto After” experiment, to minimize stress and provide better measurements, mice were maintained four days on their respective alcohol diets after ended testing, and were euthanized by decapitation and trunk blood collected. Liquid diet intake (by bottle weight) and bodyweights were recorded.

Control experiments

Effects of diet-induced ketosis per se in the light-dark transition test were tested in a separate cohort (n=6) after six days on alcohol-free regular, KD, or KME diet – admittedly a shorter diet exposure than the three weeks used in Experiment 1. The next day, the same mice were tested for effects of diet on alcohol clearance and intoxication (anesthetic/sedative effect). To this end, mice were administered 3.5 g/kg alcohol intraperitoneally (22% in saline, 20 mL/kg), and loss of righting reflex was tested as described in Supplemental methods (Fee et al. 2004; Lynch et al. 2013). The mice were euthanized 180 min after alcohol injection and trunk blood was collected for blood alcohol analysis.

Because we observed lower blood alcohol levels in the KME group in the “Keto Throughout”, we added a control experiment to assess whether KME affects voluntary intake of alcohol. In two separate cohorts of mice maintained on regular chow (n=7, 9), we tested the effect of KME pretreatment on voluntary intake of, and operant behavior maintained by 30μl of, oral alcohol (20% in water, never sweetened) and liquid food (vanilla flavor nutridrink, Nutricia, Denmark), respectively. Mice were trained to nose-poke reinforced under a fixed-ratio 1 timeout 20s schedule of reinforcement in 2h-sessions as previously described (Bornebusch et al. 2019; Thomsen et al. 2009), see Supplemental methods for details. Mice were administered 3 g/kg KME by gavage 60 min before session start. As a control for effect of satiety, an isocaloric amount of oil emulsion (Calogen unflavored, Nutricia) was tested, within-subjects, in a counterbalanced order with KME. Chow was removed 2h before gavage.

Data analysis

Light-dark transition test data, HIC AUC, and blood alcohol were analyzed for each experiment by one-way ANOVA with diet as between-subjects factor and planned posthoc comparisons (KD and KME vs. alcohol-regular) adjusted for false discovery rate (Benjamini, Krieger and Yekutieli procedure, limit 5%). Food intake, bodyweight, blood BHB and glucose were analyzed by two-way ANOVA with time as a repeated measure and diet as between-subjects measure, Greenhouse-Geisser corrected where appropriate and with false discovery rate-adjusted posthoc comparisons. Food intake was averaged per cycle component (alcohol/abstinence); bodyweight was averaged per cycle. Loss of righting reflex duration was analyzed by Logrank test. Reinforcers earned per session were analyzed by two-tailed paired-sample t-test with false discovery rate correction. No data were excluded based on statistical outlier values or other criteria. Analyses were performed using GraphPad Prism version 8. A priori power calculations were not performed for these studies.

Results

Both KD and KME increased blood BHB levels

Blood BHB levels reflected the diet ([F_3,94=116.1, p<0.0001], [F_1,21=18.5, p=0.0003]) and time ([F_1.4,64.3=11.4, p=0.0004], [F_2.1,35.8=16.9, p<0.0001]) in the “Keto Throughout” and “Keto After” experiment, respectively, with an interaction in “Keto After” [F_6,94=11.8, p<0.0001]. In the “Keto Throughout” Experiment (test diets provided during alcohol exposure and abstinence), both KD and KME produced BHB levels higher than the regular diets (p<0.0001; Figure 2A). In the “Keto After” Experiment, KME produced higher BHB than KD at all time points (p<0.05; Figure 2B). Blood glucose was measured in the “Keto After” experiment and was related to diet only [F_2,30=13.6, p<0.0001], with KD and KME producing lower glucose levels than regular diet (p<0.001; Figure 2C).

Fig. 2. Blood ketone and glucose levels.

Blood BHB as a function of time in the “Keto Throughout” experiment (a, n=12) and the “Keto After” experiment (b, n=10–12), blood glucose as a function of time in “Keto After” (c, n=10–12). *p<0.05, **p<0.01 vs. alcohol-regular diet. Data are group means ± s.e.m.

Alcohol withdrawal testing: somatic symptoms

In both experiments, alcohol-regular diet mice showed HIC scores consistent with previous reports from this strain (Anji and Kumari 2008; Metten and Crabbe 2005), reaching peak scores of around 3 from 3h to 8–10h after alcohol removal, declining by 20–28h. The use of a relatively long observation period with fewer time points in the last half may have resulted in higher calculated AUC relative to studies with shorter observation or higher temporal resolution. In the “Keto Throughout” Experiment, HIC score was related to diet group as 3h-AUC [F_2,33=3.74, p=0.03] and as 8h-AUC [F_2,33=26.1, p<0.0001]. Consistent with our previous report, mice in KD group had lower scores relative to alcohol-regular diet for 3h-AUC (p=0.03; Figure 3A) and 8h-AUC (p<0.0001; Figure 3B), while the KME group had reduced score only for 3h-AUC (p=0.047). In the “Keto After” Experiment, HIC score was related to diet group as 3h-AUC [F_2,30=6.85, p<0.004] but not as total AUC; only KME-fed mice had lower scores relative to regular diet (p=0.007; Figure 3A). Scrutiny of the cage averages suggested a consistent effect rather than random cage differences; for instance all three cages in the KME group had lower average 3h-AUC than the alcohol-regular diet cages in “Keto After”, and similarly for the 8h-AUC in alcohol-KD vs. alcohol-regular in “Ketosis Throughout”. HIC score as total AUC was negatively correlated with blood BHB levels in the KD group only, in both “Keto Throughout” (p=0.004) and “Keto After” (p=0.0006; Figure S1). Thus with the KD, higher BHB was associated with lower withdrawal symptoms, while with KME, the symptoms were improved at all BHB elevations.

Fig. 3. Effects of diets on handling-induced convulsions.

HIC score as AUC for the first three hours of abstinence (a) and total observation period (b) for no-alcohol controls, in “Keto Throughout” (n=12), and “Keto After” (n=10–12). **p<0.01, ***p<0.001 vs. alcohol-regular diet. Data are shown as individual subjects, with bars representing group means ± s.e.m. Note different ordinate scales for 0–3h and for total time.

Alcohol withdrawal testing: anxiety-like behavior

Number of light-dark compartment crossings was related to treatment group in the “Keto Throughout” Experiment [F_2,15=4.38, p=0.03]. KD-fed mice had more crossings relative to alcohol-regular diet (p=0.01; Figure 4A), interpreted as reduced anxiety-like behavior. A similar trend was seen for KME-fed mice (p=0.08). Voluntary time in the lit compartment was also affected by treatment condition in “Keto Throughout” [F_2,15=11.1, p=0.001]. KME-fed mice spent more time in the lit side relative to alcohol-regular diet (p=0.0005; Figure 4D), interpreted as reduced anxiety-like behavior. Neither measure was affected by diet in the “Keto After” experiment (Figure 4B,E), or by the diets per se (no alcohol, Figure 4C,F). Anxiety-like behaviors were not correlated with blood BHB levels (Figure S2).

Fig. 4. Effects of diets in the light-dark transition test.

Light-dark side crossings (a-c) and voluntary time in the lit side (d-f) in the “Keto Throughout” experiment (a,d, n=6), the “Keto After” experiment (b,e, n=10–11), and no-alcohol controls (c,f, n=6–12). *p<0.05, ***p<0.001 vs. alcohol-regular diet. Data are shown as individual subjects, with bars representing group means ± s.e.m.

Blood alcohol, food intake, and bodyweight

Food intake was significantly related to time in the “Keto Throughout” experiment [F_1.7,6.69=9.37, p=0.01], less so in the “Keto After” experiment (p=0.09), but in both there was an interaction of diet and time ([F_18,24=2.29, p=0.03], [F_12,36=2.24, p=0.03]), with no main effect of diet (Figure S3A,B). No post-hoc comparisons were significant. Bodyweight changed over time in both experiments ([F_1.7,61.7=12.2, p<0.0001], [F_1.4,41.6=7.40, p=0.005]; Figure S3C,D), with a diet by time interaction in both ([F_9,108=17.4, p<0.0001], [F_6,90=2.57, p=0.02]) and a main effect of diet in the “Keto Throughout” experiment only [F_3,44=14.0, p<0.0001]. In the “Keto Throughout” experiment, all alcohol groups weighed less than no-alcohol (p<0.01); KD and KME groups also weighed less than alcohol-regular (p<0.0001). In the “Keto After” experiment, the KD group weighed more than the other groups overall (p<0.05) but this seemed to reflect a preexisting difference between the cages.

Blood alcohol levels were related to diet in the “Keto Throughout” experiment [F_2,15=5.67, p=0.01], with mice consuming alcohol-KME (but not KD) diet showing lower blood alcohol levels relative to alcohol-regular diet (p=0.005; Figure 5A). In the “Keto After” experiment, in which all mice consumed the same alcohol-regular diet during alcohol exposure, alcohol levels did not differ between alcohol-exposed groups.

Fig. 5. Blood alcohol levels and control experiments.

Blood alcohol in “Keto Throughout” and “Keto After” (a, n=5–6). Blood alcohol (b) and loss of righting reflex (c, LORR) after acute 3.5 g/kg alcohol administration, in mice fed no-alcohol regular, KD, or KME diet; n=6. Number of alcohol (20% in water, d) or liquid food (e) reinforcers taken per 2h-session at baseline and after intragastric administration of 3 g/kg KME or isocaloric oil control, n=7–9. Data are shown as individual subjects, with bars representing group means ± s.e.m.

Control experiment: alcohol clearance, loss of righting reflex, and voluntary alcohol and food intake

Because we observed lower blood alcohol levels in the KME group in the “Keto Throughout” experiment, we tested whether ketosis affected alcohol clearance or acute intoxication, by measuring blood alcohol and loss of righting reflex, respectively. Diet did not significantly affect blood alcohol 180 min after administration of 3.5 g/kg alcohol (Figure 5B) or loss of righting reflex duration (Figure 5C). Bodyweight remained comparable between groups over the week on experimental diets (Figure S3E). Blood BHB levels at righting reflex test start were: control 0.28±0.03, KD 1.63±0.30, KME 2.35±0.60 mM.

We then tested the hypothesis that the difference in blood alcohol in “Keto Throughout” was due to KME decreasing voluntary alcohol intake or food intake. Acute KME administration decreased both food- and alcohol-reinforced operant responding and did so to the same degree as an isocaloric oil emulsion control (Figure 5D,E). The basis behind the impact of KME and oil remains an open and interesting question to be explored in future studies.

Discussion

When test diets were given throughout, the reduction in HIC was briefer for KME than for KD, consistent with less sustained blood BHB levels by KME ingestion relative to endogenous production. While BHB levels under the KD is likely relatively stable, the measure in the KME-fed mice would have fluctuated depending on meal patterns, and harder to “capture” a good measure of. As opposed to effect duration, the effect size in the first 3 hours was comparable between KD and KME, perhaps suggesting that once a certain threshold is reached, additional BHB may not provide further HIC reduction, although there was some correlation between blood BHB level and effect. When ketosis was not induced until abstinence, only KME reduced HIC. Again, the effect lasted only a few hours, consistent with the high blood BHB level recorded after the intragastric loading dose, followed by a lower level maintained by the diet. Blood BHB increases gradually over days after onset of KD. Here, it did not surpass 1.8 mM within the HIC testing period in the “Keto After” experiment, although in all groups, total ketone levels (BHB, acetoacetate and acetone) were probably higher than recorded, as we only measured BHB. We confirmed the expected increased effect of KME relative to KD when administered at abstinence start. The effect of KME on HIC is consistent with studies using different formulations of exogenous ketones showing reduced epileptic-like brain activity or seizures (D’Agostino et al. 2013; Kovacs et al. 2017). Taken together, the present study indicates that ketone bodies alone, including exogenous, are sufficient to reduce somatic seizure-like alcohol withdrawal symptoms in rodents, including administered at abstinence start.

Both KD and KME produced effects consistent with decreased anxiety-like behavior (i.e., normalization towards behavior in no-alcohol controls), when administered throughout the experiment. The fact that voluntary time spent in the lit compartment was normalized in the KME group at a time when any effect on HIC had ceased suggests some lasting effects, or effect on the development of anxiety-like behavior. The latter would be consistent with the lack of significant effect of KD or KME when given at abstinence, perhaps suggesting, speculatively, that ketosis can reduce the development but not expression of anxiety-like behaviors. The lack of effect could also be due to too low blood ketone levels. In the previous study in rats (Dencker et al. 2018), an anxiolytic-like effect of the KD using the common anxiety test elevated plus-maze failed to reach statistical significance. This may represent a greater sensitivity of the light-dark transition test for detecting anxiogenic-like effects of alcohol withdrawal in rodents, which is consistent with an overview of the literature (Kliethermes 2005). Also, consistent with our present results using two measures in the light-dark transition test, KD and KME may produce subtly different improvements in anxiety-like behaviors that are detected using different behavioral endpoints. Alternatively, the effects of ketosis on anxiety-like behavior may not be robust or consistent. Further investigation is needed to determine whether therapeutic ketosis can provide clinically relevant improvement in anxiety or other affective symptoms of alcohol withdrawal, which are associated with risk of relapse in alcohol-dependent patients (Brower 2003; Engel et al. 2016; Heilig et al. 2010). Nevertheless, other studies have reported moderately reduced anxiety-like behaviors in rodents fed exogenous ketone supplements (Ari et al. 2016; Kashiwaya et al. 2013), lending support to the potential utility of the approach.

The brain uptake and utilization of ketone bodies increases as a function of blood ketone concentration in rodents and humans (Courchesne-Loyer et al. 2017; Hasselbalch et al. 1996; Zivin and Snarr 1972). However, transport is a saturable and inducible process with a high K_M, and it has been shown to increase after KD or fasting in rodents and humans, concordant with increased brain monocarboxylic acid transporter 1 expression (Gjedde and Crone 1975; Leino et al. 2001; Morris 2005; Pifferi et al. 2011; Zivin and Snarr 1972). This likely explains why a higher blood BHB was apparently needed to achieve measurable HIC reduction in the “ketosis at abstinence” condition relative to the “ketosis throughout” condition: in the latter, diets had time to cause monocarboxylic acid transporter upregulation and increase brain ketone uptake. We also measured lower blood BHB levels in the KME-consuming mice in the last week of “Keto Throughout” relative to the earlier weeks. It is possible that increased brain uptake and utilization of ketone bodies contributed to this, but it is unclear why the effect should be restricted to KME, while ketone levels in the KD group increased over weeks. Interestingly, chronic intermittent alcohol exposure was recently shown to increase brain expression of monocarboxylic acid transporters in mice (Lindberg et al. 2019); it is not known whether effects of alcohol and ketosis on monocarboxylic acid transporter expression are additive. If also the case in humans, alcohol-induced monocarboxylic acid transporter upregulation would help brain uptake of ketones in detoxification in patients with AUD. Further, brain uptake of ketones may be higher in humans than in rodents (Morris 2005). Taken together, those findings support the feasibility of inducing therapeutic levels of ketosis in patients, but also caution that KME doses needed may be relatively high and/or frequent to produce clinically relevant effects.

The mice drinking the KME-containing alcohol diet had lower blood alcohol relative to alcohol-regular or alcohol-KD, at the time of testing. While that introduces the possibility that reduced withdrawal symptoms stem simply from lower alcohol exposure, the fact that total-time HIC in the KME group was comparable to regular-alcohol group makes it unlikely. Control studies indicate that KME did not alter alcohol metabolism, and our preliminary measurements of voluntary intake showed similar effects of KME and oil (i.e., likely attributable to satiety). Therefore, the apparent difference may be an artifact due to measuring time-varying blood alcohol levels at a single time point. It would be worthwhile testing whether KD or KME can reduce voluntary alcohol intake in alcohol-dependent rodents, and in particular whether it can reduce abstinence-induced escalation in alcohol intake. It would also be useful to repeat the studies using alcohol vapor exposure (Rogers et al. 1979) to obtain tighter control over blood alcohol levels. Oral exposure, the typical route of alcohol taking in humans, was chosen because changes in gut microbiota may underlie some effects of diet-induced ketosis (Ma et al. 2018; Newell et al. 2016; Olson et al. 2018), alcohol (Dubinkina et al. 2017; Peterson et al. 2017; Temko et al. 2017; Xiao et al. 2018), or their interaction, and it is unclear whether these factors are affected similarly by vapor exposure. Although alcohol permeates the body water and fat, it may also be that flow of alcohol directly through the digestive track has a different impact. The immediate effect of KME on HIC argues against this hypothesis, although it cannot be excluded that effects on anxiety-like behaviors were absent in the “Keto After” experiment because the intervention did not allow time for diet to modify gut microbiota. A potential limitation to be addressed in our future studies is a need for an oil placebo bolus to balance the KME bolus administration. Finally, it will be important to extend these studies to female subjects, as well as strains showing more pronounced withdrawal symptoms. These studies used male mice because a main focus was the acute somatic withdrawal signs, which have been more consistent in male mice in the literature, and acute withdrawal severity does not appear to be gender-biased in humans (Goodson et al. 2014). However, women may experience more prolonged and/or severe later withdrawal symptoms (Petit et al. 2017), and future studies should include both sexes (see (Kliethermes et al. 2004).

KD and BHB lower seizures and produce neuroprotective and anti-inflammatory effects via multiple mechanisms, including modulation of glutamate and GABA systems (Boison 2017; Calderon et al. 2017; Kashiwaya et al. 2010; Olson et al. 2018; Pflanz et al. 2019). For instance, KD was shown to increase hippocampal GABA/glutamate ratio in rodents (Calderon et al. 2017; Olson et al. 2018). Glutamate and GABA, among other systems, are altered by chronic alcohol intake and sudden abstinence, and increased extracellular glutamate is thought to be an important contributor to acute withdrawal symptoms and neurotoxicity (Fliegel et al. 2013; Hermann et al. 2012; Roberto and Varodayan 2017). Thus, ketone bodies might also affect withdrawal symptoms by balancing glutamate and/or GABA transmission. Unlike KD, fasting, acetone, or acetoacetate, BHB generally has not shown antiepileptic effects in (adult) laboratory animals, at least not as acute dosing (Simeone et al. 2018; Wood et al. 2018). In a pilot study, both KD and KME failed to attenuate morphine withdrawal in mice (data not shown), a syndrome that also features increased neuronal excitability (Brown and Russell 2004). Taken together, reduced neuronal excitability is less likely to be a major mechanism of action by which ketogenic manipulations reduce alcohol withdrawal symptoms.

Conclusions, implications, and future directions

First, we replicated our previous findings that a KD decreased alcohol withdrawal symptoms in rats, extending the findings to mice and to longer alcohol exposure. Second, the present study showed that ketone bodies alone, including exogenous BHB in a regular carbohydrate-containing diet, are sufficient to reduce alcohol withdrawal symptoms in rodents. This is important not only for understanding mechanism, but also in terms of potential clinical application. Taking a ketone product as medication is arguably easier than implementing a KD. Further, the use of KD may be particularly challenging in patients with AUD who often present with low bodyweight and poor nutritional status (Addolorato et al. 1998; Jeynes and Gibson 2017). Because heavy alcoholic consumption can itself induce ketoacidosis, a pathological state of profound ketosis (McGuire et al. 2006), it would nevertheless be important to ascertain ketosis and nutritional status in patients before any ketogenic intervention. Third, our findings indicate that diet-induced ketosis can decrease at least some withdrawal symptoms when administered at abstinence start rather than throughout alcohol exposure, e.g., when a patient with AUD is admitted for detoxification. Thrice-daily intake of KME for 28 days was safe and well tolerated in healthy volunteers (Soto-Mota et al. 2019), and we believe that therapeutic ketosis warrants evaluation as an adjunctive treatment to early detoxification in alcohol-dependent patients seeking to become abstinent. Clinical trials are currently ongoing (clinicaltrials.gov NCT03878225, NCT03255031). Whether KD or ketone supplements can also reduce later alcohol withdrawal symptoms such anxiety, sleep disorders, or other affective symptoms linked to relapse, should be evaluated further. Finally, mechanisms of action largely remain to be elucidated.