Small-molecule ketone esters treat brain network abnormalities in an Alzheimer’s disease mouse model

Altered brain network activity and the resulting hypersynchrony are important for the pathogenesis of cognitive decline in Alzheimer’s disease (AD) mouse models. Treatments that reduce epileptiform discharges (EDs) or network hyperactivity improve cognition in AD models and humans. We first show that ketogenic diet, but not fasting, rapidly and persistently reduced EDs in the hAPPJ20 Alzheimer’s mouse model over timescales of hours to months. Then, to identify the specific mechanism of the pleiotropic ketogenic diet, we developed small molecule ketone esters to deliver ketone bodies pharmacologically. Two ketone esters recapitulate ED suppression without other dietary manipulation, over time scales of minutes to one week. This small molecule rescue was associated with reduced low-frequency oscillatory activity similar to the recently reported mechanism of an NMDA receptor modulator molecule in this model. Long-term KD resulted in cognitive improvement and in a sex-stratified analysis also improved survival in the more severely affected hAPPJ20 males. Agents that deliver ketone bodies via small molecules or act on downstream targets may hold therapeutic promise in AD through the mechanism of improved network function and reduced epileptiform activity.


Summary 1
Altered brain network activity and the resulting hypersynchrony are important for the 2 pathogenesis of cognitive decline in Alzheimer's disease (AD) mouse models. 3 Treatments that reduce epileptiform discharges (EDs) or network hyperactivity improve 4 cognition in AD models and humans. We first show that ketogenic diet, but not fasting, 5 rapidly and persistently reduced EDs in the hAPPJ20 Alzheimer's mouse model over 6 timescales of hours to months. Then, to identify the specific mechanism of the 7 pleiotropic ketogenic diet, we developed small molecule ketone esters to deliver ketone 8 bodies pharmacologically. Two ketone esters recapitulate ED suppression without other 9 dietary manipulation, over time scales of minutes to one week. This small molecule 10 rescue was associated with reduced low-frequency oscillatory activity similar to the 11 recently reported mechanism of an NMDA receptor modulator molecule in this model. 12 Long-term KD resulted in cognitive improvement and in a sex-stratified analysis also 13 improved survival in the more severely affected hAPPJ20 males. Agents that deliver 14 ketone bodies via small molecules or act on downstream targets may hold therapeutic 15 promise in AD through the mechanism of improved network function and reduced 16 epileptiform activity. 17

Introduction 1
Links between epilepsy and Alzheimer's disease (AD) are seen in both human patients 2 and mouse models. Human patients with AD may commonly have subclinical 3 epileptiform discharges (EDs ) (Vossel et al., 2016), and overt epilepsy is associated 4 with more rapid cognitive decline (Lam et al., 2017;Vossel et al., 2013). Mechanistic 5 studies in mouse models of AD have shown that altered oscillatory activity and EDs 6 stem from dysfunctional inhibitory interneurons (Palop and Mucke, 2016), which are key 7 elements of cortical circuits underlying cognition (Jagirdar and Chin, 2019;Kvitsiani et 8 al., 2013). Pathological tau and Aβ promote network hyperexcitability, including through 9 interactions with other AD risk genes (Kazim et al., 2021;Voskobiynyk et al., 2020). 10 Pharmacological treatments or genetic manipulations that reduce EDs improve 11 cognition in these models, including an NMDA receptor positive allosteric modulator that 12 suppresses epileptiform activity by reducing low-frequency oscillations (Hanson et al., 13 2020;Martinez-Losa et al., 2018;Merlini et al., 2021;Sanchez et al., 2012;Verret et al., 14 2012). In humans with mild cognitive impairment (MCI), low doses of the antiepileptic 15 levetiracetam improves network hyperactivity and cognitive performance (Bakker et al., 16 2012). Thus, targeting subclinical epileptiform activity or network hyperactivity are 17 promising new therapeutic approaches to AD and are now being tested in early clinical 18 trials (Bakker et al., 2015;Bakker et al., 2012;Vossel et al., 2021;Vossel et al., 2017). 19 20 Ketogenic diet (KD) has long been used to treat certain forms of epilepsy, though 21 primarily inherited developmental disorders (Keene, 2006). Ketogenic diets are defined 22 by restricting carbohydrate intake sufficiently to induce endogenous synthesis of ketone 23 bodies. Ketone bodies are small molecule metabolites, primarily β-hydroxybutyrate 1 (BHB) and acetoacetate, which are synthesized in the liver from lipids in order to 2 provide an alternative, fat-derived, source of circulating energy when glucose is scarce, 3 such as during fasting or a ketogenic diet. But KD is highly pleiotropic and its 4 mechanism of action varies between epilepsy models, in some models involving 5 glycolysis, insulin, specific fatty acids, or even gut microbiome metabolites rather than 6 ketone bodies (Kim and Rho, 2008;Olson et al., 2018). Aside from clinical use in 7 childhood epilepsies, ketogenic interventions including KD (Lilamand et al., 2021b;Neth 8 et al., 2020;Phillips et al., 2021;Taylor et al., 2018) and medium chain triglyceride 9 supplements (Croteau et al., 2018;Fortier et al., 2021;Lilamand et al., 2021a) are 10 under clinical investigation for AD primarily based on the hypothesis of improved energy 11 metabolism from ketone bodies (Cunnane et al., 2020). However, a wide variety of 12 alternative relevant mechanisms in AD have been proposed, based on the complex 13 physiology of ketogenic diets and the pleotropic molecular actions of ketone bodies 14 (Newman and Verdin, 2017), including inflammatory modulation via NLRP3 (Shippy et 15 al., 2020;Youm et al., 2015), G-protein coupled receptors (Hasan-Olive et al., 2019;Wu 16 et al., 2020), epigenetics (Shimazu et al., 2013), the microbiome (Ang et al., 2020;Ma 17 et al., 2018;Olson et al., 2018), and others. 18

19
The key barrier to translation of ketogenic-related therapies for AD is a better 20 understanding of specific relevant mechanisms that can guide more targeted therapies 21 and be linked to translatable biochemical or physiological biomarkers. KD is a 22 challenging intervention to implement in cognitively impaired individuals (Taylor et al., 23 2018) and the necessarily high fat content can be associated with hyperlipidemia, 1 constipation, electrolyte disturbances and other side effects (Cervenka et al., 2017). 2 Exogenous delivery of ketone bodies is an obvious candidate for a more targeted 3 therapy, if shown to be mechanistically relevant. Yet the ketone bodies BHB and 4 acetoacetate are rapidly-metabolized organic acids, so direct exogenous delivery of 5 pharmacological quantities requires a deleteriously large salt or acid load. Ketone esters 6 are a technology to avoid this difficultly by masking the carboxylic acid moiety in an 7 ester bond with other ketone bodies or ketogenic precursor molecules. The specific 8 design of ketone esters can be targeted to deliver varying ratios of BHB and 9 acetoacetate, to provide only exogenous ketone bodies or stimulate endogenous 10 ketogenesis as well, and to optimize delivery kinetics -once these goals can be defined 11 in a particular therapeutic context. 12

13
Here we provide a comprehensive and rigorous multidimensional 14 electroencephalography (EEG) and behavioral study of the concurrent effects of a 15 ketogenic diet or pharmacological delivery of ketone bodies via novel small molecule 16 ketone esters on brain network and epileptiform activity, cognitive decline, and survival, 17 using the well-characterized hAPPJ20 mouse model of AD. We find that a ketogenic 18 diet consistently reduces epileptiform discharges over time frames ranging from minutes 19 to months. To define the specific mechanism, we show that similar reduction of ED is 20 observed using two structurally related ketone esters without ketogenic diet; and that all 21 of these interventions act independent of gamma oscillations associated with inhibitory 22 interneurons but instead the mechanism of ED suppression may be via suppression of 23 Crawford, 2021). Importantly for potential clinical applications, pleiotropic mechanisms 1 that are not required for ED suppression may underly undesired side effects associated 2 with KD. It is therefore crucial to identify the specific mechanisms of KD that are 3 responsible for ED suppression. A key hallmark of KD is its stimulation of endogenous 4 ketone body production, resulting in circulating blood BHB levels in the millimolar range. 5 To test if ketone bodies are the active component induced by KD in ED suppression, we 6 engineered novel small molecule ketone esters to deliver naturally occurring ketone 7 bodies pharmacologically and therefore bypass the complex diet intervention. In order 8 to best emulate the physiology of ketone body production in KD, we engineered ketone 9 ester molecules that comprise a BHB moiety (for immediate BHB delivery) ester-linked 10 to six-carbon medium chain fatty acids (to stimulate endogenous ketone body 11 production). Medium-chain fatty acids are rapidly metabolized to a physiological and 12 naturally occurring mix of BHB and acetoacetate in the liver regardless of diet context. 13 crossover study of hAPPJ20 mice fed only control diet, we injected both C6-BHB and 17 saline intraperitoneally on separate days and recorded 50-minute EEGs both before and 18 after each injection (Fig. 3D). Injection of C6-BHB increased plasma BHB (Fig. 3E,  19 measured post-recording) and substantially suppressed EDs compared to saline 20 injection (Figs. 3F and 3G). The effect was evident from the first minutes of EEG 21 recording, started 20 minutes after injection. These data are consistent with the model 22 Similar analysis of the cohort of mice that alternated from control diet to KD and back to 1 control over a period of 3 weeks ( Fig. 2A) revealed the same patterns (Supplementary 2 Fig. 1). Overall, these data are consistent with KD acting downstream or independent of 3 gamma oscillations. 4

5
We found similar results upon analysis of EEG data from the C6x2-BHB ketone ester 6 feeding cohort (Fig. 4). The ketone ester reduced EDs at all movement levels (Fig. 5G) 7 and at all levels of movement-adjusted gamma activity (Fig. 5H), and did not increase 8 the induction of gamma power with increased movement (Fig. 5I). There was no change 9 in the overall fraction of EEG power in the gamma range ( Supplementary Fig. 2). C6x2-10 BHB and 1,3-butanediol both increased movement during the EEGs to a similar degree 11 ( Supplementary Fig. 3). A simpler analysis of the smaller injection EEG data set (Fig. 3) 12 similarly showed that ED suppression by C6-BHB was not a result of increasing 13 exploratory movement (mice moved less after C6-BHB than saline injection) and that 14 ED suppression was greater at rest than while moving ( Supplementary Fig. 4). 15 16 Interestingly, while C6x2-BHB did not alter gamma power, it did reduce power in the 17 lower frequency bands delta, theta, beta, and alpha, spanning 0.5-20 Hz (Fig. 5J). For 18 individual mice, greater power reduction in these bands by C6x2-BHB was associated 19 with greater suppression of EDs (Fig. 5K). Recently, we reported a positive allosteric 20 NMDA receptor modulator that improved cognitive function and reduced epileptiform 21 discharges in hAPPJ20 mice via the mechanism of reducing aberrant low-frequency 22 oscillatory power (Hanson et al., 2020). C6x2-BHB may act via a similar mechanism. 23 Altogether, the ketone esters C6x2-BHB and C6-BHB alone are sufficient to elicit the 1 benefits of KD and appear to act via a similar or identical mechanism to ketogenic diet, 2 suppressing EDs independently of gamma oscillations possibly through rescue of 3 aberrant low-frequency oscillations, and these effects are not explained by alterations in 4 locomotor activity. 5 6 KD improves long-term memory and survival in hAPPJ20 mice 7 Finally, we investigated whether ED suppression by KD is associated with long-term 8 improvements in memory or survival in hAPPJ20 mice. These mice display 9 characteristic deficits in context-dependent learning, which can be observed as a failure 10 to habituate to a novel environment. During a three-month study (Fig. 2D), mice were 11 exposed to the open field four times in the first month of treatment as training, followed  To further investigate the effect of KD on memory and premature mortality, we 1 undertook a larger, 7-month study, beginning with 2-month-old hAPPJ20 and WT mice 2 ( Fig. 6E). KD was associated with significant weight gain in both genotypes (Fig. 6F) 3 and higher mean daily calorie intake ( Supplementary Fig. 5). Six blood draws obtained 4 every ~2 weeks from the start of the study showed that KD increases plasma BHB 5 levels, averaging ~1 mM over the 6-month period which is ~10-fold higher than controls 6 ( Fig. 6G). Blood glucose levels were similar in all groups ( During Morris water maze testing, hAPPJ20 mice on KD showed significantly better 16 performance in the hidden-platform training (learning) trials of the water maze than 17 hAPPJ20 mice on the control diet (Fig. 6I). This improvement remained consistent when 18 the location of the platform was moved during reversal training (Fig. 6J). However, there 19 was no difference in performance during the probe (memory) trial of the water maze, 20 either after initial hidden platform training or after reversal training ( Supplementary Fig.  21

8). There was no clear association between sex and performance in either habituation 22
to the open field or water maze ( Supplementary Fig. 9). 23 1 hAPPJ20 mice exhibit substantial early mortality, particularly in males (Davis et al., 2 2020;Verret et al., 2012) that may be due to fatal seizures. KD did not affect the 3 relatively lower mortality among females (Fig. 6K). Remarkably, however, the high early 4 mortality among males was strongly reduced (Fig. 6L). 5 6 Reports of the effect of KD or ketone bodies on Aβ deposition have been conflicting, 7 with Aβ either reduced (Kashiwaya et al., 2013;Van der Auwera et al., 2005;Xu et al., 8 2022) or unchanged (Aso et al., 2013Beckett et al., 2013;Brownlow et al., 2013). 9 While changes in plaque accumulation could not explain the very rapid kinetics we 10 observed for the effect of KD and ketone esters on epileptiform activity, we nevertheless 11 carried out an exploratory immunohistochemical analysis of mice after 7 months on KD 12 In summary, we found that KD suppresses epileptiform discharges, improves context-21 dependent learning and visuo-spatial memory performance, and increases survival in 22 the hAPPJ20 mouse model of AD. ED suppression can be attained with similar effect 23 size and electrophysiological profile by small molecule ketone ester compounds that 1 provide ketone bodies pharmacologically. These data provide a mechanistic 2 understanding of earlier observations of visuo-spatial memory improvement with KD (Xu 3 et al., 2022;Yin et al., 2016) or a ketone ester diet (Kashiwaya et al., 2013) in AD 4 models. We are the first to show that KD reduces AD-induced epileptiform activity, 5 providing a functional mechanism of action linked to rapid cognitive decline in AD 6 patients (Lam et al., 2017;Vossel et al., 2013;Vossel et al., 2016) and with translational 7 applicability in patients (Bakker et al., 2015;Bakker et al., 2012;Vossel et al., 2021;8 Vossel et al., 2017). The magnitude of ED suppression is similar to that obtained by low 9 doses of the antiepileptic drug levetiracetam (Sanchez et al., 2012) or by transgenic 10 expression of SCN1A in interneurons (Verret et al., 2012). We further show for the first 11 time that ketone bodies are the functional component of KD for this mechanism, and a 12 small molecule ketone ester alone can fully replicate ED suppression. We carefully 13 define the kinetics of ED suppression by KD and ketone esters, and we identify a 14 relevant potential intermediate mechanism in reduced low-frequency oscillatory activity. 15 A key strength of this study is its scope and replicability, with a large number of mice of 16 both sexes and varying ages analyzed in 6 separate cohorts testing different timing, 17 intervention, and behaviors, all with consistent results supporting the rigor of the 18

conclusions. 19 20
Our data argue strongly that ketone bodies are the relevant mechanism of ED 21 suppression in this model, given the close similarity in effect size and EEG 22 characteristics between the KD and ketone ester cohorts. The downstream mechanism 23 of ketone bodies in ED suppression is likely multifactorial, but the options are 1 constrained by our observed kinetics. We found the rescue to be robust, immediate, and 2 stable over time. Direct actions on neurotransmitter signal transduction could fit these 3 kinetics. The detailed EEG analysis argues against modulation of inhibitory interneuron-4 associated gamma oscillations, but interestingly the ketone ester suppressed low 5 frequency oscillatory activity similar to our recent report of ED suppression by the 6 NMDA receptor positive allosteric modulator GNE-0723 (Hanson et al., 2020). Low-7 frequency oscillations are associated with the brain resting state and network 8 hypersynchrony in AD models, while suppression in this band favors an active brain 9 state and desynchronized neuronal activity. GNE-0723 acts specifically on GluN2A- studies will be to parse the roles of the two major ketone bodies BHB and acetoacetate. 7 While BHB is the predominant ketone body generated both in KD and from the ketone 8 esters, these all produce acetoacetate as well at a physiological ratio. It is surprising 9 that 1,3-butanediol, which alone among the interventions exclusively produces BHB, 10 generated inconsistent EP spike suppression. However, (R/S)-1,3-butanediol produces 11 both the endogenous R-enantiomer of BHB as well as the non-physiologic S-12 enantiomer, and while they share many common molecular actions (Newman and 13 Verdin, 2017), it is possible S-BHB acts counter to R-BHB in this context. The 14 inconsistent efficacy of fasting, being a potent inducer of ketone bodies, is also 15 surprising and may be due to a counteracting mechanism or exacerbating factor in the 16 complex metabolic milieu of fasting. An additional difference is that KD and both ketone 17 esters provide substrates for endogenous ketogenesis, while 1,3-butanediol does not. 18 The bulk of ketogenesis occurs in the liver with resulting systemic circulation of BHB 19 and acetoacetate, and from the brain's perspective there may be no difference between 20 exogenous and liver-derived ketone bodies. But there is growing evidence for local and 21 paracrine roles for ketogenesis. For example, ketone bodies act locally in the liver to Ketogenic diet improves motor performance but not cognition in two mouse models of 33 Alzheimer's pathology. Croteau, E., Castellano, C.A., Richard, M.A., Fortier, M., Nugent, S., Lepage, M., 7 Duchesne, S., Whittingstall, K., Turcotte, E.E., Bocti, C., et al. (2018). Ketogenic 8 Medium Chain Triglycerides Increase Brain Energy Metabolism in Alzheimer's Disease. 9 Journal of Alzheimer's disease : JAD 64, 551-561. 10.3233/JAD-180202. Roy, M., Beauvieux, M.C., Naulin, J., El Hamrani, D., Gallis, J.L., Cunnane, S.C., and 20 Bouzier-Sore, A.K. (2015). Rapid adaptation of rat brain and liver metabolism to a 21 ketogenic diet: an integrated study using (1)H-and (13) Feasibility and efficacy data from a ketogenic diet intervention in Alzheimer's disease.

45
Alzheimers Dement (N Y) 4, 28-36. 10.1016/j.trci.2017.11.002. 46 Van der Auwera, I., Wera, S., Van Leuven, F., and Henderson, S.T. (2005)  Response of individual mice to KD or fasting. ED per minute are normalized to control 9 diet EEG recordings for each individual mouse. All data are presented as mean ± SEM.  Intraperitoneal injection of C6-BHB in wild-type mice acutely increases blood BHB levels 5 similarly to ketogenic diet, without affecting blood glucose levels (C). D-G, Longitudinal 6 crossover study of C6-BHB injection vs. saline in hAPPJ20 mice (12-24 months old; 7 N=17, 13M, 4F). D, Study schematic. E, Increased blood BHB levels 70 minutes after 8 C6-BHB injection. F, C6-BHB injection reduces ED throughout the 50 minute recording 9 and in total (G) compared to after saline injection. ED per minute are normalized to the 10 pre-injection EEG recordings for each individual mouse injection. All data are presented 11 as mean ± SEM. P-values by paired T-tests (B, G) or Wilcoxon test (E). week reduces epileptiform discharges in hAPPJ20 mice. A, The C6x2-BHB ketone 2 ester, optimized for delivery in food, generates one molecule of BHB directly and 3 3 molecules of either BHB or acetoacetate via fatty acid oxidation (FAO) of its 6-carbon 4 medium chain fatty acid and medium chain alcohol moieties. B, Schematic of the 5 longitudinal study of hAPPJ20 mice (5-14 months old; N=9, 3M, 6F), which included a 6 cross-over feature with an additional control diet period for C6x2-BHB. C, Per calorie 7 macronutrient diet composition of the study diets. C6x2-BHB and 1,3-butanediol were 8 provided at 10% w/w. D, Both compounds increase blood BHB levels vs. control diet, 9 and have minimal effects on blood glucose (E). F, Body weight change over the 1 week 10 diet period. G, C6x2-BHB, but not 1,3-butanediol, reduces ED compared to control-fed 11 EEGs. H, Response of individual mice. ED per minute are normalized to the control diet 12 EEG recordings for each individual mouse during each study period (C6x2-BHB or 1,3-13 butanediol, as in the schematic). All control diet periods were combined for the 14 metabolic measurements. All data are presented as mean ± SEM. P-values by one-way 15 ANOVA with (D, E) or without (F, G, H) matching and Šídák or Holm-Šídák multiple 16 comparisons test.      month study of KD in hAPPJ20 mice starting at 6 weeks old. At start of study, APP Con 6 n=21 (9M, 12F), APP KD n=23 (9M, 14F), WT Con n=26 (12M, 14F), WT KD n=23 7 (16M, 7F). For water maze, APP Con n=12 (3M, 9F), APP KD n=14 (5M, 10F), WT Con 8 n=11 (6M, 5F), WT KD n=11 (6M, 5F). E, Study schematic. Mice gained weight on KD 9 (F), had persistently elevated blood BHB (G), and normal blood glucose (H). KD 10 improved visuo-spatial learning performance for APP mice during both forward (I) and 11 reverse (J) training in the Morris water maze. KD did not affect survival of female 12 hAPPJ20 mice (K) but markedly improved male survival (L). Arrows show the start of 13 study diets (45 days old). All data are presented as mean ± SEM. P-values via one-way 14 (B-D) or two-way (I, J) ANOVA with Tukey's multiple comparisons test, or log rank 15 survival test (K, L). littermates that do not carry the APP transgene. Some ancillary experiments involving 16 effects of ketogenic diet or novel compounds were performed using wild-type C57BL/6 17 male mice from the National Institute on Aging Aged Rodent Colony, usually obtained at 18 11 months of age, with experiments carried out between 11 and 16 months of age. 19 These mice were tested and found to carry the Nnt partial gene deletion common to 20 certain C57BL/6 substrains. 21 22

Experimental diets 23
Customized ketogenic and control diets were obtained from Envigo. The control diet is 1 based on AIN-93M, and contains per-calorie 10% protein, 13% fat, and 77% 2 carbohydrates (TD.150345). The ketogenic diet contains per-calorie 10% protein and 3 90% fat (TD.150348). The primary fat sources are Crisco and corn oil in both diets. 4 Fatty acids in the ketogenic diet are, by weight, approximately 24% saturated, 39% 5 monounsaturated, and 37% polyunsaturated. The ketogenic diet is matched to the 6 control diet on a per-calorie basis for micronutrient content, fiber, and preservatives. 7 The compound diets contained 10% w/w of either the ketone ester C6x2-BHB (~8 8 kcal/g) or 1,3-butanediol (~6 kcal/g). They were otherwise also matched to the control 9 diet on a per-calorie basis for protein content (10% of kcal), fat content (12% of kcal), 10 micronutrients, fiber, and preservatives. Note that the standard vivarium chow (PicoLab 11 5053) contains per-calorie 24% protein, 13% fat, and 62% carbohydrates. The 12 ketogenic diet is of dough-like texture that permits it to be placed in the food well of the 13 cage-top wire lid, in the same manner as pellets. All other diets are firm pellets. All diet 14 were fed ad libitum to avoid confounding effects from partial fasting, with caloric intake 15 and body weights tracked. All custom diets were changed weekly. 16 17

Mouse cohort descriptions 18
In all figures, "APP" is hAPPJ20; "WT" are wild-type littermates that do not carry the 19 hAPP transgene; "Con" or "con diet" is the control diet; "KD" is ketogenic diet. 20 Figure 1E-I: N=8 (3M, 5F), all hAPPJ20. 21 Blood for plasma BHB testing was obtained via minimal distal tail snip, with mice placed 12 into a whole-body restrainer (Braintree Scientific) for comfort. ~40 µL whole blood was 13 drawn for a BHB assay, and collected into microvettes coated with Li-Heparin 14 (Sarstedt). Plasma was separated by centrifugation at 1500 x G for 15 min at 4 °C, and 15 kept frozen at -20 °C until use. Unless noted, blood draws were done in the morning 16 shortly after lights-on (8-11am). 17 18

Blood assays 19
BHB concentrations were determined from plasma using a BHB enzymatic detection kit 20 (Stanbio Laboratory, Boerne, TX). Reactions were run with 3 uL of plasma each, in 21 triplicate. Background absorbance recorded before substrate addition was subtracted 22 form the final absorbance to correct for any hemolysis in the plasma samples. 23