Empagliflozin Induced Ketosis, Upregulated IGF-1/Insulin Receptors and the Canonical Insulin Signaling Pathway in Neurons, and Decreased the Excitatory Neurotransmitter Glutamate in the Brain of Non-Diabetics

Sodium-glucose cotransporter-2 inhibitors (SGLT2is), such as empagliflozin, lower blood glucose in type 2 diabetes mellitus and improve cardiorenal outcomes regardless of diabetes presence. Whether SGLT2is exert any effects on the brain’s metabolism has not been studied. We conducted a single-arm clinical trial to investigate the effects of once daily administration of oral empagliflozin (25 mg) for 14 days on systemic and brain metabolism in 21 non-diabetics aged 55 years old or older. Empagliflozin lowered circulating insulin and elevated β-hydroxybutyrate over 34-h periods, both following its first administration and after 14 days of daily administration, with minor alterations in glucose homeostasis. Levels of phosphorylated insulin-like growth factor-1 receptor (pIGF-1R), phosphorylated insulin receptor (pIR), phosphorylated-in-tyrosine insulin receptor substrate-1 (pY-IRS-1), and phosphorylated protein kinase B or AKT (pAKT) were increased in extracellular vesicles enriched for neuronal origin (NEVs) following the first empagliflozin administration, but not after 14 days. Our finding of IGF-1R upregulation in NEVs is promising because several post-mortem and epidemiological studies support the idea that upregulation of IGF signaling may protect against Alzheimer’s disease (AD). Moreover, our finding showing activation of insulin signaling and, in particular, the canonical pathway (pIR, pY-IRS-1, pAKT) in NEVs is important because such changes have been repeatedly associated with neuronal survival. Using brain magnetic resonance spectroscopy (MRS), we detected decreased concentrations of the excitatory neurotransmitter glutamate and its precursor glutamine after empagliflozin administration. This finding is also encouraging since glutamatergic excitotoxicity has long been implicated in AD pathology. Overall, our findings may motivate the repurposing of SGLT2is for use in AD and other, related diseases that are characterized by downregulation of IGF-1/insulin signaling in neurons and excitotoxicity.


Introduction
Sodium-glucose cotransporter-2 inhibitors (SGLT2is) were originally approved for treating type 2 diabetes mellitus (T2DM) because they lower blood glucose and HbA1c levels. These effects result from the inhibition of sodium and glucose reabsorption at the proximal tubules of the kidneys leading to their spillage in the urine [1]. The SGLT2i empagliflozin has also recently been approved for the treatment of adults suffering from heart failure with reduced ejection fraction (HFrEF). This approval came after a large multicenter trial reported a reduction in cardiovascular mortality or hospitalization due to heart failure and a reduction in kidney function decline, regardless of the presence of T2DM [2]. The mortality and cardiorenal benefits of SGLTIs may be partially explained by an induction of a state that mimics the adaptive cellular response to starvation or fasting,

Study Procedures during Visits
Study procedures are shown in Figure 1b. Subjects were admitted for 34 h during visits 1 (baseline), 2 (first dose) and 3 (last/14th dose). Participants were admitted the evening before each visit to facilitate study procedures. After a 12-h overnight fast, participants underwent intravenous (IV) placement for a fasting blood draw (t o ), followed by drug ingestion (visits 2 and 3) and repeated blood draws (all visits) over a 34-h window (initially every 1.5 h × 2 times, then every 1 h × 1 time, then every 1.5 h × 8 times, then every 2 h × 4 times, then every 1.5 h × 2 times, then every 1 h × 1 time, and finally every 1.5 h × 4 times). For serum collection, blood was collected into BD Vacutainer serum separation tubes (gold top, SKU: 367983). For plasma collection, blood was collected into BD Vacutainer EDTA tubes (purple top, SKU: 367856). All blood samples were immediately centrifuged at 4 • C on site and plasma and serum were stored in frozen aliquots at −80 • C in less than 1 h. Each aliquot was thawed only once. Serum metabolites and plasma hormones were measured from each blood draw (t 0 to t 34 ). TEV and NEV biomarkers were measured at fasting timepoints only (visit 1 t o , visit 2 t 24, visit 3 t 24 ) to limit variation in lipoprotein concentrations that interfere with EV isolation. The standardization of the timing of blood draws, the fact that these were performed in the fasting condition, and the processing of plasma samples in < 1 h with immediate storage at −80 • C are steps that satisfy key recommendations regarding pre-analytical factors affecting EV isolation [16]. Participants had breakfast around 8.30 am (30 min after drug ingestion), lunch around 12.30 pm and dinner around 5.30 pm. MRS was performed around 4 pm on each visit. Timings of blood draws, meals and MRS were similar across all visits.
For 13 days after visit 2 (first day on medication), the participants ingested empagliflozin every morning at home about 30 min before breakfast. The participants had a continuous glucose monitoring system (FreeStyle Libre Pro 1.1.1) placed between visits to further safeguard against hypoglycemia. The end of visit 3 concluded all study procedures. . Study procedures include: fasting blood draws for isolation of EVs (visit 1 t o , visit 2 t 24 and visit 3 t 24 ); empagliflozin administration from visit 2 to visit 3; breakfast, lunch and dinner; brain magnetic resonance spectroscopy (MRS); repeated blood draws for serum metabolites and plasma hormones throughout every visit [17].

Isolation of Total and Neuronal Extracellular Vesicles from Blood Plasma
An extensive description of the methodology used to isolate NEVs from plasma has been previously reported by our group [20]. Plasma aliquots (0.5 mL) were defibrinated using Thrombin (System Biosciences, Inc., Palo Alto, CA, USA). Plasma was then treated with particle precipitation solution, Exoquick (System Biosciences, Inc., Palo Alto, CA, USA). The resulting pellet containing crude EVs was resuspended in 0.7 mL of distilled water supplemented with protease and phosphatase inhibitors and was centrifuged at 400× g for 5 min at +4 • C to remove insoluble contaminants. After transfer to a clean tube, the supernatant was incubated with 4 µg of mouse anti-human CD171 (L1CAM; clone 5G3) biotinylated antibody (Thermo Scientific, Inc., Waltham, MA, USA) for 2 h at +4 • C, followed by incubation with 25 µL of Pierce Streptavidin Plus UltraLink Resin (Thermo Scientific, Inc., Waltham, MA, USA) for 1 h at +4 • C. After centrifugation at 800× g for 10 min at +4 • C, the supernatant was removed, and the pellet containing NEV-bead conjugates was further purified of soluble material by adding 500 µL of distilled water (with inhibitors) and centrifugation at 800× g for 10 min at +4 • C. Following removal of the supernatant, NEVs were eluted with 200 µL of 0.1 M glycine and beads were separated by centrifugation at 4500× g for 5 min at +4 • C. The resulting supernatant containing NEVs was transferred to a clean tube and the pH was immediately neutralized using 1 M tris-HCl. NEVs were lysed and protein was extracted using M-PER (Thermo Scientific, Inc., Waltham, MA, USA) and two freeze-thaw cycles. The NEV lysate was stored at −80 • C. A variation of the same methodology was used to capture TEVs utilizing 4 µg of a mix of antibodies against three canonical EV markers: CD9 (clone KMC8; BD Pharmingen, Inc., San Diego, CA, USA), CD63 (clone TEA3/18; Abnova, Taipei City, Taipei, Taiwan), and CD81 (clone 1.3.3.22; Ancell, Inc. Bayport, MN, USA).
To demonstrate the ultrastructural properties of NEVs, cryogenic transmission electron microscopy (Cryo-TEM) of intact NEV was performed at the Nanoscale and Microscale Research Centre of the University of Nottingham, UK. We used Holey carbon TEM grids (EM resolutions, Sheffield, UK). NEV samples were left to adsorb onto the grids (5 µL/grid) for 2 min, and then excess solution was removed using filters. The NEV samples were blotted for 1 s and were frozen in liquid ethane using a Gatan CP3 plunge freezing unit (Ametek, Leicester, UK). The frozen samples were loaded to a Tecnai G2 Spirit BioTWIN, a 20-120 kV/LaB6 Transmission Electron Microscope, with Cryo-TEM carried out with an accelerating voltage of 100 kV. We obtained the images using an inbuilt Gatan SIS Megaview IV digital camera.
Recently, we published additional evidence on the characterization of NEVs, demonstrating by WBs that L1CAM+ NEVs display the full length 220 kD characteristic band for L1CAM, which is present in brain lysate. This is unlike other sub-populations of plasma EVs that display solely the 200 kD band, a band which corresponds to soluble L1CAM [21]. Moreover, L1CAM+ NEVs contain many-fold higher L1CAM (normalized to canonical EV marker CD9) compared to other plasma EV sub-populations, suggesting the specificity of the anti-L1CAM immunoprecipitation [21]. Finally, using confocal fluorescence microscopy after double immunolabeling with L1CAM and neuronal marker VAMP2, we demonstrated the co-existence of L1CAM and VAMP2 on particles at the size range of single EVs [21].

Magnetic Resonance Spectroscopy
To measure in vivo brain metabolite concentrations, single-voxel 1 H MRS data were acquired on a Philips Achieva 3T whole-body MR scanner equipped with an 8-channel SENSE head coil. A 25 × 18 × 20 mm 3 MRS voxel was placed within the posteromedial cortex, centered to ensure maximal coverage of bilateral precuneus, as previously undertaken [22]. Point-Resolved Spectroscopy (PRESS) with non-suppressed water reference was used to acquire metabolite concentrations, including standard water-scaled metabolites Ala, Asp, Cr, PCr, GABA, Glc, Gln, Glu, GPC, GSH, Ins, Lac, NAA, NAAG Scy, Tau, Pch, as well as the combined signals for Glu & Gln (Glx), GPC & PCh, and NAA & NAAG [23,24]. The ketone-related metabolites BHB, AcAc, and Acetone (Ace) were detected via a modified basis set provided in July of 2019 by LCModel's developer, Dr. Stephen Provencher. This modification consisted of a supplemental scanner specific LCModel "control.txt" file that allowed us to measure BHB, AcAc, and Ace in addition to the standard set of metabolites in the basis set (Supplemental Othor Materials S2). PRESS parameters were TE = 35 msec, TR = 2000 msec, 256 averages, direct dimension bandwidth = 2 kHz, and 2048 sample points. Reliability of the measurements and fitting procedure for each metabolite was assessed using Cramer-Rao lower bounds (CRLB) [25], and line widths for water resonance were monitored for intra-subject scan reliability and were previously observed at a stable (mean ± SD) 7.3 ± 1.7 Hz.

Statistical Methods
Biofluid biomarkers were analyzed using linear mixed models with repeated measures in SPSS (Build 1.0.0.1508). For circulating metabolites/hormones assessed repeatedly over the course of 34 h during each visit, the models included participant ID as a random effect, "Visit" and "Timepoint" as repeated-measures variables, and the factors "Visit", "Timepoint", "Sex", and the interaction "Visit*Sex" as terms for fixed effects. For NEV/TEV biomarkers and calculated HOMA-IR, the models included "Visit" as the repeated-measures variable. The terms for fixed effects included the factors "Visit", "Sex", the covariate "BMI", and the interactions "Visit*Sex" and "Visit*BMI". We included the interactions "Visit*Sex" and "Visit*BMI" in our models because of evidence that SGLT2 inhibitors may have different clinical effects depending on sex and BMI [26][27][28]. To correct skewness, EV biomarker measures were natural logarithm (ln)-transformed before being implemented into models. 1 H MRS data was processed in LCModel, an automated fitting routine that quantifies the concentration of select metabolite resonances within the MRS voxel [29]. Water-scaled metabolite concentrations (mmol/L) obtained via LCModel were then analyzed using R v.3.6.3. To minimize unreliable measurements, metabolites were censored using a strict CRLB% of <20 for the standard metabolites and a lenient exploratory < 100 for ketone metabolites (BHB, AcAc, Ace). Ala, Asp, GABA, Glc, PCh, Lac, NAAG, Scyllo, and Tau were excluded from the analysis set due to having 50% or more CRLB-censored values. Additionally, a set of weights was derived from the inverse of the CRLB for each measurement to reduce the influence of less reliable (i.e., high CRLB) measurements [30]. This ensured that more reliable (i.e., low-CRLB) measures had more impact on the statistical model than less reliable ones. We then derived the fraction of CSF (fCSF) in the MRS voxel to correct for the potential confounding effects of varying CSF volume in the precuneal voxel. To do this, we used the T 1 -weighted anatomical images (MP-RAGE) and a custom MATLAB script adapted from Partial Volume Code for Philips MRS data provided by Dr. Nia Goulden and Dr. Paul Mullins of Bangor University [31]. This script output the gray matter (GM), white matter (WM) and CSF volumes used to calculate the fractional volume of cerebrospinal fluid (fCSF = CSF vol/ (GM vol + WM vol + CSF vol )) within the MRS voxel. Finally, we used a linear mixed effects model (nlme v.3.1) with participant ID as the random effect, MRS "Visit" (1-3) as repeated measure, each metabolite as the dependent variable, and "Visit", "Age", "Sex", and "fCSF" as factors or covariates. "Age" was included in the model given strong evidence that it affects MRS metabolites [32]. This model included the aforementioned weighting term for the inverse CRLB for each metabolite, restricted maximum likelihood (REML) as the variance estimator, and first order autoregression (AR(1)) for autocorrelation.

Participant Flow and Baseline Characteristics
Of the 39 participants that underwent screening, 18 were excluded prior to drug administration for eligibility reasons, 21 took at least one drug dose, and 20 took all drug doses completing all study visits (Figure 1a). Baseline characteristics are presented in Table 1.

Outcomes
In this study, we implemented repeated-measures analyses as appropriate for each outcome. Biomarkers directly assayed in plasma or serum (glucose, insulin, NEFAs, glucagon, ketone bodies) were measured multiple times over 34 h to determine their diurnal variation and relationship to meals. To demonstrate the effects of empagliflozin both acutely and after chronic administration, all outcomes were assessed at three timepoints (baseline, following the first dose, following 14 days of daily administration). We provide three different metrics (% change from baseline, F test, mean difference) for each outcome. For outcomes repeated over 34-h periods we report results for the overall pattern of repeated measurements rather than a single timepoint.

Empagliflozin Acutely Upregulated IGF-1/Insulin Cascade-Associated Proteins in Plasma-Derived NEVs, but Not in TEVs
Cryo-TEM pictures of NEV preparations demonstrate multiple examples of small and medium size EVs with typical morphology (single membranous nanoparticles) and size distribution. These pictures are highly consistent with the isolation of a mixed population of exosomes and microvesicles (Supplemental Figure S1; scale bars included for reference). Figure 3 and Table 3 show phosphorylated levels of IGF-1/insulin cascade mediators measured in TEVs and NEVs at visit 1 (t o ), visit 2 (t 24 ), and visit 3 (t 24 ). No changes of pIGF-1R levels were identified in TEVs after the first or last empagliflozin doses compared with baseline. However, levels of pIGF-1R in NEVs significantly increased after the first empagliflozin dose compared with baseline (F 1, 18 = 7.706, p = 0.012; MD = 0.247; 95% CI = 0.075, 0.419). In addition, there was a significant interaction between Visit and BMI (F 1, 18 = 9.482, p = 0.06), with lower-than-mean BMI being associated with no change in NEV pIGF-1R levels (F 1, 18 = 0.013, p = 0.911), mean BMI being associated with NEVs pIGF-1R increase (F 1, 18 = 9.076, p = 0.007), and higher-than-mean BMI being associated with a sharper NEV pIGF-1R increase (F 1, 18 = 18.381, p < 0.001). After the last dose, NEV pIGF-1R levels were similar to baseline (  . means (SE) of phosphorylated (p) and total (t) levels of IGF-1/insulin cascade-associated proteins measured in extracellular vesicles enriched for neuronal origin (NEVs) at baseline (visit 1 t0 (V1)), 24 h after the first dose (visit 2 t24 (V2)), and 24 h after the last dose (visit 3 t24 (V3)). Statistical analysis involved repeated-measures linear mixed models with "Visit" as the repeated-measures variable and additional variables as . means (SE) of phosphorylated (p) and total (t) levels of IGF-1/insulin cascade-associated proteins measured in extracellular vesicles enriched for neuronal origin (NEVs) at baseline (visit 1 t 0 (V1)), 24 h after the first dose (visit 2 t 24 (V2)), and 24 h after the last dose (visit 3 t 24 (V3)). Statistical analysis involved repeated-measures linear mixed models with "Visit" as the repeated-measures variable and additional variables as factors and covariates (see Section 2.9 for details). Significance is indicated as *, corresponding to p ≤ 0.05. (b). IGF-1/insulin signaling in neurons reflecting acute biomarker changes detected in NEVs. Significant increases are marked with * [17].
Significant results for the omnibus test across all three visits are indicated by a black horizontal bar above the individual plot, with the p-value above. Columns from left to right include the standard abbreviation for each metabolite, mean overall Cramer-Rao lower bounds and its standard deviation (SD), mean and SD for each water-scaled metabolite concentration for the three visits, degrees of freedom (numerator, denominator), F-value, and p-value. Statistical analysis involved repeated-measures linear mixed models with "Visit" as the repeated-measures variable and water-scaled concentration as the dependent variable. Additional factors and covariates are detailed in the Statistical Methods Section 2.9.

Discussion
In this clinical study, we found that empagliflozin decreased insulin, increased glucagon, and increased NEFAs and BHB blood levels in individuals without diabetes [33]. While the first dose of empagliflozin lowered circulating glucose levels, after 14 days of daily empagliflozin (visit 3), glucose levels were similar to baseline (visit 1). These findings suggest a rapid diminution of the glucose-lowering effect of empagliflozin in individuals without diabetes and a restoration of glucose homeostasis, which facilitates any potential repurposing of the drug for medical indications besides diabetes. Interestingly, we found that empagliflozin acutely elevated pIGF-1R, pIR and common downstream mediators of IGF and insulin signaling such as pY-IRS-1 and pAKT in NEVs but not in TEVs, indicating a neuronal-specific upregulation of these pathways. Although no elevation of ketone bodies was directly observed in the brain using MRS, empagliflozin reduced the excitatory neurotransmitter glutamate and its precursor glutamine, metabolites that have been shown to respond to ketogenic interventions.
Overall, empagliflozin induced metabolic changes that overlap with those of fasting and calorie restriction, such as a metabolic shift towards lipolysis and ketogenesis [3,33,34]. Also analogous to calorie restriction, chronic empagliflozin administration decreased circulating insulin levels without significantly altering glucose homeostasis [34,35]. Empagliflozin's insulin-reducing capability is of great importance for neurodegenerative diseases, as chronically elevated insulin levels downregulate autophagy and inhibit clearance of abnormal proteins [4,36,37]. Evidence from other clinical studies suggests that empagliflozin induces beneficial cellular responses regardless of presence or absence of hyperglycemia [2,4]. Such benefits may be due to empagliflozin-induced activation of SIRT1 and AMPK, potentially as a response to glucose loss in the urine, which constitutes a form of nutrient deprivation. Activation of SIRT1 and AMPK has been associated with increased autophagy and reduced inflammation and oxidative stress. These downstream effects may promote cellular resilience in the brain and other organs [4,38,39].
We showed that empagliflozin elevated circulating BHB levels after acute and chronic dosing, in agreement with a previous study of empagliflozin that showed a trend towards BHB elevation in a non-diabetic population [33]. Accumulating evidence suggests that ketones are an efficient alternative to glucose as an energy fuel, and so may be potentially useful as an intervention in chronic neurodegenerative disorders such as AD, in which the brain has a decreased ability to efficiently utilize glucose [14,15,40]. Therefore, BHB elevation by empagliflozin may be beneficial for such disorders. Interestingly, despite the observed BHB increase, AcAc was decreased and the ratio AcAc/BHB was also decreased. The decreased AcAc/BHB ratio may reflect increased conversion of AcAc to BHB, typically associated with an elevated NAD + /NADH ratio. Evidence from animal models indicates that NAD + induces enzymes involved in promoting neuronal health, such as sirtuins, and therefore a potential NAD + increase by empagliflozin would be beneficial in AD and other neurodegenerative disorders [41].
Following the first dose, empagliflozin induced significant elevations of pIGF-1R, pIR, and common downstream mediators of the IGF and insulin signaling pathways such as pY-IRS-1 and pAKT (with the caveat of high CV for the internal control) in NEVs. These changes indicate an activation of the IGF-1 and insulin signaling pathways in neurons. No NEV biomarker differences were observed after the last empagliflozin dose compared with baseline, perhaps due to a potential adaptation to repeated dosing. Our findings on the acute elevation of proteins of the IGF-1 and insulin signaling pathways detected in NEVs are important because these NEV biomarkers have been linked to AD diagnosis [42], age-related cognitive decline [43], grey matter volume and volume of white matter hyperintensities [44]. Post-mortem studies, have shown that AD brains are characterized by decreased IGF-1R and IGF-1 mRNA levels (that worsen with increasing Braak stage) [45], as well as decreased IRS-1 levels [46]. In the Framingham study, decreased serum IGF-1 levels predicted the development of future AD, providing additional evidence for the involvement of IGF-1/IGF-1R signaling disturbances in AD [47]. In contrast, evidence mainly from animal models indicates that inhibition of IGF-1R is associated with longevity and neuronal resilience to AD pathologies [48][49][50]. Therefore, the effects of empagliflozin on the IGF-1 signaling cascade and its interplay with AD pathologies requires further investigation in vitro and in vivo. Of note, empagliflozin induced elevations of proteins of the canonical pathway of insulin signaling (pIR, pY-IRS-1, pAKT). Interestingly, activation of the canonical pathway is known to promote neuronal survival [51,52]. Empagliflozin did not alter proteins of the alternative pathway (i.e., (i) ERK which has been associated with both neuronal survival (acute activation) and death (chronic activation), and is additionally involved in synaptic plasticity [53][54][55][56]; (ii) JNK which has been associated with neuronal death and synaptic plasticity [57]; and (iii) p38 which has been associated with neurotoxicity or neuroprotection, depending on the p38 subtype activated or neuron type involved [58]). Prior NEV studies have shown that drugs with different mechanisms of action can affect the cascade downstream of IRS-1 differently; for example, exenatide acts on AKT [59], but infliximab acts on JNK, p38, ERK1/2 [60].
The primary MRS finding was a significant decrease in brain glutamate and glutamine after empagliflozin administration, which is consistent with earlier findings in ketogenic intervention studies. BHB has been shown to decrease glutamate availability in vitro, ostensibly due to reduced malate-aspartate shuttle activity in neurons reliant on BHB [61]. This finding is also consistent with the effect of ketogenic diets on glutamine levels as a treatment for glioblastoma, in which ketogenic diets inhibit both glycolysis and glutaminolysis [62]. This decrease in glutamate is also interesting in the context of both epilepsy [63] and AD [64,65], in which glutamatergic excitotoxicity has long been implicated in their respective pathological cascades. In AD, the presence of Tau tangles and Aβ plaques appears to exacerbate the effects of glutamatergic excitotoxicity [66,67]. Unfortunately, the ketone metabolites (BHB, AcAc, Ace) were difficult to detect reliably in this study due to their low concentration.
The main limitation of this study was its small sample size. Nevertheless, the study had many strengths. First, it included repeated assessments of metabolites and hormones over 34 h to provide detailed time courses of their change. Second, we gained insight into how drug effects may change with meals, the sleep-wake cycle, and duration of treatment. Third, our study included equal numbers of men and women, and therefore our findings have added value, demonstrating that both sexes have increased circulating BHB, upregulated IGF-1/insulin receptors and proteins of the canonical pathway of insulin signaling in neurons, and decreased brain Gln and Glu. Fourth, our study population reflects the age at which this class of drugs may be studied as potential therapies for cognitive impairment. Additional strengths were the integration of multiple complementary methodologies to probe brain effects, particularly NEV isolation, and brain MRS. This enabled us to derive a plethora of biologically relevant outcomes, including circulating metabolites and hormones, brain metabolites, and EV biomarkers. This study also distinguished between neuronalspecific and systemic effects on biomarkers of the IGF-1/insulin cascade by isolating different EV populations.
Overall, our findings confirm the known ketogenic effect of empagliflozin and additionally provide novel evidence that empagliflozin potentially promotes neuronal survival through activation of the canonical pathway of insulin signaling and may act as a neuroprotectant through upregulation of IGF-1R (based on studying NEVs). We also showed that empagliflozin decreases potentially harmful excitatory neurotransmission in the brain by decreasing glutamate and glutamine (based on brain MRS). These neuronal and brain effects may be beneficial in relevance to AD and related dementias and motivate the repurposing of empagliflozin for these disorders. Future double-blind placebo-controlled randomized clinical trials that may include these and additional biomarkers, as well as cognitive and functional brain outcomes, are needed to test this hypothesis.