Metabolomic Analysis Identifies Lactate as an Important Pathogenic Factor in Diabetes-associated Cognitive Decline Rats*

The study reveals in chronic diabetes conditions, excess lactate secretion in hippocampus play an inhibitory function on cognition ability, through lactate receptor (GPR81)-depended mechanism. Besides, we proposed hypothesis of “lactate shuttle in chronic diabetic state.” And lactate dehydrogenase-A may be one of drug’s targets for cognitive dysfunction. Graphical Abstract Highlights Quantitative metabolomics identified elevated lactate levels in diabetic brains. Levels and enzymatic activity of LDH-A were also found significantly up-regulated. GPR81 dependent PKA-CREB regulated cognition decline in the diabetic rats. Mechanistic insights into role of lactate in diabetes-associated cognitive decline. Diabetes mellitus causes brain structure changes and cognitive decline, and it has been estimated that diabetes doubles the risk for dementia. Until now, the pathogenic mechanism of diabetes-associated cognitive decline (DACD) has remained unclear. Using metabolomics, we show that lactate levels increased over time in the hippocampus of rats with streptozotocin-induced diabetes, as compared with age-matched control rats. Additionally, mRNA levels, protein levels, and enzymatic activity of lactate dehydrogenase-A (LDH-A) were significantly up-regulated, suggesting increased glycolysis activity. Importantly, by specifically blocking the glycolysis pathway through an LDH-A inhibitor, chronic diabetes-induced memory impairment was prevented. Analyzing the underlying mechanism, we show that the expression levels of cAMP-dependent protein kinase and of phosphorylated transcription factor cAMP response element-binding proteins were decreased in 12-week diabetic rats. We suggest that G protein-coupled receptor 81 mediates cognitive decline in the diabetic rat. In this study, we report that progressively increasing lactate levels is an important pathogenic factor in DACD, directly linking diabetes to cognitive dysfunction. LDH-A may be considered as a potential target for alleviating or treating DACD in the future.

Diabetes mellitus (DM) is the most common metabolic disease in the world and can adversely affect multiple organs. Both type 1 and type 2 diabetes patients are more likely to develop cognitive decline than healthy people (1)(2)(3)(4). The reported cognitive dysfunctions include impairments in memory, executive function, language, and processing speed (4 -7). Diabetes-associated cognitive decline (DACD) 1 was first reported in 2006 (8) and received increasing acceptance and attention since then. Efforts to understand the pathophysiological changes that underlie the development and progression of DACD are of vital importance for the development of novel treatments or prevention measures.
Many studies have associated glucotoxicity with learning and memory dysfunction in diabetes, as it is known that hyperglycemia can induce neuronal dysfunction through augmented oxidative stress, excessive release of cytokines, activation of protein kinase C, and increased flux of the polyol pathway (9). Hyperglycemia is an important factor in the development of cognitive decline in diabetic patients (10). Moreover, altered levels of metabolites, including free fatty acids, lipids, and advanced glycation end products generated by nonenzymatic glycation, are also known to initiate and aggravate pathological damages (11,12). However, the key pathogenic factors of the disease are still unclear, and the metabolic changes in specific brain regions need to be further investigated.
Metabolomic analysis is an important platform used to measure and identify key metabolites related to pathological conditions. It has been extensively applied for the exploration of potential biomarkers and has provided crucial insights into the pathogenesis of diseases (13)(14)(15)(16). Using this technology, Prentice K. J. et al. originally identified the furan fatty acid metabolite 3-carboxy-4-methyl-5-propyl-2-furanpropanoic acid (CMPF) to be significantly elevated in gestational diabetic patients. In a subsequent study, they further identified CMPF as a link between ␤ cell dysfunction and diabetes that could be targeted therapeutically (17,18). By nuclear magnetic resonance (NMR)-based metabolomics, we have previously shown that the development of DACD is related to alterations in glucose metabolism and to the impaired glutamate-glutamine cycle in the hippocampus of the diabetic dyslipidemia mouse model (db/db mice) (19). Furthermore, by employing 13 C NMR with labeled glucose and acetate as substrates, we observed an altered rate of particular metabolic pathways (i.e. increased glycolysis) in the diabetic animals (20,21). By direct analysis of the metabolic features of purified astrocytes cultured in a high-glucose environment, time-dependent lactate production and the utilization of specific amino acids were identified (22,23). Nevertheless, determining the causal link between these metabolic alterations and the disease requires further exploration.
In the present study, we found that the anaerobic glycolysis metabolite lactate is significantly elevated in a time-dependent manner in hippocampal extracts of rats with streptozotocin (STZ)-induced diabetes. Furthermore, we confirmed elevated glycolysis in the hippocampus of diabetic rats by measuring the activity and the expression levels of lactate dehydrogenase-A (LDH-A). In addition, the possibility that astrocytes release excess lactate in diabetic conditions, which could affect learning and memory, was also explored.

EXPERIMENTAL PROCEDURES
Animal Treatment-Male Sprague-Dawley rats weighing 220 Ϯ 15 g (eight weeks old) were purchased from SLAC Laboratory Animal Co. Ltd. Shanghai, China. All animals were kept in a specific pathogen-free facility of the Laboratory Animal Center of Wenzhou Medical University (Wenzhou, China) with regulated temperature and humidity. During the whole experimental procedure, the rats were fed with certified standard rat chow and tap water ad libitum. All animal treatments were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.
After a 12-h fast, rats were randomly selected and injected intraperitoneally (i.p.) with STZ (Sigma-Aldrich), which was freshly prepared in citrate buffer (0.1 M, pH 4.5) at a single dose of 70 mg/kg body weight. Control rats were injected with the same volume of vehicle (citrate buffer). Two days after STZ administration, blood glucose concentrations were measured and the rats that had glucose concentrations higher than 16.70 mmol/l were defined as diabetic (21). Diabetic rats were analyzed two weeks, eight weeks, 12 weeks, and 15 weeks after STZ administration, while age-matched rats without STZ administration were used as control rats.
Male 15-week db/db (BKS.Cg-mϩ/ϩ Leprdb/J, n ϭ 11) and WT (C57BLKS/J-mϩ/ϩdb, n ϭ 15) mice were purchased from the Model Animal Research Center of Nanjing University. All animals were kept in a specific pathogen-free colony of the Laboratory Animal Center of Wenzhou Medical University (Wenzhou, China) with regulated temperature and humidity. During the whole experimental procedure, rats were fed with certified standard rat chow and tap water ad libitum. All animal treatments were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. db/db mice were sacrificed by decapitation, and tissue samples were immediately dissected, snap-frozen in liquid nitrogen, and stored at Ϫ80°C until use. The induction of hypoglycemia in STZ-induced diabetic rats was carried out by insulin (10 ml: 400 IU, i.p. Becton Dickinson & Company, China) at a dose of 40 IU/kg body weight (24).
Morris Water Maze Test-Rats were subjected to the Morris water maze (MWM) test following previously published methods (25,26). Briefly, the test was conducted in a circular pool (pool diameter, 110 cm; pool height, 30 cm), filled with water that was made opaque with nontoxic paint and had a temperature of 26 Ϯ 1°C. The circular escape platform (platform diameter, 7 cm) was submerged 1 cm below the water surface. Cues were hung at four locations (north, west, south, and east corners) of the swimming pool wall. The rats were trained for four consecutive days, and on each training day, they swam four trials (starting from a different initial placement each time) for 60 s or until they located and climbed onto the hidden escape platform (within 60 s). The rats that failed to find the platform within 60 s were guided to the platform by the operator. In addition, the rats were tested in a single 90-s probe trial without the platform on the last training day. The swimming path length, the escape latency, and the swimming velocity were recorded by a computer system.
Magnetic Resonance Imaging-Rats were anesthetized by mechanically administered gas anesthesia (1-3% isoflurane mixed with 0.5 l/min oxygen) for magnetic resonance imaging (MRI). MRI was performed in a 3.0 Tesla MR scanner (Philips Achieva, Netherlands) by placing the rats into a micro-imaging coil. T2-weighted images were acquired by rapid acquisition with the following OAx T2 TSE spin echo sequence: TR/TE ϭ 3,198/80 ms, field of view ϭ 4 cm, matrix ϭ 200 ϫ 216, section thickness ϭ 0.5 mm, intersection gap ϭ 0.14 mm, NEX ϭ 8. MRI was performed at 12 and 15 weeks after the STZ treatment.
Sample Collection and 1 H NMR Spectra Acquisition-The Hippocampi and other tissues were dissected immediately after euthanasia of the rats, snap-frozen in liquid nitrogen, and stored at Ϫ80°C, until use. The preparation of brain samples and extracts and the acquisition of 1 H-NMR spectra were performed as described previously (21,27,28). High-resolution 1 H-NMR spectra were obtained on a Bruker AVANCE III 600 spectrometer operating at 600.13 MHz equipped with a triple resonance probe. The lyophilized samples were dissolved in 450 l D 2 O (which included 0.1 mM TSP as a reference for chemical shift) and then transferred into 5-mm NMR tubes. Both spectra were acquired with an 8.0-s relaxation delay and a 32-K data point. The acquisition time was 5.44 s, and in total 2,048 scans for intracellular samples and 256 scans for extracellular samples were performed. A one-dimensional ZGPR pulse sequence was applied to achieve satisfactory water suppression in aqueous extracts.
Multivariate Pattern Recognition Analysis-The 1 H NMR spectra were phase-and baseline-corrected and integrated to binning data with a size of 0.01 ppm, from 0.4 to 10.0 ppm, using the Bruker Topspin 2.1 software package. For the NMR spectra recorded in hippocampal extracts, the region 4.69 -5.04 of ␦ was removed to eliminate artifacts related to the residual water resonance. The remaining spectral segments were normalized to the total sum of the spectral intensity to compensate for variations in the total sample volume. The normalized integral values were then subjected to multivariate pattern recognition or quantitative analysis using the SIMCA-P ϩ V12.0 software package (Umetrics, Umea, Sweden). Data were visualized by the score plots of the first two principal components (t [1] and t [2]) to provide the 2D information, where the position of each point represents one sample.
Primary Cultures of Rat Cortical Neurons and Astrocytes-Primary neurons were cultured as previously described with some modifications (29). Briefly, neuron cultures were prepared from less than 24-h-old Sprague-Dawley rat pups. The cerebral cortical tissues were dissected under sterile conditions and the meninges were removed under the dissecting microscope. Tissues were minced and then digested with 0.25% trypsin for 10 min at 37°C. Digestion was 1 The abbreviations used are: DACD, diabetes-associated cognitive decline; CREB, cAMP-response element-binding protein; NMR, nuclear magnetic resonance; STZ, streptozotocin; PLS-DA, partial least squares-discriminant analysis; PC, principal components, MRI, magnetic resonance imaging; LDH, lactate dehydrogenase; MWM, Morris water maze; BDNF, brain-derived neurotrophic factor; MCTs, monocarboxylate transporters; Arc; activity-regulated cytoskeleton-associated protein; EGR, early growth response; GAP, growth-associated protein; SYP, synaptophysin; PSD, postsynaptic density protein; TCA, tricarboxylic acid; HBSS, Hank's balanced salt solution. terminated by DMEM, which contained 10% fetal bovine serum. The cell suspension was passed through a 200-mesh cell strainer to obtain a single-cell suspension. The cells were plated in poly-D-lysine coated Falcon flasks (75 cm 2 ) at a density of 2 ϫ 10 6 cell/ml after counting the cells by using a hemocytometer before culturing them in a humidified incubator with 5% CO 2 at 37°C (Thermo Fisher Scientific, USA). After allowing the cells to settle down for 12 h, they were washed once with HBSS and the culture medium was replaced by fresh Neurobasal medium that was supplemented with B27 (1ϫ), 2 mM glutamine, 100 U/ml penicillin, and 0.1 mg/ml streptomycin. To prevent excessive proliferation of astrocytes, cytosine arabinoside was added to a final concentration of 10 M for 12 h after 48 h in culture. After that, the neurons were cultured for another four days. Half of the medium was exchanged twice a week during the culturing period. The neurons were assessed by immunostaining for the neuronal marker microtubule-associated protein-2 and the astrocyte marker glial fibrillary acidic protein. Glucose concentration in the culture media was 5.5 mM.
Primary astrocytes were cultured as previously described with some modifications (30). Briefly, cultures were prepared from 2-3day-old Sprague-Dawley rat pups. Cerebral cortical tissues were dissected under sterile conditions. Astrocytes were seeded in 24 culture flasks (10 6 cells/flask) and five 24-well plates (10 5 cells/well) for examination of intracellular and extracellular lactate levels, respectively. Cells were washed once with HBSS 24 h after seeding, and the culture medium was replaced by fresh DMEM containing 5.5 mM glucose for the control group or 25 mM for the high-glucose group (31)(32)(33).
Cell Metabolite Extraction-Cells were harvested after rapid quenching at the end of high-glucose stimulation. Quenched cells were pelleted by centrifugation and rinsed with HBSS before they were transferred to a centrifuge tube. Subsequently, cell pellets were resuspended in 450 l of ice-cold CH 3 OH:CHCl 3 (v/v 2:1) and sonicated on ice for 30 min. Afterward, another 450 l ice-cold CHCl 3 : H 2 O (v/v 1:1) were added to form an emulsion, which was left on ice for 15 min before being centrifuged at 12,000 rpm for 20 min (4°C). The upper phase was collected, lyophilized, and stored at Ϫ80°C until NMR analysis (two flasks/sample, n ϭ 6/group).
The five plates corresponded to the five time intervals (0, 2, 24, 48, and 72 h). Culture medium (1 ml) was transferred from each well into individual 15 ml centrifuge tubes at each time point after exposure to high glucose. Then, 3 ml ice-cold CH 3 OH:CHCl 3 (v/v 2:1) were added followed by 1 ml ice-cold CHCl 3 . This solution was vigorously mixed for 60 s on ice before being centrifuged at 8,000 rpm for 20 min at 4°C. Finally, the collected supernatants were lyophilized and stored at Ϫ80°C for further NMR analysis.
LDH Activity Determination-Isolated rat hippocampi were homogenized in protein lysis buffer and debris was eliminated by centrifugation at 12,000 rpm for 10 min at 4°C. The protein concentrations in all samples were determined by the Bradford protein assay kit (Beyotime). Spectrophotometric assays were then used to determine LDH activity converting pyruvate (P)3lactate (L) and L3 P in a tunable microplate reader (Spectramax M5; Molecular Devices) with the corresponding software (Soft Max Pro) (34). For determination of LDH L3 P activity, an LDH kit (Nanjing Jiancheng Bioengineering Institute, Nanjing, China) was used according to the manufacturer's directions. LDH activity was detected at 25°C by measuring sample absorbance at 430-nm wavelength. For LDH P3 L activity determination, an assay was designed based on the procedure of Krieg et al. (35). Protein lysates were first diluted to a concentration of 0.05 mg/ml in 500 mM potassium phosphate buffer and then added to the LDH assay reagent containing 100 mM pyruvate and ␤-nicotinamide adenine dinucleotide (NADH) in 500 mM potassium phosphate buffer (5 mg/20 ml of assay buffer, pH 7.5). The changes in NADH absorbance were measured at 25°C, at one-minute intervals for 10 min at 340-nm wavelength. LDH activity was calculated in UI/mg protein and was expressed as P:L ratio (LDH P3 L/LDH L3 P).
Determination of Lactate Levels-Lactate concentrations in the hippocampus and cells extracts were measured using commercial kits (Sigma). The total reaction volume for each reaction was 2 ml and 100 ml of the sample were added.
Real-Time Reverse Transcription-Polymerase Chain Reaction-The isolated rat hippocampi (50 -70 mg) were homogenized in TRIzol reagent (Invitrogen, Carlsbad, CA) for RNA extraction, according to the manufacturer's protocol. The reverse transcription step was carried out using the M-MLV RT kit (Invitrogen). CFX96 Touch™ Real-Time PCR Detection System (Bio-Rad, Hercules, CA) was used for q-PCR analysis. The complete set of primers is listed in Supplemental Table S1.
Western Blotting-Following addition of sample loading buffer, protein samples were electrophoresed and then electrotransferred onto PVDF membranes. The blots were blocked for one hour at room temperature with fresh 5% nonfat milk in Tris-buffered saline containing Tween 20 (TBST) and were then incubated with specific primary antibodies diluted in TBST, for 16 -18 h, at 4°C. Following three washes with TBST, the blots were incubated with horseradish peroxidase-conjugated secondary antibodies for 1 h and the immunoreactive bands were visualized using an enhanced chemiluminescence kit. The bands were quantified using Image-Pro Plus 6.0 software.
Tissue Preparation for Immunohistochemistry-For preparation of fresh frozen sections for immunohistochemistry, the rats were killed by cervical dislocation and their brains were dissected, rapidly frozen on dry ice, and stored at Ϫ80°C. The frozen tissues were embedded in mounting medium (Tissue-Tek; Sakura Finetek). Sections of 14 m were cut on a Microm Model HM 500 M Cryostat (Microm), collected onto Super Frost slides (Menzel Glä ser), and stored until use.
Intracerebroventricular Injection-Ten-week-old diabetic or control rats received intracerebroventricular injections of the LDH inhibitor oxamate (10 g dissolved in ultrapure water at a final concentration of 0.4 mM, Sigma Aldrich) once daily for 14 days. The oxamate-injected rats were subjected to the MWM test at 12 weeks after the initial STZ treatment.
Statistics-Data are presented as means Ϯ standard deviation, unless stated otherwise. The statistical significance was analyzed using a two-tailed Student's t test or one-way ANOVA followed by Dunn's test for multiple group comparison where appropriate (SPSS, Inc.). Statistical significance was defined as p Ͻ 0.05.

Determination of Cognitive Decline in Diabetic Rats-To
determine the occurrence of cognitive decline in rats with STZ-induced diabetes, MWM, immunohistochemistry, MRI, and RT-PCR were carried out at different time points. Fig.  1A-1C show the swimming behavior data of the diabetic and age-matched control rats. We found that at the time points of eight and 12 weeks after STZ treatment, the diabetic rats had longer escape latency during the original MWM test and fewer platform crossings during the probe trial. Cross-sectional MRI studies showed that 12-week diabetic rats displayed larger hyperintensities in the lateral ventricle regions, compared with age-matched control rats, suggesting the appearance of brain atrophy (Fig. 1D). Moreover, hippocampus expression levels of glial fibrillary acidic protein, an indicator of astrocyte reac-tivity, were up-regulated at all time points checked (Figs. S1 and S2).
In addition, the mRNA levels of biomarkers like brain-derived neurotrophic factor (BDNF), c-Fos, activity-regulated cytoskeleton-associated protein (Arc), presynaptic and postsynaptic plasticity proteins, i.e. early growth response-1 (EGR-1), growth-associated protein 43 (GAP43), synaptophysin (SYP), postsynaptic density protein-93 (PSD-93), and PSD-95, that underlie the molecular basis of learning and memory, were markedly down-regulated in 12-week diabetic rat samples compared with samples from age-matched controls, but not in two-week diabetic rats (except c-Fos and BDNF, which were also altered at two weeks) indicating the occurrence of DACD (Fig. 1E, Fig. S3).
Metabolomic Analysis-We used 1 H NMR spectrometry to examine 16 metabolites in hippocampal extracts from STZ and vehicle-treated rats at two, eight, and 12 weeks post STZ administration (representative spectra are shown in Fig. 2A). To characterize the metabolic changes that occur at different stages of diabetes progression, the mean trajectory was calculated based on partial least squares discriminant analysis (PLS-DA) of the spectra. The trajectory exhibited a distinct separation between the diabetic and control groups along the principal component 1 (PC1) and PC2 direction (Figs. 2B and 2C). Additionally, the diabetic rats showed a clear trajectory plot during the period from two weeks to 12 weeks, in contrast to the age-matched controls. The corresponding loading plot showed that the contents of myo-inositol, fumarate, and RT-PCR determination of mRNA levels of brain-derived neurotrophic factor (BDNF), c-fos, activity-regulated cytoskeleton-associated protein (Arc), and early growth response-1 (EGR-1) in hippocampal extracts of diabetic and control rats at two and 12 weeks after treatment. Data represent mean Ϯ S.D. from five rats in each group. Significance was determined by two-tailed unpaired Student's t test of diabetic rats versus controls and denoted with *p Ͻ 0.05, **p Ͻ 0.01. lactate were significantly higher, while succinate, glutamine, glutamate, and citrate were decreased in 12-week diabetic rats (Fig. 2D). In the hippocampus samples, lactate levels were increased by 30% in eight-week diabetic rats, and a nearly twofold increase of lactate was apparent in 12-week diabetic rats, as compared with age-matched controls (Fig.  3). In addition, levels of most of the cerebral intermediates or products related to the Krebs cycle were decreased with the exception of fumarate. These results suggest that high anaerobic glycolysis and low tricarboxylic acid (TCA) cycle activity are pathological features of the DACD rat hippocampus.
Next, we measured hippocampal lactate concentrations in two-week and 12-week diabetic rats using ELISA to confirm the NMR findings. Our results showed that lactate levels are higher in 12-week but not in two-week diabetic rats as compared with age-matched controls (Fig. S4). Due to the strong similarities between DACD and diabetes, we examined whether lactate levels were also elevated in other tissues or in other diabetic animal models (Fig. S5). In different brain regions of the 12-week STZ diabetic rats, such as the cortex and the striatum, lactate levels were also significantly elevated. A similar increasing trend of lactate levels was found in hippocampal extracts of 17-week old db/db mice (type 2 DACD model). However, lactate levels in urine, serum, and heart tissue of rats with STZ-induced diabetes were decreased or not altered. Interestingly, in hypoglycemic rats treated with insulin, hippocampal levels of lactate displayed a trend for reduction (p ϭ 0.062, n ϭ 6, Fig. S5). The abovementioned results show that the elevation of lactate levels in diabetic animals is independent of the tissue, organ, animal age, and STZ toxicity.

Assessment of LDH Activity and Expression-
The functioning of neurons in the brain depends on glycolysis and as a consequence, neurons must replenish NAD ϩ , which is accomplished by increasing pyruvate (P) to lactate (L) conversion by lactate dehydrogenase (LDH; EC 1.1.1.27) (36). There are five isoenzymes, LDH-1, LDH-2, LDH-3, LDH-4, and LDH-5. Each isoform is composed of M and H subunits, which are the gene products of LDH-A and LDH-B, respectively (Fig. 4A). LDH-A is mainly responsible for the P3 L conversion, while LDH-B is responsible for the L3 P conversion. Using a microplate reader to measure enzymatic activity, we found in 12-week diabetic rats a 31% increase of P3 L conversion and a 13% increase of L3 P conversion, as compared with age-matched controls. This corresponds to a total of 20% increase of P3 L conversion in anaerobic glycolysis. Interestingly, in two-week diabetic rats, L3 P Our RT-PCR data showed that LDH-A levels are up-regulated by 90% in 12-week diabetic rats, while they are not altered in two-week diabetic rats, as compared with the corresponding controls (Fig. 4D). Quantitative Western blot anal-ysis confirmed that 12-week diabetic rats had a significant increase in hippocampal LDH-A expression compared with age-matched control rats (Figs. 4E and 4F). These results, coupled with the LDH gene expression and enzymatic activity changes, indicate that the observed higher lactate levels in the hippocampi of diabetic rats were the result of a metabolic shift to anaerobic glycolysis, where large amounts of lactate are being produced from pyruvate in an environment of consistent hyperglycemia. Furthermore, we determined by immunohistochemistry staining that LDH-A was highly expressed mainly in CA1 and CA2 regions of the 12-week diabetic rats, while it was neither detected in the CA3 and dentate gyrus regions of 12-week rats, nor in the hippocampi of the twoweek diabetic rats (Fig. 4G, Figs. S6 -S9).
Determination of PKA-CREB Expression-The activation of G protein-coupled receptors 81 (GPR81) by lactate has been shown to modulate CREB through the inhibition of cAMP and PKA-CREB activity (37). To determine the mechanism through which lactate decreases c-Fos and Arc biosynthesis, we examined the protein expression levels of c-Fos and Arc in hippocampal extracts from different time points of diabetic rats and control rats. In 12-week diabetic rats, PKA and CREB phosphorylation activity was significantly inhibited compared with age-matched controls (Fig. 5), which was not the case in two-week diabetic rats, indicating inhibited PKA-CREB transduction under chronic diabetic conditions.
Lactate Transport Examination-It is known that lactate is transported through the cell membrane by proton-linked monocarboxylate transporters (MCTs) (38). In the brain, MCT4 is expressed on astrocytes that are glycolytic and can supply other cells with lactate, while MCT1 is mainly detected in endothelial cells of the brain's blood vessels and in oligodendrocytes (39). MCT2 is found in the cell bodies, dendrites, and axons of neurons (39,40). In our study, we performed RT-PCR to determine the mRNA levels of MCT1, MCT2, and MCT4 in the hippocampi of the different diabetic groups. The results showed significantly decreased MCT1 and MCT4 levels in 12-week diabetic rats, compared with age-matched controls, which may reflect the compensatory response of astrocytes in an environment of excess lactate. Conversely, no changes in MCT2 were detected either in two-week or in 12-week diabetic rats, also indicating no alterations of lactate transport into neurons (Figs. 6A-6C).
To confirm the transport process, lactate contents were determined in the intra-and extracellular compartments of astrocytes and neurons, respectively, under high-glucose challenge. We found significantly elevated lactate levels in the extracellular compartments of astrocytes, as well as a faster rate of lactate efflux (Fig. 6D). Interestingly, we determined reverse directions of lactate transport into neurons (influx), under high-glucose conditions. These results suggest that in a high-glucose environment, or under diabetic conditions, excess lactate originates from astrocytes and not from neurons.
Pharmacological Intervention-To determine the causal role of lactate signaling in cognition impairment, LDH-A inhibitor oxamate was intracerebroventricularly injected into ten-week diabetic rats once daily for 14 days. This treatment recovered the levels of lactate in the cortex or midbrain regions to control values (Figs. 7A-7C). Furthermore, we found that the increased latency of the diabetic rats in the MWM test was significantly recovered by oxamate treatment, compared with vehicle-injected diabetic rats (Figs. 7D-7H). Additionally, mRNA and protein levels of molecular markers related to cognition, such as BDNF, Arc, PSD-95, and Syp were accordingly recovered (Figs. 7I-7M). DISCUSSION In the central nervous system, lactate is considered as equivalent to glucose regarding its access to the TCA cycle in neurons and it also serves as a precursor for the synthesis of neurotransmitters (41). Furthermore, lactate also plays a signaling role in the brain (42). Recently, it was revealed that Montreal cognitive assessment scores of clinical subjects were negatively correlated to lactate levels in the cerebrospinal fluid but not in the blood, strongly suggesting that excess cerebral lactate is related to cognitive decline (43). In this study, we proposed one model that during early stages of diabetes in rodents, there is increased gliosis, acceleration of glycolysis, and elevated extracellular lactate, which is available to the surrounding neurons. However, at later stages of diabetes, due to mitochondrial dysfunction, lactate utilization is reduced and it accumulates in the extracellular space. This excess of lactate causes neuronal dysfunction, by modulating the GPR81-PKA-CREB signaling pathway, which finally leads to decrease of synaptic plasticity, down-regulation of proteins related to memory function (i.e. c-Fos, Arc, BDNF), and cognitive decline (Fig. 8).
Employing NMR-based metabolomics, we detected that lactate levels increase over time in hippocampal extracts from diabetic rats. By determining the activity of the key proteins of glycolysis, LDH-A and LDH-B, we identified L3 P conversion to be dominant at the two-week stage of diabetes, which switched to the opposite direction at the 12-week stage. These results are consistent with results from our previous 13 C NMR studies on the same diabetic model, in which lactate C 2 was enriched and the pyruvate recycling pathway was enhanced in one-week diabetic rats, but it was found weakened in 15-week diabetic rats (21). Our studies suggest that at the early stages of diabetes (one to two weeks), the L3 P procedure is accompanied by enhanced pyruvate recycling and TCA cycle. Conversely, in chronic diabetes (12 weeks or more), increased glycolysis and inhibited pyruvate recycling occur, resulting in the accumulation of lactate in the brain.
Lactate secreted from astrocytes is available for transport by MCT2 into neurons for energy supply or it may accumulate in the extracellular space and exert pathological function. In peripheral adipocytes, it was reported that the effects of lactate in lipolysis involve GPR81 (44) and PKA-CREB signal transduction (37). CREB is one of the most important transcription factors, activated by its phosphorylation on serine 133. This activation regulates the expression of regulatory immediate early genes that are responsible for long-term synaptic plasticity and memory formation (45). In this study, we confirmed that the expression levels of PKA-pCREB and of molecules implicated in memory function are down-regulated in 12-week diabetic rats, while being normal in two-week diabetic rats. These results strongly suggest GPR81-mediated memory impairment in chronic diabetic conditions, which is indicative of DACD. The results are consistent with previous in vitro experiments of neuronal activity suppression (46,47). However, here we provide evidence that GPR81 mediates lactate signaling by inhibiting the PKA-CREB pathway in the diabetic rat brain, which has not been documented in previous studies.
In fact, lactate is not just a metabolic substrate for neurons, but it also functions as a gliotransmitter (46), an autocrine hormone (48), and a receptor agonist (47). For example, it was reported that glycolysis leads to the production of ATP, which modulates selective neuronal membrane conductance through ATP-sensitive potassium channels (K ATP ) (49,50). Lactate is also known to stimulate synaptic plasticity-related gene expression in neurons by the modulation of N-methyl-D-aspartate receptor activity and its downstream signaling cascade comprising extracellular signal-regulated kinases 1 and 2 (ERK1/2) (51). Moreover, it was reported that lactate shuttling between astrocytes and neurons is probably required for longterm memory formation (52). The activation of different signal transduction pathways, the different time course, and doserelated effects may contribute to the contradictory results on the role of lactate in brain function, which needs to be further explored in future studies.
According to the 'lactate shuttle' hypothesis, lactate is a metabolic substrate released under physiological conditions by glial cells (mainly astrocytes) and is taken up by the adjacent neurons (53). In our previous study using [2-13 C]acetate as a tracer substrate, we demonstrated in the same diabetic rat model reduced lactate supply to neurons, which was accompanied by lactate accumulation in astrocytes (21). In this study, we employed an in vitro primary cell culture system in high-glucose conditions to recapitulate the in vivo diabetic environment. Indeed we detected higher extracellular lactate levels. Lactate was also increased in astrocytes but not in neurons. Accordingly, in hippocampal extracts of 12-week diabetic rats, we found decreased levels of MCT1 and MCT4, but not MCT2, also suggesting that changes in lactate transport was only related to astrocytes, and not to neurons.
Due to its critical role in sustaining a high rate of glycolysis, LDH-A was suggested as a potential therapeutic target for DACD. By pharmacological treatment, we showed that cognition decline in diabetic rats was reversed by the LDH-A inhibitor oxamate, suggesting that lactate is implicated in DACD. Interestingly, previous studies have shown that lactate is involved in many physiological and pathological processes in the brain, such as appetite (54), mental disorders (55), neurotransmitter transporter function (56), and protection from brain damage (57), as well as cognition (38,47,51,52).
In conclusion, our study reveals that glycolysis activity is elevated in the hippocampal region during the late stages of diabetes progression, but not during early stages. Consistently, the expression levels and activity of LDH-A are increasing during disease progression in the diabetic rat brain, suggesting an important role of glycolysis and LDH-A in DACD and the potential for LDH-A to be examined as an early diagnostic marker or as a therapeutic target. Nonetheless, the exact molecular mechanisms that can induce the lactate-mediated cognitive dysfunction need further elucidation.

DATA AVAILABILITY
The NMR spectrometry data have been deposited to the KiMoSys repository (http://kimosys.org) with the dataset identifier Data EntryID 91. In the later stages of diabetes, due to gliosis and glycolysis enhancement, excess lactate levels are transported out of the astrocytic cytoplasm by MCT1 and MCT4. This is accompanied by reduction of lactate uptake by adjacent neurons. Thus, lactate accumulates in the extracellular space. This leads to neuronal dysfunction by affecting the GPR81-PKA-CREB signal transduction pathway, which finally results in decreased levels of immediate early genes (IEGs, i.e. c-Fos, Arc), BDNF, and other proteins related to learning and memory and finally resulting in cognitive decline in diabetic rats.