Epigenetic Mechanisms Linking Diabetes and Synaptic Impairments

The prevalence of dementia increases with age. It is estimated that >36 million people worldwide are currently living with this devastating disease, and the prevalence is expected to increase to 66 million by 2030 and 115 million by 2050 (1). Dementia can be caused by a number of progressive diseases such as Alzheimer disease (AD) and vascular dementia. Mounting evidence suggests that cognitive impairments associated with dementia may be traced back to neuropathological conditions initiated several decades before clinical onset. Therefore, clinical development of novel interventions for dementia is now based on mechanisms associated with primary or secondary prevention.

Diabetes increases the risk of both AD dementia and vascular dementia (2–9). Based on the National Diabetes Health Fact Sheet, ~8.3% of Americans have diabetes and, as with dementia, the prevalence of diabetes increases with age; currently, 26.9% of people ≥65 years old have diabetes (10). Despite national and international campaign efforts to counter type 2 diabetes (T2D), its prevalence is still rising. T2D and hyperglycemia are associated with cognitive dysfunction and an increased risk for dementia, including AD-type dementia. Three longitudinal studies conducted by Yaffe et al. (11), Komulainen et al. (12), and Dik et al. (13) independently showed that T2D is related to a higher risk of cognitive decline. A meta-analysis performed by Profenno, Porsteinsson, and Faraone (14) reviewed thirteen longitudinal epidemiological studies that examined the association between diabetes and dementia and found that diabetes significantly and independently increases one’s risk for dementia.

Emerging evidence has demonstrated that diabetes-associated chromatin modifications pertinent to epigenetic mechanisms may play an important role in the pathogenesis of diabetes (15,16). For example, epigenetic alterations of histone deacetylases (HDACs) cause decreases in H3 and H4 acetylation at the proximal promoter of pancreatic and duodenal homeobox factor 1 (Pdx-1) in pancreatic β-cells, leading to defects in glucose homeostasis and insulin resistance and eventually leading to T2D (17). Similarly, histone modifications of glut4 mediated by DNA methyltransferase (Dnmt) and HDACs result in reduced transcription of GLUT4 in skeleton muscle in intrauterine growth–restricted offspring (18).

Epigenetic mechanisms have also been widely studied in the brain. Histone methylation and histone acetylation in the hippocampal formation play an important role in modulating synaptic strength after neuronal activation (19–26). Mice that lack both Dnmt1 and Dnmt3a exhibited abnormalities in long-term potentiation (LTP) and long-term depression after stimulations in CA1 synapses, and these dysfunctions were associated with impaired learning and memory (19). HDAC2 has emerged as a major histone modification enzyme in modulating synaptic plasticity and long-term memory. Overexpression of HDAC2 is associated with structural dysfunction including reduced number of dendritic spine and synapses as well as functional deficits such as impaired synaptic plasticity and memory function (27). Moreover, pharmacological inhibition of HDACs by HDAC inhibitors such as trichostatin A resulted in increased LTP responses, further confirming the role of histone acetylation in brain function (23).

In the current study, we tested whether the diabetic condition might alter the epigenetic mechanisms in the brain, which may lead to structural or functional changes that ultimately influence the ability of the brain to cope with conditions associated with neurological disorders, such as AD.

Human postmortem brain specimens from T2D and age-matched nondiabetic (control) case subjects were obtained from the Department of Psychiatry Brain Bank, Icahn School of Medicine at Mount Sinai. The brains were donated by deceased residents of the Jewish Home and Hospital (Manhattan, NY, and Bronx, NY). Diabetic or nondiabetic case subjects were identified using criteria previously described (28). Specifically, patients were diagnosed with diabetes either by a geriatrician or by an internist based on the American Diabetes Association criteria symptoms of diabetes plus casual plasma glucose concentration >200 mg/dL, fasting plasma glucose >126 mg/dL, and 2-h plasma glucose >200 mg/dL during an oral glucose tolerance test. We only included the subjects with T2D. Detailed demographic information is presented in Table 1. The Clinical Dementia Rating scale was used to assess cognitive and functional impairments associated with dementia. The cohort used in this study was devoid of any significantly cerebrovascular disease neuropathology.

Total RNA from brain tissues (hippocampus) was isolated using an RNeasy Mini Kit (Qiagen, Valencia, CA) and reverse transcribed using SuperScript III first-strand synthesis supermix for qRT-PCR (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions. Epigenetic chromatin modification enzyme PCR array for human or mouse (Qiagen) was used to evaluate the levels of gene expression. Expression of synaptic plasticity gene PSD95 and synaptophysin in the brains of T2D mice was measured in four replicates by quantitative RT-PCR using Maxima SYBR Green master mix (Fermentas) in ABI Prism 7900HT with primers amplifying the target genes: PSD95 (forward, CGGGAGAAAATGGAGAAGGAC, and reverse, GCATTGGCTGAGACATCAAG) and synaptophysin (forward, AGTGCCCTCAACATCGAAG, and reverse, GCCACGGTGACAAAGAATTC). Mouse hypoxanthine phosphoribosyltransferase (forward, CCCCAAAATGGTTAAGGTTGC, and reverse, AACAAAGTCTGGCCTGTATCC) expression level was used as an internal control. Data were normalized using the 2^−ΔΔCt method (29). Levels of target gene mRNAs were expressed relative to those in control groups and plotted in GraphPad Prism.

Brain samples (the same hippocampal tissue as used in the gene expression studies) were extracted with radio immunoprecipitation assay buffer and boiled in Laemmli sample buffer for 5 min. Proteins were separated by gel electrophoresis using 10% sodium dodecyl sulfate–polyacrylamide gels and electrotransferred onto nitrocellulose membranes, blocking with 5% nonfat dried milk (Sigma-Aldrich, St. Louis, MO) in Tris-buffered saline. Membranes were probed with primary antibodies anti-PSD95 (1:2,500; Millipore, Billerica, MA) or anti-synaptophysin (1:5,000; Sigma-Aldrich) at 4°C overnight, followed by horseradish peroxidase–conjugated secondary antibodies. Specific protein signals were detected using an enhanced chemiluminescence system (ECL plus; PerkinElmer LAS, Inc., Boston, MA), and signals were quantified by densitometric scanning (GS-800; Bio-Rad, Hercules, CA) and analyzed using Quantity One software (Bio-Rad). β-Actin (1:3,000; Sigma-Aldrich) was served as the loading control.

Female C57BL/6J mice were purchased from The Jackson Laboratory (Bar Harbor, ME) and housed in the centralized animal care facility of the Center for Comparative Medicine and Surgery at Icahn School of Medicine at Mount Sinai. The C57BL/6J mouse develops hyperinsulinemia and hyperglycemia when fed a high-fat diet (30). Mice were randomly grouped into two groups and received the following treatment starting at ∼2 months of age: control mice were treated with a regular diet (CTRL group) (10% kcal fat, cat. no. D12450B; Research Diets), T2D mice were treated with a high-fat diet (T2D group) (60% kcal fat, D12492; Research Diets). All animals were maintained on a 12:12-h light:dark cycle with lights on at 0700 h in a temperature-controlled (20 ± 2°C) vivarium, and all procedures were approved by the Mount Sinai School of Medicine Institutional Animal Care and Use Committee.

Intraperitoneal glucose tolerance test was performed as previously reported (31,32). Specifically, mice were given a single dose of glucose (2 g/kg body wt i.p.) postprandially, and blood was collected from the tail vein periodically over a 3-h period. Blood glucose content was assessed using the CONTOUR blood glucose system (Bayer, Mishawaka, IN). Insulin was measured using the mouse serum adipokine multiplex MAP kit from Millipore (Billerica, MA).

Mice were killed by decapitation, and the brains were quickly removed. Hippocampal slices (350 μm) were placed into oxygenated artificial cerebrospinal fluid at 29°C for a minimum of 90 min to acclimatize. Slices were then transferred to a recording chamber and perfused continuously with oxygenated artificial cerebrospinal fluid in the presence or absence of 60 nmol/L oligomeric Aβ (oAβ) at 32°C. For extracellular recordings: CA1 field excitatory postsynaptic potentials (fEPSPs) were recorded by placing stimulating and recording electrodes in CA1 stratum radiatum. Basal synaptic transmission was assayed by plotting the stimulus voltages against slopes of fEPSP. For LTP experiments, a 15-min baseline was recorded every min at an intensity that evokes a response ∼35% of the maximum evoked response. LTP was induced using θ-burst stimulation (4 pulses at 100 Hz, with the bursts repeated at 5 Hz and each tetanus including three 10-burst trains separated by 15 s), and fEPSPs were monitored for 60 min to assess the magnitude of potentiation.

The Bruker Biospec 70/30 system, a 7T micro-magnetic resonance imaging (MRI) scanner, was used. Prior to scanning, mice were anesthetized with 2% isofluorane in NO₂ (75%) and O₂ (23%). They were then positioned onto the scanner bed. Anesthesia was maintained at 1.5% isofluorane throughout the scan time. Data acquisition was done using a four-channel mouse brain-dedicated phased array coil as follows: after a localizer, we manually shimmed the brain region for optimal signal and signal distortion. The resting state functional magnetic resonance imaging (fMRI) was acquired using an echo planar imaging blood oxygen level–dependent sequence with the following protocol: segmented echo planar imaging (×4), repetition time 1,500 ms, echo time 18.9 ms, matrix 96 × 96, field of view 12.8 mm, slice thickness 1.1 mm, 11 slices. Total acquisition time was ~5 min per scan; three scans were acquired for each mouse.

Preprocessing of the functional images began by excluding scans where there was movement. Manual brain extraction was then performed on the mean functional image. A study-specific template was created based on all the mean functional images using the FMRIB software library (FSL). Independent component analysis was used to identify unique networks of resting-state activity using MELODIC (33) as implemented in FSL. We visually identified the default mode network (DMN) from the networks produced by FSL. The DMN was defined as containing orbital cortex, cingulate cortex, and hippocampal regions according to the literature (34). Group independent component analysis was performed, and the DMN and sensory motor were visually identified for dual regression analysis. General linear modeling and permutation-based inference testing (RANDOMIZE) were used to test for group differences and symptom severity correlates of the DMN (35,36).

Embryonic-day 16 cortico-hippocampal neuronal cultures were prepared from C57BL/6J mice as previously described (32). Neurons were seeded on poly-d-lysine–coated sixwell plates and cultured in Neurobasal medium supplemented with 2% B27, 0.5 mmol/L l-glutamine, and 1% penicillin streptomycin (Gibco-BRL) in the tissue culture incubator at 37°C with 5% CO₂. After 5 days in vitro culturing, the neurons were treated either with 200 nmol/L MC1568 (Selleckchem) or with vehicle for 16–18 h. Cells were washed once with cold PBS before RNA isolation.

In these studies, all values are expressed as mean ± SEM. Differences between means were analyzed using either two-way repeated-measures ANOVAs or two-tailed Student t test. In all analyses, the null hypothesis was rejected at the 0.05 level. All statistical analyses were performed using the Prism Stat program (GraphPad Software, Inc.).

To test whether there are epigenetic changes in the brains of diabetic patients, we measured the expressions of epigenetic chromatin modification enzymes in the brains of diabetic patients and normal control subjects using a PCR-based array. We found that there are significant changes in the expression of select chromatin modification enzymes, particularly HDAC class IIa (Fig. 1A), while there are no changes with regard to the expression of HDAC class I (Fig. 1B). Specifically, >50% upregulation of HDAC4, -7, and -9 was observed in the brains of diabetic subjects, as well as an almost twofold increase in HDAC5 expression compared with the nondiabetic control subjects.

We next tested whether diabetes-induced epigenetic modifications in the brain might be associated with the modulation of neuronal activity at the level of synapse. We examined the levels of proteins involved in synaptic plasticity in the pre- and postsynaptic compartments as indices for synaptic integrity in the control and T2D subjects. Western blot was used to examine the levels of synaptophysin, an important presynaptic protein responsible for synaptic vesicle biogenesis and recycling, and postsynaptic density protein (PSD)-95, the most abundant scaffolding protein involved in recruiting and anchoring glutamate receptor subunits in PSD. We found that the levels of synaptophysin and PSD-95 were both significantly lower in postmortem brain specimens from the T2D subjects compared with the control subjects (Fig. 1C [inset]). Moreover, correlation studies showed that the expression of HDAC5 was significantly reverse correlated with PSD-95 (P < 0.05) and synaptophysin (P < 0.01) (Fig. 1D). The expression of HDAC7 was significantly reverse correlated with synaptophysin (P < 0.05) but had no correlation with PSD-95. We also found that the expression of HDAC9 showed nonsignificant reverse correlations with the expression of PSD-95 (P = 0.085) and synaptophysin (P = 0.084). We did not find any correlation between HDAC4 and the expression of synaptic proteins.

Based on the specific HDAC changes observed in the postmortem brains from diabetic subjects, we tested whether diabetic conditions can causally induce HDAC IIa alteration in the brain. We used the C57BL/6J mouse as a model of human T2D because it develops hyperinsulinemia and hyperglycemia when fed a diabetogenic diet but remains physically normal when fed low-fat chow (30). We treated mice with a regular low-fat diet or the diabetogenic diet starting at 2 months of age. After 6 months of chronic diet treatment, the T2D mice developed severe diabetes including increased baseline fasting glucose and impaired insulin sensitivity and glucose tolerance compared with the CTRL mice (Fig. 2A–C). Examination of chromatin modification enzyme expressions in the brains after 6 months of the diabetogenic diet treatment revealed that, similar to the human data, two of the class IIa HDACs are upregulated in the brains of T2D mice. Most notably, HDAC5 was significantly upregulated, fourfold, and HDAC9 was upregulated twofold (Fig. 2D). We did not find any changes in the expression of class I HDACs including HDAC1, -2, -3, and -8 in the brain. HDAC10, which belongs to HDAC IIb, was significantly reduced in the T2D mouse brains, while there was no change in the expression of HDAC6 and HDAC11.

We next tested whether the epigenetic changes in the brain of T2D mice coincide with synaptic impairments. Electrophysiological studies were performed on the hippocampal slices isolated from the T2D mice. We found that hippocampal slices derived from the T2D mice after 6 months of diabetogenic diet treatment had normal electrophysiological responses after tetanus stimulation (data not shown). However, these hippocampal slices had a much lower threshold to oAβ_1–42–induced LTP deficits. Normally, healthy hippocampal slices show compromised LTP when exposed to 200 nmol/L oAβ (data not shown). However, hippocampal slices derived from T2D mice showed limited potentiation after tetanic stimulation in the presence of 60 nmol/L oAβ, while slices from strain-, age-, and sex-matched mice on the control diet showed robust LTP under the same conditions (Fig. 3A). Consistent with previous observations, oAβ had no effect on basal synaptic transmission (37) (Fig. 3B). Synaptic deficits were exacerbated with prolonged diabetes conditions. While hippocampal slices isolated from T2D mice, after 6 months of diabetogenic diet treatment, showed normal LTP response in the absence of oAβ, slices isolated from mice after 10 months of treatment showed compromised LTP responses (Fig. 3C). In the diabetic mice, the LTP responses were 153.48 ± 5.05% 60 min after the θ-burst stimulus, while in the CTRL mice the responses were 233.97 ± 14.82% (P < 0.01), further stressing that diabetes conditions might adversely influence brain synaptic function and may eventually lead to cognitive impairments.

To test whether diabetes condition–induced epigenetic alterations/synaptic impairments in the brain might be associated with brain biology changes, we used MRI imaging to objectively measure the biological effects of the diabetic condition on the brain. We used fMRI with blood oxygen level–dependent contrast to examine the brain connectivity in T2D and control mice, following 10 months treatment. We found significant difference between the CTRL and T2D mice in the DMN.

Compared with CTRL mice, T2D mice had significantly stronger signal intensities in the DMN (Fig. 4A) as well as in the sensory/motor cortex regions (Fig. 4B). The green color represents the group resting-state network and the dark blue color (red arrows) represents significant difference between CTRL and T2D.

To confirm whether HDAC IIa might be responsible for the synaptic impairment we observed in the electrophysiology study on the T2D mice treated with diabetogenic diet for 10 months (Fig. 3C), we treated hippocampal slices isolated from these mice with an HDAC IIa–specific inhibitor, MC1568 (38,39), and recorded the LTP responses. We found that ex vivo treatment with MC1568 significantly improved LTP responses compared with the vehicle-treated slices. In the vehicle-treated diabetic brain slices, 60 min after the θ-burst stimulus, the LTP responses were 151.85 ± 8.64% while in the MC1568-treated diabetic slices, the responses were 202.51 ± 12.08% (P < 0.01) (Fig. 5A, left panel). There was no difference in basal synaptic transmission (Fig. 5A, right panel). Similarly, we also found that ex vivo MC1568 treatment promoted the LTP in brain slices from T2D mice following 6 months diabetogenic diet treatment when challenged with 60 nmol/L of oAβ (from 156.58 ± 4.48 to 198.02 ± 10.28%, P < 0.01). In control studies, we found that 200 nmol/L MC1568 treatment did not affect LTP responses in hippocampal slices from mice treated with regular diet.

We also tested whether pharmacological inhibition of HDAC IIa could influence some of the indices of neuronal activity. In primary neurons derived from C57BL/6J mice, treatment with MC1568 significantly increased the transcription level of the synaptic structure proteins PSD-95, synaptophysin, and brain-derived neurotrophic factor (Fig. 5B), which are important molecules for learning and memory, thus confirming the role of HDAC IIa in the regulation of synaptic activity.

Diabetes mellitus has become an epidemic, and the prevalence in the U.S. is projected to reach 39 million in 2050. As one of the major causes of functional disability, this disease poses tremendous burden on the health system including medical expenses, costs for caregivers, and reduced productivity. Diabetes is one of the major causes of morbidity and mortality and is associated with long-term complications that affect the peripheral as well as the central nervous system.

Diabetes has been linked to cognitive impairment and an increased risk factor for dementia, including AD-type dementia. Hyperglycemia-induced metabolic and vascular disturbances are well acknowledged, however, the exact mechanisms underlying cognitive dysfunction in diabetes are unclear. We found that select chromatin modification enzymes, specifically HDAC class IIa, were upregulated in the brains of diabetic patients and this increase was associated with reduced expression of synaptic proteins that are important for synaptic structural integrity and function. Using an animal model of T2D, we found similar HDAC IIa changes in the brains and these alterations coincided with increased susceptibility of synaptic function to amyloid toxicity. In addition, prolonged diabetic condition exacerbated the observed synaptic dysfunction. Since MRI has been widely used to assess hemodynamic responses related to neural activity in the brain, we used resting-state fMRI to investigate the regionally connected functional networks, e.g., motor/sensory cortex, hippocampus, amygdala, striatum, visual, etc., in the T2D mouse brains. We found that T2D mice exhibited increased signal intensity both in the hippocampal formation and the sensory motor in the cortex, suggesting that T2D brain might not be functioning efficiently and requires much more energy to accomplish a task compared with the normal efficient brain (40,41). It is also possible that a T2D brain may not be signaling coherently and the increase in signaling is a result of haphazard signaling. Our findings are consistent with the human imaging and epidemiology studies demonstrating that progressive cerebral atrophy is associated with prolonged diabetes condition (42–46) and diabetic individuals have an increased risk of developing cognitive dysfunction, including mild cognitive impairment and dementia (6–8,47).

HDACs are a family of enzymes that remove the acetyl moiety of lysine residues from the N-terminus of histone proteins, leading to chromatin compaction and thus transcription repression. Both class I and class II HDACs are abundantly expressed in the brains and have been shown to regulate neuronal activity that is essential for synaptic function, as well as for learning and memory. Our study for the first time demonstrates that the diabetes condition can causally induce changes in HDAC levels in the brain, which, in turn, trigger a detrimental molecular cascade that leads to compromised neuronal structure integrity and synaptic plasticity. Our study also demonstrates that diabetic brains are more susceptible to oAβ-induced synaptic toxicity. oAβ is increasingly accepted as one of the major culprits for AD dementia. Cognitive deterioration in AD is directly linked to the brain accumulation of extracellular soluble oAβ species rather than amyloid plaque deposition (48,49). Previous epidemiological studies showed that the age of onset, duration, and severity of diabetes can influence the incidence of mild cognitive impairment (50). This is consistent with our findings that a 6-month diabetogenic diet treatment resulted in increased susceptibility to oAβ, while prolonging the diabetic condition for an additional 4 months resulted in reduced synaptic response in the absence of exogenous insults. Thus, our results may provide mechanistic explanations of the epidemiological observation that diabetic subjects have an increased risk of developing AD dementia. Notably, HDACs are capable of regulating many cellular and molecular pathways involved in synaptic function and learning and memory. It is possible that modulation of other molecular mechanisms, such as brain-derived neurotrophic factor, cAMP response element–binding (CREB) signaling, and protein kinase–mediated hyperphosphorylation of microtubule-binding protein tau, can all contribute to the phenomenon observed in our study. These mechanisms need to be further investigated.

Diabetes is a multifaceted metabolic disorder that affects both the peripheral and the central nervous system. Based on our findings, it is plausible that the epigenetic alterations in diabetic brains may make the brain not only more susceptible to oAβ but also to other insults such as aberrantly aggregated neurofibrillary tangles made of hyperphosphorylated tau, the other pathological hallmark of AD. It is also possible that these changes make the brains more vulnerable to infarction and small-vessel disease, which may lead to vascular dementia and other insults such as inflammation and advanced glycation end products that are prevalent in diabetes. Future research on the relationship of diabetes-induced alteration of epigenetic mechanisms and the susceptibility to these potential diabetes-associated pathologic changes will further elucidate the molecular mechanisms underlying the increased risk of diabetic patients developing dementia.

Acknowledgments. The authors thank Dr. Vahram Haroutunian, Department of Psychiatry and Neurology from Mount Sinai School of Medicine, for providing the samples from the brain bank. The authors acknowledge the editorial assistance of Samara Levine of the Department of Neurology at the Icahn School of Medicine for help preparing the manuscript.

Funding. This study was supported by discretionary funding to G.M.P.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.W. designed the studies, wrote the manuscript, and performed experiments. B.G., W.Z., C.T., M.V., T.N., W.B., A.B., S.B., P.V., and L.K. performed experiments. C.T. and L.H. designed the studies. G.M.P. designed the studies and wrote the manuscript. G.M.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.