Seizures exacerbate excitatory: inhibitory imbalance in Alzheimer’s disease and 5XFAD mice

Abstract

Approximately 22% of Alzheimer’s disease (AD) patients suffer from seizures, and the co-occurrence of seizures and epileptiform activity exacerbates AD pathology and related cognitive deficits, suggesting that seizures may be a targetable component of AD progression. Given that alterations in neuronal excitatory:inhibitory (E:I) balance occur in epilepsy, we hypothesized that decreased markers of inhibition relative to those of excitation would be present in AD patients. We similarly hypothesized that in 5XFAD mice, the E:I imbalance would progress from an early stage (prodromal) to later symptomatic stages and be further exacerbated by pentylenetetrazol (PTZ) kindling.

Post-mortem AD temporal cortical tissues from patients with or without seizure history were examined for changes in several markers of E:I balance, including levels of the inhibitory GABA_A receptor, the sodium potassium chloride cotransporter 1 (NKCC1) and potassium chloride cotransporter 2 (KCC2) and the excitatory NMDA and AMPA type glutamate receptors. We performed patch-clamp electrophysiological recordings from CA1 neurons in hippocampal slices and examined the same markers of E:I balance in prodromal 5XFAD mice. We next examined 5XFAD mice at chronic stages, after PTZ or control protocols, and in response to chronic mTORC1 inhibitor rapamycin, administered following kindled seizures, for markers of E:I balance.

We found that AD patients with comorbid seizures had worsened cognitive and functional scores and decreased GABA_A receptor subunit expression, as well as increased NKCC1/KCC2 ratios, indicative of depolarizing GABA responses. Patch clamp recordings of prodromal 5XFAD CA1 neurons showed increased intrinsic excitability, along with decreased GABAergic inhibitory transmission and altered glutamatergic neurotransmission, indicating that E:I imbalance may occur in early disease stages. Furthermore, seizure induction in prodromal 5XFAD mice led to later dysregulation of NKCC1/KCC2 and a reduction in GluA2 AMPA glutamate receptor subunit expression, indicative of depolarizing GABA receptors and calcium permeable AMPA receptors. Finally, we found that chronic treatment with the mTORC1 inhibitor, rapamycin, at doses we have previously shown to attenuate seizure-induced amyloid-β pathology and cognitive deficits, could also reverse elevations of the NKCC1/KCC2 ratio in these mice.

Our data demonstrate novel mechanisms of interaction between AD and epilepsy and indicate that targeting E:I balance, potentially with US Food and Drug Administration-approved mTOR inhibitors, hold therapeutic promise for AD patients with a seizure history.

See Blum et al. (https://doi.org/10.1093/brain/awae146) for a scientific commentary on this article.

Introduction

Seizures can be a comorbidity in patients with Alzheimer’s disease (AD), with 10%–22% showing clinically identifiable seizures and up to 64% displaying subclinical epileptiform activity.¹ Notably, the co-occurrence of seizures and epileptiform activity in AD patients enhances disease progression and worsens cognitive performance.^1-5 We and others have demonstrated bidirectional interactions, with AD and epilepsy sharing similar pathology and convergence upon common underlying cellular mechanisms. Human temporal lobe epilepsy brain tissue shows increased amyloid-β (Aβ) plaques and increased hyperphosphorylated tau (pTau),^6-8 and the accumulation of these neuropathological proteins in AD animal models is associated with elevated neuronal hyperexcitability and seizure susceptibility.^9-14 Furthermore, we recently demonstrated that a history of seizures is associated with enhanced Aβ and pTau pathology in human AD.¹⁴

While mechanisms of seizure generation in AD are not fully elucidated, evidence points to a synaptic excitatory:inhibitory (E:I) imbalance in chronic epilepsy, including altered expression and function of certain neurotransmitter receptors and ion cotransporters.^15-17 Chronic epilepsy human and animal model studies show evidence of reduced inhibitory neurotransmission, with decreases in the ratio of α1/α3 inhibitory gamma-aminobutyric acid type A receptor (GABA_AR) subunits that diminish inhibitory postsynaptic currents (IPSCs).^18-22 Inhibitory GABAergic transmission is further attenuated by paradoxical depolarizing GABA_AR currents due to reversal of the chloride (Cl⁻) gradient caused by increases in the ratio of the sodium (Na⁺) potassium (K⁺) Cl⁻ co-transporter 1 (NKCC1), which imports Cl⁻, to that of the potassium Cl⁻ cotransporter 2 (KCC2), which exports Cl⁻.^23-26 Commensurate with decreased inhibitory drive in epilepsy, there is an increase in glutamate-receptor mediated excitability, in part due to alterations in α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) and N-methyl-D-aspartate receptor (NMDAR) subunit expression, with increases in the ratios of GluA1/GluA2 AMPAR subunits and GluN2B/GluN2A NMDAR subunits resulting in GluA2-lacking, Ca²⁺-permeable AMPAR^27-29 and extended decay times and elevated Ca²⁺ influx, respectively.^30-32

Pathological E:I alterations are not unique to epilepsy, as evidence for neuronal hyperexcitability has been seen in other neurodevelopmental and neurodegenerative disorders.^33-36 In AD patients, altered GABA_AR subunit regulation has been observed,³⁷ including decreased α1/α3 expression.³⁸ Furthermore, evidence for paradoxically depolarizing GABA currents has been found, with elevations of NKCC1 after Aβ_1–42 hippocampal injection³⁹ and reduced KCC2 in AD mouse models, with behavioural rescue after restoration of membrane KCC2.^40,41 Aβ_1–42 has also been shown to increase Ca²⁺-permeable AMPARs which can result in memory deficits via reduced long-term potentiation and synapse loss.^42-45 However, given the prevalence of network excitability and seizures in AD, and the evidence that seizures can alter E:I balance in epilepsy, understanding the impact of seizures and the evolution of neuronal hyperexcitability in AD progression may lead to new mitigable therapeutic E:I targets. To understand the progression of E:I balance, we examined the 5XFAD mouse model at two stages: prodromal and symptomatic. To understand seizure impact on E:I imbalance in AD, we examined human AD with known seizure history and 5XFAD mice at a symptomatic stage after seizure induction during the prodromal stage.

Preclinical and clinical epilepsy research has revealed several candidate signalling pathways that might link E:I imbalance and network hyperexcitability in epilepsy to neurodegeneration. One example is the substantial evidence for the activation of the mammalian target of rapamycin complex 1 (mTORC1) as a key common element between epilepsy and neurodegeneration. mTORC1 mediates pathological tau and APP processing^46-50 and can regulate excitatory and inhibitory receptors and network function.^51,52 Models of tuberous sclerosis complex, characterized by constitutive mTORC1 overactivation commonly resulting in epilepsy,^53,54 have demonstrated that blockade of mTORC1 with rapamycin can rescue GABAergic neurotransmission and epilepsy.^55-58 Rapamycin treatment has also rescued neuronal hyperexcitability in models of autism spectrum disorder⁵⁹ and neuropathic pain.⁶⁰ Together, these results indicate that targeting mTORC1 activity might restore E:I balance across multiple disease models, including AD.

Considering these data and that rapamycin and other rapalogues are approved by the US Food and Drug Administration for other indications, mTORC1 is a promising target for therapeutic intervention for patients with AD and co-morbid epilepsy. Indeed, we recently found that AD patients with a history of seizure display enhanced tau and Aβ pathology, associated with increased mTORC1 activity.¹⁴ In addition, we observed that the induction of seizures in the 5XFAD mice exacerbates Aβ pathology and cognitive deficits, and that these changes are reversible with chronic low-dose rapamycin treatment.¹⁴

Given that E:I imbalance may contribute to the cognitive impairments observed in AD, we hypothesized that key components of E:I balance may be dysregulated in AD and that these alterations emerge early in the disease course and may be exacerbated by seizures. First, we found worsened cognitive and functional performance in AD patients with a seizure history compared with those without and found seizures were associated with greater alterations in several components of the GABAergic system in largely the same human control and AD temporal neocortex from patients with and without known seizure history as our previous study.¹⁴ Next, we examined the 5XFAD mouse model for protein and electrophysiological indications of hyperexcitability at the prodromal stage (4 months of age) and directly assessed the effects of chemoconvulsant [pentylenetetrazol (PTZ)]-induced seizures at this stage on E:I markers at a later, symptomatic time point (7 months of age). Finally, given that we found protective effects on pathology and cognition with rapamycin in this model,¹⁴ we used the tissue from our prior study to examine whether these therapeutic benefits extend to markers of E:I balance. Overall, we found that E:I imbalance began at a prodromal stage in 5XFAD mice and that seizures significantly exacerbated excitatory and inhibitory alterations, while chronic rapamycin treatment significantly attenuated these effects.

Materials and methods

Human subjects

All protocols and procedures were approved under the ethical standards of the Institutional Review Board of the University of Pennsylvania (Philadelphia, PA). The AD cognitive study population consisted of 105 patients, selected by querying the integrated neurodegenerative disease database⁶¹ of the Center for Neurodegenerative Disease and Research (CNDR) at the University of Pennsylvania for available global Clinical Dementia Rating (CDR) score and CDR Sum of Boxes (CDR-SOB) closest to death. All patients were enrolled in observational research at the University of Pennsylvania, with standardized assessments and medical history that included assessment of seizure history.^62,63 Of the AD cases, 18 (6 males and 12 females; mean age at death = 77.38 years) had a reported clinical seizure history (AD+Sz). The remaining 87 AD patients (43 males and 44 females; mean age at death = 79.7 years) had no known seizure history (AD−Sz). AD clinical diagnosis was established during life based on the clinical history, neurological and neuropsychological assessment, and confirmed by post-mortem histopathological staging of AD neuropathological markers (Aβ₄₂ and tau pathology), according to the National Institute on Aging-Alzheimer’s Association (NIA-AA) guidelines.⁶²

For a subset of the population detailed above, post-mortem brain specimens and region matched controls were acquired from CNDR and consisted of superior/mid-temporal cortex frozen samples (control: n = 13, AD−Sz: n = 19, AD+Sz: n = 11) and formalin-fixed paraffin-embedded sections (control: n = 5, AD−Sz: n = 7, AD+Sz: n = 8).

Detailed methods for neurocognitive assessments and clinical characteristics for subjects included in both neurocognitive and biochemical studies can be found in the Supplementary material and Supplementary Table 1.

Mice

All animal procedures were approved and performed in accordance with the guidelines of the Institutional Animal Care and Use Committee (IACUC) Office of Animal Welfare of the University of Pennsylvania (Philadelphia, PA). Male and female 5XFAD and wild-type (WT) mice on a mixed B6SJL background second generation (F2) generated from mice originally purchased from Jackson Laboratory (MMRC stock #3840) were used in these studies. 5XFAD mice harbour the human APP and PSEN1 genes under the Thy1 promoter containing five mutations for aggressive amyloid development (APP: Swedish K670N/M671L, Florida I7516V and London V717I; PSEN1: M146L and L286V). Genotyping was conducted as specified by Jackson Laboratory. Mice were allocated randomly to treatment groups, which were counterbalanced for sex and litter. Additional methodology, including PTZ kindling, rapamycin treatment, electrophysiology and western blot are provided in the Supplementary material.

Statistics

For cognitive and functional performances, comparisons between AD+Sz and AD−Sz were made using the Mann–Whitney test. For global CDR, we determined the potential effects of clinical characteristics by separate ordinal regressions with seizure history and age at test, disease duration, sex or education as independent variables. For the CDR-SOB and Dementia Severity Rating Scale (DSRS), we examined these same clinical characteristics by separate analysis of covariance (ANCOVA), with seizure history as the independent variable and age at test, disease duration, sex or education as a covariate. For the human tissue studies, all AD to control comparisons were made with unpaired t-tests or Mann–Whitney tests for non-normal data. Control, AD−Sz and AD+Sz comparisons were made using ANOVA or the Kruskal–Wallis test for non-normal data. For human biochemical analyses, age at death, disease duration and sex were examined by multiple linear regression. For the prodromal mouse studies, 5XFAD to WT comparisons were made using unpaired t-tests and Kolmogorov–Smirnov (K-S) tests for cumulative probability comparisons. For the non-therapeutic PTZ study, comparisons were made using two-way ANOVA, with genotype and kindling as independent variables. For the therapeutic PTZ study, 5XFAD and WT mice were analysed separately using two-way ANOVAs, with PTZ and rapamycin treatment as independent variables. All ANOVA analyses were followed by Tukey’s post hoc test to examine group differences when main effects or interactions were found. Multiple linear regressions were performed for mouse biochemical studies to determine potential sex effects. Where sex effects were present, data-points are displayed as squares (female) or triangles (male) to visualize these differences, and histograms separated by sex can be found in the Supplementary material. Correlations were calculated by Pearson correlation coefficients. ANCOVA and ordinal regressions were performed in R (version 4.0), and the remaining statistical analyses were performed in GraphPad Prism 9 software (San Diego, CA). Histogram error bars represent the standard error of the mean. We considered the results to be significant at P < 0.05.

Results

Alzheimer’s disease patients with seizure history have worsened cognitive and functional performance

To establish potential associations between seizures and cognitive and functional deficits in AD, we examined scores from clinical (global CDR and CDR-SOB) and caretaker questionnaires (DSRS) for subjects with known seizure history. No significant differences between AD−Sz and AD+Sz were found in years since onset to final CDR assessment (mean AD−Sz = 8.3 years; mean AD+Sz = 9.9 years), nor in age at final CDR assessment (mean AD−Sz = 77.1 years; mean AD+Sz = 76.0 years) (Supplementary Table 1). The AD+Sz group showed elevated final global CDR compared with AD−Sz (P < 0.01, n = 18–87; Fig. 1A). Separate ordinal regressions revealed that age at test [Sz+age: age P < 0.05 (0.0034, 0.078)] and disease duration [Sz+duration: duration P < 0.001 (0.18, 0.44)] were both significant predictors of CDR, while no significant effects of sex or education were found (Supplementary Table 2). Importantly, seizures were significant predictors of CDR in all models [Sz+age: Sz P < 0.01 (0.59, 3.32); Sz+duration: Sz P < 0.05 (0.87, 5.57); Sz+sex: Sz P < 0.05 (0.48, 3.19); Sz+education: Sz P < 0.05 (0.53, 3.25)] (Supplementary Table 2). AD+Sz also showed worsened CDR-SOB compared with the AD−Sz group (P < 0.05, n = 18–87; Fig. 1B). Separate ANCOVAs revealed that seizures significantly impact CDR-SOB when age [F(1,102) = 4.81, P < 0.05], disease duration [F(1,94) = 6.23, P < 0.05] and education [F(1,102) = 4.1, P < 0.05] are adjusted for, and there was a significant interaction between seizures and sex [F(1,102) = 355.17, P < 0.0001] (Supplementary Fig. 1 and Supplementary Table 2). AD+Sz patients also showed significantly worsened overall functional rating (DSRS) compared with those in the AD−Sz group (P < 0.01, n = 14–19) (Fig. 1C), including significantly worse ratings for memory, recognition of family members, mobility, orientation to place, social and community activities and incontinence (P < 0.05, n = 14–19) (Supplementary Fig. 2). Separate ANCOVAs revealed that seizures significantly impact DSRS when age [F(1,30) = 6.68, P < 0.05], sex [F(1,30) = 6.78, P < 0.05] and education [F(1,30) = 5.95, P < 0.05] were adjusted for, and we found a significant interaction between seizures and disease duration [F(1,27) = 62.39, P < 0.0001] (Supplementary Table 2). Overall, these data indicated that seizures are associated with worsened cognitive performance and overall function, corroborating previous studies examining epileptiform activity in AD^2-5 and suggesting heightened neuronal dysfunction.

Alzheimer’s disease with seizure history shows worse dysregulation of proteins that mediate E:I balance

We hypothesized that the worsened functional and cognitive deficits found in AD+Sz patients would be associated with worsening of markers of E:I imbalance. Thus, we examined the ratios of α1/α3 GABA_AR subunits¹⁹ and of Cl⁻ cotransporters, NKCC1/KCC2.^23,24,64 We found a significant decrease in GABA_ARα1 expression in AD+Sz compared with controls (F = 3.52, Tukey’s post hoc: P < 0.05) (Fig. 2B) that was not found in AD−Sz and which drove the reduction in GABA_ARα1 found across all AD patients (P < 0.05) (Fig. 2B). This alteration also resulted in a significantly decreased GABA_ARα1/GABA_ARα3 ratio in all AD patients when compared with controls (P < 0.01) (Fig. 2B). The ratio of NKCC1/KCC2 was also significantly increased in the AD+Sz group when compared with the AD−Sz group (Tukey’s post hoc: P < 0.05) (Fig. 2C). This difference was largely attributable to downregulation of KCC2 in the AD+Sz group compared with the AD−Sz group (Tukey’s post hoc: P < 0.05) (Fig. 2C). In addition, sex was a significant predictor of KCC2 dysregulation, with decreased levels in males [P < 0.05, (0.031, 0.230)] (Fig. 2C, Supplementary Fig. 3 and Supplementary Table 3). Taken together these results suggest AD+Sz patients had greater GABAergic dysfunction due to reduced tonic inhibition and indications of depolarizing GABA_AR-mediated signals.

To explore inhibitory dysfunction further, we compared levels of the predominant, parvalbumin (PV)-expressing interneuron subtype⁶⁵ in tissues from AD+Sz and AD−Sz, given that there is a well-established association between PV cell loss, cognition⁶⁶ and epilepsy.^67,68 Immunohistochemistry revealed that the AD+Sz group had significantly decreased PV+ cells compared with controls (Tukey’s post hoc: P < 0.01) (Supplementary Fig. 4A and B) that was not found in AD−Sz, which drove the significant decrease in PV+ cells found between all AD and control samples (P < 0.01) (Supplementary Fig. 4A and B).

To determine if neuronal excitation may be heightened in AD and whether seizures cause further alterations, we examined the expression of AMPA and NMDA glutamate receptor subunit proteins, namely ratios of GluA1/GluA2 and GluN2B/GluN2A subunits^27,28,30 (Fig. 2D and E). While no differences were found between AD+Sz and AD−Sz in AMPAR subunits or subunit ratios (Fig. 2D), we found increased expression of GluA2 subunits in AD temporal cortex overall compared with controls (P < 0.05) (Fig. 2D), and the corresponding GluA1/GluA2 ratio was decreased in the AD cases compared with controls (Tukey’s post hoc: P < 0.05) (Fig. 2D). In examining NMDAR subunits, there was no difference in GluN2B or GluN2A levels nor in the ratio of GluN2B/GluN2A between the AD+Sz and AD−Sz groups (Fig. 2E). However, GluN2A and GluN2B subunits were both expressed at significantly higher levels in AD temporal cortex when compared with controls (P < 0.05) (Fig. 2E).

Finally, we determined correlations between clinical outcomes of AD and markers of E:I balance. We found that CDR-SOB was inversely correlated with GluN2A levels across all AD subjects (r = −0.39, P < 0.05) (Fig. 2F), while gross brain atrophy ratings at autopsy were associated with decreased levels of GABA_ARα1 in AD+Sz (r = 0.90, P < 0.01) (Fig. 2G), indicating an interaction between these markers of E:I balance and cognitive decline and AD severity, respectively. No correlations were found between markers of E:I balance and pathology previously measured in these samples.¹⁴ While AD patients were significantly older than controls at time of death [AD: mean = 75.6, standard deviation (SD) = 9.7; control: mean = 62.4, SD = 8.0; P < 0.001], the post-mortem interval (PMI) did not differ between the control and AD groups, and multiple linear regression revealed no significant relationships between age at death, disease duration or PMI and any marker of E:I balance (Supplementary Table 3). These results provided strong evidence of decreased inhibition and increased excitation in AD patients overall, regardless of seizure history, and an exacerbation of selective regulators to worsen GABAergic dysfunction in the AD+Sz group.

Excitatory:inhibitory imbalance in prodromal stage 5XFAD mice

Given that our human tissue samples are limited by the fact they are from patients with late-stage AD (Supplementary Table 1), we used the 5XFAD mouse model to examine earlier stages. Indeed, there is evidence of E:I imbalance in AD mouse models at early stages.^69-71 We determined whether baseline E:I imbalance is present in prodromal stage (4 months of age) 5XFAD mice, when mild behavioural impairment,⁷² AD pathology,⁷³ subclinical epileptiform electroencephalographic activity⁷⁴ and decreased thresholds for chemoconvulsant-induced seizures¹⁴ are present.

To examine inhibitory synaptic function, we measured miniature inhibitory postsynaptic currents (mIPSCs) in CA1 pyramidal neurons in hippocampal slices from prodromal 5XFAD mice using whole-cell patch clamp recordings. We found decreased mIPSC amplitudes [P < 0.0001, n = 11 694 (WT) and 10 948 (5XFAD) events, 20 cells/group, 3 mice/group] (Fig. 3A–C) and increased mIPSC decay time [P < 0.001, n = 7477 (WT) and 6857 (5XFAD) events, 20 neurons/group, 3 mice/group] (Fig. 3A, D and E) in 5XFAD mice compared with WT mice, suggesting that decreased GABA_ARα1/GABA_ARα3 ratios^21,22 underlie inhibitory synaptic dysfunction. In addition, we found increased inter-mIPSC intervals (reduced frequency) in 5XFAD compared with WT [P < 0.0001, n = 11 673 (WT) and 10 929 (5XFAD) events, 20 neurons/group, 3 mice/group] (Fig. 3A, F and G), indicating reduced inhibitory presynaptic inputs. We also examined the excitatory synaptic function of CA1 neurons by recording miniature excitatory postsynaptic currents (mEPSCs). mEPSC amplitudes were significantly increased in 5XFAD [P < 0.0001, n = 15 619 (WT) and 11 082 (5XFAD) events, 19 (WT) and 21 (5XFAD) neurons, 3 mice/group] (Fig. 3H–J), suggesting increased postsynaptic glutamate receptor densities. mEPSCs frequencies were significantly decreased in 5XFAD, as demonstrated by increased inter-mEPSC intervals [P < 0.0001, n = 15 600 (WT) and 11 060 (5XFAD), 19 (WT) and 21 (5XFAD) neurons, 3 mice/group] (Fig. 3H, K and L), consistent with decreased excitatory presynaptic drive. These changes in synaptic function were also associated with increased intrinsic excitability of CA1 pyramidal neurons in 5XFAD mice as measured by recording evoked action potentials (APs) in current clamp mode using depolarizing step currents (Supplementary Fig. 5). Together, these data demonstrated reduced inhibitory activity and overall hyperexcitability, which may underlie the increased seizure susceptibility we previously found in prodromal 5XFAD mice.¹⁴

To investigate the protein changes involved in E:I balance, we performed western blot on hippocampi and cortices from prodromal 5XFAD and WT mice (Fig. 4). Consistent with reduced mIPSC amplitude and increased decay times, we found decreased GABA_ARα1/GABA_ARα3 ratios in the hippocampus of 5XFAD compared with WT mice (P < 0.05, n = 8/group) (Fig. 4B). We also found a trend towards decreased KCC2 in the cortex (P = 0.09, n = 14–15) (Fig. 4G). Notably, excitatory glutamate AMPAR and NMDAR subunits in hippocampus and cortex remained unchanged at this stage (Fig. 4).

Kindling exacerbates the progressive dysregulation of proteins involved in E:I balance in 5XFAD mice

Given the findings of enhanced E:I imbalance in human tissue from AD patients with seizures, we sought to determine how E:I balance changes with disease progression in 5XFAD mice and whether chronic seizures are associated with further dysregulation of E:I imbalance. We performed PTZ seizure kindling or control protocols in prodromal (3.5-month-old) 5XFAD and littermate WT mice and analysed the expression levels of markers of E:I imbalance in the symptomatic (7-month-old) hippocampus (Fig. 5) and cortex (Supplementary Fig. 6). 5XFAD mice had decreased hippocampal GABA_ARα1/α3 ratios compared with WT mice, regardless of treatment group [genotype effect: F(1,61) = 22.76, P < 0.0001] (Fig. 5A), with significant decreases in both non-kindled (Tukey’s post hoc: P < 0.05) and PTZ-kindled 5XFAD mice (Tukey’s post hoc: P < 0.01). No differences in GABA_ARα1/α3 ratios were found due to kindling in either 5XFAD or WT mice. The changes in the hippocampal GABA_AR ratio seem to be driven largely by decreases in the GABA_ARα1 subunit in 5XFAD mice [genotype effect: F(1,61) = 40.09, P < 0.0001] (Fig. 5A). In addition, sex was a significant predictor of hippocampal GABA_ARα1/α3 with greater decreases found in males [P < 0.05 (0.0156, 0.1591)] (Supplementary Fig. 7). Decreased GABA_ARα1/α3 ratios were also found in the cortex [interaction: F(1,43) = 4.25, P < 0.05], with a significant reduction in non-kindled 5XFAD mice compared with WT (Tukey’s post hoc: P < 0.05) (Supplementary Fig. 6A). These GABA_AR changes were similar to patterns seen in human AD brain (Fig. 2) and prodromal (4-month) 5XFAD tissue (Fig. 4), demonstrating persistent GABA_AR alterations into symptomatic disease stages in 5XFAD mice.

We next tested whether there were age-related and/or kindling-induced alterations to Cl⁻ cotransporters in 5XFAD mice, given their critical role in regulating GABAergic inhibition. Kindling resulted in an increased NKCC1/KCC2 ratio in the hippocampus of symptomatic 5XFAD mice [interaction: F(1,61) = 5.78, P < 0.05] compared with WT and non-kindled 5XFAD mice (Tukey’s post hoc: P < 0.01, 0.05) (Fig. 5B), due to a significant increase in expression of the Cl⁻ importer, NKCC1 [interaction: F(1,61) = 7.93, P < 0.01, Tukey’s post hoc: P < 0.01] (Fig. 5B). 5XFAD mice showed decreased hippocampal expression of the Cl⁻ exporter, KCC2, regardless of PTZ treatment [genotype effect: F(1,61) = 12.35, P < 0.001], which was not seen in prodromal 5XFAD mice, suggesting worsened inhibitory deficits commensurate with pathological progression (Table 1). In the cortex (Supplementary Fig. 6), we also found a significant increase in NKCC1/KCC2 ratios due to PTZ kindling [kindling effect: F(1,43) = 4.49, P < 0.05], which was largely driven by a reduction of KCC2 in kindled WT mice [interaction: F(1,43) = 6.75, P < 0.05]. Interestingly, while cortical NKCC1/KCC2 ratios were not significantly different, there was a decrease in NKCC1 expression [genotype effect: F(1,43) = 5.2, P < 0.05] in non-kindled 5XFAD compared with non-kindled WT mice (Tukey’s post hoc: P < 0.05) (Supplementary Fig. 6B), which may indicate a compensatory mechanism. These data suggest that depolarizing GABA⁷⁵ may contribute to the exacerbation of GABAergic dysfunction in the kindled 5XFAD hippocampus.

We next determined whether glutamate receptors may be dysregulated as a function of disease progression and by kindled seizures in 5XFAD mice. In the hippocampus, we found a significant effect to increase GluA1/GluA2 ratios in PTZ-kindled hippocampus [Genotype × Kindling: F(1,61) = 4.622, P < 0.05, Tukey’s post hoc: P < 0.01] (Fig. 5C), which contributes to E:I imbalance in epilepsy.³⁰ In addition, both subunits were decreased in the hippocampus of 5XFAD mice regardless of kindling status [GluA1 genotype effect: F(1,61) = 7.722, P < 0.01; GluA2 genotype effect: F(1,61) = 30.28, P < 0.0001] (Fig. 5C), and similar alterations in AMPAR subunit expression and ratios were found in the cortex (Supplementary Fig. 6C). The NMDAR GluN2A subunit was also decreased in 5XFAD hippocampus [genotype effect: F(1,61) = 6.31, P < 0.05] (Fig. 5D) and cortex [genotype effect: F(1,43) = 9.0, P < 0.01, Tukey’s post hoc: P < 0.05] (Supplementary Fig. 6D) without change GluN2B and subunit ratios. These data demonstrate dysregulation to glutamate receptor expression beginning by 7 months of age in 5XFAD mice and an exacerbation of this effect due to PTZ kindling.

In addition, we used our retrospective measures of disease severity in this same cohort of mice¹⁴ and found that NKCC1/KCC2 was positively correlated with Aβ₄₂ (r = 0.57, P < 0.05) and pTau AT100 (Thr212, Ser214)/Tau (r = 0.64, P < 0.05) in the hippocampus of PTZ-kindled 5XFAD mice and with performance on the Y-maze (r = −0.61, P < 0.01) (Fig. 5F–H), which we previously demonstrated was worsened by PTZ kindling.¹⁴ No correlations were found in non-kindled mice nor between GABA_AR, AMPAR or NMDAR in kindled mice. These findings suggest a relationship between inhibitory dysregulation, AD pathology and cognitive decline.

In summary, 5XFAD mice showed a worsening of markers of inhibitory imbalance, which was associated with worsened pathology and cognitive decline, and the emergence of dysregulated glutamate receptors as pathology progresses from prodromal to later stages (Table 1). Furthermore, PTZ kindling exacerbated imbalance in both excitatory and inhibitory markers (Table 2).

mTORC1 inhibition modifies select seizure-induced E:I markers in the hippocampus of 5XFAD mice

Rapamycin treatment has rescued neuronal hyperexcitability in various disease models.^55-60 Thus, we sought to determine whether chronic low-dose rapamycin treatment beginning at the cessation of PTZ kindling (prodromal stage ∼4 months, 2.24 mg/kg daily), previously used to rescue seizure-induced neuropathology and cognitive deficits in 5XFAD mice,¹⁴ could ameliorate PTZ-exacerbated E:I imbalance. The 5XFAD (Fig. 6) and WT (Supplementary Fig. 8) mice used in our previous report were analysed separately by two-way ANOVA, with PTZ kindling and rapamycin treatment as independent variables.

In 5XFAD mice, rapamycin showed a strong trend (Tukey’s post hoc P = 0.058) to increase GABA_ARα1/GABA_ARα3 in non-kindled mice compared with the non-kindled, vehicle-treated group [kindling effect: F(1,45) = 6.1, P < 0.05; rapamycin effect: F(1,45) = 4.1, P < 0.05] (Fig. 6A). These shifts corresponded with GABA_ARα1 subunit expression [kindling effect: F(1,45) = 4.2, P < 0.05; rapamycin effect: F(1,45) = 6.1, P < 0.05] (Fig. 6A). Rapamycin also reversed the elevation of the NKCC1/KCC2 ratio found in PTZ-kindled 5XFAD mice [interaction: F(1,45) = 9.8, P < 0.01] compared with control 5XFAD mice (Tukey’s post hoc: P < 0.05) (Fig. 6B). The changes in Cl⁻ cotransporter ratio was largely driven by changes in the Cl⁻ importer, NKCC1 [interaction: F(1,45) = 7.95, P < 0.01] (Fig. 6B). These results likely reflect a rescue of GABA_AR function in kindled 5XFAD mice by rapamycin.

The analysis of AMPAR subunits showed that PTZ kindling increased GluA1/GluA2 ratios in symptomatic 5XFAD mice (Tukey’s post hoc: P < 0.01) (Fig. 6C), consistent with the non-therapeutic cohort, which was reversed by rapamycin treatment [interaction: F(1,45) = 19.83, P < 0.01] (Fig. 6C). Rapamycin increased GluA1/GluA2 ratio in non-kindled 5XFAD mice (Tukey’s post hoc: P < 0.05), largely driven by increased GluA1 expression [interaction: F(1,45) = 4.3, P < 0.05] (Fig. 6C). Additionally, rapamycin treatment reduced GluN2A and GluN2B levels in untreated and PTZ-kindled mice [GluN2A rapamycin effect: F(1,45) = 5.2, P < 0.05; GluN2B rapamycin effect: F(1,45) = 4.1, P < 0.05] (Fig. 6D). PTZ-kindled, rapamycin treated 5XFAD mice had elevated GluN2B/GluN2A ratio compared with control 5XFAD mice [kindling effect: F(1,45) = 5.66, P < 0.05; Tukey’s post hoc: P < 0.05] (Fig. 6D).

In WT mice, no significant differences were found due to PTZ kindling or rapamycin treatment for GABA_ARs, AMPARs or NMDARs (Supplementary Fig. 8), although kindling did increase NKCC1 expression in WT mice across rapamycin and vehicle treatments [kindling effect: F(1,47) = 5.2, P < 0.05] (Supplementary Fig. 8B). Overall, these data suggested that rapamycin can ameliorate PTZ-exacerbated E:I imbalance in 5XFAD mice.

Discussion

Here we identified novel mechanisms of neuronal dysfunction that may underlie seizure-exacerbated cognitive and functional deficits in AD. Human AD temporal cortex showed dysregulation of E:I balance proteins, which was worsened in those with seizure history. Similarly, physiological and protein indications of neuronal E:I imbalance were found in prodromal 5XFAD mice, and further perturbations to synaptic protein regulators were seen at symptomatic stages, with additional exacerbations by PTZ-kindling (Tables 1 and 2). Together with our recent study demonstrating enhanced AD pathology in AD+Sz patients and PTZ-kindled mice,¹⁴ our data suggest that E:I imbalance begins at prodromal stages and may play a role in seizure-induced exacerbation of neuropathology and cognitive decline in AD. The most consistent E:I alterations include decreased GABA_ARα1 subunit expression and shifts in the balance between the Cl⁻ cotransporters NKCC1/KCC2 and the AMPAR GluA1/GluA2 subunit ratios as factors in this interaction between AD and seizures. Furthermore, our data indicate that using mTORC1 inhibitors such as rapamycin to therapeutically target the E:I imbalance may prove to be a promising therapy to alleviate AD progression due to neuronal hyperexcitability.

Our rare autopsy-confirmed AD cases with prospective collection of seizure history data suggest that seizures are associated with worsened cognitive function and E:I imbalance. We also found that markers of E:I balance correlated with clinical measures of decline and disease severity in these tissue samples. Importantly, with a limited sample size, we found seizures were significant predictors of clinical outcomes across separate statistical models when either sex, age at test, disease duration or education level were accounted for, justifying follow-up studies to further examine clinical and biochemical outcomes associated with seizure history in AD with more comprehensive statistical modelling that was not possible in the dataset here. It is important to note that the reports of seizure history for patients in this study did not contain information on the aetiology of seizures, age of onset (whether seizures began before or after AD diagnosis), extent of seizure severity, nor whether the seizures were controlled. Even with these limitations, when taken with our prior reports in these patients,¹⁴ our data suggest that seizures significantly worsen AD disease progression, indicating that AD patients should be more closely monitored for seizures, thus providing more comprehensive descriptions of seizure activity and history that were not available for the current study.

Here we found a specific decrease in GABA_ARα1/GABA_ARα3 ratios in AD patients compared with controls, supporting previous reports of reduced GABA sensitivity in membranes isolated from the temporal cortex of AD brains.³⁸ Indeed, previous studies have found marked alterations in synaptic GABAergic signalling in AD, including downregulation of GABA_AR subunits^37,38,76,77 and decreased expression of inhibitory synapses overall.^78,79 We show that AD patients with seizure history expressed significantly lower GABA_ARα1 and GABA_ARα3 subunits with GABA_ARα1 levels correlating with brain atrophy in AD patients with seizures. In addition, NKCC1/KCC2 ratios were increased in AD patients with a history of seizures as compared with those without, consistent with collapse of the neuronal Cl⁻ gradient. We also found a loss of PV-containing interneurons in the temporal cortex of AD patients compared with controls, consistent with the reported reduction of PV cells within the AD parahippocampal gyrus,⁸⁰ which was driven by the AD+Sz group. Fast-spiking PV interneurons are major contributors to generation of gamma oscillations, which are disrupted in AD patients.^81,82 Taken in the context of prior studies, these data suggest that a history of seizures in AD is associated with diminished GABAergic transmission via reduced GABA_AR synapses, loss of interneurons and possibly a depolarizing GABA response at those synapses that remain due to elevated intracellular Cl⁻. Greater loss of these GABAergic interneurons in AD patients with epilepsy may promote network hyperexcitability and contribute to the worsened cognitive and functional performance found in these subjects (Fig. 1).

Electrophysiological recordings in AD models, and 5XFAD mice in particular, are sparse, especially at stages in which plaque pathology has developed. Our electrophysiological data from the CA1 of prodromal 5XFAD mice demonstrated significantly increased network excitability as well as deficits in inhibitory synaptic transmission as evidenced by reductions in mIPSC amplitude and frequency as well as increased decay times. Consistent with the functional inhibitory changes, we found decreased GABA_ARα1/GABA_ARα3 ratios in the hippocampus of prodromal 5XFAD mice, GABA_AR configurations previously shown to alter IPSC amplitude²¹ and decay time.²² These results align with prior studies showing reductions in GABARα1 and α5 transcripts in patients with mild cognitive impairment, suggesting GABA_AR vulnerability at initial disease stages.⁸³

When taken with the prodromal 5XFAD data, non-kindled 7-month-old 5XFAD mice showed a worsening of GABA_AR dysregulation with reduced GABA_ARα1/GABA_ARα3 in the cortex in addition to the hippocampus (Table 1). We and others have also found decreased PV immunoreactivity,^14,84-87 suggesting a broader deficit of inhibition and reflecting our results in AD temporal cortex. Indeed, GABA_AR loss has been linked to seizures and cognitive impairment in mouse models of AD.^88-90 This mechanism may be responsible for such changes in the 5XFAD model. Consistent with our human data, we found that induced seizures may exaggerate GABAergic dysfunction through increased NKCC1/KCC2 ratios, which were reversed in rapamycin-treated mice.

With respect to glutamatergic neurotransmission, studies of AD hippocampus highlighted decreased expression of GluA2 compared with controls,^91-93 a configuration known to increase AMPAR Ca²⁺ permeability.^{27,28,30,94,95} Notably, the human AD temporal cortex we examined was late stage (Supplementary Table 1) and we instead found an increase in GluA2 with a decrease in corresponding GluA1/GluA2 ratio in AD compared with controls (Fig. 1E). These changes may represent a compensatory mechanism to neuronal hyperexcitability, but the significant GluA2 elevation may actually result in further impairments, given that Ca²⁺-permeable AMPAR is critical for the induction of late-phase long-term potentiation, essential for the formation of long-term memory,^96,97 as seen in a neurodevelopmental disorder model.⁹⁸ We also found an increase in NMDAR subunits in AD patients, which is consistent with studies of the AD parietal cortex demonstrating an increased excitatory to inhibitory synaptic ratio.⁹⁹ Elevated GluN2A in AD patients may again represent a compensatory mechanism, as lower GluN2A was associated with worsened CDR-SOB scores across all AD patients in our dataset, consistent with correlations in working memory performance and GluN2A in aged rats.¹⁰⁰ Elevated GluN2B is associated with epileptogenesis^15,30 and extrasynaptic excitotoxic signalling in AD.¹⁰¹ Furthermore, elevations in AMPAR and NMDAR subunits are suggestive of elevated glutamatergic synapses overall, which may further tip E:I balance towards excitation.

While changes in AMPAR and NMDAR were not found in prodromal 5XFAD mice, we found elevated EPSC amplitude and increased intrinsic excitability of CA1 neurons, which are the primary outputs of hippocampus. Together with decreased GABAergic signals, this suggests increased excitability of the hippocampal network overall. At later, symptomatic stages we found increased expression of GluA1/GluA2 AMPAR subunit in PTZ kindled 5XFAD mice which was ameliorated with rapamycin treatment. Overall, these results suggest that AMPAR and NMDAR subunit composition is dynamic through AD progression and certain changes could be compensatory or neuroprotective from excitotoxicity, as often seen in acute epilepsy models.^102,103

Recent epidemiologic research has underscored strong associations between epilepsy, late-onset seizures and AD.^1,104-108 We and others have provided increasing preclinical and clinical evidence that seizures significantly contribute to neuropathology and cognitive decline and may also be a treatable component of this complex disease. Indeed, in a recent phase 2a clinical trial, levetiracetam, an antiepileptic drug, was shown to improve memory and executive function in AD patients with epileptiform activity.¹⁰⁹ Beyond seizure suppression, our data suggest that targeting specific E:I imbalances also holds therapeutic potential. Notably, recent studies have identified neuronal Cl⁻ cotransporters as candidates to treat AD. Our data demonstrated significant exacerbation of NKCC1/KCC2 in both AD+Sz and seizure-kindled 5XFAD mice, although the contributing factor differed between these datasets, with AD+Sz showing significantly decreased KCC2, while seizure-kindled 5XFAD mice had elevated NKCC1. These discrepancies may be due to the brain region examined or disease stage, with our post-mortem tissue being late disease stage (Supplementary Table 1) and symptomatic, 7-month 5XFAD mice being an intermediate stage of neurodegeneration. The therapeutic strategy may depend on which alteration is found: the clinically available agent, bumetanide, antagonizes NKCC1, while CLP290 has been shown to stabilize membrane KCC2 in mouse models.^41,110 Of note, both have been used to restore physiological neuronal activity and rescue cognitive deficits in AD models.^41,111 Furthermore, NKCC1/KCC2 ratios were positively correlated with AD pathology and negatively correlated with performance related to spatial working memory in kindled 5XFAD mice, supporting previous literature linking neuronal excitability and AD progression.¹¹² Overall, these data suggest that restoration of the Cl⁻ gradient may be therapeutically relevant, particularly for AD patients with seizure history.

We recently demonstrated that seizures exacerbate mTORC1 in AD patients and 5XFAD mice and that chronic low-dose rapamycin treatment is sufficient to ameliorate AD pathology and cognitive dysfunction in PTZ-kindled 5XFAD mice.¹⁴ Rapamycin is FDA-approved and has been proven to benefit both AD and epilepsy models.^14,113,114 While rapalogues may affect numerous signalling pathways downstream of mTORC1 to attenuate the disease course, including autophagy and neuroinflammation,^115,116 growing evidence demonstrates efficacy in regulating neuronal hyperexcitability.^55-60 In the studies presented here, rapamycin reduced NKCC1/KCC2 and GluA1/GluA2 ratios in PTZ-kindled 5XFAD mice and showed a trend towards increased GABA_ARα1/GABA_ARα3 in non-kindled 5XFAD mice, suggesting a reversal of inhibitory dysfunction, which may also indicate restoration of the proper circuitry involved in memory rescue in these mice.¹⁴ However, the chronic rapamycin treatment induced small but significant elevations in GluN2B/GluN2A in PTZ-kindled 5XFAD mice and increased GluA1/GluA2 in non-kindled 5XFAD mice. In contrast, studies that used acute rapamycin treatments demonstrated that mTORC1 blockade decreases GluN2B and GluA1.^117-120 Thus, the relative increases that we found in GluN2B and GluA1 in later stage mice may represent a compensatory mechanism in response to chronic rapamycin administration. Indeed, one prior study demonstrated that chronic rapamycin induces increased surface GluN2B, and that these changes are associated with amelioration of age-dependent cognitive decline.¹²¹ These data indicate that targeting E:I balance with rapamycin should be explored for clinical efficacy in AD patients with seizure history.

In summary, the data presented here identify novel mechanisms of neuronal dysfunction in AD that are amplified by seizures and may play a role in worsened cognitive outcomes associated with seizures in AD. Importantly, our data indicate the importance of taking a seizure history at a minimum in patients with AD. In addition, these data suggest that targeting E:I imbalance, perhaps with rapamycin as well as other antiepileptic agents, may hold therapeutic promise in AD patients with comorbid seizures.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgements

The authors thank the CNDR at the University of Pennsylvania for supplying the human tissue used in these studies as well as the University of Maryland Brain and Tissue Bank, part of the NIH Neuro Biobank for additional control brain samples. We would also like to thank Dr Robert Vassar for supplying some of the 4-month 5XFAD mouse tissue.

Funding

These studies were supported by the National Institutes of Health (NIH) National Institute of Neurological Disorders and Stroke (NINDS): R21NS105437 (F.E.J.), R37NS115439 (F.E.J.), R01NS101156 (D.M.T.), National Institute on Aging (NIA): R01AG077692 (F.E.J. and D.M.T.) and T32AG000255 (A.J.B.), Alzheimer’s Association AARF-22-972333 (A.J.B.). Clinical and post-mortem tissue and data from Alzheimer’s disease patients were obtained from National Institutes of Health studies, including P30-AG-072979 (formerly AG010124), U19 AG062418 (formerly P50 NS053488), P01-AG-066597 and R01-AG-054519.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

References

Author notes