Restoring neuronal chloride extrusion reverses cognitive decline linked to Alzheimer’s disease mutations

Abstract

Disinhibition during early stages of Alzheimer's disease is postulated to cause network dysfunction and hyperexcitability leading to cognitive deficits. However, the underlying molecular mechanism remains unknown. Here we show that, in mouse lines carrying Alzheimer's disease-related mutations, a loss of neuronal membrane potassium-chloride cotransporter KCC2, responsible for maintaining the robustness of GABA_A-mediated inhibition, occurs pre-symptomatically in the hippocampus and prefrontal cortex.

KCC2 downregulation was inversely correlated with the age-dependent increase in amyloid-β 42 (Aβ42). Acute administration of Aβ42 caused a downregulation of membrane KCC2. Loss of KCC2 resulted in impaired chloride homeostasis. Preventing the decrease in KCC2 using long term treatment with CLP290 protected against deterioration of learning and cortical hyperactivity. In addition, restoring KCC2, using short term CLP290 treatment, following the transporter reduction effectively reversed spatial memory deficits and social dysfunction, linking chloride dysregulation with Alzheimer's disease-related cognitive decline.

These results reveal KCC2 hypofunction as a viable target for treatment of Alzheimer's disease-related cognitive decline; they confirm target engagement, where the therapeutic intervention takes place, and its effectiveness.

Keramidis et al. report that neuronal chloride balance is impaired in mice carrying Alzheimer's disease-related mutations. Restoring chloride extrusion reversed cognitive decline in the mice, validating a novel target for the treatment of dementia associated with Alzheimer's disease.

Introduction

Aberrant brain activity has been reported in pre-symptomatic Alzheimer's disease patients¹ and in rodents carrying Alzheimer's disease-linked mutations.² This abnormal brain activity appears to arise from neural hyperexcitability due to disrupted GABA_A-mediated inhibitory signaling³ near amyloid-β (Aβ) plaques.⁴ Conversely, overexpression of tau suppresses neuronal activity and counteracts the Aβ-dependent neuronal hyperactivity.^5,6

While hyperexcitability has been associated with cognitive decline,^7–9 loss of inhibition per se, especially through loss of synaptic specificity due to impaired chloride homeostasis, arises as a potential mechanism of memory deficits.¹⁰ The potassium chloride cotransporter KCC2 is responsible for maintaining a low intracellular chloride concentration ([Cl⁻]_i) in neurons by extruding chloride, preserving the robustness of GABA_A signalling.¹¹ Notably, it has been reported that KCC2 levels are reduced in the frontal lobe of patients with sporadic Alzheimer's disease,¹² and that amyloid fibril injection in the rodent hippocampus can downregulate KCC2 expression.¹³ Here we address the hypothesis that KCC2 hypofunction underlies cognitive deficits in mice carrying Alzheimer’s disease-linked mutations that result in Aβ pathology.

Materials and methods

Animal subjects

Experimental protocols were approved by the committee for animal protection of Université Laval (CPAUL) and the University of Lethbridge Animal Care Committee in accordance with the guidelines from the Canadian Council on Animal Care.

Male B6SJL-Tg(APPSwFlLon,PSEN1*M146L*L286V)6799Vas/Mmjax (5xFAD)¹⁴ mice were purchased from the Jackson Laboratory (#34840-JAX) and bred with female B6SJLF1/J mice (The Jackson Laboratory, #100012). These breeders were used to produce both 5xFAD mice and littermate non-transgenic (NonTg) control mice. Two separate 5xFAD colonies were maintained at the CERVO Brain Research Centre and the Canadian Centre for Behavioural Neuroscience (CCBN). Breeding pairs of C57BL/6-App<tm3(NL-G-F)Tcs> knock-in mice (App^{NL-G-F/NL-G-F})¹⁵ were provided by RIKEN Center for Brain Science, Japan (RBRC #06344). The App^{NL-G-F/NL-G-F} colony was maintained at the CCBN. Mice were housed on a 12 h light/12 h dark cycle, and all experiments were performed during the light cycle. Food and water were provided ad libitum. Both sexes were included in each group tested for any experiment conducted. Details on the age, sex and treatment of the mice are provided in Supplementary material.

Drugs and viral constructs

The viral constructs were produced by the Canadian Neurophotonics Platform Viral Vector Core (RRID: SCR_016477). The AAV2/9.CaMKIIa.SuperClomeleon was purified on an iodixanol gradient from cell culture at 1.8 × 10¹³ GC/ml. The AAV2/PHP.N.CaMKIIa.HA-KCC2 (gift from Zhigang He and subcloned by the CNP Viral Vector Core) and the AAV2/PHP.N.CaMKIIa.eGFP were purified at 1.5 × 10¹³ GC/ml.

CLP290 was dissolved in 20% 2-hydroxypropyl-β-cyclodextrin (HPCD) at 10 mg/ml. A fresh solution of CLP290 was prepared daily. The administration was performed per os (P.O.) by gavage 1–4 h prior to each experiment. A 20% HPCD solution was used as vehicle treatment. For the short-term treatment (Groups 1–3), mice received a single dose of CLP290 (100 mg/kg) or vehicle daily for the duration of the behavioural testing. For the long-term CLP290 treatment [Group 4 and local field potential (LFP) recordings group], the mice received CLP290 (100 mg/kg) or vehicle daily starting at 4 months of age and continuing until 9 months of age.

VU0463271 (Sigma-Aldrich SML2333) was solubilized in dimethyl sulfoxide (Sigma-Aldrich 67-68-5) at a concentration of 5 mM. A fresh solution was then prepared in PBS at a concentration of 20 μM before each experiment. For the in vivo electrophysiology, 500 nl of VU0463271 or vehicle (PBS) were injected bilaterally into the hippocampus through implanted cannulae.

Stereotaxic surgeries and viral injections

For all the recovery surgeries, mice were anaesthetized with 1.5–4% isoflurane and placed in a stereotaxic apparatus (Stoetling Co.) on a heating pad.

For the ex vivo chloride imaging experiments, 100 nl of AAV2/9.CaMKIIa.SuperClomeleon was injected into the medial prefrontal cortex (mPFC) at rostro-caudal (R/C) +1.8 mm, medial-lateral (M/L) ±0.3 mm and dorso-ventral (D/V) −2.0 mm from Bregma of 3-month-old 5xFAD and NonTg mice or into the hippocampal CA1 region at R/C −2.5 mm, M/L ±2.0 mm and D/V −1.4 mm of 5-month-old 5xFAD and NonTg mice using a glass capillary and a NANOLITER2020 injector (World Precision Instruments LLC). Imaging started 4 weeks after viral injection.

For the in vivo chloride imaging experiments, 300 nl of AAV2/9.CaMKIIa.SuperClomeleon was injected unilaterally into the mPFC at +2.3 mm R/C, +0.3 mm M/L and −1.5 mm D/V from Bregma of 3-month-old mice, as described above. Imaging started 4 weeks post viral injection.

For the selective KCC2 overexpression experiments, the AAV2/PHP.N.CaMKIIa.eGFP and AAV2/PHP.N.CaMKIIa.HA-KCC2 were injected intravenously via the retro-orbital sinus.¹⁶ Behavioural testing started 9 weeks after viral injections.

For the in vivo LFP recordings, mice were implanted with neocortical, hippocampal and muscular electrodes made from Teflon-coated stainless-steel wire. For neocortical LFP recording, bipolar electrodes (vertical tip separation = 0.6 mm, bare diameter 50.8 µm) were implanted in the left retrosplenial cortex (RSC) at R/C −2.5 mm, M/L 0.7 mm and D/V −1.1 mm, and in the right barrel cortex at R/C −0.1 mm, M/L 3.0 mm and D/V −1.4 mm. For hippocampal LFP recordings, a monopolar electrode (bare diameter 50.8 µm) was implanted in the right hippocampal CA1 region at R/C −2.5 mm, M/L 2.0 mm and D/V −1.1 mm. For electromyography (EMG) recording, a multi-stranded wire (gauge 40) was implanted into the neck musculature. The reference electrode was placed under the skull overlying the cerebellum. The exposed electrode wires were clamped between two receptacle connectors (Mill-Max Mfg. Corp.), and the headpiece and connectors were secured to the skull using C&B Metabond (Parkell Inc.). After the final recording, electrode tip positions were marked by passing current through the electrodes (0.1 mA for 5 s) and tip placement was verified by histology.

For the VU0463271 experiments, a 24-gauge hypodermic needle attached to a 30-gauge stainless steel wire was implanted bilaterally as a guide cannula in the hippocampus at R/C −2.5 mm, M/L 0.45 mm and D/V 1.2 mm. Cannulae were implanted bilaterally and secured to the skull using C&B Metabond. Bipolar electrodes were also implanted bilaterally at R/C −2.0 mm, M/L ±1.5 mm and D/V −1.2 mm.

Preparation of brain slices

Mice were anaesthetized with 30% urethane in saline, the brain was excised and immersed briefly in ice-cold cutting saline containing: 204.68 mM sucrose, 2.5 mM KCl, 1 mM CaCl₂, 2 mM MgCl₂, 1.25 mM NaH₂PO₄, 25 mM NaHCO₃, 25 mM glucose, 0.4 mM Na-ascorbate and 3 mM Na-pyruvate. Coronal slices (300 μm) were obtained with a Leica vibratome. Slices were recovered for 30 min at 32°C in an immersion chamber in an oxygenated 1:1 mix of cutting saline and artificial CSF (126 mM NaCl, 5 mM KCl, 2 mM CaCl₂, 2 mM MgCl₂, 10 mM Glucose, 26 mM NaHCO₃ and 1.25 mM NaH₂PO₄; pH 7.35 and 295 mOsm), followed by 30 min recovery at room temperature in oxygenated artificial CSF.

For the CLP290 experiments, slices were obtained 3 h post CLP290 or vehicle administration. Slices were recovered for 15 min at 32°C in an oxygenated 1:1 mix of cutting saline and artificial CSF followed by 15 min recovery at room temperature in oxygenated artificial CSF.

Ex vivo chloride imaging

Coronal brain slices expressing SuperClomeleon¹⁷ were transferred to a perfusion chamber and perfused with artificial SCF (2 ml/min) at room temperature. Fluorescence lifetime imaging microscopy (FLIM) was performed as previously described¹⁸ using two optical set-ups, a Zeiss 880 laser-scanning microscope and a polygonal mirror video-rate microscopy system (VMS; Bliq Photonics) coupled with an 80 MHz femtosecond-pulsed Ti-Sapphire laser set at 800 nm. This excitation wavelength minimized the FRET acceptor excitation. For the Zeiss microscope, FLIM images (1 per 10 s) were acquired with a 20× Plan-Apochromat water immersion objective (Zeiss, 1.0 NA), and FLIM images were generated onto a TCSPC module PMC-100-1 detector (Becker & Hickl GmbH). For the VMS, images were acquired with a 16× objective (Nikon, 0.8 NA) and FLIM images (31 Hz) were generated onto a SPC-150 N TCSPC module with a HPM-100-40 detector (Becker & Hickl GmbH). The VMS allowed concurrent intensity detection [Chroma ET473/24 m (donor channel) and Chroma ET530/30x (green channel)]. All FLIM acquisitions were achieved using a bandpass filter (Chroma, ET473/24 m) to collect only the donor emission.

Fluorescence lifetime values of CaMKIIa positive neurons were acquired under three conditions: (i) artificial CSF with low extracellular KCl (5 mM); (ii) artificial CSF with low extracellular KCl (5 mM) and 5 μM bicuculline, 10 μM CNQX and 50 μM APV; and (iii) artificial CSF with high extracellular KCl (15 mM) with bicuculline, CNQX and APV. For the CLP290 experiments, fluorescence lifetime values were acquired under the same conditions within 5 h following the CLP290 or vehicle administration. Each FLIM image corresponds to 10 s acquisitions. The experiment was divided into the three conditions described above. We acquired 2.5 min using the artificial CSF with low extracellular KCl (5 mM); then either 15 min or 10 min of acquisition, for the mPFC and the CA1 experiments, respectively, using artificial CSF with low extracellular KCl (5 mM) including blockers; and, finally, 15 min of acquisition using the artificial CSF with high extracellular KCl (15 mM) including blockers.

FLIM analysis

For each acquisition, the user delineated cell bodies of neurons in individual regions of interest (ROIs). The instrument response function (IRF) was acquired by using an 80 nm gold nanoparticle to generate a second-harmonic signal. Using a custom script, the FLIM histogram was generated by pooling all the pixels in the ROIs for each time point. Every time point (10 s) of the experiment was fitted using a mono-exponential decay, and the fluorescence lifetime (τ) was estimated.

For each neuron analysed, a Hill-Slope was fitted into the temporal evolution of lifetimes to estimate the initial and final lifetime values (for 5 and 15 mM KCl artificial CSF, respectively). The maximum slope of the lifetime (units of ns per min) was estimated upon 15 mM KCl application using a plateau followed by an exponential plateau function. Neurons that did not have enough photons in the FLIM acquisition (300 photons), a stable lifetime during the 5 min before the potassium increase, or did not respond, meaning no significant increase of lifetime was observed after the potassium increase, were excluded from the analysis (32 and 93 NonTg neurons and 41 and 62 5xFAD neurons were excluded from the mPFC and the CA1 analysis, respectively) (Supplementary Fig. 1).

In vivo chloride imaging

Mice were anaesthetized intraperitoneally with a mix of 100 mg ketamine, 15 mg xylazine and 2.5 mg acepromazine per kg, and placed in a head fixing microscope platform (Stoelting Co). The scalp was removed, and a chamber was formed using dental cement. The skull was thinned above the PFC using a dental drill, and the remaining bone was removed to open a cranial window over the imaged region. The chamber was filled with HEPES-buffered artificial CSF (126 mM NaCl, 2.5 mM KCl, 2 mM CaCl₂, 2 mM MgCl₂, 10 mM D-glucose and 10 mM HEPES; pH 7.0; 290 mOsm), and the dura was carefully removed. Following the surgery, the mice were transferred under a multiphoton VMS (Bliq Photonics). The Ti:sapphire laser was set to 840 nm to excite the FRET donor of SuperClomeleon¹⁷ (Supplementary Fig. 2A). The emitted signals were separated by a long-pass dichroic mirror (Semrock, FF509-Di01). Additional band-pass emission filters were used for Cerulean (CRF; Chroma, ET480/40 m) and YFP (Chroma, ET540/40 m). A 40× NIR Apo water immersion objective (Nikon, 0.8 NA) was used for imaging. Panoramic images of 1000 × 500 pixels at a rate of 32 Hz were acquired up to 200 μm in depth from the visible surface and with 0.204 μm pixel size. For each field of view, an image stack of 5 s acquisition (∼160 frames) was recorded.

In vivo ratiometric chloride analysis

Acquired image stacks were registered with a custom MATLAB script as previously described.¹⁹ Both the YFP and CRF channels were used to correct for movement. The out-of-focus frames were automatically deleted during registration. An average projection of the registered image stack was generated as the single field representative image. Rectangular ROIs were assigned manually in the cytoplasm of individual neurons. A background ROI was defined in a region without any visible neurons. The average ROI intensity was measured for YFP and CRF separately. The ratiometric chloride concentration of individual neurons, R, was calculated using the formula R = (YFP − YFP_b) / (CRF − CRF_b), where YFP and CRF are the intensities for each neuron, and YFP_b and CRF_b the background intensities from the same field. The neurons with low signal-to-noise ratio (YFP/YFP_b < 5 or CRF/CRF_b < 5) were excluded to ensure data consistency (195 and 242 neurons were excluded for the 5xFAD and the NonTg groups, respectively). The dependency of the calculated Rs on the laser intensity was validated by imaging the same ROIs in a field at varying laser powers (Supplementary Fig. 2B). The R was independent of the laser intensity at a given imaging depth. Furthermore, the depth distribution of the imaging fields for each mouse did not vary (Supplementary Fig. 2C).

In vivo local field potential recordings

The set-up for chronic LFP recording in freely-moving mice consisted of a large plexiglass box, into which the lidless home cage was placed. A tethered recording probe connected to a commutator (NeuroTek, Inc.) was inserted into the implanted connector. After at least 1 week of post-surgery recovery, mice were habituated to the recording setup and probe for 5 days, with the habituation time gradually increasing from 1 to 3 h per day. After, LFP/EMG activity was recorded on three consecutive days for 3 h per day, at approximately the same time each day, during the light phase. Recordings were conducted longitudinally at 6 and 9 months of age in mice that received long-term daily treatment with CLP290 or vehicle starting from 4 months of age. For the habituation/recording days, drug administration was performed immediately before the start of habituation/recording.

LFP/EMG activity was amplified, filtered (0.1–4000 Hz) and digitized at 16 kHz using a Digital Lynx SX Electrophysiology System (Neuralynx, Inc.). Data were recorded and stored on a local computer using Cheetah software (Neuralynx, Inc.). Analysis was performed offline. Sleep scoring was performed in 6-s intervals to define periods of wakefulness, REM sleep, and non-REM (NREM) sleep. For sleep scoring, the raw EMG activity was filtered (90–1000 Hz), rectified and integrated using a 4-s moving window, and thresholded to detect periods of mobility. Power in the slow wave/delta band (0.5–4 Hz) was calculated for the RSC LFP signal and thresholded using values between 0.04–0.1 mV²/Hz to detect slow-wave activity. NREM sleep was scored when the mouse was immobile and exhibited slow-wave activity. For detecting REM sleep, the ratio of theta power (6–10 Hz) to total power of the hippocampal LFP signal was calculated. When this ratio was above 0.4–0.6 and the mouse was immobile, REM sleep was scored. All remaining periods were considered wakefulness. After sleep scoring, spectral power of the neocortical and hippocampal LFP signals during NREM sleep was calculated for different frequency bands (30–60, 60–120 and 120–200 Hz) using Welch's method. Sleep structure (i.e. the percentage of total recording time spent in each state) and spectral power were averaged across the three recording days for each mouse.

Immunohistochemistry

Mice were anaesthetized with 30% urethane in saline and perfused intracardially with 0.1 M PBS followed by 4% paraformaldehyde in PBS. Brains were excised, stored in the same fixative for 15 h and then cryoprotected in 30% sucrose in PBS. Coronal slices of the mPFC and the hippocampus were cut at 40 μm on a Leica vibratome and permeabilized in PBS with 0.2% Triton X-100 (PBST). Sections were incubated for 15 h with the primary antibody (anti-KCC2, 1:1000, Millipore 07–432) diluted in PBST containing 4% normal goat serum. After washing in PBST, slices were incubated for 2 h with the secondary antibody (Cy3-conjugated goat anti-rabbit, 1:500, Jackson Labs 111-165-003). Finally, slices were rinsed in PBS, mounted on slides (Superfrost^TM Plus Microscope Slides, Thermo Fisher Scientific) and cover-slipped using an aqueous mounting media (Abcam ab104139). Negative control staining was performed by incubating sections with only the secondary antibody following the same protocol (Supplementary Fig. 3A).

Confocal microscopy and image acquisition

Confocal images were acquired with a Zeiss LSM710 laser-scanning microscope using an MBS 405/488/555/639 beam splitter or a Zeiss LSM880 laser-scanning microscope using an MBS 488/555/639 beam splitter and the same oil immersion objective (Plan-Apochromat Oil 63x, 1.4 NA). Each channel was imaged sequentially to avoid crosstalk, and unique emission filters were used for each fluorophore. The laser power and the Photo-Multiplier Tube voltage were set to optimal values to limit pixel saturation and avoid photobleaching. The laser power, polarization voltage, filters, dichroic mirrors and scan speed were kept constant for each unique dataset and between controls and comparable samples. Images were acquired at 12 bits and 2048 × 2048 pixels, with 0.103 μm pixel size.

Global KCC2 intensity and KCC2 subcellular profile analysis

Fluorescence confocal images of the mPFC and CA1 were acquired. To measure the global KCC2 expression, we quantified the mean KCC2 fluorescence intensity within each image. The intensity of the immunostaining background was defined for several ROIs where KCC2 is known to be absent (i.e. in the forceps minor of the corpus callosum) (Supplementary Fig. 3B) and subtracted from the KCC2 mean intensity values of each image as previously described.^20,21 A custom MATLAB algorithm, coined MASC-π,²¹ was utilized to perform the membrane analysis of subcellular intensity, as previously described.²² Briefly, for each confocal image, the membrane of randomly and blindly selected neurons (for which a continuous membrane across the entire cell body circumference was identified) was manually delineated. For each pixel in the ROIs defining individual neurons, the distance to the closest membrane segment was calculated to generate a distance map with the neuronal membrane defined as zero. The mean intensity per pixel and the standard deviation of the KCC2 fluorescence was computed as a function of the distance to the membrane. This approach provides an unbiased estimate of the membrane expression and excludes any potential effect of neuronal loss in the protein level measurements within the brain ROIs.

Morris water task

The test was performed in a circular tank (diameter: 154 cm; depth: 50 cm), with three distinct visual cues placed outside the tank to facilitate spatial navigation. The tank was filled with water (22 ± 1°C) to a height of 40 cm. Non-toxic white tempera paint was added to the water to make it opaque. A hidden escape platform (diameter: 11 cm) was situated 1 cm below the water surface in one of the tank quadrants. The platform's location remained fixed throughout training. For acquisition training, mice performed four trials per day for eight consecutive days. On a given day, mice were released into the pool from each of the four cardinal directions, with the order of the start locations pseudorandomly determined. A trial was terminated when the mouse located the hidden platform, after which the mouse was left on the platform for 10 s before being removed from the tank. If a mouse did not locate the platform within 60 s, it was guided to the platform. Twenty-four hours after the completion of acquisition training, a probe test was conducted. This consisted of a single 60 s trial with the platform absent from the tank. Tracking software (Water 2100, HVS Image) was used to quantify the following measures for each acquisition training day: average latency to find the platform, average path distance to the platform, and average swim speed. For the probe test, the average proximity to the platform's previous location was measured.

Y-maze spatial memory test

The Y-maze apparatus consisted of three identical arms (each 36.2 × 8.25 cm) in a Y configuration (placed at 120° to one another) connected by a centre polygonal area. The walls of the arms were transparent so the animals could identify two distinct visual cues placed outside the apparatus to facilitate spatial navigation. During the training phase, the mice were placed individually in the maze and left freely to explore two arms, for 10 min, while the third arm was blocked by a removable door. An hour later, the mice were returned to the maze and allowed to explore all arms. The latency to enter the new (third) arm was measured for each mouse and used to score spatial memory performance. Entry into the arm was defined as a mouse placing all four paws on the arm.

Unconditioned social preference

The unconditioned social preference test used a three-chambered apparatus, with two side compartments (35 × 20 cm) with differently coloured walls (black or white), separated by a smaller middle compartment with large horizontal white and black gratings on its walls. Mice were habituated to the apparatus containing an inverted metallic cup in each large compartment for 10 min. After, a sex- and age-matched conspecific was placed in one of the two inverted cups and an object in the other. The test mouse was then transferred to the apparatus and left to explore the compartments for an additional 10 min. Mice were assigned to a random testing order. Movement was recorded by a camera located on the top of the apparatus and analysed offline by ANY-maze (Stoetling Co.). Behavioural analysis was performed blind to the treatment. The times in the social and neutral chambers were calculated as the time spent by each mouse in the social-partner-containing and the object-containing (neutral) compartments, respectively.

Statistical analysis

The statistical analysis was performed with GraphPad Prism 9 (GraphPad Software, San Diego, CA). The two-group null-hypothesis significance was tested with Student's t-test, and the normality of the samples’ distributions was tested with an Anderson-Darling or a Shapiro-Wilk test. One-way ANOVA was used to analyse the difference of independent groups, followed by a Fisher's LSD test. Multi-group designs with two categorical variables were analysed by two-way ANOVA with a post hoc Tukey test for multiple comparisons. A repeated-measures ANOVA with a Dunnett's post hoc comparison was utilized to compare values obtained from the same subjects at different time points. For the datasets not showing a normal distribution, differences were tested with a Mann-Whitney test. Cumulative distributions were compared by a Kolmogorov-Smirnov test. Data are reported as mean ± standard error of the mean (SEM) with N (or n) indicating the number of mice, unless otherwise specified. P-values < 0.05 were considered statistically significant.

Results

KCC2 is downregulated in the medial prefrontal cortex and CA1 of 5xFAD and App^NL-G-F mice

In brains from 2, 4 and 6-month-old 5xFAD mice and their age-matched NonTg littermates, we measured via immunohistochemistry global and membrane KCC2 expression in both the mPFC layer 2/3 and the hippocampal CA1 pyramidal layer (Fig. 1A and B). We found that KCC2 global labelling intensity was reduced in the mPFC at all ages (Fig. 1C), while membrane KCC2 was significantly reduced in 4- and 6-month-old, but not 2-month-old, 5xFAD mice compared to NonTg mice (Fig. 1D). In the CA1, both global and membrane KCC2 levels were reduced at 6 months but not 4 months in 5xFAD mice (Fig. 1E and F). Comparable downregulation of KCC2 was observed in 9-month-old App^{NL-G-F/NL-G-F} mice for both the mPFC and the CA1 (Fig. 1G and H). In contrast, when we measured global and membrane KCC2 levels in the nucleus accumbens, a region with lower Αβ plaque load compared to the neocortex and hippocampus, at 6 months of age,²³ there was no KCC2 reduction in the 5xFAD mice as compared to NonTg mice (Supplementary Fig. 4). In addition, we quantified the mean area of the individual neurons we analysed, and we did not find any difference in neuronal surface area, which could have accounted for the decrease in membrane KCC2 expression (Supplementary Fig. 5).

Figure 1.

Changes in global and membrane KCC2 expression in the mPFC and hippocampal CA1 regions of 5xFAD and App^NL-G-F mice. (A and B) Confocal images showing KCC2 immunostaining in medial prefrontal cortex (mPFC) LII/III (A) and the CA1 (B) of 5xFAD versus non-transgenic (NonTg) mice (KCC2 in green, DAPI in blue). Scale bar = 50 μm. Insets show KCC2-expressing neurons from ROIs. Scale bar = 10 μm. On the right, examples of KCC2 subcellular profile intensities. The colour-coded maps illustrate the distance from the membrane. The graphs represent the KCC2 intensity as a function of the distance to the membrane (scale bars = vertical, 1000 intensity units; horizontal, 1 μm). (C and E) Global index KCC2 analysis of the mean KCC2 pixel intensity in mPFC LII/III of 2, 4 and 6-month-old 5xFAD versus NonTg mice (C) and the CA1 of 4 and 6-month-old 5xFAD versus NonTg mice (E). (D) Mean KCC2 intensity profiles across the plasma membrane of individually identified 5xFAD (blue) and NonTg (grey) neurons in the mPFC (5xFAD: n₂ = 1163, n₄ = 1537, n₆ = 912 from N₂ = 12, N₄ = 15, N₆ = 10 mice, respectively; NonTg: n₂ = 932, n₄ = 1164, n₆ = 974 neurons from N₂ = 10, N₄ = 11, N₆ = 13 mice, respectively). (F) Similar to D but in the CA1 (5xFAD: n₄ = 625, n₆ = 608 from N₄ = 10 and N₆ = 10 mice, respectively; NonTg: n₄ = 360, n₆ = 523 from N₄ = 8 and N₆ = 7 mice, respectively). (G) Pixel intensity of KCC2 immunostaining in 9-month-old App^{NL-G-F/NL-G-F} versus App^w/w mice in the mPFC (left) and the CA1 (right). (H) Average KCC2 intensity profiles in App^{NL-G-F/NL-G-F} (cyan) versus App^w/w (orange) mice in the mPFC (left) and CA1 (right) (App^{NL-G-F/NL-G-F}: n_mPFC = 742 and n_CA1 = 746 neurons from N_mPFC = 10 and N_CA1 = 10 mice, respectively; App^w/w: n_mPFC = 880, n_CA1 = 594 neurons from N_mPFC = 9 and N_CA1 = 7 mice, respectively). (I) Scheme of the KCC2 membrane analysis of subcellular profile intensity (MASC-π). Circles in C, E and G represent single mice. Data are presented as mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001. ns = non-significant.

We also measured the levels of KCC2 protein and of soluble amyloid-β 42 (Aβ42) in frontal lobe lysates from 2, 4 and 6-month-old 5xFAD and NonTg mice. We found that the levels of soluble Αβ42 increased while KCC2 levels decreased with age in the 5xFAD mice, but not in NonTg mice (Fig. 2A and B). At 6 months of age, KCC2 levels were inversely correlated with levels of Aβ42 (Fig. 2C). In contrast, the protein levels of the sodium-potassium-chloride cotransporter NKCC1 were unaltered in 6-month-old 5xFAD mice as compared to NonTg mice (Supplementary Fig. 6). Finally, we found a significant reduction in membrane KCC2 levels in neurons incubated with 10 μM Aβ42 as compared to neurons incubated with scrambled peptide (Fig. 2D–G), indicating that KCC2 downregulation is downstream of the amyloid pathology.

Figure 2.

Aβ42 reduces membrane KCC2 in cultured neurons, and its levels are negatively correlated with KCC2 levels in 5xFAD mice. (A) The levels of soluble Aβ42 and Αβ40 in the frontal lobe of 2, 4 and 6-month-old 5xFAD (blue) and non-transgenic (NonTg) (grey) mice. The dashed lines represent the fitted linear regression equations of the Αβ42 or Αβ40 levels versus the age of the mice. (B) Total KCC2 protein levels, quantified with an ELISA, in the frontal lobe of 2, 4 and 6-month-old 5xFAD (blue) and NonTg (grey) mice. The dashed lines represent the fitted linear regression equations of the KCC2 levels versus age. (C) Pearson r correlation of Aβ42 and KCC2 levels in the frontal lobe of 6-month-old mice (dots represent individual mice). (D) Representative confocal images showing KCC2 (grey) and NeuN (red) immunostaining in cultured primary hippocampal neurons incubated with Aβ42 or scramble peptide (Scramble; scale bars = 10 μm). (E) Mean membrane KCC2 intensity of individual neurons (Scramble - CLP257: n = 27 neurons from two coverslips; Scramble + CLP257: n = 25 neurons from two coverslips; Αβ42 - CLP257: n = 56 neurons from four coverslips; Αβ42 + CLP257: n = 50 neurons from four coverslips). (F) Mean KCC2 intensity profiles across the plasma membrane of individually identified neurons. (G) The average area of neurons analysed for membrane KCC2 expression. The R² and P-values of the linear regressions are shown above the graphs. Whiskers in box plots show the 5–95 percentiles. Data in E, F and G are presented as mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. ns = non-significant.

Impaired neuronal chloride extrusion capacity in 5xFAD mice

To test for accompanying deficits in neuronal chloride extrusion capacity, we virally transduced the Cl⁻ sensor SuperClomeleon¹⁷ under the CaMKIIa promoter in 5xFAD and NonTg mice, targeting the mPFC and CA1 in 4-month-old and 6-month-old mice, respectively (Fig. 3A and D and Supplementary Fig. 1A and D). In brain slices taken from these mice we performed time-resolved FRET measurements of Cl⁻ while we stepped extracellular K⁺ ([K⁺]_e) from 5 to 15 mM (Fig. 3B and E and Supplementary Fig. 1C and F) to reverse the KCC2 transporter. The rate of Cl⁻ accumulation resulting from the raised [K⁺]_e was significantly slower in the mPFC of 4-month-old 5xFAD mice compared to NonTg mice (Fig. 3C) and at 6 months in the CA1 (Fig. 3F), reflecting weaker Cl⁻ transport capacity in the 5xFAD mice. We did not find a difference in neuron diameters following the increase in [K⁺]_e (Supplementary Fig. 7), indicating that the observed [Cl⁻]_i changes were not secondary to a change in cell volume. Also, to ensure the difference measured was not due to an effect of the slice preparation, we performed SuperClomeleon-based ratiometric measurements of [Cl⁻]_iin vivo (Fig. 3G and H) in the mPFC of NonTg and 5xFAD mice (Supplementary Fig. 2A). We found that the median FRET ratio of pyramidal neurons from 5xFAD mice was significantly lower as compared to NonTg neurons, indicating higher [Cl⁻]_i (Fig. 3I), while the estimated imaging depth did not differ between 5xFAD and NonTg mice (Fig. 3I).

Figure 3.

Impaired chloride transport in 5xFAD mice. (A and D) FLIM images of 5xFAD and NonTg CaMKIIa-positive neurons expressing SuperClomeleon in mPFC (A) and hippocampal (D) slices. Scale bars = 10 μm. (B) Timelapse recording of Cl⁻ accumulation in the cell body of selected NonTg (grey) and 5xFAD (blue) neurons from the ROI in A upon 15 mM KCl extracellular application (dashed line). Insets show examples of fitted photon-distribution histograms during the low KCl condition (left) and upon 15 mM KCl application (right). Scale bars = vertical, 50 photons; horizontal, 5 ns. (C) The mean lifetime of NonTg and 5xFAD neurons during the baseline condition (5 mM KCl) and upon 15 mM KCl application. The distribution of the mean rate of Cl⁻ changes measured in 4-month-old 5xFAD and NonTg mPFC slices. The inset presents the mean slope (±standard deviation, SD) measured in individual neurons from 5xFAD and NonTg mice. Scale bar = 0.1 ns/min. (5xFAD: n = 73 neurons from N = 10 mice; NonTg: n = 46 neurons from N = 7 mice). (E) Similar to B but for NonTg and 5xFAD CA1 neurons within ROIs shown in D. (F) Similar to C but for CA1 neurons from 6-month-old 5xFAD and NonTg mice. Scale bar = 0.1 ns/min. (5xFAD: n = 133 neurons from N = 6 mice; NonTg: n = 122 neurons from N = 5 mice). (G) Graphical illustration of the SuperClomeleon in vivo steady state imaging experimental set-up, and representative images of the mean intensity of YFP and Cerulean merged (left) and split for the areas selected in dashed squares (right) in the prefrontal cortex of 4-month-old 5xFAD and NonTg mice. Scale bars = 20 μm and 10 μm for the selected areas. (H) Images of the sum of the CFP and YFP fluorescence intensity of the selected areas in G and intensity profiles of CFP (blue) and YFP (green) for the lines in cyan. (I) The median FRET ratio of 5xFAD versus NonTg neurons (left; 5xFAD: n = 361 neurons from N = 5 mice; NonTg: n = 638 neurons from N = 5 mice) and the estimated mean depth of imaging (right; mean ± SEM). *P < 0.05; ***P < 0.001. ns = non-significant.

Enhancing KCC2 improves memory retention and social behaviour in 5xFAD mice

Previous reports identified the CLP257 family of drugs, including CLP290, as KCC2 enhancers,²⁴ which effectively restore KCC2 expression in various pathological conditions.^20,25–27 To validate whether CLP290 can enhance membrane KCC2 levels in 6-month-old 5xFAD mice, we analysed KCC2 expression in the mPFC (Fig. 4A). Short-term daily treatment with CLP290 (100 mg/kg) significantly increased global and membrane KCC2 in 5xFAD mice as compared to vehicle-treated mice (Fig. 4B). Acute treatment of hippocampal neurons with 1 μM CLP257 also restored membrane KCC2 after its reduction by Aβ42 (Fig. 2D–F). Moreover, CLP290 treatment restored the rate of Cl⁻ accumulation resulting from the rise in [K⁺]_e in CA1 neurons from 5xFAD mice (Fig. 4C–E).

Figure 4.

CLP290 enhances membrane KCC2 and restores chloride transport in 5xFAD mice. (A) Confocal images of medial prefrontal cortex (mPFC) slices stained for KCC2 from CLP290 versus vehicle treated 5xFAD mice (KCC2 in green, DAPI in blue). Scale bar = 50 μm. Insets show KCC2-stained neurons from ROIs. Scale bar = 10 μm. Colour-coded maps illustrate the distance from the membrane and graphs represent the subcellular KCC2 intensity profile. Scale bars = vertical, 1000 intensity units; horizontal, 2 μm. (B) On the left, the mean global index KCC2 pixel intensity. In the centre, the mean KCC2 intensity profiles across the plasma membrane of CLP290-treated (filled dark blue dots; n = 919 neurons from N = 9 mice) and vehicle-treated (blue circles; n = 985 neurons from N = 10 mice) 5xFAD mice. On the right, the mean area of neurons selected for the analysis of membrane KCC2 expression. Data are presented as mean ± SEM. (C) FLIM images of CaMKII-positive neurons expressing SuperClomeleon in the CA1 of vehicle- and CLP290-treated 5xFAD mice. Scale bars = 25 μm. (D) Timelapse recording of Cl⁻ accumulation in the cell body of selected vehicle-treated 5xFAD (blue circles) and CLP290-treated 5xFAD (dark blue dots) neurons from the ROI in C upon 15 mM KCl extracellular application (dashed line). Insets show examples of fitted photon-distribution histograms during the low KCl condition (left) and upon 15 mM KCl application (right). Scale bars = vertical, 100 photons; horizontal, 4.8 ns. (E) Distribution of the average rate of Cl⁻ changes (slope) measured in the CA1 of 6-month-old CLP290-treated 5xFAD mice and vehicle-treated 5xFAD and NonTg mice. Inset displays the mean slope (±SD) measured in individual neurons from CLP290-treated 5xFAD mice and vehicle-treated 5xFAD and NonTg mice. Scale bar = 0.1 ns/min. (NonTg + Veh: n = 1028 neurons from N = 6 mice; 5xFAD + Veh: n = 639 neurons from N = 4 mice; 5xFAD + CLP290: n = 675 neurons from N = 6 mice). *P < 0.05; **P < 0.01; ****P < 0.0001. ns = non-significant.

Having verified that CLP290 can restore KCC2 levels and Cl⁻ transport in 5xFAD mice, we next tested whether CLP290 also rescues behavioural function in these mice. Vehicle and CLP290-treated 5xFAD mice were tested for memory, anxiety-like behaviour and social behaviour (Fig. 5A). Vehicle-treated NonTg mice were also tested, to verify that 5xFAD mice are impaired in the tests used. In the Morris water task, a test of spatial learning and memory, all groups showed a comparable learning curve during the acquisition phase, with performance improving across days (Fig. 5B). Averaging across all days, the NonTg mice performed significantly better than both 5xFAD groups, and there was no significant difference between the vehicle and CLP290-treated 5xFAD mice. The swim speed during the acquisition phase was not significantly different between the three groups (Supplementary Fig. 8A). On the following probe test, the vehicle-treated 5xFAD mice performed significantly worse than the NonTg mice, searching further away from the platform's previous location, indicating a spatial memory impairment, and CLP290 treatment significantly reduced this spatial memory deficit (Fig. 5C and D).

Figure 5.

CLP290 augments spatial memory and social behaviour in 5xFAD mice. (A) The chronological order of behavioural testing for the three CLP290 and vehicle-treated experimental groups. (B) Learning curves measured by the path length during the acquisition phase of the Morris water task in non-transgenic (NonTg) vehicle-treated mice and CLP290- and vehicle-treated 5xFAD mice. (C) The average proximity to the platform's previous location during the Morris water task probe trial. (D) Heat maps indicating the average time spent in each spatial bin (bin size = 1 cm²) during the probe trial of the Morris water task. The black circle in the top right quadrant represents the platform's position during acquisition training. (E) The latency to explore the novel arm during the Y-maze spatial memory test. (F) The time spent in the social versus neutral chamber during the unconditioned social interaction test. (G) Heat maps representing the location of the animal's head during the social interaction test. (H) The swim path length during the acquisition phase of the Morris water task (left), and the average proximity to the platform's previous position during the probe trial test (right) of 9-month-old 5xFAD mice receiving long-term (100 mg/kg per day for 5 months) treatment with CLP290 versus vehicle. Inset shows the learning curve for the four first days of the acquisition phase (box-whisker plots with the 5–95 percentile; scale bars = vertical, 2 m; horizontal, 1 day). Circles or dots in C, E, F and H represent single mice. Data are presented as mean ± SEM. *P < 0.05; **P < 0.01. ns = non-significant.

In the Y-maze spatial memory test, the vehicle-treated 5xFAD mice performed significantly worse than the NonTg mice, showing higher latency to explore the novel arm of the Y-maze, indicating a spatial memory deficit, which was rescued by CLP290 treatment (Fig. 5E). In tests of cued and contextual fear memory, we also observed a significant impairment in the 5xFAD mice, with the 5xFAD mice exhibiting less freezing than the NonTg mice, suggesting worse memory (Supplementary Fig. 8B). However, unlike the spatial memory impairments in the Morris water task and the Y-maze, this fear memory impairment was not improved by CLP290 treatment.

In the elevated plus-maze (EPM) test of anxiety-like behaviour, the ratio of time spent in the open arms to the closed arms was significantly lower in the NonTg mice than the vehicle-treated 5xFAD mice but not the CLP290-treated 5xFAD mice, although the vehicle- and CLP290-treated mice did not differ significantly (Supplementary Fig. 8C). The total number of arm entries was comparable between all the groups (Supplementary Fig. 8D).

In the three-chamber unconditioned social preference task, NonTg mice showed a preference for the social chamber over a neutral chamber, but the vehicle-treated 5xFAD mice did not (Fig. 5F and G). CLP290 treatment restored the social preference in the 5xFAD mice (Fig. 5F and G). As early as 4 months of age, the 5xFAD mice displayed social preference deficits in this behavioural paradigm, as compared to age-matched NonTg mice (Supplementary Fig. 8E). When we measured the maximum speed and the distance travelled during the social preference test, there were no significant differences between the three groups, suggesting that there were no motor deficits in 6-month-old 5xFAD mice (Supplementary Fig. 8F). CLP290 did not alter social preference in NonTg mice (Supplementary Fig. 8G). In conjunction with our data showing a KCC2 deficit already at 4 months of age in the mPFC of 5xFAD mice (Figs 1D and 3C), these results suggest that social preference deficits may be an early behavioural sign of Cl⁻ transport dysregulation in the mPFC. In contrast, the reported lack of spatial memory deficits in 5xFAD mice at 4 months of age²⁸ is consistent with our finding that KCC2 levels are normal in the hippocampus at this age in 5xFAD mice (Fig. 1F).

In turn, we tested whether selective overexpression of KCC2 in CNS neurons had similar effects to CLP290 on social behaviour of 5xFAD mice. An AAV2/PHP.N vector expressing haemagglutinin (HA)-tagged KCC2 (HA-KCC2) under the CaMKIIa promoter was used. Injecting this virus into the retro-orbital sinus of 9-month-old 5xFAD mice, allowing for brain-wide expression (Supplementary Fig. 9A and B), enhanced KCC2 levels 9 weeks after the injection (Supplementary Fig. 9C and D). The 5xFAD mice overexpressing HA-KCC2 exhibited significant social preference, while control 5xFAD mice similarly overexpressing GFP did not (Supplementary Fig. 9E).

Restricting KCC2 downregulation prevents cortical hyperactivity and spatial learning deficits in 5xFAD mice

Finally, we asked whether long-term treatment with CLP290 is sufficient to prevent KCC2 downregulation, as well as cortical hyperactivity and spatial learning and memory deficits measured late in the progression of the disease. 5xFAD mice were treated daily with CLP290 or vehicle, beginning at 4 months of age. At 6 months, this treatment had effectively protected against KCC2 downregulation in the mPFC (Supplementary Fig. 10) but did not reduce the number or size of plaques nor the levels of soluble Αβ (Supplementary Fig. 11). In another experiment, mice were treated daily for 5 months, and their spatial learning and memory was assessed at 9 months of age in the Morris water task. The swim speed during the acquisition phase was not significantly different between the CLP290- and vehicle-treated groups (Supplementary Fig. 8H). Even at this late stage of disease, the 5xFAD mice were able to acquire the task, with performance improving significantly from the first to the final day in both groups (Fig. 5H). The performance of the two groups converged by Day 5 of acquisition, with both groups maintaining a stable plateau thereafter. Thus, we focused our analysis on the early phase of acquisition (Days 1–4). Averaging across days, spatial learning was significantly better in the CLP290-treated mice relative to the vehicle-treated mice, in terms of both path length and latency (Fig. 5H and Supplementary Fig. 8H). Examining within groups, performance in the CLP290 group showed a significant improvement by Day 2 of acquisition, whereas the vehicle group showed no significant improvement even by Day 4. On the probe test, there was no significant difference between groups, consistent with the convergent performance of the two groups across the late phase of training (Fig. 5H). In summary, even at an advanced stage of disease progression, long-term CLP290 treatment was effective at preventing spatial learning impairments.

Another group of mice was also treated daily with CLP290 or vehicle for 5 months, as before. Spontaneous LFP activity in the neocortex and hippocampus was recorded longitudinally at 6 and 9 months while the mice were freely moving in their home cages (Fig. 6A and B). At 6 months, there was no significant difference between the vehicle- and CLP290-treated mice in the power of gamma (30–120 Hz) or high-frequency (120–200 Hz) oscillations in the RSC during NREM sleep. However, by 9 months of age, gamma and high-frequency power were significantly greater in the vehicle-treated mice than the CLP290 treated mice (Fig. 6C). Similar results were observed in the barrel cortex, though the effects were less pronounced and did not reach statistical significance at 9 months (Supplementary Fig. 12). The greater effect in the RSC could be due to this region's status as a hub of the brain's default mode network.²⁹ This network exhibits elevated activity at rest, which has been linked to increased regional Aβ burden.³⁰ In the hippocampus, we observed a significant interaction between age and CLP290 treatment on the high-frequency band, with power increasing in the vehicle-treated mice but decreasing in the CLP290 treated mice. (Fig. 6D). CLP290 treatment did not affect the sleep structure of the mice (Supplementary Fig. 13). In contrast to the CLP290 treatment, intrahippocampal infusion of the selective KCC2 antagonist VU0463271³¹ (20 μM) increased the power of fast gamma and high-frequency oscillations as compared to vehicle infusion in 4-month-old wild-type mice (Supplementary Fig. 14), similar to the phenotype observed in 5xFAD mice. These results show that long-term CLP290 treatment prevents the age-dependent emergence of cortical hyperactivity in 5xFAD mice and that inhibition of KCC2 is sufficient to reproduce this form of hyperactivity.

Figure 6.

Long-term CLP290 treatment prevents cortical hyperactivity in 9-month-old 5xFAD mice. (A) Traces of local field potential (LFP) activity in retrosplenial cortex (RSC), both raw and filtered in the 60–120 Hz (fast gamma) frequency band and electromyography (EMG) activity from representative 5xFAD mice treated with either vehicle (top) or CLP290 (bottom) for 5 months. The traces depict activity during NREM sleep. (B) Left: LFP wavelet spectrograms of RSC activity during concatenated NREM sleep episodes from the same mice as in A. Right: spectrograms generated from the same data but scaled to highlight the difference in power in the fast gamma band. (C) The mean power of RSC activity in the slow gamma (30–60 Hz), fast gamma and high frequency (120–200 Hz) bands during NREM sleep in mice treated with either vehicle or CLP290 for 5 months, beginning at 4 months of age. LFP recordings from freely moving mice were conducted longitudinally at 6 and 9 months of age. For each figure, power is normalized to the mean of the vehicle-treated group at 6 months of age. (D) Same as C, but for LFP recordings from the hippocampus. Circles or dots in C and D represent single mice. Data are presented as mean ± SEM. ^#P < 0.05 for the interaction between age (6 versus 9 months) and drug treatment (vehicle versus CLP290). *P < 0.05; **P < 0.01.

Discussion

We present evidence for KCC2 hypofunction in the mPFC and CA1 of both 5xFAD and App^{NL-G-F/NL-G-F} mice, which translates into cognitive and social dysfunctions. This KCC2 hypofunction appears to be downstream of Aβ accumulation, and is in line with reports associating Aβ with deficits in inhibition.³² Indeed, soluble Aβ has been reported to induce cortical³³ and hippocampal hyperactivity³⁴ due to disrupted inhibitory input.³ The conclusion that KCC2 hypofunction is downstream of Aβ accumulation is consistent with several lines of evidence from clinical research suggesting that Aβ pathology is not directly responsible for producing the cognitive decline associated with Alzheimer's disease.^35,36 First, deposition of Aβ pathology in the brain begins long before cognitive symptoms become pronounced.^37,38 Second, some people exhibit abundant Aβ pathology with no detectable cognitive impairment.³⁹ Third, treatments that effectively reduce amyloid in the brain have largely proven ineffective at reversing cognitive symptoms and disease progression.^40–42 These findings suggest that Aβ initiates a pathological process that, once sufficiently advanced, can no longer be reversed by targeting Aβ itself. This is a key reason why it is necessary to develop treatments that target components of the cascade downstream of Aβ, such as KCC2.

The global increase in the power of gamma and high-frequency oscillations during NREM sleep shown here conforms with observations of elevated gamma power in the neocortex or hippocampus of other Alzheimer's disease-like transgenic lines^32,43,44 and in patients⁴⁵ with impaired inhibitory function resulting from diminished KCC2. Our findings of cognitive deficits are also consistent with the observation that loss of long-term potentiation (LTP) synaptic specificity in ageing is directly associated with diminished KCC2.¹⁰ While impaired Cl⁻ homeostasis by itself can be a substrate of the hyperexcitability associated with some Alzheimer's disease syndromes through both disinhibition¹¹ and ensuing enhanced NMDA receptor function,⁴⁶ elevated neural activity has been reported, in turn, to reduce KCC2 function.⁴⁷ This positive feedback loop has the potential for yielding an unclamped spiral of escalating hyperexcitability, eventually resulting in excitotoxicity. Attempting to restore Cl⁻ homeostasis through blocking the Cl⁻ importer NKCC1 in Alzheimer's disease⁴⁸ appears a less promising target, first because it is not brain-specific, but more importantly because it does not restore Cl⁻ extrusion capacity from the cells and thus does not protect the cells from Cl⁻ overload nor spatially constrains inhibition.^{11,20,21,49,50}

In conclusion, enhancing KCC2 function^24,51,52 represents a particularly promising therapeutic target with the multipronged advantages of restoring normal neural excitability, synaptic plasticity precision and cognitive function associated with Alzheimer's disease pathophysiology.

Data availability

All data and custom software/algorithms necessary to interpret and replicate the findings and methods of this article are available upon request.

Funding

Weston Brain Institute Transformation Research grant TR192089 (M.H.M., Y.D.K.). Canadian Institutes of Health Research (CIHR) grant FDN—159906 (Y.D.K.). Canada Research Chair program (Y.D.K.). Canadian Institutes of Health Research (CIHR) grants 390930 and 156040 (M.H.M.). Fonds de recherche du Québec—Santé (FRQS) Junior 1 Scholar, 269555 (A.G.G.). Sentinel North (Canada First Research Excellence Fund) Partnership Research Chair on Probing Life and the Environment with Light (A.G.G.). Natural Sciences and Engineering Research Council of Canada (NSERC) grant 06507 (A.G.G.). Université Laval Norampac Research grant on Alzheimer’s and Related Diseases (I.K.). Université Laval Fondation de la famille Lemaire Research grant on Alzheimer’s and Related Diseases (I.K.). NSERC Postdoctoral Fellowship (B.B.M.)

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.