Abstract
The cortex immediately surrounding a brain ischemic lesion, the peri-infarct cortex (PIC), harbors a large part of the potential to recover lost functions. However, our understanding of the neurophysiological conditions in which synaptic plasticity operates remains limited. Here we hypothesized that the chronic imbalance between excitation and inhibition of the PIC prevents the normalization of the gamma rhythm, a waveband of neural oscillations thought to orchestrate action potential trafficking. Probing the local field potential activity of the forelimb primary sensory cortex (S1FL) located in the PIC of male adult mice, we found a constant, deep reduction of low-gamma oscillation power (L-gamma; 30–50 Hz) precisely during the critical time window for recovery (1–3 weeks after stroke). The collapse of L-gamma power negatively correlated with behavioral progress in affected forelimb use. Mapping astrocyte reactivity and GABA-like immunoreactivity in the PIC revealed a parallel high signal, which gradually increased when approaching the lesion. Increasing tonic inhibition with local infusion of GABA or by blocking its recapture reduced L-gamma oscillation power in a magnitude similar to stroke. Conversely, the negative allosteric modulation of tonic GABA conductance using L655,708 or the gliopeptide ODN rescued the L-gamma power of the PIC. Altogether the present data point out that the chronic excess of ambient GABA in the PIC limits the generation of L-gamma oscillations in the repairing cortex and suggests that rehabilitative interventions aimed at normalizing low-gamma power within the critical period of stroke recovery could optimize the restitution of lost functions.
Significance Statement
After a stroke, the recovery of lost motor function depends on the reorganization of surviving neural networks. However, the excitation/inhibition balance in the repairing area is suboptimal as it leans excessively toward inhibition. In this work, using an in vivo approach, we demonstrate here that this imbalance, occurring during the critical window for recovery, leads to the collapse of gamma oscillations, a crucial cerebral rhythm for organizing neural communication. This study reinforces the concept of timely therapeutic interventions aimed at correcting the pathological oscillatory regimen of stroke recovery to enhance plasticity.
Introduction
The peri-infarct cortex (PIC) defines the neuroanatomically intact cortex, extending from the edge of a consolidated stroke lesion, where neurons are well alive but neural circuits are partially damaged (Stroemer et al., 1995; Brown et al., 2007, 2008, 2009; Murphy and Corbett, 2009). In rodent models of stroke, the PIC extends radially by 1–2 mm from the border of the lesion (Ohab et al., 2006; Brown et al., 2007; Murphy and Corbett, 2009; Hazime et al., 2021), and in humans, it potentially extends up to 1 inch (Cramer et al., 2006; Funck et al., 2017). Neuroplastic processes taking place in the PIC are crucial for the magnitude of functional recovery (Heiss et al., 1999; Dancause et al., 2005; Heiss and Thiel, 2006; Grefkes and Fink, 2014) but remain suboptimal amid an unfavorable cellular and molecular environment (e.g., heightened glial reactivity; Overman et al., 2012; Hazime et al., 2021). Furthermore, these processes peak during a limited time window in rodents and humans (Krakauer et al., 2012; Zeiler and Krakauer, 2013; Ballester et al., 2019; Dromerick et al., 2021). As a result, the PIC constitutes a reservoir of plasticity whose potential may never fully be reached. The functional state of the PIC and how the neurobiological disorders of the PIC hinder the process of synaptic reconnection and ultimately the regain of a lost function is not fully understood. During the past decade, an array of studies has brought the understanding that an excessive tonic GABAergic inhibition, located in the PIC, lasting several weeks, limits sensorimotor recovery in rodents (Clarkson et al., 2010; Clarkson, 2012; Alia et al., 2016; Cirillo et al., 2020; Nam et al., 2020; Lamtahri et al., 2021; van Nieuwenhuijzen et al., 2021). Considering the parallel mechanisms of normal learning and poststroke spontaneous recovery, it is reasonable to think that this imbalance between the excitation and the inhibition (E/I) of the PIC antagonizes neuronal activity, information flow, and, eventually, synaptic Hebbian processes (Joy and Carmichael, 2021).
An increasing number of studies also show the potential importance of neural oscillations in synaptic plasticity (Zarnadze et al., 2016; Womelsdorf and Hoffman, 2018). The fundamental role of these rhythmic fluctuations of neuronal excitability is to regulate action potential trafficking by synchronizing the firing of neuronal populations (Buzsáki, 2006). In particular, gamma wavebands (30–80 Hz) are precisely tuned to dictate a timing of collective neuronal discharges that match the temporality of synaptic plasticity mechanisms, such as long-term potentiation (LTP) or spike timing-dependent plasticity (STDP; Engel et al., 1992; König et al., 1996; Magee and Johnston, 1997; Bi and Poo, 1998; Buzsáki and Draguhn, 2004; Wespatat et al., 2004; Harris, 2005; Harris et al., 2003; Nishiyama et al., 2010). Indeed, in various neurological and psychiatric conditions, the gamma rhythm is specifically disturbed. Logically, the modification of network connections allowing functional recovery after a stroke requires that homeostatic mechanisms stick to their operating range, including the maintenance of gamma oscillatory activity (Llinás et al., 2005; Schnitzler and Gross, 2005; Buzsáki, 2006). Although the oscillatory activity of the brain has been probed after stroke in both animals and humans using electroencephalography (EEG), the latter is ineffective to precisely map the oscillatory changes in the PIC. Interestingly a recent work recording the local field potential (LFP) in the PIC of mice recovering from a focal ischemic stroke found a massive, specific, and long-lasting (several weeks) deficit of the power of low-gamma oscillations (L-gamma; 30–50 Hz; Hazime et al., 2021). A pending issue is therefore to understand the origin of this deficit and whether gamma oscillations play a role in poststroke recovery. We used a combination of behavioral testing, LFP recordings, and pharmacological modulations to examine the link between gamma power, functional recovery, and tonic GABA neurotransmission. We show that the PIC remains in severe deficit of gamma power, inversely correlated with the restitution of paretic skilled limb function. Furthermore, we found that correcting the excessive tonic inhibition of the PIC can safely restore the power of gamma oscillations.
Material and Methods
Animals and approvals
Eight- to 12-week-old (20–25 g) male C57BL/6J mice were purchased from Janvier Labs. Experiments, approved by the Ethics Committee for Animal Research of Normandy (Approval No. 76-451-04), were conducted by authorized investigators in accordance with the recommendations of the European Communities 86/609/EEC. All procedures were undertaken, and reporting was done in accordance with the ARRIVE (Animal Research: Reporting In Vivo Experiments) guidelines. All in vivo procedures were carried out between 8 A.M. and 6 P.M. in specific experimental rooms.
Stroke model: distal middle cerebral artery occlusion
Anesthesia was induced with 5% isoflurane in an induction chamber. Following the loss of the righting reflex, the mouse was rapidly transferred to a nose cone mask and maintained with isoflurane (2–2.5%) delivered at 1 L/min in oxygen-enriched air and 0.15 mg/kg of buprenorphine (Par Sterile Products) was injected intraperitoneally (i.p.). Spontaneous breathing frequency was kept over 0.5 Hz, and body temperature was strictly maintained at 37 ± 0.2°C with a feedback-controlled heating pad (Harvard Apparatus) throughout the surgery. Focal ischemia was induced by the permanent occlusion of the distal branch of the left middle cerebral artery as reported previously (Llovera et al., 2014). Under an operating microscope, a skin incision was made between the orbit and ear. Another incision dividing the temporal muscle was made and the left lateral aspect of the skull was exposed by gently parting the temporal muscle and surrounding soft tissue. The distal course of the middle cerebral artery was then visible through the translucent skull. A small burr-hole craniectomy was performed with a dental drill. The left middle cerebral artery was electrocoagulated by bipolar diathermy (KLS Martin). The muscle and soft tissue were replaced, the incision was sutured, and mice were transferred to a recovery box. In this model, ischemia is restricted to the neocortex. Buprenorphine was administered on the day of the surgery (0.015 mg/kg, s.c.) and on subsequent 2 d.
Infarct size determination
Infarct volumes were determined 48 h, 1 week, and 3 weeks poststroke. Extracted brains were flash frozen in isopentane and stored at −80°C. Using a cryostat (NX70, Leica Microsystems), 20-µm-thick slices were taken from every 10th section and mounted on slides. Thionin staining was performed, and images were taken with an upright light microscope (Axioscope 7, Zeiss). ImageJ analysis was used to quantify infarct volume as follows: infarct volume (mm3) = areas (mm2) × (section thickness (mm) + section interval (mm)). Infarct volumes were corrected for edema as described previously (Lin et al., 1993). All analyses were performed by an observer blind to the treatment groups.
Drug preparation
GABA (Sigma-Aldrich; 24278167) was dissolved in artificial cerebrospinal fluid (aCSF; in mM: 124 NaCl, 3 KCl, 1.3 MgCl2, 2.6 CaCl2, 1.25 NaH2PO4, 26 NaHCO3, 10 d-glucose, pH 7.4). Tiagabine (Cayman Chemical; 145821-59-6) was dissolved in absolute ethanol and then diluted with NaCl for use. L655,708 (Sigma-Aldrich; 24278527) was dissolved in DMSO and then diluted with NaCl. Mouse/rat ODN (H-Gln-Ala-Thr-Val-Gly-Asp-Val-Asn-Thr-Asp-Arg-Pro-Gly-Leu-Leu-AspLeu-OH) was synthesized as previously described (Leprince et al., 2001) and diluted in aCSF.
In vivo electrophysiological recording
Seven days, 21 d, and 4 months after distal middle cerebral artery occlusion (dMCAO), anesthesia was induced with 5% isoflurane in an induction chamber. Following the loss of the righting reflex, the mouse was rapidly transferred to a nose cone mask and maintained with isoflurane (2–2.5%) delivered at 1 L/min in air. A calibrated isoflurane dispenser with consistent and constant oxygen pressure was used to ensure reliable and reproducible delivery of specific isoflurane concentrations. Animal was placed in a digital stereotaxic frame (World Precision Instruments), 0.15 mg/kg of buprenorphine was injected intraperitoneally, and the eyes were protected with ophthalmic gel. Breathing frequency was maintained over 0.5 Hz, and body temperature was strictly maintained at 37 ± 0.2°C throughout the surgery using a low-noise feedback-controlled heating pad (World Precision Instruments). The head was shaved and disinfected with iodine, the skull was exposed, and two small holes were drilled in the parietal bone and the dura mater was incised with care to leave the cortex intact. Using digital stereotaxic arms, two recording tungsten microelectrodes (World Precision Instruments) of 2–3 µm tip diameter and 2.0 MΩ impedance were gently lowered into layer 4 of the exposed cortices. The first and second window corresponded to the ischemic ipsilateral hemisphere (PIC) and contralateral hemisphere, respectively (coordinates from the bregma: ML, 2.1 mm; AP, 0.26 mm; DV, 0.4 mm; and ML, −2.1 mm; AP, 0.26 mm; DV, 0.4 mm, respectively). Two Teflon-coated silver wire electrodes were implanted in the cerebellum for reference and ground. The electrodes were secured to the skull with some dental cement (Paladur, DentalMedical). For drug microdelivery, a glass micropipette (10-µm-diameter opening) was placed 50–100 µm away from the peri-infarct recording pipette's tip. After the surgical procedure, isoflurane was reduced to 1.1 ± 0.1% for a resting period of 60 min followed by the recording period. After recording baseline for 15 min, 0.5 µl of drugs was locally infused into the cortex at a rate of 0.15 µl/min, using a syringe pump (KD Scientific) coupled to a Hamilton syringe (Sigma-Aldrich) and the injection glass pipette. At the end, anesthesia was elevated to 5% for a 5 min period followed by an intracardiac injection of saturated KCl solution inducing heart arrest.
Data acquisition and analysis
All recordings were done in a Faraday chamber. A four-channel amplifier DP-314 (Warner Instruments) with active headstages coupled to a PowerLab 8/35 (ADInstruments) running Labchart Pro software (ADInstruments) was used to acquire the data. Extracellular signals were amplified (1,000×), bandpass filtered (0.1–100 Hz) for the LFP, and acquired continuously at 20 kHz. In addition, a mains filter was added to suppress 50 Hz electrical noise. For spectral analysis of oscillatory patterns (epoch of 15 min), a modified version of the multitaper FFT MATLAB package by Mitra and Pesaran (1999) was used. To obtain power spectral density (PSD), FFTs were performed with the following parameters: window size of 4 s, three to five tapers, frequency bins of 0.15 Hz, no overlap between successive windows, time bandwidth of 3 (Mitra and Pesaran, 1999; Quilichini et al., 2010). Discrete FFTs were run with the LabChart toolbox to obtain PSD of L-gamma. FFT size of 16,000 data points with a Hann data window was applied with no overlap between successive windows. To derive L-gamma power, we applied a Butterworth filter to the LFP data, isolating the L-gamma band (30–50 Hz) with an autoadjusted transition width. This ensured exclusion of frequencies outside the L-gamma band. Morlet wavelet spectrograms of the LFP oscillations were calculated using a wavelet convolution of the wavelet toolbox (MATLAB, The MathWorks), built with a Morlet wavelet with 10 voices per octave and frequency bounds of 20–70 Hz.
Power spectrum parameterization
Power spectra were decomposed into periodic and aperiodic components using the open source, python-based “fitting oscillations & one over f,” FOOOF (Donoghue et al., 2020). The model was fit between 1 and 100 Hz, and the parameters used were as follows: peak width limit, [1,50]; max n peaks, 5. A two-step fitting process was used; a “knee” mode was initially employed and after obtention of the knee parameter, the power spectra were refit using the fixed model. The values for minimum peak height were adjusted to optimally fit the power spectra of each condition
Immunofluorescent labeling of GFAP and GABA
Reactive astrogliosis and GABA-like immunoreactivity were assessed by immunofluorescent labeling for GFAP and GABA, respectively, at 7 d, 21 d, and 4 months poststroke. Mice were anaesthetized with a ketamine/xylazine (Par Sterile Products) cocktail (50 and 20 mg/kg) and perfused transcardially with 1× phosphate-buffered saline (PBS) solution followed by 4% paraformaldehyde (PFA) in 1× PBS at pH 7.4. Whole brains were extracted and postfixed in 4% PFA for 10 h and cryoprotected in 30% sucrose solution. Free-floating 40-µm-thick coronal sections were obtained using a vibratome (Leica Microsystems). For immunofluorescence, sections were incubated for 30 min in a blocking solution (0.3% Triton X-100, 3% donkey serum, and 5% BSA in 1× PBS). An extra blocking step was applied using a mouse on mouse (MOM) kit (Abcam; ab269452) overnight at 4°C. Immunostaining was then performed with a mixture of primary antibodies in the blocking solution at 4°C on a shaker for 12 h. Sections were rinsed in PBS three times and were incubated with corresponding fluorescent secondary antibodies for 2 h and then further rinsed with PBS three times. When needed, nuclear counter stain DAPI (Sigma-Aldrich, 1:2,000) was done during the second rinsing procedure. Finally, sections were mounted on slides with Mowiol and dried. Primary antibodies used for fluorescent immunostaining are as follows: goat anti-GFAP (1:1,000; Abcam ab53554), mouse anti-GABA (1:500; Abcam ab86186). Fluorescent secondary antibodies were purchased from Thermo Fisher Scientific and used in 1:500 dilutions.
Image acquisition and analysis
Whole coronal slice fluorescent mosaic images were taken with Leica Thunder Imager tissue, and computational tissue clearing was applied to systematically reduce noise. Zoom of regions of interest (ROIs) were obtained using a Leica SP8 confocal microscope. Then,10 µm Z stack images in 1 µm Z-step size was processed for further analysis FIJI (NIH) program. For mosaic images, quantification of the signal intensity of GFAP- and GABA-like immunoreactivity was assessed with ImageJ using the plot profile tool over ROIs drawn from the interhemispheric fissure to the lesion border. The limit of the glial-defined PIC corresponds to the beginning of a statistical difference (p < 0.05) between the fluorescence signal intensity in the ipsilateral cortex and the homotopic contralateral cortex. Astrocyte morphology was assessed using the Sholl analysis neuroanatomy plugin (Ferreira et al., 2014) of the FIJI program. Binary masks were created from thresholds of 10 µm maximum intensity z projections containing visible astrocytes. The Sholl analysis was performed using the following parameters: radius step size of 3 µm, primary branches were automatically inferred from starting point. Ramification indices were then extracted. In total, slices from four mice were used and at least 10 astrocytes per animal were analyzed when possible.
Behavioral assay
Mice were tested on capellini handling task, 1 week before surgery to establish baseline performance levels (Tennant et al., 2010). Herein, mice were trained for a period of 2 weeks, and they were subsequently tested at Days 7, 14, and 21 after stroke induction before in vivo electrophysiological recordings. The parameter observed was the number of adjustments made during pasta consumption (adjustments being paw flexions and adductions, contacts, and recontacts with pasta piece; Tennant et al., 2010). Behavior was scored by observers who were blind to the treatment or experimental group. Here, the main parameter used to evaluate functional recovery was the fine adjustments made by the affected paw. A functional recovery score was established as shown: (Progress between D14 and D21) × (Performance at D14) / (Potential of recovery between D14 and D21).
Statistics
Sample size was calculated using the JavaScript utilities available at www.stat.ubc.ca/∼rollin/stats/ssize/index.html with parameters determined from prior work. For experiments presented in Figures 1A, 1D, 2B, and 3D, the knowledge of the variability was too uncertain to reliably calculate a sample size (>80% power). For those experiments, we relied on reasonable assumptions to determine a priori the number of animals used. Data are presented as mean ± SEM or as box-and-whisker plot showing the median (box, first and third quartiles; whisker, range). Statistics were performed using Prism software (GraphPad). Normal distribution of the datasets was tested by a Kolmogorov–Smirnov test. When comparing three or more groups, data from histology, behavioral test, and in vivo electrophysiology were all analyzed using one-way and two-way ANOVA followed by post hoc Tuckey's or Bonferroni's test when needed. Pairwise mean comparisons were performed using t test for normally distributed data, or Mann–Whitney test otherwise. Sample size in behavioral studies was assessed by power analysis using a significance level of α = 0.05 with 80% power to detect differences in ANOVA. No animals were excluded from analyses and for recovery studies, two-way ANOVA followed by post hoc Tukey's or Bonferroni's test for multiple comparisons was performed.
Results
Peri-infarct L-gamma oscillation power is reduced at 7 and 21 d poststroke
Cerebral ischemia was induced in mice using a variation of the well-characterized Tamura model consisting of the permanent occlusion of a distal trunk of the middle cerebral artery. This procedure generated, 48 h after the ischemia, a focal lesion of 8.5 ± 1.3 mm3 centered on the barrel field cortex (Fig. 1A,B). The histopathological examination showed that, despite the slight variation of the cortical infarction size, the sensory-motor cortical network of the forepaw was always located in a 2 mm marginal zone surrounding the infarction (Fig. 1B). Seven or 21 d after the occlusion, two LFP recording electrodes, one in each hemisphere, were implanted in layer 4 of the primary somatosensory forelimb cortex (S1FL) under general anesthesia to probe the production of L-gamma oscillations (30–50 Hz; Fig. 1B). Short duration of L-gamma bursts essentially appears during up-state phases as exemplified in Figure 1D–E and I,J. To further confirm the periodic nature of the gamma rhythm, a power spectrum parameterization with the FOOOF fitting method was carried out. In all animals, the spectrum was not only composed of an aperiodic component but also a consistent periodic component. Therefore, the presence of L-gamma power above the aperiodic component confirms the presence of L-gamma oscillations within the frequency range 30–50 Hz (Fig. 1F,K). Generally, sham animals (SC) did not display noticeable change in L-gamma power at 7 and 21 d postsurgery in and between S1FL cortices of both hemispheres. In comparison, ischemic animals displayed a severe decrease of L-gamma power in the ipsilateral S1FL cortex (located in the PIC) at 7 d poststroke (SC: 43.03 ± 4.41 µV2/Hz, PIC: 27.01 ± 1.77 µV2/Hz, p < 0.01; Fig. 1C–G). L-gamma power remained significantly decreased at 21 d (SC: 59.01 ± 13.59 µV2/Hz, PIC: 24.02 ± 4.05 µV2/Hz, p < 0.01; Fig. 1H–L). The contralateral S1FL cortex (CC) of ischemic mice showed a transient increase of L-gamma power at 7 d. Looking further into the high gamma spectrum (70–300 Hz), we did not find any difference in peak power at 7 and 21 d (p > 0.05; data not shown) showing that the collapse of the gamma rhythm is centered on the L-gamma range.
Deficit of low-gamma oscillation power at 7 and 21 d poststroke. A, Experimental arrangement. The permanent occlusion of a distal branch of the middle cerebral artery (MCAO) by electrocoagulation resulted in a lesion majorly occupying the barrel field of the somatosensory cortex surrounded by a cortical zone defined as the peri-infarct cortex (PIC). At 7 and 21 d, two electrodes were implanted for recordings in the ipsilateral PIC and in the homotopic contralateral cortex corresponding to the sensory-motor circuit of the forelimb (FL) as schematized in B; MsFL, supplementary motor cortex of the FL; M1FL, primary motor cortex of the FL; S1FL, primary somatosensory cortex of the FL. C, Left (top and bottom), respectively, Thionine-stained coronal slice showing lesion area and lesion map at 48 h (incidence based on 20 mice). Most of lesions were circumscribed in a cortical territory of 2 mm. C, Right, Size and extent of a representative infarct obtained from distal middle cerebral occlusion. D, I, Representative electrophysiological traces showing unfiltered and low-gamma (30–50 Hz bandpass filtered) LFP activity of the sham cortex (SC, green), contralateral cortex (CC, blue), and peri-infarct cortex (PIC, red) at 7 and 21 d, respectively, arranged from top to bottom in each case. E, J, 4 s representative wavelet transforms of LFPs in SC, CC, and PIC at 7 and 21 d, respectively, from top to bottom. F, K, Parameterized power spectra (1–100 Hz) obtained at 7 and 21 d comparing SC, PIC, and CC LFP activity and showing FOOOF-derived aperiodic fits. G, L, L-gamma (filtered 30–50 Hz) power spectra of SC, PIC, and CC. H, M, Quantifications of L-gamma power in SC, CC, and PIC at 7 and 21 d, respectively, are reported in the respective histograms. Comparisons between hemispheres or groups are reported by (*) when significant (*p < 0.05; **p < 0.01; ns, nonsignificant; one-way ANOVA followed by Tuckey's correction, 7 d: n = 14 recordings from 7 ischemic animals at 7 d poststroke and n = 16 recordings from 8 animals at 21 d poststroke). Sham animals: n = 10 recordings from 5 animals and n = 12 recordings from 6 animals at 7 and 21 d, respectively.
L-gamma power interhemispheric asymmetry negatively correlates with recovery
To test the hypothesis of a relationship between the L-gamma power collapse of the PIC and a behavioral function involving the activity of the S1FL cortex, we used the Capellini handling test (Tennant et al., 2010). This test measures dexterous forepaw function based on the number of adjustments of each paw necessary to eat a 2 cm pasta piece (Fig. 2A). We found a typical stroke-induced partial recovery of the left paw function (contralateral to the ischemic cortex) over the 21 d postischemia (Fig. 2B; number of adjustments: 24.71 ± 2.37, 11.91 ± 2.37, 15.21 ± 1.35, 22.31 ± 3.57 at pre-stroke, 7, 14, and 21 d, respectively). A recovery score was calculated to take into account the individual rate of progress during Week 3 (Days 14–21) in relation to the remaining potential improvement. The recovery score at 21 d showed no correlation (R = 0.362; p > 0.05) with the initial deficit amplitude at 7 d poststroke (Fig. 2C). This indicates that the progress measured by the recovery score is not influenced by the severity of the initial deficit. However, we found a strong negative correlation between L-gamma power of the ipsilateral S1FL cortex relative to the homotopic one and the recovery score (Fig. 2D; R = 0.797; p < 0.01). In other words, animals with a deeper collapse of L-gamma power at 21 d were those who recovered the least during the 7 preceding days. Conversely, mice had a greater handling recovery when L-gamma power of the S1FL was closer to baseline level.
Interhemispheric L-gamma power predicts recovery. A, Assessment of skilled forepaw function using the capellini handling test: the mouse was videotaped while eating a 2-cm-long pasta piece. The sensorimotor sequence comprises the number of adjustments made with the paretic paw (left paw; red) compared with the unaffected paw (right paw; green). B, Evaluation of functional recovery using dexterous forelimb movement before and after (7, 14, and 21 d) dMCAO. C, Linear regression analysis showing the absence of correlation between the recovery score and the initial poststroke deficit (Day 7). (n = 10; R = 0.362; p > 0.05). D, Linear regression analysis showing the correlation of the recovery score (the progress made between Days 14 and 21 relative to the highest possible progression) with gamma power difference between the PIC and the homotopic contralateral cortex. (n = 11; R = 0.797; p < 0.01). All linear regression data were analyzed using Pearson's correlation coefficient (R).
Reactive astrocytes over express GABA in the PIC
In search for the origin of the collapse of L-gamma power in the PIC, we hypothesized that a shift of the E/I balance caused by an aberrant inhibitory conductance could well compromise the production of a L-gamma rhythm. Several groups have indeed reported that reactive astrocytes in the PIC lose their ability to regulate extracellular GABA concentration (Clarkson et al., 2010; Carmichael, 2012; Jo et al., 2014; Nam et al., 2020), leading to an increased tonic inhibitory conductance. The aim was therefore to map the expression of GFAP and GABA by immunohistochemistry at 7 and 21 d (Fig. 3). The specificity of the GABA immunoreactivity was confirmed by a test of extinction consisting of loading the preparation with exogenous GABA, as already reported (Inokawa et al., 2010). The contralateral and remote ipsilateral (>2 mm) cortex showed homogeneous baseline fluorescence signal intensity for GFAP and GABA-like immunoreactivity (Fig. 3A). In contrast, the PIC showed a gradient of fluorescence intensity, increasing when approaching the border of the lesion, slightly more extended at 7 d than at 21 d (Fig. 3A,D). Remarkably, the two gradients of GFAP and GABA-like immunoreactivity were very much parallel, the intensity of one strongly correlating with the intensity of the second (Fig. 3E). GFAP expression was most intense in the glial scar with polarized astrocytes showing a long leading process extended in the direction of the infarct. Beyond the scar, astrocytes displayed a marked protoplasmic shape without preferred orientation of their processes (Fig. 3A,B). Further analysis of the cellular morphology revealed that at 7 and 21 d, astrocytes remain highly ramified in the PIC (Sholl analysis; Fig. 3C). At higher magnification, the colocalization of GFAP and GABA-like immunoreactivity was often obvious with a clear punctiform enrichment of the latter in proximity with the astrocytic GFAP + cytoskeleton (Fig. 3B). These observations agree with previous demonstrations of reactive astrocytes as a main contributor to abnormal GABA production in the PIC. None of sham animals did present an infarction; however, we observed a slight increase of astrocyte reactivity in the cortical area under the craniotomy site, distant from the site of recording.
Excessive astrocytic production of GABA in the peri-infarct cortex. A, Measurement of astrogliosis (GFAP, green) and GABA-like immunoreactivity (red) was done at 7 and 21 d post dMCAO. Representative mosaic scans (top 3 rectangles; scale bar, 1 mm) and confocal images of the three ROIs: the contralateral cortex (CC), remote cortex (RC), and peri-infarct cortex (PIC; scale bar, 20 µm). B, High-resolution confocal imaging of two reactive astrocytes located in the PIC showing a high level of costaining for GFAP and GABA-like immunoreactivity. C, Sholl analysis of representative astrocytes in the CC and PIC at 7 and 21 d (top and bottom images, respectively) showing the characteristic cellular morphology of reactive astrocytes and their corresponding ramification indices. Each point represents an astrocyte (7 d, PIC; n = 55 astrocytes, CC; n = 25 astrocytes, 21 d, PIC; n = 50 astrocytes, CC; n = 34 astrocytes for both, n = 5 mice; Mann–Whitney U test). D, Topological profiles of GFAP and GABA-like immunoreactivity at 7 d (left) and 21 d (right) quantified by signal intensity from the stroke border toward the RC. GFAP and GABA overexpression extends until the end of statistical significance of signal intensity relative to the CC (n = 5 mice and n = 4 mice at 7 and 21 d post dMCAO, respectively). E, Linear regression showing the correlation of GFAP signal intensity with GABA signal intensity in square ROI of 200 µm located in the PIC at 7 (left) and 21 d (right); Pearson's correlation coefficient R = 0.97, p < 0.001, and 0.70, p < 0.01, respectively. (*p < 0.05, **p < 0.01, and ***p < 0.001; ns, nonsignificant). Data was presented as mean ± SEM (C, D; *p < 0.05, **p < 0.01, and ***p < 0.001, ns, nonsignificant).
Gamma power and reactive astrogliosis are normalized at 4 months poststroke
To finalize the electrophysiological and immunohistological characterization of the PIC during stroke recovery, we checked on the state of GFAP/GABA and L-gamma power on the long run (Fig. 4A). At 4 months after focal ischemia, in the chronic phase, we found that GFAP expression returns to sham level except in the immediate boundary of the residual lesion formed by the glial scar and that GABA-like immunoreactivity also returned to baseline levels (Fig. 4B). Interestingly, L-gamma power in the PIC also returned to sham-like levels (Fig. 4C–G, SC: mean power: 37.02 ± 2.32 µV2/Hz, CC: 35.01 ± 2.06 µV2/Hz, PIC: 39.04 ± 3.11 µV2/Hz). These results show that the L-gamma deficit occurs during the period of increased astrocyte reactivity and excessive GABA suggestive of a mechanistical coupling between reactive astrocytes, tonic inhibition, and L-gamma wave production.
Gamma power normalization coincides with reduction of astrocyte reactivity at 4 months post stroke. A, Experimental arrangement. LFPs were recorded in the PIC and CC at 4 months poststroke. B, Representative mosaic scans and confocal images of GFAP, GABA, and merged immunofluorescence in the CC and PIC at 4 months post dMCAO. C, Representative electrophysiological traces showing spontaneous unfiltered and L-gamma (30–50 Hz bandpass filtered) LFP activity of sham cortex (SC), peri-infarct cortex (PIC), and contralateral cortex (CC), respectively, arranged from top to bottom at 4 months. D, 4 s representative wavelet transform of LFPs in SC, PIC, and CC, respectively, arranged from top to bottom at 4 months. E, Power spectra (0.5–100 Hz) obtained at 4 months comparing SC, PIC, and CC LFP activity. F, Corresponding L-gamma (filtered 30–50 Hz) power spectra of SC, PIC, and CC. G, Quantification of L-gamma power in SC, CC, and PIC at 4 months are reported in histograms. Comparisons between hemispheres or groups are reported by (*) when significant (*p < 0.05; **p < 0.01; one-way ANOVA followed by Tuckey's correction, n = 14 recordings from 7 ischemic animals and n = 10 recordings from 5 sham mice.
Increasing tonic inhibition reduces L-gamma power in vivo
We next investigated whether an increase of extracellular GABA in vivo would affect L-gamma oscillation power as observed in the repairing cortex after stroke. To that end, we designed a protocol of microinfusion in the vicinity of the recording electrode implanted in the layer 4 of a healthy mouse. Electrode to injection pipette distance was stereotaxically maintained at 50–100 µm. We managed to find the parameters of speed, volume, and pressure for which the infusion of the vehicle would cause no change in the recorded neuronal activity, in term of oscillation power (between 1 and 100 Hz) or spiking (data not shown). In these conditions, like for other frequency bands, L-gamma power remains stable for at least 20 min after vehicle microinfusion (Fig. 5A–D). We provoked a rise of extracellular GABA in two ways: first, by directly infusing GABA (10 mM; 500 nl), and second, by infusing tiagabine (5 mM; 500 nl), a well-characterized blocker of the astrocytic GABA transporter GAT-1 (Borden et al., 1994). In both cases, in comparison with vehicle injection, the treatment dramatically reduced the power of L-gamma oscillations and in a magnitude similar to the PIC (power change vehicle: 0.31 ± 0.78 µV2/Hz, GABA: −23.04 ± 3.02 µV2/Hz, tiagabine: −15.02 ± 5.31 µV2/Hz; p < 0.01; Fig. 5D,G,J,K). Therefore, increasing tonic GABA conductance diminishes L-gamma oscillation power in a magnitude comparable with stroke.
Heightened tonic inhibition diminishes low-gamma power in vivo. A, Experimental setup. Vehicle, GABA, or tiagabine were slowly microinjected in the vicinity of the recording electrode in the healthy cortex. B, E, H, Representative electrophysiological traces showing spontaneous unfiltered and L-gamma (30–50 Hz bandpass filtered) LFP activity of healthy mice cortex before and after micro infusion of vehicle, GABA, and tiagabine, respectively, arranged from top to bottom in each case. C, F, I, 4 s representative wavelet transforms of LFPs before and after injection of vehicle, GABA, and tiagabine, respectively, from top to bottom. D, G, J, L-gamma (filtered 30–50 Hz) power spectra before and after injection of the three compounds from top to bottom. K, Quantification of L-gamma power change after administration of each compound is depicted on whisker plots. Comparisons between groups are reported by (*) when significant (*p < 0.05; **p < 0.01; two-way ANOVA and Bonferroni's multiple comparisons; vehicle: n = 12 recordings from 6 mice, GABA: n = 16 recordings from 8 mice, and tiagabine: n = 10 recordings from 5 animals). Extended Data Figure 5-1 A,B shows the effect of GABA microinjection on the aperiodic component and shift in exponent. Exponent changes in the CC and PIC of stroke animals at 7 and 21 d are indicated in Extended Data Figure 5-1C.
Figure 5-1
Local GABA injection shifts aperiodic exponent similarly to the peri-infarct cortex. A, Parameterized full power spectra (1–100 Hz) of a representative animal obtained at pre-and- post GABA injection; left to right respectively, showing FOOOF derived aperiodic fits and periodic L-gamma activity. B, Quantification of exponent change pre and post GABA depicted on whisker plots. C, Quantification of exponent change in CC and PIC of stroke animals at 7- and 21-days post stroke; left and right respectively). Pairwise comparisons by paired t test (* p < 0.05; ** p < 0.01) n = 12 recordings from 6 mice for GABA, 7 days: n = 14 recordings from 7 mice at 7 days post-stroke and n = 16 recordings from 8 mice at 21 days post-stroke. Download Figure 5-1, TIF file.
Reducing tonic inhibition rescues gamma power in the PIC
Following this logic and using the same technical approach, we then hypothesized that the diminished L-gamma oscillation power of the PIC can be rescued by blocking the extrasynaptic GABAA receptors mediating tonic inhibition. We choose to focus on alpha-5 (α5) containing GABAA receptors because it has been repeatedly shown that reducing tonic inhibition with a negative allosteric modulator (NAM), such as L655,708, improves functional recovery after stroke (Clarkson et al., 2010; Lake et al., 2015; Orfila et al., 2019; Lamtahri et al., 2021). In healthy mice, in comparison with vehicle, L655,708 (50 µM, 500 nl) could not significantly boost L-gamma oscillation power (Fig. 6B,D,R; power change vehicle: −1.03 ± 0.57 µV2/Hz, L655,708: −0.41 ± 0.99 µV2/Hz, p < 0.01). However, 21 d after stroke, L655,708 increased the L-gamma oscillation power of the PIC (Fig. 6L–N; power change, L655,708: 10.01 ± 1,66 µV2/Hz, p < 0.01). We then tested another recently characterized NAM, the gliopeptide ODN (10 µM, 500 nl) known to be synthesized by astrocytes in the cortex (Lamtahri et al., 2021). In healthy mice, ODN could not increase L-gamma power (Fig. 6H–J; power change ODN: −0.12 ± 0.11 µV2/Hz). At 21 d after stroke, we found the same effect as L655,708 as ODN boosted the L-gamma power of the PIC (Fig. 6O–Q; power change ODN: 13.04 ± 2.24 µV2/Hz; p < 0.01). These results reinforce the idea that poststroke overload of extracellular GABA is causal in gamma oscillations collapse during recovery.
GABAAR NAMs reduce low-gamma power deficit in the PIC. A, K, Experimental arrangements. LFPs were recorded in healthy mice and in the PIC at 21 d after dMCAO and vehicle, L655,708 or ODN were microinfused into the PIC in proximity of the recording electrode. B, E, H, L, O, Representative raw traces showing spontaneous L-gamma (30–50 Hz bandpass filtered) LFP activity in healthy mice cortex and in the PIC before and after micro infusion of vehicle, L655,708, or ODN, respectively, arranged from top to bottom in each case. C, F, I, M, P, 4 s representative wavelet transform of LFPs in healthy cortex and in the PIC before and after injection of vehicle, L655,708, or ODN, respectively, from top to bottom. D, G, J, N, Q, Low-gamma (filtered 30–50 Hz) power spectra before and after injection of vehicle, L655,708, or ODN in healthy mice and in the PIC from top to bottom. R, Quantification of low-gamma power change after administration of vehicle, L655,708, or ODN in healthy mice and PIC is depicted on whisker plots. Comparisons between groups are reported by (*) when significant (*p < 0.05; **p < 0.01; in healthy mice, vehicle: n = 10 recordings from 5 mice, L655,708: n = 10 recordings from 5 mice, and ODN: n = 12 recordings from 6 mice). In the PIC, L655,708: n = 12 recordings from 6 mice, and ODN: n = 16 recordings from 8 mice.
Discussion
The results of the present study show that, after a stroke, the PIC goes through a prolonged dearth period of L-gamma oscillations power precisely during the critical time window of plasticity-induced functional recovery. Our observations point out that excessive ambient GABA, likely derived from reactive astrocytes, is a major cause of this L-gamma oscillation collapse. Considering the strong correlation we found between L-gamma power and the magnitude of the recovery, and the increasing evidence linking gamma oscillations with synaptic plasticity, our study argues in favor of therapeutic solution aimed at restoring the gamma rhythm of the repairing cortex.
Gamma oscillations collapse in the PIC during the critical window
In human, the highest potential for recovery is observed within the first 6 months after stroke (Kwakkel et al., 2004; Prabhakaran et al., 2008). In rodent models of focal cerebral ischemia, this critical period of plasticity corresponds to the first 4 weeks, after which functional recovery gains are no more significant (Dijkhuizen et al., 2001; Biernaskie et al., 2004; Brown et al., 2007; Joy and Carmichael, 2021). We assessed functional recovery using skilled forelimb movement, a behavior governed by an intricate but coordinated interaction of multiple neocortical circuits and subcortical feedback loops (Alstermark and Isa, 2012; Conner et al., 2021; Yang et al., 2023). Targeting a distal branch of the middle cerebral artery created a lesion topology for which the PIC comprised most of the sensory-motor network of the contralateral forelimb, namely, the primary somatosensory cortex (S1), the associated motor cortex (AM), and the motor cortex (M1). The S2 cortex of the forelimb was systematically located in the lesion core. Therefore, with this configuration, most of plasticity efforts necessary for the restitution of the forelimb function were located inside the peri-infarct area.
We chose to measure gamma oscillation of the repairing cortex in absence of external stimuli under anesthesia to favor the stability and the reproducibility of an internally driven regimen and to avoid the stochastic changes of the oscillatory pattern associated with the awake state (Lustig et al., 2016). Isoflurane is a well-known depressor of cortical activity (Land et al., 2012); however, at ∼1%, we systematically observed a rich LFP activity with a consistent gamma rhythm peaking ∼35 Hz.
The collapse of L-gamma power in the PIC confirms our previous observation (Hazime et al., 2021). Here our analysis focused on L-gamma rhythm within the conceptual framework of their role in synaptic plasticity (Galuske et al., 2019; Hadler et al., 2024). L-gamma oscillations appear in bursts of 4–8 cycles, generally during up-state phases. Using spectral parameterization (Donoghue et al., 2020), we have shown that genuine L-gamma oscillations are clearly distinguishable from the underlying aperiodic component. Although its neurobiological basis remains unclear, several recent studies suggest reconsidering the aperiodic component of the power spectrum as a dynamic element differing from stationary noise (Donoghue et al., 2020). The L-gamma power change we observed in the PIC does not imply the absence of change in the aperiodic component of the spectrum. Indeed, the exponent of the aperiodic component systematically increased 7 and 21 d after stroke (Extended Data Fig. 5-1C), in line with the interpretation of an increased inhibition (Gao et al., 2017; but see Brake et al., 2024). Interestingly, the recent study by Biskamp et al. (2022) shows that the aperiodic component of the poststroke EcoG spectrum in awake mice is also a good biomarker of motor recovery. The periodic gamma component was not the focus of the aforementioned study, making it difficult to determine the contribution of true gamma oscillations to both the power reduction in the 30–60 Hz band and the 1/f shape of the spectrum. Further reflections and consensus-based methods are needed to carefully weigh the relative contribution of a gamma oscillatory peak to the trending of the broadband LFP or EcoG power spectrum (Gerster et al., 2022).
Testing the recovery of each mouse to use their paretic arm during 3 weeks after stroke, we found a strong correlation between the forearm motor performance and L-gamma power in the PIC. A straightforward interpretation is that L-gamma is instrumental to the wiring of new functional circuits. This is in line with an in-progress corpus of results tending to demonstrate that gamma oscillations act as a facilitator for synaptic plasticity (Galuske et al., 2019; Li et al., 2021). Synchronized oscillation of neuronal membrane potential at the gamma frequency assures coincidence between presynaptic and postsynaptic activity within a temporal window that is optimal for synaptic plasticity process such as STDP (Engel et al., 1992; König et al., 1996; Magee and Johnston, 1997; Bi and Poo, 1998; Harris et al., 2003; Buzsáki and Draguhn, 2004; Wespatat et al., 2004; Harris, 2005). In other words, it is postulated that, under a gamma rhythm of oscillation, the probability to associate pertinent information to assemble a neuronal circuit increases. However, one limit of this interpretation in our study is the fact that the lesion, although very reproducible in size and topography, are close, by ±300 µm from the site of LFP recording. Knowing that L-gamma power decreases when approaching the edge of the lesion (Hazime et al., 2021), one may argue that the correlation reflects a link between the size of the lesion (the bigger the closer) and the motor performance of the paretic arm (the bigger the more deficit). We believe that this possibility does not herald the link between L-gamma power and forearm performance since we found no clear correlation between the size of the lesion at 7 d and L-gamma power in the S1FL. Further gain or loss experiment testing the causal role of gamma oscillations will be necessary to confirm that L-gamma power is not just a biomarker predicting good functional recovery but one of the neurophysiological actors of recovery.
The reorganization of the cortical oscillatory activity following a stroke has been keenly investigated during both acute and chronic phases of recovery. The focus has mainly been on low frequencies (<20 Hz), with some groups reporting that the power of delta oscillations (1–4 Hz) is increased in awake or asleep animals (Gulati et al., 2015; Kim et al., 2022). In our conditions, we consistently found that low-frequency oscillations are not significantly modulated at 7 and 21 d poststroke, in line with Biskamp et al. (2022). Similarly, Carmichael and Chesselet (2002) also observed that this increase is transient and only persists up to 5 d after the onset of the stroke. But it is also possible that isoflurane masked the increase in slow oscillations. More observations are needed to conclude when and how exactly low oscillations may intervene in the process of recovery.
Here gamma oscillations were analyzed only in respect to their power. Their interactions with other frequency bands, particularly theta waves, were not examined. It is well established that gamma rhythm is modulated by theta oscillations through cross-frequency coupling. Interestingly, it has been shown that low-frequency electrical stimulation at theta frequency can enhance skilled forelimb function by enhancing neural cofiring and activity propagation in the PIC (Khanna et al., 2021). It remains to be clarified whether gamma oscillations are instrumental in this effect.
Astrocytic GABA overexpression parallels gamma oscillation collapse
The E–I balance is essential for the formation of neural oscillations (Gao et al., 2017). In search for the origin of L-gamma collapse in the repairing cortex, we postulated that excessive ambient GABA from reactive astrocytes affects the production of L-gamma. We found strickling parallel gradients of GABA immunoreactivity and astrocyte reactivity along the extent of the PIC. In line with previous observations made in several neuropathological contexts including stroke (Jo et al., 2014; LI et al., 2019; Nam et al., 2020), reactive astrocytes were strongly positive for GABA, 3 weeks after stroke. Several reports have recently demonstrated that the buildup of extracellular GABA by reactive astrocytes is both due to a deficiency in their ability for GABA uptake (Clarkson et al., 2010; Lie et al., 2019) and an excessive production and release of GABA (Nam et al., 2020). L-gamma power follows the same gradient, i.e., a loss of power along with the increase of GABA immunoreactivity (Hazime et al., 2021). Outside the PIC, ∼2 mm off the lesion, GABA immunoreactivity and L-gamma power retrieve baseline values. Correspondingly, 4 months after stroke, both GABA immunoreactivity and L-gamma power retrieve their baseline values. We propose that during the critical time window of recovery, there is an uncoupling between the requirement for synaptic plasticity and the ability to generate gamma oscillations. A limit of our observation is the lack of quantification of ambient GABA. Recent advancements have led to the development of brain-implantable multifunctional probes capable of detecting GABA in real time in vivo (Moldovan et al., 2021). The use of these probes will help quantify the spatiotemporal evolution of extracellular GABA concentration in the PIC.
Ambient GABA levels in vivo modulate gamma oscillation power
To explore the causal link behind this strong spatiotemporal correlation between GABA and L-gamma, we pharmacologically manipulated GABA signaling by in vivo microinfusion into the cortex, in the vicinity of the recording pipette. We mimicked increased tonic GABAA conductance by microinfusion of GABA or tiagabine to block a major GABA reuptake pathway. Both lowered L-gamma power. Of note, also very similar to the PIC, the effect of GABA not only flattened the L-gamma band but also tilted the broadband power spectrum, i.e., increasing the exponent of the aperiodic component of the power spectrum (Extended Data Fig. 5-1A,B). This agrees with recent observations linking an increase of aperiodic exponent with an increase in inhibition within a neural network (Gao et al., 2017). Conversely, limiting tonic GABAA conductance with two NAMs (L655,708 and ODN) increased L-gamma power of the PIC. This is very consistent with previous in vitro observations demonstrating a negative relationship between GABA tonic inhibition and gamma oscillation power (Towers et al., 2004; Mann and Mody, 2010) in vivo (Zanos et al., 2017) or in human (Bolognani et al., 2015). Interestingly, L655,708 and ODN could not significantly increase L-gamma power in the healthy cortex. L655,708 has already been demonstrated to exert a minimal effect on tonic current in healthy mice ex vivo (Clarkson et al., 2010). Given the robust GABA transport system, in the healthy cortex, it is likely that any significant reduction in tonic inhibition is promptly detected and returned to homeostatic levels to re-establish the narrow E–I balance. It is also possible that, in the healthy cortex, L-gamma power reaches a plateau, insensitive to the physiological ambient concentration of GABA.
Furthermore, the involvement of astrocytes in the control of gamma oscillation (but not the other frequency bands) have recently been proposed (Lines et al., 2020). Interestingly, the gliopeptide ODN that we recently characterized as a partial α5-GABAAR NAM (Lamtahri et al., 2021) was more efficient than L-655,708 to boost PIC's L-gamma power but also ineffective at increasing them in the healthy cortex. Both L655,708 and ODN have been shown to improve functional recovery in various models of rodent focal cortical ischemia (Clarkson et al., 2010; Lake et al., 2015; Lamtahri et al., 2021). Thus, the present result suggests that the recovery enhancing effect of α5 GABAAR NAMs could involve the normalization of L-gamma power of the repairing cortex.
The origin of L-gamma dysfunction in the periphery of the stroke lesion could also be searched in the light of cellular network properties. Gamma oscillations are generated sporadically by local groups of cortical neurons spatially confined into a volume of ∼250 µm in diameter (Tischbirek et al., 2019). According to the interneuron-gamma (ING) and pyramidal-interneuron-gamma (PING) models, parvalbumin interneurons are the key cells of their generation (Bartos et al., 2007; Cardin et al., 2009). Indeed, a loss of function as demonstrated by optogenetic inhibition or selective genetic deletions of PV interneurons results in gamma oscillation suppression (Sohal et al., 2009). It remains to be determined if PV interneurons are selectively vulnerable in the surviving cortex, as suggested in some reports (Guadagno et al., 2008; Baron et al., 2014).
Altogether the present data point out that the chronic excess of ambient GABA in the PIC impairs the production of L-gamma oscillations in the repairing cortex and suggest that rehabilitative interventions aimed at normalizing L-gamma power after stroke could optimize the restitution of lost functions.
Footnotes
Fondation pour la recherche sur les AVC (AVC20180600390).
The authors declare no competing financial interests.
- Correspondence should be addressed to Julien Chuquet at julien.chuquet{at}univ-rouen.fr.
