All animal experiments were performed according to the protocol approved by the Institutional Animal Care and Use Committee of Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences. Adult (90 ± 2 days old) male C57BL/6 mice were used in this study.

The Rosa-CAG-Flag-Kir2.1-2A-tdTomato-WPRE targeting vector (Figure 1A) was designed with a CMV-IE enhancer/chicken-actin/rabbit β-globin hybrid promoter (CAG), FRT site, loxp-flanked STOP cassette (with stop codons in all three reading frames and a triple poly(A) signal), Flag-Kir2.1-2A-tdTomato sequence, woodchuck hepatitis virus post-transcriptional regulatory element (WPRE, to enhance mRNA transcript stability), a poly(A) signal, and attB/attP-flanked PGK-FRT-Neo-poly-A cassette. The entire construct was inserted between exons 1 and 2 of the Gt (ROSA)26Sor locus via electroporation into C57BL/6-derived embryonic stem (ES) cells. Targeted ES cells were selected and injected into C57BL/6 blastocysts, and chimeric animals were bred into C57BL/6 mice to generate Kir2.1loxp/loxp mutant mice. The conditional mutant strain of theKir2.1 mice was developed by crossing Kir2.1loxp/loxp mice with Tg (CaMK2α-Cre/ERT2) mice (Stock number: 012362, Jackson Laboratory, Bar Harbor, Maine, USA). Male adult (90 ± 2 days old of age) Kir2.1 (+) mice and their Kir2.1 (–) littermates were treated with tamoxifen (Tam) dissolved in core oil (100 mg/kg, i.p., once per day for 5 consecutive days). After recovery for 2 days, the animals were used for the following experiments.

The water maze task was performed with a circular tank (120 cm diameter) filled with opaque water (21–23°C) and a hidden platform (6 cm diameter) submerged 1 cm below the surface of the water. The devices (tank and platform) and software [WMT-100] were purchased from Tai Meng Technology Co., Ltd. (Chengdu, China). Before the start of the training trials, the mice were allowed to acclimate to the testing room for 30 min. Mice were trained to find the invisible platform within 70 s on six consecutive days, with four trials per day. If a mouse failed to find the platform within 70 s, it was guided to find the platform and allowed to remain there for 15 s. Escape latency to find the hidden platform, path length, and swim velocity were recorded. After 2 days of rest, the platform was removed, and the mouse was allowed to search for the pool for 70 s (probe tests). The time spent in each quadrant was analyzed. No training means no learning and probe process in MWM, but the animals still swim in the water, while training means an intact MWM experiment. And mice sacrificed at day 6 after probe test.

Locomotion activity was measured in clear boxes measuring 42 cm × 42 cm, outfitted with photo-beam detectors for monitoring horizontal and vertical activity. Data were collected via a PC. Mice were placed in a corner of the open-field apparatus and left to move freely. Mice were not exposed to the chamber before testing. Data were individually recorded for each animal during 30 min.

As described previously (Gilbert et al., 2016; Yang et al., 2018), slices (350 μm) of the hippocampus were obtained from male mice at 90 days of age and were placed in a holding chamber for at least 1 h, and saturated with 95 and 5% CO₂. Artificial cerebrospinal fluid (ACSF, in mM): 124 NaCl, 3 KCl, 1.25 NaH₂PO₄, 2 MgCl₂, 2 CaCl₂, 26 NaHCO₃, and 10 glucose. The internal solution contained (in mM): 140 potassium gluconate, 10 HEPES, 0.2 EGTA, 2 NaCl, 2 MgATP, and 0.3 NaGTP, 5 QX-314was used for sEPSCs or mEPSCs, and 140 CsCl, 10 HEPES, 0.2 EGTA, 2 NaCl, 2 MgATP, and 0.3 NaGTP, 5 QX-314 was used for sIPSCs or mIPSCs. The temperature for the patch clamp recording is maintained at about 30 degrees, the electrode enters the liquid surface and corrects the liquid junction potential, and the cell with a resistance less than 20 MΩ after breaking through the cell membrane is considered to be usable. The current-voltage relations for transfected and control neurons were calculated by recording whole-cell currents under a voltage clamp. Voltage steps in 10 mV increments were applied every 250 ms. To measure the amount of current required to reach the action potential threshold, a series of current steps (2 ms or 500 ms duration, 0–1,000 pA range with 100 pA step increments) were injected into the cell until an action potential was generated. Under voltage clamp, the patched cell was held at −70 mV. A bipolar stimulating electrode (FHC Inc., Bowdoin, ME, USA) was placed in the Schaffer collaterals to deliver the stimuli, and the evoked excitatory postsynaptic currents (EPSCs) or inhibitory postsynaptic currents (IPSCs) were recorded under 30 μM bicuculline (Tocris Bioscience, Bristol, UK) or 10 μM CNQX (Tocris Bioscience, Bristol, UK), respectively. Without TTX (Tocris Bioscience, Bristol, UK), spontaneous EPSCs or spontaneous IPSCs were recorded with an external solution containing 30 μM bicuculline or 10 μM CNQX (Tocris Bioscience, Bristol, UK). Miniature EPSCs or miniature IPSCs were recorded with an external solution containing 1 μM TTX, 30 μM bicuculline (Tocris Bioscience, Bristol, UK), or 10 μM CNQX (Tocris Bioscience, Bristol, UK). The synaptic currents were monitored using an Axon 200 B amplifier (Molecular Devices). Spontaneous events were handpicked and analyzed using the Clampfit 10.2 software (Molecular Devices, Sunnyvale, CA, USA), using template matching and a threshold of 5 pA. All data were acquired at 10 kHz and filtered using a low-pass filter at 2 kHz.

Uncaging of MNI-glutamate, calcium imaging, and whole-cell recordings were performed under a DIC/fluorescence Olympus microscope (FV1000-BX61WI). Slices with a thickness of 300 μm were prepared as described above. The slices were kept at room temperature for at least 1.5 h before being transferred to the recording chamber. Twenty ml of oxygenated magnesium-free ACSF containing 0.2 mM caged glutamate (Tocris Bioscience, Bristol, UK), 1 μM TTX, 30 μM bicuculline, and 10 μM CNQX at 30°C was perfused into the slice recording chamber through a custom-designed flow system driven by pressurized 95% O₂–5% CO₂ at roughly 2 ml/min. Slices were examined under a 20 × objective for proper targeting of tdTomato-expressed CA1 pyramidal neurons. To target whole-cell recordings, cells were visualized at a high magnification (60 × objective, 1.0 NA; LUMPLFLN60XW, Olympus). Kir2.1-positive neurons, which were selected on the basis of their pyramidal somata detected under DIC, were RFP-positive on the monitor. Patch pipettes (3–5 MΩ resistance) made of borosilicate glass were filled with an internal solution containing (in mM): 140 potassium gluconate, 10 HEPES, 2 NaCl, 2 MgATP, and 0.3 aGTP, with 20 μM Alexa594 and 50 μM Fluo-5F. For calcium imaging, the neurons were filled via the patch electrode for 10–20 min before imaging. The dendrites of the recorded neurons were 30–60 μm below the surface of the slice. Laser scanning and photo stimulation were performed using a 60 × objective lens. One laser was tuned to 405 nm (FV5-LD405-2) to activate MNI-glutamate. The second was tuned to 800 nm (Ti: sapphire laser; 100 fs pulses; Mai Tai HP DeepSee, Spectra-Physics) for the excitation of Alexa Fluor-594 and Fluo-5F. Calcium signals were used to locate dendritic spines 30–80 μm from the soma. The selected dendrite was scanned repeatedly with a series of 200 scans (scan dimensions: 30 μm × 30 μm, 5 μm × 5 μm; 100 ms per scan). The uncaging laser spot was located ∼1 μm away from the spine in Figure 1E. In the first scan, acquisition of NMDAR currents (Axon200B and FV10-ANALOG in our BX61WI microscope) and calcium signals began, and in the 10th frame, uncaging occurred with a constant area 0.25 μm². Glutamate was uncaged using 5–30 ms pulses to generate the I-O curve of the laser strength calcium-NMDAR current under voltage clamp (Vh = −70 mV). To ensure that uncaging was performed in a constant location, image drift was corrected before each line scan acquisition by collecting a frame scan and calculating the cross-correlation to a reference image region of interest (ROI) analysis from these scans, as explained below. Fluorescence changes were analyzed off-line with laboratory-written software using the IGOR-Pro programming environment (Wavemetrics). To study the time course and amplitude of Ca²⁺ rise, the average fluorescence was measured in small “ROIs” (2.25-4 μm²) and converted to the percentage change in fluorescence: Δ⁢FF⁢o=100×(F-F⁢r)/(F⁢r-B), where F is the measured fluorescence signal at any given time, Fr is the average fluorescence from the scans preceding the stimuli, and B is the average value of the background fluorescence in the scanned field that does not contain any part of the dye-filled cell.

Total RNA from the hippocampus and cortex was purified using the RNeasy Plus Micro Kit (Qiagen, Hilden, Germany). RNA quality (RIN 8–9) and quantity were analyzed on a 2,100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA) using RNA 6,000 Pico chips; ds-cDNA was produced using the Ovation RNA-seq system V2 (NuGEN, San Carlos, CA, USA) and fragmented using a Covaris S-Series System (Covaris, Woburn, MA, USA). DNA fragments in the 150–300 bp size range were recovered to construct a sequencing library using the Encore NGS Library System I (NuGEN, San Carlos, CA, USA) for 100 bp paired-end RNA-seq using the Illumina HiSeq-2500 sequencer. Raw data were processed using the cutadapt (Martin, 2011) software to remove joints and filter low-quality reads, and 77–121 M clean reads were obtained for each sample with an average Q30 > 95%. RNA-sequencing data alignment and differential gene expression analysis were performed using Tophat2 (Kim et al., 2013) and edgeR (Robinson et al., 2010). Specifically, RNA-sequencing data were aligned to the reference genome (mm10 NCBI buid 38.1) using Tophat2, using default parameters. Uniquely mapped and properly paired alignments were used for further analyses. FeatureCounts (Liao et al., 2013) were used to count the number of reads mapped to each gene. Prior to differential gene expression analysis, for each sequenced library, the read counts were adjusted using the edgeR program package through one scaling normalized factor. Differential expression analysis under the two conditions was performed using the edgeR R package with bcv = 0.1. P-values were adjusted using the Benjamini and Hochberg method. A corrected P-value of 0.05 and absolute fold change of 2 were set as the thresholds for significantly differential expression. Sample cluster analysis of all detected genes was performed using the pheatmap R package.¹ Gene Ontology (GO) enrichment analysis of differentially expressed genes was performed using the clusterProfiler R package (Yu et al., 2012; Wu et al., 2021), in which the gene length bias was corrected.

As described previously (Shi et al., 2021; Tian et al., 2021), brain tissue was perfused and fixed with 4% paraformaldehyde (PFA) solution, dehydrated with 30% sucrose, and frozen to obtain 30 μm sagittal sections with a cryostat microtome (Leica CM1950). Fluorescence images were obtained using a FV1000-BX61WI microscope.

Statistical analysis of the transcriptomics can be found in the RNA-sequencing and data analysis sections. Other statistical analyses were performed using Graphpad Prism version 8. The amount of data collected is described in the corresponding figure legend. The statistical variables in Figures 1–3 are all continuous variables, described using mean and standard deviation, represented as mean ± SEM. In Figures 1, 2, one way ANOVA were used to compare the differences between the Tam (+) group and the other three groups. In Figures 3A, B, an independent samples t-test was used to compare the differences between the Kir2.1 (+) and Kir2.1 (-) groups. One-way analysis of variance (ANOVA) was used in Figures 3C, D, and the Bonferroni correction was used for multiple comparisons between different groups. The description of the significance levels of p-values is also given in the corresponding figure legends.

To study the homeostatic regulation of E-I balance in the forebrain, we genetically suppressed forebrain excitatory neural activity by conditional expression of the inward rectifying potassium channel. Targeting strategies and corresponding genotypic assays are described in detail in the (Section “Materials and Methods). Adult mice received 5 days of tamoxifen (Tam) or vehicle (Veh) treatment, showing successful expression of Kir2.1 channels by morphological and functional confirmation (Figure 1, A–F). First, we showed the presence of a tdTomato-tagged Kir2.1 protein in the forebrain excitatory neurons (Figure 1A). Second, The Kir2.1 currents were plotted against the holding potentials in CA1 pyramidal neurons from Kir2.1 (+) and Kir2.1 (–) mice. The (+)/Tam group mice exhibited large inward rectifying potassium currents (green) recorded from a 250 ms voltage clamp from −160 to −10 mV with Kir2.1 overexpression compared to other groups (Figure 1B). Third, the (+)/Tam group mice had an increased threshold for action potential firing with a 2 ms current step injection compared to the other groups (Figure 1C). The CA1 pyramidal neurons in the Kir2.1 (+) mice required more currents to elicit firing with a 2 ms current step and less firing probabilities in vitro with 500 ms current step injection in the (+)/Tam group mice than in others (Figure 1D). Finally, the (+)/Tam group mice showed less calcium influx in Kir2.1 (+) neurons (Figures 1E, F) compared to others with MNI-glutamate uncaging (>10 ms) under a two-photon microscope. Together, these data demonstrate that excitatory neural activity in the forebrain is selectively inhibited by Kir2.1 overexpression.

To study the changes in E-I homeostatic balance after inhibition of forebrain excitatory neural activity, we evaluated the changes in spontaneous and miniature neural transmission in the hippocampus, which is one of the most important areas of the brain for spatial cognition. We observed a reduction in spontaneous EPSCs, but not spontaneous IPSCs in the hippocampal principal neurons (Figures 2A, B). However, the frequency of both miniature mEPSCs and mIPSCs decreased after Kir2.1 overexpression in forebrain pyramidal neurons, and the amplitude of mIPSCs decreased after Kir2.1 overexpression in the forebrain (Figures 2C, D). Furthermore, both evoked IPSCs and evoked EPSCs were reduced after Kir2.1 overexpression in the forebrain of Kir2.1 (+)/Tam group mice compared to those of the other groups, but the E-I currents remained balanced (Figure 2E). These data indicate that reduction in GABA transmission plays an important role in compensating for the E-I imbalance in CA1 elicited by pyramidal neural inhibition.

In an ever-changing world, experiences regulate E-I homeostatic balance, and rebalanced E-I networks are often thought to optimize learning and memory (Mizumori and Jo, 2013; Huber, 2018; Girardeau and Lopes-dos-Santos, 2021). To study the association between E-I homeostatic balance and cognition, we evaluated the cognitive level in Kir2.1 (+) mice and tested the changes in miniature postsynaptic currents in the hippocampus with or without spatial training. As many of the previous tests showed no differences in neuronal activity-related parameters among the negative control groups, we only examined the parameter changes between Kir2.1 mice with (Kir2.1 (+)) or without tamoxifen (Kir2.1 (–)) induction treatment in the follow-up experiments. First, we showed that compared with Kir2.1 (–) mice, Kir2.1 (+) mice spent more time reaching the escape platform during the 5-day training session (Figure 3A) and less time in the target quadrant during probe testing (Figure 3B), indicating learning and memory impairments in Kir2.1 (+) mice, while no obvious difference in swimming speed in MWM and travel distance in the open field was found between Kir2.1 (–) and Kir2.1 (+) mice (Supplementary Figure 1). Next, we scarified them and compared the miniature postsynaptic currents in trained mice or untrained Kir2.1 (+) and Kir2.1 (–) animals. We found that both excitatory and inhibitory transmission increased after 5-day spatial training in the control group Kir2.1 (–) mice (Figure 3C). However, only GABA transmission, that is, the amplitude of hippocampal mIPSCs, was enhanced after the 5-day spatial training in Kir2.1 (+) mice (Figure 3D). These data indicate that homeostatic modulation of forebrain E-I balance via GABA transmission may be related to cognitive level and experience.

The changes of mIPSC amplitude could be correlated to reduction in inhibitory/GABA receptors in response to cognitive experiences and E-I balance, we determined the underlying genome-wide transcriptome difference in the cortex and hippocampus of trained or untrained Kir2.1 (+)/Kir2.1 (-) mice via RNA sequencing. Details of sample grouping and naming, as well as sequencing data, can be found in Figure 4A and Supplementary Table 1; 77–121 M clean reads were obtained for each sample, with an average Q30 > 95%. The count expression matrix for all samples is shown in Supplementary Table 2. Unsupervised clustering of all gene expression profiles between samples showed that spatial training induced more severe transcriptome perturbation in the hippocampus than the overexpression of Kir2.1 in the cortex (Figure 4B). This indicates that spatial training has a more direct effect on the hippocampus, while Kir2.1 overexpression has a more direct effect on the cortex. A detailed list of differentially expressed genes (DEGs) in each group is shown in Supplementary Table 3. From the results of the GO analysis of DEGs caused by Kir2.1 overexpression, we found that the Kcnj2 gene (Kcnj2 is the gene symbol name of Kir2.1) was significantly upregulated, this was expected given the fact that they are overexpressing Kir. Hence it served as a control. However, GABA and glycine receptors were significantly downregulated in both the cortex and hippocampus (Figure 4C). However, in the spatial training group, GABA and glycine receptors were significantly upregulated in both the cortex and hippocampus (Figure 4D), particularly in the Kir2.1 (+) group (Figure 4E). This indicates that the transcription mechanism for inhibitory transmission or chloride homeostasis is an important strategy for homeostatic control of the E-I balance in the forebrain network and also suggests many gene candidates for cognition-related homeostasis in the forebrain.