E4BP4 helps regulate daily brain rhythms that affect thinking during delirium

Mice were treated with lipopolysaccharide (LPS) plus midazolam (named LM treatment) or subjected to a simple laparotomy under isoflurane anesthesia (named AS treatment) to induce delirium‐like syndrome (Figure1A).^[27, 28, 29^] Delirium‐associated cognitive deficits were assessed based on behavioral tests with novel object recognition (NOR) and Y maze. As expected, mice with LM or AS treatment showed reduced novel object preference in the NOR test and decreased spontaneous alternation in Y maze test, indicating impairment of cognitive function (Figure 1B,C; Figure S1, Supporting Information). Based on an open field test (OFT), they also had a reduced locomotor activity (decreased total distance), reflecting of delirium‐related behavioral disruption, such as lethargy and anxiety (Figure 1D,E). It was noted that the behavioral deficits recovered 6 days after LM treatment and 12 days after AS treatment (Figure S2, Supporting Information). We next tested whether circadian rhythms are affected in the delirium conditions. Mice with LM‐ and AS‐induced delirium showed attenuated circadian rhythms in wheel‐running activities with an increased activity in the subjective daytime and a reduced activity in the subjective nighttime (Figure 1F,G; Figure S3A, Supporting Information). The activity amplitudes of delirious mice were reduced to 15–34% of normal mice (Figure 1G). Moreover, circadian sleep–wake behavior was disrupted in delirious mice (Figure 1H,I; Figure S3B, Supporting Information). Delirious mice showed shortened wake time and prolonged sleep (i.e., NREM and REM) time in the dark phase (Figure 1I), and had altered sleep architecture with increased sleep fragmentation in the dark phase (Figure 1J). Supporting this, the circadian pattern of plasma corticosterone, a sleep‐related hormone, was disrupted in delirious mice (Figure S3C, Supporting Information). As the hippocampus has a major regulatory role in cognitive function, we examined circadian expression of clock genes in delirious hippocampus. Delirium resulted in overall disruption of core clock genes in mouse hippocampus, including E4bp4, Bmal1, Rev‐erbα, Cry1, and Per2 (Figure 1K; Figure S3D, Supporting Information). Of note, E4bp4 was one of the genes altered the most and its amplitude was reduced to 10–20% of the control (Figure 1K). In addition, we observed disrupted rhythms in E4BP4, BMAL1, and REV‐ERBα in the whole blood of patients with delirium (Figure S3E,F, Supporting Information). Altogether, delirium is associated with disrupted circadian rhythms and clock dysfunction in the hippocampus.

Given that delirium is associated with disrupted circadian rhythms, it was of interest to investigate a potential role of circadian rhythms in the development of delirium. To this end, jet‐lagged mice (a model of physiological disruption of circadian rhythms) were established using a published protocol (i.e., 8 h light advance every 2 days for 10 days, Figure2A),^[30^] and subjected to delirium assessment (on day 12). Jet lag resulted in perturbance of clock genes such as E4bp4, Bmal1, and Rev‐erbα in mouse hippocampus (Figure 2B). Intriguingly, jet‐lagged mice showed delirium‐like syndrome as evidenced by reduced novel object preference in the NOR test and decreased spontaneous alternation in Y maze test (Figure 2C,D; Figure S4, Supporting Information). We further tested the effects of constant lighting (for 14 days, another model of physiological circadian disruption) on delirium development. Constant lighting led to impaired hippocampus‐dependent cognitive performance as measured by the NOR and Y maze tests (on day 15) (Figure 2E,F; Figure S4, Supporting Information). Moreover, we examined the effects of Per2 deletion (a genetic model of circadian disruption) on the development of delirium. Per2‐deficient (Per2^−/−) mice showed an increased susceptibility to cognitive impairment according to the NOR and Y maze tests (Figure 2G,H; Figure S4, Supporting Information). Taken together, physiological and genetic circadian disruptions sensitize mice to delirium‐like syndrome with respect to cognitive dysfunction.

As shown above (Figure 1K; Figure S3E,F, Supporting Information), E4bp4 expression was severely compromised in the hippocampus of delirious mice and in the whole blood of humans with delirium. On the other hand, we observed marked disruption of E4bp4 in the hippocampus of jet‐lagged mice (Figure 2B). It was thus hypothesized that E4bp4 has a critical role in connecting delirium to circadian clock. To test this hypothesis, we examined the regulatory effects of E4bp4 on delirium using mice with global deletion of E4bp4 (E4bp4^−/− mice). E4bp4^−/− mice showed exacerbated delirium‐associated cognitive decline after LM or AS induction, as evidenced by a slower recovery from behavioral deficits based on NOR and Y maze tests (Figure3A,B; Figure S5, Supporting Information). Furthermore, application of SR8278, a chemical that can increase E4BP4 expression in the hippocampus by antagonizing its repressor REV‐ERBα,^[31^] rescued LM‐induced delirium‐associated cognitive deficits (Figure 3C–E). These findings indicated a regulatory role of E4bp4 in the development of delirium in mice. As E4bp4 is rhythmically expressed in mouse hippocampus with higher expression in the subjective nighttime (Figures 1K and 2B), it was of interest to test circadian time‐varying severity in delirium symptoms. The severity of delirium (or the extent of cognitive impairment, negatively correlated with the novel object preference in the NOR test and with spontaneous alternation in Y maze test) was assessed after LM or AS induction at six different circadian time points (i.e, CT2, CT6, CT10, CT14, CT18, and CT22). Delirium severity displayed a robust circadian rhythm (most severe at CT6 and least severe at CT18) (Figure 3F,G). However, the circadian time‐dependency of delirium severity was lost in E4bp4^−/− mice (Figure 3H,I). Altogether, E4bp4 regulates the development of delirium with respect to cognitive impairment, accounting for circadian rhythm in the disease severity.

To understand the pathological basis underlying the delirium‐associated cognitive impairment, we performed transcriptomic analyses of the hippocampus from delirious mice (induced by LM) and controls collected at CT6 when the disease is the most severe (Figure 3). A total of 1032 differentially expressed genes (DEGs) were identified between delirious and control hippocampus (Figure4A). Of these DEGs, 728 genes were up‐regulated and 302 genes down‐regulated (Figure 4B). Gene ontology (GO) enrichment analyses suggested an association of delirium with activation of immune and inflammatory responses (Figure 4C). As microglia and astrocytes are key regulators of neuroinflammation,^[32, 33^] we wondered whether they are affected in delirious mice. LM‐ and AS‐induced delirium led to microglial activation in the hippocampus as evidenced by up‐regulation of microglial activation‐related genes (such as Il6ra, Tyrobp, Cd68, Aif1, Csf1r, and Trem2) and of relevant inflammatory factors (such as Il‐1β, Tnfα, Ccl2, Ccl5, and Ccl8) (Figure 4D; Figure S6A, Supporting Information). This was supported by an increased number of Iba1⁺ cells according to immunofluorescent staining (Figure 4E; Figure S6B,C, Supporting Information). By contrast, astrocytes were unaffected in the hippocampus of delirious mice as illustrated by no significant changes in astrocyte activation markers (such as Gfap and Timp1) or in the number of GFAP⁺ cells (Figure S6D–F, Supporting Information). Given that the delirious condition was associated with microglial activation, we further tested whether microglia have a role in delirium‐associated cognitive impairment. To this end, microglia were depleted by feeding mice (wild‐type) with PLX3397 for 14 days and cognitive function was assessed after delirium induction. Unsurprisingly, mice treated with PLX3397 showed a marked loss of microglia (Figure 4F; Figure S6G, Supporting Information). We found no significant inflammation in the hippocampus of PLX3397‐treated mice after LM induction (i.e., no changes in inflammatory factors such as Tnfα, Il‐1β, Il‐6 and Ccl2; Figure S6H,I, Supporting Information). Intriguingly, microglia depletion protected mice from cognitive decline associated with LM‐induced delirium, as evidenced by their normal cognitive behaviors (i.e., novel object preference and spontaneous alternation) in the NOR and Y maze tests (Figure 4G,H). Activation of microglia has been demonstrated in the literature to impair the hippocampal long‐term potentiation (LTP), a neural basis of cognitive function.^[34^] We found that LTP was attenuated in the hippocampus of delirious mice (Figure 4I,J). However, microglia depletion by PLX3397 prevented delirium‐induced LTP impairment in mice (Figure 4I,J). Altogether, these findings indicated that microglial activation in the hippocampus contributes to delirium‐associated cognitive impairment in mice through impairing hippocampal LTP.

As microglial activation is involved in pathogenesis of delirium and E4bp4 has a potential role in immune responses,^[19, 20^] we next examined whether E4bp4 regulates delirium via modulation of microglial activation. As noted before, E4bp4^−/− mice fed a normal chow diet showed exacerbated delirium‐associated cognitive impairment (Figure 3A,B). This was accompanied by microglial activation in the hippocampus as the levels of functional markers (e.g., Aif1, Csf1r, and Trem2) were elevated and the number of Iba1⁺ cells were increased after delirium induction (Figure5A–D; Figure S7, Supporting Information). However, E4bp4^−/− mice fed a chow diet containing PLX3397 (used to deplete microglia) were deficient in microglial activation and lack of inflammatory responses in the hippocampus (Figure S8, Supporting Information), and were protected from delirium‐associated cognitive impairment (Figure 5E,F). These findings suggested involvement of microglial activation in E4bp4 regulation of cognitive dysfunction in delirium‐like syndrome. Supporting this, mice with E4bp4 specifically deleted in microglia (generated by breeding mice carrying a conditional E4bp4 allele with mice expressing Cx3cr1^CreER, named microglia^E4bp4‐KO mice^[35^]) showed aggravated delirium‐associated cognitive decline, as evidenced by a slower recovery based on the NOR and Y maze tests (Figure 5G,H). In addition, the circadian time‐dependency of delirium severity was lost in microglia^E4bp4‐KO mice (Figure 5I,J). In contrast, mice lacking in E4bp4 in hippocampal CA1 neurons (E4bp4^cKO mice) had a normal phenotype (Figure S9, Supporting Information). Moreover, E4bp4 showed inhibitory effects on LPS‐induced inflammation in a microglial cell line (BV2 cells) as it decreased Tnfα, Il‐1β, Il‐6 and Ccl2 mRNAs as well as TNFα and IL‐1β proteins (Figure 5K,L). Prior studies suggest that neuronal loss, astroglial activation, down‐regulation of brain‐derived neurotrophic factor (BDNF), disruption of blood–brain barrier, and brain hypoxia contribute to delirium pathophysiology.^[36, 37, 38, 39, 40^] We tested whether these factors are involved in E4bp4 regulation of delirium. The numbers of neurons and astrocytes did not differ between E4bp4^−/− and control mice (Figure S10A,B, Supporting Information). We found no changes either in BDNF and its receptor TrkB, tight junction‐associated proteins (ZO‐1 and occludin), as well as Hif1α (hypoxia‐induced factor) and its target genes such as Phd2 (oxygen‐sensing prolyl hydroxylase domain protein 2) and Vegfa (vascular endothelial growth factor A) (Figure S10C–E). Taken together, microglial activation is likely the main mechanism for regulation of delirium‐associated cognitive decline by E4bp4.

To explore the mechanisms by which E4bp4 regulates microglial activation and inflammation in the hippocampus, we performed single‐cell RNA‐sequencing of hippocampus tissues from E4bp4^−/− and control mice (Figure6A). A total of 22 259 cells were sequenced based on 10× Genomics libraries of E4bp4^−/− and control hippocampus, of which 14 803 cells passed the quality filters and the average number of detected genes per cells was about 1800 (Figure S11A–D, Supporting Information). 25 clusters were identified and grouped into six distinct cell types (i.e., microglia, astrocytes, oligodendrocytes, endotheliocytes, neurons, and pericytes) based on known cell identity markers (Figure 6B; Figure S11E, Supporting Information).^[41^] Each of the cell clusters from E4bp4^−/− and control libraries had a similar proportion of cells, suggesting that E4bp4 ablation did not affect the cell types (Figure 6C; Figure S11F, Supporting Information). Notably, microglia were one of the most abundant cell populations, present in clusters 0, 2, 6, 18, and 22 (Figure 6B). Thus, we analyzed the DEGs in hippocampal microglia between E4bp4^−/− and control mice. Of 52 DEGs, the inflammatory factors such as Ccl2 and Ccl4 as well as the microglial activation transcript Csf1 were significantly increased, suggesting microglial activation in the gene knockout mice (Figure 6D). Supporting this, qPCR analyses and immunofluorescent staining demonstrated increases in microglial activation transcripts and in Iba1⁺ cell number in the hippocampus of E4bp4^−/− mice (Figure S12, Supporting Information). According to GO enrichment analyses, E4bp4‐associated DEGs converged on ERK1/2 signaling and related pathways (such as ERK1/2 cascade, regulation of ERK1/2 cascade, negative regulation of MAPK cascade, and negative regulation of protein phosphorylation), suggesting a potential role of ERK1/2 signaling in E4bp4 regulation of hippocampal microglial activation (Figure 6E). In fact, it has been established that ERK1/2 signaling pathway contributes to microglial activation and ensuing inflammatory responses.^[42, 43^] The underlying mechanism involves activation of NF‐κB signaling, an essential regulator of microglial activation and neuroinflammation.^[44, 45^]

We next examined whether E4bp4 does regulate ERK1/2 signaling pathway in the hippocampus. E4bp4^−/− mice after delirium induction showed increases in total phosphorylated proteins of ERK1/2 and p65 (an NF‐κB subunit and a downstream target of ERK1/2) in the hippocampus (Figure7A,B), and an elevation in nuclear phosphorylated ERK1/2 and p65 (Figure 7C). Furthermore, E4bp4 ablation led to an increase in phosphorylation of both ERK1/2 and p65 in BV2 cells and primary mouse hippocampal microglia (Figure 7D). Consistently, E4bp4 overexpression resulted in reduced phosphorylation of ERK1/2 and p65 in the cells (Figure 7E). As expected, U0126, an ERK phosphorylation inhibitor, markedly alleviated LPS (an NF‐κB activator)‐induced inflammation (measured by TNFα and IL‐1β levels) in BV2 cells and primary mouse hippocampal microglia (Figure 7F,G). Intriguingly, in the presence of U0126, the effects of E4bp4 ablation on LPS‐induced inflammation were attenuated in BV2 cells and primary mouse hippocampal microglia (Figure 7H,I). Additionally, SR8278 (a chemical that can increase E4BP4 expression, Figure 3C) decreased the levels of phosphorylated ERK1/2 and p65 in mouse hippocampus, accompanied by alleviated inflammation (Figure 7J). Altogether, these findings indicated that E4bp4 regulates ERK1/2 signaling pathway and inflammation in the hippocampus.

Next, we investigated the mechanisms by which E4bp4 regulates ERK1/2 signaling pathway in the hippocampus. We first tested whether E4BP4 affects the phosphorylation of MEK1/2, the MAPKK kinases that phosphorylate and activate ERK1/2.^[46^] We observed unchanged phosphorylation of MEK1/2 in the hippocampus of E4bp4^−/− mice (Figure S13A, Supporting Information). Autophagy proteins as cellular scaffolds play a regulatory role in ERK phosphorylation.^[47^] We found no changes in hippocampal expression of autophagy proteins such as ATG7 and LC3 in E4bp4^−/− mice (Figure S13B, Supporting Information). Thus, the roles of MEK1/2 and autophagy proteins in E4bp4 regulation of ERK1/2 can be excluded. As E4bp4 is known as a transcriptional repressor, we further examined whether E4BP4 regulates the transcription and expression of Mapk1 and Mapk3 (two genes encoding ERK2 and ERK1 protein, respectively). Loss of E4bp4 in mice (without delirium induction) significantly increased Mapk1/3 mRNAs and ERK1/2 proteins in the hippocampus (Figure8A). Similar alterations in both Mapk1/3 mRNAs and ERK1/2 proteins were observed in E4bp4^−/− mice with delirium (Figure 7A,B; Figure S13C, Supporting Information). Notably, other genes involved in ERK1/2 cascade, including Tlr4, Braf, Raf, Rem1, Kras, Hras1 and Map2k1, were unaffected in E4bp4 knockout mice (Figure 8A; Figure S13C, Supporting Information). Moreover, knockdown of E4bp4 (by a specific siRNA) led to increases in Mapk1/3 mRNAs and ERK1/2 levels in BV2 cells (Figure 8B), whereas overexpression of E4bp4 resulted in reduced expression of Mapk1/3 and their proteins (Figure 8C). These data support a negative role of E4BP4 in regulation of ERK1/2 expression in hippocampal microglia.

Parallel changes in Mapk1/3 mRNAs and in their proteins suggested a transcriptional mechanism for E4BP4 regulation of ERK1/2. In luciferase reporter assays, E4BP4 dose‐dependently inhibited the promoter activities of both Mapk1 and Mapk3 (Figure 8D). Sequence analysis revealed two potential D‐box (a putative DNA motif for E4BP4 binding and function) elements in Mapk1 promoter and three potential D‐boxes in Mapk3 promoter (Figure 8E). Furthermore, truncation and mutation assays demonstrated that −1142 bp D‐box in the Mapk1 and −1865 bp D‐box in the Mapk3 promoter were responsible for E4BP4 effects (Figure 8E). According to chromatin immunoprecipitation (ChIP) assays, E4BP4 protein was recruited to the −1142 bp D‐box of Mapk1 and to the −1865 bp D‐box of Mapk3 in mouse hippocampus (Figure 8F). However, recruitment of E4BP4 to Mapk1/3 ceased to exist in E4bp4‐deficient hippocampus (Figure 8F). Taken together, E4BP4 trans‐repressed Mapk1 and Mapk3 via direct binding to a D‐box element in their promoters.

LPS and granulocyte‐macrophage colony‐stimulating factor (GM‐CSF) were purchased from Sigma–Aldrich (St. Louis, MO). U0126 was purchased from MCE (Monmouth Junction, NJ). Midazolam was obtained from the First Affiliated Hospital of Jinan University (Guangzhou, China). PLX3397 chow (290 mg kg⁻¹, formulated in AIN‐76A) was obtained from Plexxikon Inc. (Berkeley, CA). Mapk1 luciferase reporters (−1977/+123 bp and two D‐box‐mutated versions) and Mapk3 luciferase reporters (−1990/+100 bp and three D‐box‐mutated versions) were obtained from Tsingke Biotech (Beijing, China). pRL‐TK, pcDNA3.1, pcDNA3.1‐E4bp4, siE4bp4 (siRNA targeting E4bp4), and a negative control for siRNAs (sequences provided in Table S1, Supporting Information) were obtained from Transheep Technologies (Shanghai, China).

E4bp4^−/− and Per2^−/− (C57BL/6 background) mice had been established in the authors’ laboratory.^[62, 63^] Microglia^E4bp4‐KO mice were generated by breeding mice carrying a conditional E4bp4 allele (i.e., exon 2 floxed allele, E4bp4^fl/fl) mice with mice expressing Cx3cr1^CreER. Genotyping of mice harboring the conditional E4bp4 allele was performed with primers F: 5’‐TCAAAGTGGAGGCTTTGGAC‐3’ and R: 5’‐CACAAGGACACCCAGACAGA‐3’. Genotyping of mice expressing Cx3cr1^CreER was performed with primers Cx3cr1^CreER‐F: 5’‐AAG ACTCACGTGGACCTGCT‐3’, Cx3cr1^CreER‐R1: 5’‐CGG TTATTCAACTTGCAC CA‐3’ and Cx3cr1^CreER‐R2: 5’‐GG ATGTTGACTTCCGAG TTG‐3’. Mice were treated with tamoxifen (i.p., 20 mg mL⁻¹) once daily for 3 days to delete the E4bp4 allele via cre‐mediated recombination.^[35^] Mice with E4bp4 specifically deleted in the CA1 pyramidal cell layer of the hippocampus (E4bp4^cKO) mice were generated by breeding E4bp4^fl/fl mice with mice expressing Camk2a‐iCre. Genotyping of mice expressing Camk2a‐iCre was performed with primers Camk2a‐iCre‐F: 5’‐AGGGTATTGCCTTAGTAGAGGGGCAT‐3’, Camk2a‐iCre‐R1: 5’‐GGTGGCAATGGTAGGGTGATCTGAC‐3,’ and Camk2a‐iCre‐R2: 5’‐GCACACAGACAGGAGCATCTTC‐3’. Mice were housed in a specific pathogen‐free facility under a 12 h light/dark cycle (unless otherwise specified) with free access to water and food. Male mice (3–4 months) were used for experiments, and experimental procedures were approved by the Institutional Animal Care and Use Committee of Guangzhou University of Chinese Medicine (ZYD‐2020‐145).

Whole blood samples were collected from five ICU patients with delirium, five ICU patients without delirium, and five healthy volunteers at different times of the day (i.e., 02:00 AM, 06:00 AM, 10:00 AM, 14:00 PM, 18: 00 PM, and 22:00 PM). Samples were subjected to RNA extraction and qPCR analyses. The study protocol was approved by the Institutional Review Board of the First Affiliated Hospital of Jinan University (KY‐2020‐003). An informed consent was obtained from every participant prior to the study.

Drug‐induced delirium: Knockout (E4bp4^−/−, microglia^E4bp4‐KO, and E4bp4^cKO) mice and controls were treated with LPS (200 µg per kg) plus midazolam (10 mg per kg, i.p., named LM treatment) to induce delirium‐like syndrome as previously described.^[27^] 24 h, 72 h, or 6 days (time points selected based on pilot experiments evaluating the duration of induced delirium) later, mice were subjected to behavioral testing.^[27^] In another set of experiments (for circadian expression profiling), mice were kept under a constant dark condition for 2 days and then treated with LM. 24 h later, mice were sacrificed every 4 circadian hours to collect plasma and hippocampus samples, followed by corticosterone measurement (by using an ELISA kit, Meimian Biotechnology, Jiangsu, China), qPCR, and/or Western blotting analyses.

Anesthesia/surgery‐induced delirium: Mice were subjected to a simple laparotomy under anesthesia (named AS treatment) as previously described.^[28, 29^] In brief, mice were anesthetized and a longitudinal midline incision was made through the skin, abdominal muscles, and peritoneum from the xiphoid to 0.5 cm proximal to the pubic symphysis. Abdominal organs were partially exposed for 2 min and the incision was sutured with 5‐0 Vicryl thread. 1, 3, 6, or 12 days (time points selected based on pilot experiments evaluating the duration of induced delirium) later, mice were subjected to behavioral testing. In another set of experiments (for circadian expression profiling), mice were treated with AS and then kept under a constant dark condition for 2 days. On the next day, mice were sacrificed every 4 circadian hours to collect plasma and hippocampus samples, followed by corticosterone measurement, qPCR, and/or Western blotting analyses.

Jet‐lagged mice were established according to a published method.^[30^] In brief, mice were subjected to a jet lag schedule of 8 h light advance of the light/dark cycle every 2 days for 10 days. Control mice were kept under a standard 12 h light/dark cycle. All mice were then transferred to constant darkness. On the next day, mice were subjected to behavioral testing. After another 24 h, mice were sacrificed to collect hippocampus tissues for qPCR analyses.

Mice were housed in individual cages equipped with running wheels (Lafayette Instrument, Lafayette, IN) in light‐tight ventilated chambers. After acclimation to the system for 7 days under a 12 h light/dark cycle, mice were treated with LM or AS, and transferred to constant darkness for continuous recording (3–5 days). Data (day 2 after LM or AS treatment) were collected and analyzed using the ClockLab software (Actimetrics, Wilmette, IL).

Mice were anesthetized and mounted in a stereotaxic apparatus. Screw electrodes were inserted into the skull of mice to measure cortical EEG using the following coordinates: +2 mm Bregma, +1 mm midline for first recording electrode; +2 mm Bregma, −1 mm midline for second recording electrode; −2 mm Bregma, −1 mm midline for a reference electrode; and −2 mm Bregma, +1 mm midline for a ground electrode. Stainless steel electrodes were implanted in dorsal neck muscle to measure EMG. After recovery for 6 days, mice (maintained under a 12 h light/dark cycle) were treated with LM. On the next day, mice were subjected to EEG and EMG recordings. In another set of experiments, mice were treated with AS. On day 3, mice were subjected to EEG and EMG recordings. Data were acquired using a tethered data acquisition system with a resolution of 500 Hz (Medusa, Biosignal technologies, Nangjing, China), waveforms visualized using Sirenia Sleep Pro software (Pinnacle technologies, Lawrence, KS). The EEG signals were high‐pass filtered (>0.5 Hz) using a digital filter and EMG was band‐pass filtered between 5 and 45 Hz. Power in the δ (0.5–4 Hz), θ (5–8 Hz), α (9–14 Hz) bands, and θ to δ band ratio were calculated, and EMG signal was scored in 4 s epochs. All these data were used to define the vigilance states of wake, NREM (non‐rapid eye movement), and REM (rapid eye movement) sleep by an automatic script (Lunion Stage software, LunionData, Nangjing, China).

Mice were anesthetized, placed in a stereotaxic frame, and treated with SR8278 or vehicle via microinjections into the hippocampus bilaterally (anteroposterior: −2.9 mm from the bregma, ventral: −3.0 mm below the dura, and lateral: ±3.0 mm from the midline). After 12 h, mice were treated with LM. 24 h later, mice were subjected to behavioral testing.

Hippocampal slices (300 µm thick) were prepared from mouse brain, incubated at 32 °C for 30 min and maintained at 26 °C for 1 h as described previously.^[64^] After recovery, slices were placed in a recording chamber at 25 °C and perfused with oxygenated artificial cerebrospinal fluid (ACSF) containing 126 m m NaCl, 2.5 m m KCl, 1.25 m m NaH₂PO₄, 26 m m NaHCO₃, 10 m m glucose, 2 m m CaCl₂, and 2 m m MgSO₄ at a rate of 1 mL min⁻¹. Extracellular field EPSPs (fEPSPs) were recorded from the CA1 area using a glass electrode filled with ACSF (2–3 MΩ). Schaffer collateral pathway was stimulated every 30 s using concentric bipolar electrodes. A theta burst stimulation (TBS) protocol (four pulses of 100 Hz repeated three times at 5 Hz, and a 20 s inter‐train interval) was used to induce LTP.^[64, 65^] Field potentials were amplified, low‐pass filtered (MultiClamp 700B, Axon Instruments), digitized, and data were analyzed using the Clampex software (Axon Instruments).

Microglia were depleted by feeding mice a chow diet containing PLX3397 for 14 days as previously described.^[66, 67^] Mice were then subjected to LM treatment. After 24 h, behavioral testing was performed and hippocampus samples were collected.

NOR, Y maze, and OFT tests were performed with mice as previously described^[68, 69^] and animal behaviors were recorded by using a SMART video tracking system and analyzed with a SMART 3.0 software (Panlab Harvard Apparatus, Barcelona, Spain). In the NOR test, object preference% = the number of novel or old object exploration/the total number of object exploration. In the Y maze test, spontaneous alternation was calculated as the number of actual alternations divided by the maximum number of alternations.

Brain samples were fixed, paraffin‐embedded, and cut into 20 µm coronal sections. Sections were blocked with 5% bovine serum albumin in phosphate‐buffered saline (PBS) containing 0.1% Triton X‐100, and then incubated with anti‐Iba1, anti‐GFAP, or anti‐NeuN antibody (Table, Supporting Information). After washing with PBS, sections were incubated with a fluorescent secondary antibody and DAPI (4’,6‐diamidino‐2‐phenylindole). Images were captured using a Nikon Optiphot fluorescent microscope (Tokyo, Japan). S3

Primary mouse microglia were isolated from newborn mice as previously described.^[32, 70^] The newborn mice were sacrificed and hippocampus was dissected and incubated in a digestion buffer (Dulbecco's modified Eagle's medium (DMEM) containing 8 u mL^‐1 papain and 125 u mL^‐1 DNase) at 37 °C for 20 min. The dissociated cells were collected and resuspended in DMEM with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin, and 5 ng mL⁻¹ GM‐CSF. Cells were then seeded into poly‐Dlysine (PDL)‐coated flasks. Flasks were shaken at 200 rpm at 37 °C for 2 h. Cell suspension was centrifuged for 10 min at 200 g and the pellets (microglia) were collected.

BV2 cells and primary mouse microglia were cultured in DMEM supplemented with 10% FBS at 37 °C in a 5% CO₂ humidified atmosphere. Cells were transfected with overexpression plasmid or siRNA or control using JetPRIME (Polyplus Transfection, Ill kirch, France). After 24 h, cells were collected for qPCR and Western blotting. In a different set of experiments, transfected cells were stimulated by LPS for 8 h, and cells were collected for qPCR and Western blotting. For inhibition studies, BV2 and primary mouse microglia were pretreated with the selective MEK1/2 inhibitor U0126 (10 µ m ) for 1 h and then treated with LPS or vehicle. After 8 h, cells were collected for qPCR and Western blotting analyses.

Total RNA was extracted with TRIzol reagent following the manufacturer's instructions (Invitrogen, Carlsbad, CA). Total RNA (1 µg) was reversely transcribed into cDNA using PrimeScript RT Master Mix (Takara Bio., Shiga, Japan). qPCR reactions were performed with GoTap qPCR Master Mix (Promega, Madison, WI) using a Biometra Toptical Thermocycler (Analytik Jena, Goettingen, Germany) as previously described.^[56^] Data were normalized to a housekeeping gene (mouse β‐actin or human GAPDH). Primers are listed in Table S2, Supporting Information.

Proteins separated by sodium dodecyl sulfate‐polyacrylamide gel electrophoresis were transferred to polyvinylidene difluoride membranes (Millipore, Bedford, MA). The membrane was blocked in a blocking solution of 2% (w/v) skimmed milk in TBST (0.1% Tween20, 50 m m Tris‐HCl, 140 m m NaCl, and 1 m m MgCl₂; pH=7.6) for 1 h at room temperature. Then, the membrane was incubated overnight at 4 °C with a primary antibody diluted in the blocking solution. Bands were visualized with enhanced chemiluminescence using an Omega LumG Imaging System (Aplegen, Pleasanton, CA) and band densities were analyzed with FluorChem 5500 (Alpha Innotech, San Leandro, CA). GAPDH, β‐tubulin and LaminA/C were used as internal controls. Information of primary antibodies is provided in Table S3, Supporting Information.

RNA was isolated from hippocampus samples using Trizol (Invitrogen, Carlsbad, CA) and quantified using Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA). RNA quality was checked using Bioanalyzer 2100 RNA 6000 Nano Kit (Agilent Technologies, Santa Clara, CA), and samples were considered qualified when RIN > 7.7. RNA‐sequencing were performed as described in the authors’ previous reports.^[50, 71^] Genes were defined as differentially expressed when p < 0.05 and fold change >1.5.

Four E4bp4^−/− mice and four control mice were sacrificed and the hippocampus tissues were isolated. Hippocampus samples of each genotype were pooled. Pooled samples were dissociated using an Adult Brain Dissociation Kit according to the manufacturer's instructions (Miltenyi Biotec, Bergisch Gladbach, Germany). Single‐cell RNA sequencing libraries were prepared using the Chromium Single Cell Controller (10× Genomics, Pleasanton, CA) as previously described.^[72^] Hippocampal cells, diluted appropriately, were mixed with reverse transcription reagent and loaded to a single cell chip, followed by the addition of gel beads and partitioning oil. After generation of emulsion droplet, reverse transcription was performed at 53 °C for 45 min and inactivated at 85 °C for 5 min. cDNA was purified and amplified for 12 cycles on a Bio‐Rad C1000 Touch thermocycler. Amplified cDNA was fragmented, end repaired, A‐tailed and ligated to sequencing adaptor. Final libraries were sequenced on Illumina NovaSeq 6000 (Illumina, San Diego) with a read length of 150 bp. To ensure high‐quality data, only genes expressed in three or more cells, and cells with more than 200 and less than 4500 detected genes, were retained in the dataset. For clustering, signals were cleaned to keep cells with a mitochondrial content of <30% and data normalized to total UMIs and mitochondrial content.

BV2 cells were co‐transfected with Mapk1/Mapk3 luciferase reporter (50 ng), pRL‐nTK (a renilla luciferase reporter, 10 ng), and E4bp4 overexpression plasmid (or blank pcDNA3.1, 200 ng) using JetPRIME (Polyplus Transfection, Ill kirch, France). 24 h later, cells were collected and luciferase activities were measured using Dual‐Luciferase Reporter Assay System (Promega, Madison, WI). Firefly luciferase activity was normalized to renilla luciferase activity and expressed as a relative luciferase unit (RLU).

ChIP assays were performed using a SimpleChip plus Enzymatic Chromatin IP kit (Cell Signaling Technology, Beverly, MA) as described in the authors’ previous publication.^[56^] Specific primers are provided in Table S1, Supporting Information.

Data are recorded as mean ± standard errors of the mean (SEM), and outliers were included in data analysis. The amplitudes of rhythmic genes and behaviors were obtained by performing cosinor analysis (https://cosinor.online↗). Student's t‐test (two‐sided) was used to analyze a statistical difference between two groups. One‐way or two‐way ANOVA followed by Bonferroni post hoc test was used for multiple group comparisons. All statistical analyses were performed with GraphPad Prism 8.0 (GraphPad Software Inc., San Diego, CA). The specific tests and number of replicates (n) are indicated in the figure legends. The level of significance was set at p < 0.05 (*).

All data associated with this study are present in the manuscript or the Supporting Information.