Abstract

   Protein translation is crucial for fear extinction, a process vital for
   adaptive behavior and mental health, yet the underlying cell-specific
   mechanisms remain elusive. Using a Tet-On 3G genetic approach, we
   achieved precise temporal control over protein translation in the
   infralimbic medial prefrontal cortex (IL) during fear extinction. In
   addition, our results reveal that the disruption of cytoplasmic
   polyadenylation element binding protein 1 (Cpeb1) leads to notable
   alterations in cell type–specific translational programs, thereby
   affecting fear extinction. Specifically, Cpeb1 deficiency in neurons
   activates the translation of heterochromatin protein 1 binding protein
   3, which enhances microRNA networks, whereas in microglia, it
   suppresses the translation of chemokine receptor 1 (Cx3cr1), resulting
   in an aged-like microglial phenotype. These coordinated alterations
   impair spine formation and plasticity. Our study highlights the
   critical role of cell type–specific protein translation in fear
   extinction and provides an insight into therapeutic targets for
   disorders with extinction deficits.
     __________________________________________________________________

   Cpeb1 promotes fear extinction by inhibiting neuronal Hp1bp3
   translation and activating microglial Cx3cr1 translation.

INTRODUCTION

   Fear extinction learning is an evolutionarily conserved and flexible
   process crucial for an organism’s survival and its ability to
   appropriately respond to dangerous or aversive stimuli ([48]1). In the
   classical Pavlovian fear conditioning paradigm, a widely used
   experimental model for studying fear extinction learning ([49]2,
   [50]3), this process unfolds as the conditioned stimulus (CS) is
   repeatedly presented in the absence of the unconditioned stimulus (US),
   prompting the organism to swiftly reverse previously acquired
   contingencies ([51]4). Dysfunctions in fear extinction learning have
   been associated with various neuropsychiatric conditions. For instance,
   individuals with posttraumatic stress disorder (PTSD) often exhibit
   decreased fear extinction abilities, leading to heightened fear
   responses to trauma-related cues ([52]5). Similarly, those grappling
   with anxiety and addiction withdrawal frequently experience challenges
   with extinction learning, perpetuating cycles of anxiety and substance
   use ([53]6). Currently, the primary effective treatment for these
   disorders is extinction-based exposure therapy, which mitigates fear
   responses by subjecting individuals to repeated exposure to feared
   stimulus in a safe environment ([54]7). Consequently, a deeper
   comprehension of fear extinction learning holds promise for developing
   innovative therapeutic approaches tailored to individuals afflicted by
   these disorders, thereby enhancing their daily functioning and overall
   quality of life.

   Protein synthesis is a fundamental cellular process used by cells to
   respond to signals originating from both inside and outside the cell,
   enabling the reshaping of cell function ([55]8) and facilitating the
   formation of new neural connections, as well as the restructuring of
   existing ones ([56]9). Deregulation of protein translation plays a
   vital role in memory by affecting protein expression, function, and
   cellular homeostasis. Protein translation dysfunction primarily stems
   from disruption in the three key processes of translation initiation,
   elongation, and termination ([57]10). For example, knockdown of
   neuronal eIF4E knockdown in the lateral amygdala, leading to inhibition
   of the neuronal translation initiation process, has been shown to
   impair memory consolidation ([58]11). Conversely, phosphomutation of
   the translation initiation factor eIF2α in excitatory or
   somatostatin-expressing inhibitory neurons enhances overall mRNA
   translation, thereby strengthening synaptic plasticity and improving
   long-term memory ([59]12). Some drugs, such as anisomycin,
   cycloheximide, and rapamycin, impair memory consolidation by inhibiting
   the translation elongation ([60]13–[61]15). Moreover, protein
   translation is crucial for the consolidation of the extinction memory,
   particularly in the infralimbic subregion (IL) of the prefrontal cortex
   ([62]9). Previous studies have demonstrated that infusion of the
   protein translation inhibitor anisomycin into the IL during extinction
   training not only reduced the activity of medial prefrontal cortex
   (mPFC) neurons but also notably hindered fear extinction in rats
   ([63]16). Similarly, extinction learning with social support was
   impaired by injecting the protein synthesis inhibitors, such as
   anisomycin and rapamycin, into the IL during extinction training
   ([64]17). Extinction of conditioned taste aversion was impeded by
   microinfusing anisomycin into the IL immediately following a
   nonreinforced extinction session ([65]18). These findings suggest that
   deregulation of the protein translation process is closely associated
   with impaired extinction memory, and the extinction process may be
   compromised in the absence of proper protein synthesis in the IL.
   However, little is known about the translation function in specific
   cell populations, and the underlying mechanism of cell type–specific
   protein translation in fear extinction remains to be elucidated.

   In this study, the genetically encodable protein synthesis inhbitor
   (gePSI) and Tet-on 3G system were used to inhibit protein synthesis in
   the IL. It was observed that fear extinction learning was impaired
   because of disruption of protein translation. Subsequently, a fear
   extinction learning protocol was implemented ([66]19), and screening of
   mice was conducted to identify groups with Impaired-extinction
   (Impaired-ext) and Normal-extinction (Normal-ext) using the principal
   components analysis (PCA) method. Our findings, combined with the
   proteomic and transcriptomic data, indicated that posttranscriptional
   translational regulation plays a critical role in fear extinction.
   Furthermore, we demonstrated that the involvement of neuron- and
   microglia-specific translational profile driven by cytoplasmic
   polyadenylation element binding protein 1 (Cpeb1) simultaneously.
   Specifically, the translational activation of Hp1bp3 in neurons
   promoted a complex network of microRNAs (miRNAs) to regulate the
   disruption of postsynaptic proteins, hindering spine formation and
   plasticity, while the suppression of translation of Cx3cr1 disturbed
   the homeostasis in microglia, resulting in an aged phenotype and
   impairing their capability to engulf the extracellular matrix (ECM)
   Aggrecan, thereby promoting the ECM deposition and reduction of the
   dendritic spine. Overall, our findings suggest that the reduction of
   Cpeb1 remodels the neuron- and microglia-specific translational
   program, leading to the disruption of synaptic structure and
   plasticity, ultimately impairing fear extinction.

RESULTS

The protein translation programs are required for fear extinction

   To ascertain whether fear extinction memory depends on protein
   translation in the IL, the gePSI and Tet-on 3G systems were used to
   achieve temporal regulation of protein synthesis ([67]20).
   Immunofluorescence and surface sensing of translation (SUnSET) assay
   ([68]21) effectively demonstrated the inhibition of translation in the
   IL (fig. S1, A to F). Subsequently, mice injected with gePSI-expressing
   virus or control virus in the IL were subjected to a fear extinction
   paradigm, the doxycycline (Dox) and Tet-on 3G system were introduced to
   inhibit the protein translation at any time. Our results suggested that
   the gePSI-expressing group with Dox treatment (gePSI-On) did not impair
   the acquisition of fear memory but exhibited a high freezing level
   compared to the gePSI-expressing group without Dox treatment
   (gePSI-Off) and control group (Con-On, Con-Off) during fear extinction
   ([69]Fig. 1, A and B, and fig. S1G). In addition, no disparity in
   freezing levels during fear extinction between control mice with
   (Con-On) or without Dox (Con-Off) was observed (fig. S1G). Moreover, in
   the gePSI (On-Off) mice, fear extinction inflexibility was reversed
   when the Dox administration was stopped ([70]Fig. 1, A and B, and fig.
   S1, H and I). Thus, transient protein synthesis inhibition before
   extinction learning blocked the shift from retrieving the original fear
   memory to initiating the formation of a new fear-extinction memory,
   suggesting that the protein translation is necessary for fear
   extinction flexibility.

Fig. 1. The protein translation programs are required for fear extinction.

   [71]Fig. 1.
   [72]Open in a new tab

   (A) Schematic diagram of the behavioral process for fear conditioning
   training, Tet-on system, and fear extinction experiment. (B) Percentage
   of freezing levels is shown in blocks of two sessions across all phases
   of the experiment (trial 4 during conditioning and trial 14 during
   extinction). Detailed experimental procedures are shown in the
   “Behavioral assays” section of Materials and Methods (*P < 0.05, ***P <
   0.001, n = 5 to 10). (C) Percentage of freezing level in fear memory
   acquisition (Acq) and average percentage of freezing of 14 trials in
   the second stage of extinction learning from No-ext, Normal-ext, and
   Impaired-ext group mice (***P < 0.001, n = 21). (D) The average
   percentage of freezing was ranked in the 14 trials during the second
   stage of extinction learning compared to the percentage of freezing in
   the last three trials for the mice. (E) PCA for the average percentage
   of freezing in the 14 trials during the second stage of extinction
   learning (n = 19; Nor, Normal-ext; Imp, Impaired-ext). (F and G)
   Analysis of proteome and transcriptome on Impaired-ext and Normal-ext
   group. Proteome and transcriptome comparison between them was performed
   in (F). Scatter plot of the fold change of RNA and protein comparing
   Impaired-ext to Normal-ext is shown in (G) (mRNA, fold change >1.5;
   Protein, fold change >1.2; group A exhibits an inverse correlation
   between protein and mRNA levels; group B represents protein levels
   change while mRNA remains constant; group C represents a positive
   correlation between protein and mRNA levels). (H) Functional enrichment
   analysis of gene populations from groups A and B in (F) or (G) (fold
   change >1.5, P < 0.05, n = 3 per group). (I) Heatmap showing the
   expression profiles of protein levels (groups A and B) between
   Impaired-ext and Normal-ext mice (fold change >1.2, P < 0.05, n = 2 per
   group).

   To identify the critical translational events associated with impaired
   fear extinction, we first used a classic example of the Pavlovian fear
   conditioning protocol and identified the Impaired-ext group and the
   Normal-ext group by analyzing the mean percentage of freezing across
   the 14 trials in session 2 of extinction learning, as well as the
   average freezing percentage during the last three trials. PCA, as
   referenced in our previous report ([73]19), further validated the
   feasibility of our classification on the basis of the freezing levels
   observed during session 2. With this approach, we selected two
   proportions of mice that displayed the quite different freezing time.
   We therefore named these two groups as Impaired-ext and Normal-ext
   groups ([74]Fig. 1, C to E, and fig. S2A). We also included a control
   group, which did not receive CS stimuli during extinction learning,
   named as No-extinction (No-ext) (fig. S2B). It is known that the mPFC
   played an important role in fear extinction ([75]22), so we then
   performed quantitative proteomics and transcriptomic analysis in the
   mPFC of Impaired-ext, Normal-ext, and No-ext groups (fig. S2C).
   Proteomes in different groups were compared, and functional enrichment
   analysis was conducted on differentially expressed proteins (fold
   change > 1.2, P < 0.05). We observed that the pathways enriched in the
   Impaired-ext group were mainly related to synaptic translation and
   ribosome assembly pathway, which did not show notable enrichment
   between the Normal-ext and No-ext groups (fig. S2, D to G). These data
   suggested that some specific protein translation programs may play an
   important role in impaired fear extinction.

   To further identify the specific translational programs implicated in
   fear extinction, we compared the proteomic and transcriptomic
   characteristics between the Normal-ext and Impaired-ext groups. All the
   matched genes/proteins were categorized into three categories: Group A
   exhibits an inverse correlation between protein and mRNA levels. Group
   B shows changes in protein levels while mRNA remains constant. Group C
   displays a positive correlation between protein and mRNA levels. We
   found that the proportion of genes in groups A and B was much higher
   than in group C ([76]Fig. 1, F and G). Functional enrichment analysis
   was conducted on the genes in groups A and B (fold change >1.2, P <
   0.05), revealing notable enrichment in pathways such as “translation”
   and “ribosome biogenesis.” Combined with the discordant proteomic and
   transcriptomic signatures, we propose that translational regulation
   mechanisms may play an important role in the impaired extinction
   learning ([77]Fig. 1H). Among the changed proteins in groups A and B
   (fold change >1.2, P < 0.05), Cpeb1 is particular of interest, because
   (i) Cpeb1 plays important role in the posttranscriptional translation
   regulation ([78]23); (ii) Cpeb1 was significantly down-regulated in
   Impaired-ext mice, especially in the IL subregions of the mPFC
   ([79]Fig. 1I and fig. S3, A to D); and (iii) Cpeb1 was implicated in
   the reduced extinction of hippocampal-dependent memories ([80]24).

Cell type–specific Cpeb1-dependent translational signature are necessary for
fear extinction

   We then tried to understand the cell type–specific changes of Cpeb1
   among different groups. To this end, we performed
   double-immunofluorescence staining using antibodies to Cpeb1 and
   different cell markers and found that the decrement of Cpeb1 was mainly
   localized in the neurons and microglial rather than in astrocytes
   ([81]Fig. 2, A and B). We subsequently investigated the specific cell
   type in which the down-regulation of Cpeb1 is involved in fear
   extinction. We injected neuronal promoter–driven shCpeb1 virus
   (AAV/9-hSyn-mCherry-5′miR-30a-shCpeb1-3′miR-30a, N-shCpeb1) and the
   adeno-associated virus (AAV) vectors containing the double-floxed
   inverse orientation (DIO) element with a short hairpin RNA (shRNA)
   (shCpeb1) (AAV2/6-CMV-DIO-EGFP-U6-shCpeb1, M-shCpeb1) into the IL of
   Cx3cr1-Cre mice ([82]Fig. 2C and fig. S4, A and B) to knock out the
   Cpeb1 gene in neuron and microglia, respectively. Obviously, the
   intensity of Cpeb1 was noticeably decreased in the mCherry-positive
   neurons and the enhanced green fluorescent protein (EGFP)–positive
   microglia of IL, confirming the effectiveness of the shRNA-based
   knockdown strategies ([83]Fig. 2C). Nissl staining revealed that
   neuron- and microglia-specific knockouts of Cpeb1 did not result in any
   substantial changes in cell density (fig. S4, C and D). We then
   subjected these mice to the fear conditioning and fear extinction
   learning ([84]Fig. 2D). We found that all mice exhibited comparable
   freezing behavior during the acquisition stage of cued fear memory
   ([85]Fig. 2E). However, silencing of Cpeb1 in either neuron or
   microglia led to a higher freezing level during the extinction stage
   ([86]Fig. 2E). Knocking out Cpeb1 in neurons showed higher freezing
   levels than knocking out Cpeb1 in microglia during the initial phase of
   extinction learning ([87]Fig. 2E). Furthermore, mice with Cpeb1
   attenuation in neurons and microglia simultaneously displayed much
   higher freezing levels in the second phase of extinction learning stage
   ([88]Fig. 2, E and F). Knockdown of Cpeb1 in the prelimbic region (PL)
   did not find any deficits in the fear extinction, suggesting the
   crucial role of IL-specific Cpeb1 down-regulation in fear extinction
   (fig. S4, E to I). No substantial changes were found in locomotor and
   anxiety-like behavior in these mice when subjected to the open-field
   test and elevated plus maze task test (fig. S4, J to L). Collectively,
   these data suggested that both neuronal and microglial Cpeb1 were
   critical for fear extinction.

Fig. 2. Cpeb1-dependent translational signatures in neuron and microglia are
crucial for fear extinction.

   [89]Fig. 2.
   [90]Open in a new tab

   (A and B) Immunofluorescence staining was performed in the IL region
   from Impaired-ext and Normal-ext groups by using anti-Cpeb1 (red) and
   Neun, Iba1, and glial fibrillary acidic protein (GFAP) (green)
   antibodies. The representative images are shown in (A), and relative
   intensity of Cpeb1 and the colocalized pixels in different cell
   populations were calculated in (B) (*P < 0.05, **P < 0.01, n = 6 cells
   from five mice per group). (C) Representative confocal image (right)
   for virus injection into IL of the mPFC from Cx3cr1-Cre mice. Coronal
   slices from virus infected (mCherry) neurons were stained with
   anti-Cpeb1 (cyan) antibody, and coronal slices from virus infected
   (EGFP) microglia cells were stained with anti-Cpeb1 (cyan) antibody.
   White dashed lines indicate neurons or microglia cells with virus
   infection, and yellow dashed lines indicate neurons or microglia cells
   without virus infection. (D) Schematic illustration of the workflow for
   the fear conditioning training and fear extinction learning assay after
   viral intervention of Cx3cr1-Cre mice. Hab, habitation; Cond,
   conditioning; Acq, acquisition. (E) Average freezing percentage of fear
   memory acquisition and average freezing percentage in the first and
   second sessions of fear extinction learning from M-shCon+N-shCon,
   M-shCon+N-shCpeb1, M-shCpeb1+N-shCon, and M-shCpeb1+N-shCpeb1 groups
   (*P < 0.05, **P < 0.01, n = 5 to 9 per group). (F) Percentage of
   freezing of 14 trials in the second session of extinction learning from
   M-shCon+N-shCon, M-shCon+N-shCpeb1, M-shCpeb1+N-shCon, and
   M-shCpeb1+N-shCpeb1 groups (n = 5 to 9 per group).

Cell type–specific downstream effectors of Cpeb1 regulate fear extinction

   We then wanted to identify specific Cpeb1-dependent translational
   events in impaired fear extinction. It is known that Cpeb1 regulates
   cellular function by controlling the translation of specific
   transcripts ([91]25). To identify the downstream targets of Cpeb1
   mediating fear extinction learning in neurons and microglia separately,
   we purified RNA from the pellet of mPFC homogenates from wild-type (WT)
   C57BL/6 J mice immunoprecipitated by a Cpeb1 antibody (fig. S5A) and
   subjected it to RNA sequencing ([92]Fig. 3A). Compared to
   immunoglobulin G (IgG)–immunoprecipitated pellets, 2791 transcripts
   were enriched in the anti–Cpeb1-immunoprecipitated pellets (fold change
   >2, P < 0.05) ([93]Fig. 3B). Because Cpeb1 binds to specific
   transcriptomes via a cytoplasmic polyadenylation element (CPE)
   ([94]26), we further analyzed these 2791 transcripts and identified 525
   genes containing CPEs. Subsequently, we merged the 525 Cpeb1-associated
   genes with our proteomic data and identified two genes: Hp1bp3 and
   Cx3cr1 ([95]Fig. 3C).

Fig. 3. Distinct downstream effectors of Cpeb1 in neuron and microglia
regulate fear extinction.

   [96]Fig. 3.
   [97]Open in a new tab

   (A) Schematic outlining the Cpeb1 RNA immunoprecipitation sequencing
   (RNA-IP-seq) workflow from C57BL/6 J mice (n = 2 per group). (B)
   Scatter plot displaying genes from two independent Cpeb1 RNA
   immunoprecipitation experiments. Each dot represents one gene, with x
   and y axes indicating fold changes of IP/IgG for each Cpeb1 RNA
   immunoprecipitation experiment. (C) Venn diagram illustrating the
   overlap between CPE-containing genes, RNA-IP-seq genes (fold change of
   IP/IgG > 2), and proteins identified by mass spectrometry (fold change
   >1.2, P < 0.05). (D and E) Cpeb1 RNA-IP-seq data for Hp1bp3 [top of
   (D)] and Cx3cr1 [top of (E)] transcripts. 3′UTRs of Hp1bp3 and Cx3cr1,
   with WT or mutated CPEs, were inserted downstream of Renilla luciferase
   (hRluc) gene. These constructs were cotransfected into 293 T cells with
   the empty vetor pcDNA3.1(+) or pcDNA3.1(+)-Cpeb1, along with a firefly
   luciferase (Luc) vector as an internal control. After 48 hours,
   luciferase activity was measured using a dual-luciferase assay.
   Schematic illustration of WT or mutated CPEs in 3′UTRs of Hp1bp3
   [middle of (D)] and Cx3cr1 [middle of (E)] are provided, and the
   corresponding luciferase activity [bottom of (D) and (E)] are shown (*P
   < 0.05, **P < 0.01, ***P < 0.001, n = 3 replications). (F and G) HT22
   or BV2 cells were transfected with either empty vector (Con) or
   overexpress Cpeb1 (OE). After 48 hours, RNA was isolated and subjected
   to PCR poly(A) test. (F) Bands for Hp1bp3 poly(A) tail and primers. (G)
   Cx3cr1 poly(A) tail and primers (RT+, reverse transcription; RT−, no
   RT; n = 3 replications). Arrowheads indicate poly(A) PCR product. (H to
   J) Immunofluorescence of IL sections from Impaired-ext and Normal-ext
   mice using anti-Hp1bp3/anti-Cx3cr1 (red) (H) and anti-Neun/anti-Iba1
   (green) (I) antibodies. Representative confocal images are provided.
   Normalized intensities of Hp1bp3 and Cx3cr1, Hp1bp3/Neun, and
   Cx3cr1/Iba1 colocalized pixels were calculated (J) (*P < 0.05, **P <
   0.01, n = 7 to 9 cells from four mice per group).

   Heterochromatin protein 1 binding protein 3 (Hp1bp3), recognized as a
   previously unidentified cognitive function modulator ([98]27) and
   Cx3cr1, a cell surface receptor predominantly expressed on microglia
   ([99]28), are both known to play pivotal roles in synaptic plasticity
   and cognitive function ([100]29, [101]30). Subsequently, we sought to
   determine whether Cpeb1 indeed regulated the translation of Hp1bp3 and
   Cx3cr1. We first verified that both Hp1bp3 and Cx3cr1 transcripts were
   enriched in Cpeb1-immunoprecipitated pellets ([102]Fig. 3, D and E). We
   then identified seven and four canonical CPE sites in the 3′
   untranslated region (3′UTR) of Hp1bp3 and Cx3cr1 mRNAs, respectively,
   through bioinformatics analysis ([103]Fig. 3, D and E). To further
   clarify the specific CPE that biologically regulated by Cpeb1 in these
   two targets, we constructed a series of luciferase reporters that
   contains different CPEs in Hp1bp3 (Seg.1: contains CPE1 and CPE2;
   Mut1-Seg.1: contains mutated CPE1 and WT CPE2; Mut2-Seg.1: contains
   mutated CPE2 and WT CPE1; Seg2: contains CPE3; Seg3: contains CPE4 and
   CPE5; Mut4-Seg.3: contains mutated CPE4 and WT CPE5; Mut5-Seg.3:
   contains mutated CPE5 and WT CPE4; Seg4: contains CPE6 and CPE7) and
   Cx3cr1 (Seg1: contains CPE1 and CPE2; Mut1-Seg.1: contains mutated CPE1
   and WT CPE2; Mut2-Seg.1: contains mutated CPE2 and WT CPE1; Seg2:
   contains CPE3 and CPE4; Mut3-Seg.2: contains mutated CPE3 and WT CPE4;
   Mut4-Seg.2: contains mutated CPE4 and WT CPE3) separately. Then, we
   transfected these reporters with pcDNA3.1(+)-Cpeb1 construct into the
   human embryonic kidney–293T cells and performed the luciferase
   analysis. We found that overexpression of Cpeb1 significantly reduced
   the luciferase activity in the Seg1 and Seg3 but not CPE2-mutated Seg1
   (Mut2-Seg.1) and CPE5-mutated Seg3 (Mut5-Seg.3) of Hp1bp3 constructs.
   Meanwhile, Cpeb1 enhanced the luciferase activity of WT Seg2 but not
   CPE3-mutated Seg2 of Cx3cr1 constructs ([104]Fig. 3, D and E). These
   data suggested that CPE2 and CPE5 in Hp1bp3, CPE3 in Cx3cr1, are key
   binding motifs of Cpeb1. As reported, Cpeb1 is able to bind with target
   mRNAs with CPEs, thereby activating or suppressing their translation by
   influencing polyadenylate [poly(A)] tail elongation and removal
   ([105]25). To examine the polyadenylation state of Hp1bp3 and Cx3cr1,
   we used a poly(A) tail-length assay. Following poly(G/I) tailing and
   reverse transcription, the poly(A) polymerase chain reaction
   (PCR)–amplified products from Cpeb1 overexpression displayed a shorter
   length of the poly(A) tail for Hp1bp3 transcript but a longer length
   for Cx3cr1 transcript ([106]Fig. 3, F and G, and fig. S5, B and C).
   Moreover, compared to the control group, Cpeb1 overexpression in HT22
   cells led to an approximate twofold reduction in Hp1bp3 protein levels.
   Conversely, knockdown of Cpeb1 in HT22 cells resulted in a nearly
   1.5-fold increase in Hp1bp3 protein expression. In BV2 microglial
   cells, Cpeb1 overexpression caused an approximate 1.3-fold elevation in
   Cx3cr1 protein levels, whereas silencing Cpeb1 led to a nearly 2-fold
   decrease in Cx3cr1 protein expression. Neither overexpression nor
   silencing of Cpeb1 affected the transcription of Hp1bp3 and Cx3cr1
   (fig. S6, A to F). Collectively, these data suggest that Cpeb1
   regulates Hp1bp3 and Cx3cr1 translation by binding specific CPEs and
   affecting the poly(A) tail elongation. Having verified the direct
   translational regulation of Hp1bp3 and Cx3cr1 by Cpeb1, we then queried
   whether Hp1bp3 and Cx3cr1 are indeed involved in the impaired fear
   extinction that affected by Cpeb1. Immunofluorescence experiments
   suggest that Hp1bp3 is mainly localized in neurons but not microglia,
   while Cx3cr1 is predominantly distributed in microglia but not neurons
   ([107]Fig. 3, H to J, and fig. S6, G to J). In the Impaired-ext mice,
   the protein level of Hp1bp3 is elevated while Cx3cr1 is reduced,
   without affecting Hp1bp3 and Cx3cr1 mRNA levels (fig. S6, K to M).
   Therefore, it is possible that Hp1bp3 and Cx3cr1 are independent
   downstream effectors of Cpeb1 in neurons and microglia to mediate fear
   extinction inflexibility.

Cell type–specific Cpeb1-dependent translational signature is necessary for
synaptic maturation

   We then wanted to explore the underlying molecular mechanisms
   contributing to impaired fear extinction induced by Cpeb1-dependent
   translational signature in neurons and microglia. By conducting the
   functional enrichment analysis on the 525 Cpeb1-associated genes (fig.
   S7A), we found that they were involved in synapse-related pathways such
   as “Synapse organization,” “Axonogenesis,” “Dendrite development,” and
   “Learning or memory” (fig. S7B), confirming the critical role of Cpeb1
   in synaptic plasticity during fear extinction, aligning with our
   proteomic findings. Given the role of Cpeb1-dependent translational
   signature in regulating synaptic plasticity, we hypothesized that the
   down-regulation of Cpeb1 in both neurons and microglia might perturb
   synaptic structure and function, thereby impairing fear extinction
   learning. Comparing miniature excitatory postsynaptic currents (mEPSCs)
   in mPFC neurons, we found a decrease in mEPSC amplitude, but not
   frequency, in the IL of the Impaired-ext group without affecting the
   PL, suggesting the potential involvement of postsynaptic mechanisms
   ([108]Fig. 4, A to E). The unchanged paired-pulse facilitation ratio
   further suggested intact presynaptic functions in the IL ([109]Fig.
   4F). Golgi staining revealed a significant decrease in both dendritic
   spine density and the percentage of matured spines in the Impaired-ext
   group ([110]Fig. 4, G and H). In addition, Sholl analysis indicated a
   notable reduction in dendritic branches in the Impaired-ext group
   ([111]Fig. 4, I and J). These findings strongly suggest that the
   postsynaptic function may play an important role in mediating the
   impaired fear extinction.

Fig. 4. Neuron- and microglia-specific Cpeb1-dependent translational
signatures are necessary for synaptic function.

   [112]Fig. 4.
   [113]Open in a new tab

   (A and B) Whole-cell patch-clamp recordings of IL neurons from
   Normal-ext and Impaired-ext mice. Representative mEPSC traces are shown
   (A), with quantification of mEPSC amplitude and frequency (B) (*P <
   0.05, n = 7 to 8 neurons from four mice per group). (C to E) Whole-cell
   patch-clamp recordings of PL neurons from Normal-ext and Impaired-ext
   mice. Representative mEPSC traces are shown (C), with quantification of
   mEPSC amplitude (D) and frequency (E) (n = 8 to 9 neurons from four
   mice per group). (F) Paired-pulse ratios recordings of IL neurons from
   Normal-ext and Impaired-ext mice. mEPSC traces are shown in the right
   panel, with quantification of EPSC amplitude on the left (n = 9 neurons
   from four mice per group). (G and H) Golgi staining of IL sections from
   Impaired-ext and Normal-ext groups. Representative images are shown
   (G), with quantification of spine density and mushroom spine percentage
   (H) (*P < 0.05, **P < 0.01, n = 7 to 8 slices from four mice per
   group). (I) Golgi staining images showing the dendritic trees from
   Impaired-ext and Normal-ext groups. (J) Sholl analysis evaluating
   dendritic complexity in neurons from the Impaired-ext and Normal-ext
   groups (n = 8 to 10 neurons from four mice per group). (K to M)
   Dendritic spine in IL neurons labeled with AAV9-hSyn-mCherry and
   AAV9-hSyn-mCherry-shcpeb1 viruses from Cx3cr1-Cre mice with the
   neuron-specific, microglia-specific, and combined Cpeb1 deletion.
   Representative confocal images of spines are shown in (K), with
   quantitative analysis of spine density (L) and mushroom spine
   percentage (M) (*P < 0.05, **P < 0.01, ***P < 0.001, n = 6 neurons from
   four mice per group). (N to Q) Whole-cell patch-clamp recordings of IL
   neurons from Cx3cr1-Cre mice with neuron-specific, microglia-specific,
   and combined Cpeb1 deletion. Representative mEPSC traces are shown in
   (N), with quantification of mEPSC amplitude (O and P) and frequency (Q)
   (*P < 0.05, ***P < 0.001, n = 7 to 10 neurons from four mice per
   group).

   We then investigated whether specific knockout of Cpeb1 in neurons and
   microglia could replicate the similar postsynaptic pathology as
   observed in Impaired-ext mice. We found that artificially knockdown of
   Cpeb1 in neurons or microglia resulted in decreased dendritic spine
   density, dendritic tree maturation, and mEPSC amplitude. Compared to
   the control virus–infected group (M-shCon+N-shCon), knockdown of Cpeb1
   in both neurons (M-shCon+N-shCpeb1) and microglia (M-sh Cpeb1+N-shCon)
   induced a more pronounced decline in the density of dendritic spines,
   the percentage of mushroom-type spines, and mEPSC amplitude ([114]Fig.
   4, K to Q, and fig. S8, A and B). These data strongly suggest that the
   disorder of cell type–specific Cpeb1-dependent translational signature
   impairs synaptic transmission and structure.

Neuron-specific translational signal regulates synaptic function via a
complex miRNA network

   To understand how neuron-specific translational signal affects the
   postsynaptic function, we generated a mouse model with neuron-specific
   Hp1bp3 overexpression in the IL by injecting the
   AAV2/9-SYN-Hp1bp3-3×FLAG-P2A-EGFP virus into the IL of C57BL/6 J mice
   (OE-Hp1bp3), and control groups were injected with the control virus
   AAV2/9-SYN-3×FLAG-P2A-EGFP into the IL (Control). Immunofluorescence
   and Western blot data validated successful overexpression of Hp1bp3 in
   neurons ([115]Fig. 5A and fig. S9A). The Western blot data indicated
   that the expression level of Hp1bp3 in the OE-hp1bp3 group was nearly
   double that observed in the control group. The molecular band
   corresponding to Hp1bp3 appeared ~3 kDa higher than that of the control
   group, attributable to the fusion of Hp1bp3 with the Flag tag (fig. S9,
   B and C). We then subjected these mice to fear extinction tasks and
   found that the OE-Hp1bp3 group displayed comparable freezing time to
   control mice (Control) during fear acquisition. However, during the
   second extinction stage, the freezing percentage in OE-Hp1bp3 mice was
   nearly double that of control mice, indicating impaired extinction
   flexibility ([116]Fig. 5, B and C). Meanwhile, compared to control
   virus–infected neuron, Hp1bp3 overexpression led to a significant
   reduction in the amplitude of mEPSCs, without affecting the frequency.
   This indicated a specific impairment in
   α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor
   (AMPAR)–dependent synaptic transmission. Furthermore, dendritic
   complexity, as quantified by Sholl analysis, was notably reduced
   following Hp1bp3 overexpression. In addition, neurons infected with
   Hp1bp3 showed decreased dendritic spine density and a substantial
   reduction in the proportion of mushroom-type spines, which are
   essential for synaptic strength and plasticity ([117]Fig. 5, D to H).
   Together, these findings suggest that neuron-specific overexpression of
   Hp1bp3 disrupts synaptic transmission and structural integrity,
   paralleling the deficits observed in Impaired-ext mice. To understand
   whether Hp1bp3 overexpression acts as the downstream effector of Cpeb1
   loss in neuron, we administered a specific shRNA targeting Hp1bp3
   (Sh-Hp1bp3) into the IL of Cpeb1^fl/fl mice (Cpeb1 WT) with Cpeb1
   specific knockout in neurons (Cpeb1 cKO) (fig. S9, D to F, injected
   with AAV2/9-Syn-Cre virus to IL) (fig. S9G). We examined whether
   inhibition of Hp1bp3 could alleviate the extinction learning impairment
   caused by neuronal Cpeb1 loss. We found that Sh-Hp1bp3 not only
   attenuated the increase in Hp1bp3 due to neuronal Cpeb1 loss (fig. S9,
   H and I) but also restored the impaired fear extinction resulting from
   neuronal Cpeb1 loss ([118]Fig. 5, I and J). In addition, Sh-Hp1bp3
   ameliorated the dendritic spine loss and reduction of mEPSC amplitude
   in Cpeb1 knockout neurons ([119]Fig. 5, K to Q). On the contrary,
   overexpression of Hp1bp3 in neurons disrupted the spine density and
   maturation and weakened the mEPSCs responses, especially the amplitude
   ([120]Fig. 5, D to H). Knockdown Hp1bp3 in WT mice also facilitated
   fear extinction ([121]Fig. 5, I and J) and reinforced the spine density
   and maturation, as well as boosting the postsynaptic electric responses
   ([122]Fig. 5, K to Q). These data strongly suggest that up-regulation
   of Hp1bp3 in neurons leads to the synaptic disorder, which mediates the
   impaired fear extinction induced by loss of Cpeb1.

Fig. 5. Neuronal Cpeb1-Hp1bp3 signaling regulates fear extinction by
affecting synaptic function.

   [123]Fig. 5.
   [124]Open in a new tab

   (A) Confocal image for Hp1bp3- infected virus in the IL from C57BL/6 J
   mice (right), and image of Hp1bp3-infected neurons stained with
   anti-Hp1bp3 (red) is shown (left). White dashed lines mark
   virus-infected neurons, and yellow dashed lines mark noninfected ones.
   (B) Freezing percentages during fear memory acquisition and the first
   and second sessions of extinction learning were compared between
   control and OE-Hp1bp3 groups (**P < 0.01, n = 6 to 7 per group). (C)
   Freezing percentages across 14 trials in the second extinction session
   of control and OE-Hp1bp3 groups (n = 6 to 7 per group). (D) Whole-cell
   patch-clamp recordings of the IL neurons from OE-Hp1bp3 and control
   mice. Representative mEPSC traces (top), with quantified amplitude
   (left) and frequency (right) (**P < 0.01, n = 6 to 7 neurons from four
   mice per group). (E and F) 2D reconstructions of IL neurons from
   OE-Hp1bp3 and control mice (E), with Sholl analysis of dendritic
   complexity (F) (n = 5 neurons from four mice per group). (G and H)
   Dendritic spine density and mushroom spine percentage in IL neurons
   from the OE-Hp1bp3 and control groups. Representative images (G) and
   quantifications (H) are shown (***P < 0.001, n = 10 neurons from four
   mice per group). (I) Freezing percentages during fear memory
   acquisition and the first and second sessions of extinction learning
   were analyzed in Cpeb1 WT+Scr, Cpeb1 cKO+Scr, Cpeb1 WT+Sh-Hp1bp3, and
   Cpeb1 cKO+Sh-Hp1bp3 groups (*P < 0.05, **P < 0.01, ***P < 0.001, n = 5
   to 7 per group). (J) Freezing percentages across 14 trials in the
   second extinction session in these groups (n = 5 to 7 per group). (K to
   M) Confocal images of dendritic spine in IL neurons from same groups
   (K), with quantified dendritic spine density (L) and mushroom spine
   percentage (M) (*P < 0.05, **P < 0.01, ***P < 0.001, n = 6 neurons from
   four mice per group). (N to Q) Representative mEPSC traces from these
   groups (N), with quantified amplitude (O and P) and frequency (Q) (*P <
   0.05, **P < 0.01, n = 7 neurons from four mice per group).

   We then queried how neuronal Hp1bp3 translation activation could induce
   synaptic disorder. It is known that Hp1bp3 is a histone H1–like
   chromatin protein that can promote global miRNA biogenesis ([125]31).
   We then performed a miRNA array in the neuronal Hp1bp3 overexpression
   mPFC tissues. We found that overexpressing of Hp1bp3 indeed induced the
   up-regulation of numerous miRNAs ([126]Fig. 6, A and B). We further
   validated these deregulated miRNAs in the Impaired-ext mice and found
   that 10 of them (let-7a-5p, miR-122-5p, miR-130b-5p, miR-146a-5p,
   miR-28a-5p, miR-30a-5p, miR-335-5p, miR-342-3p, miR-433-5p, and
   miR-7a-5p) were increased ([127]Fig. 6C). The up-regulation of these 10
   miRNAs in neuron-specific Cpeb1 knockout mice could be fully restored
   when neuron-specific shRNA targeting Hp1bp3 was administered (fig.
   S10A). We then chose these 10 miRNAs for further study. Quantitative
   PCR (qPCR) analysis suggested that the primary transcripts (pri-) of
   these 10 miRNAs were not up-regulated, while all the precursors (pre-)
   were increased upon Hp1bp3 overexpression, which further verified that
   the boost of miRNA biogenesis by Hp1bp3 was due to promoting the
   processing rather than transcription of pri-miRNAs ([128]Fig. 6, D and
   E). We then used TargetScan to analyze and identify target genes for
   the 10 selected miRNAs and found that most of the targets are enriched
   in the “Synapse organization” and “Dendrite development” pathways (fig.
   S10B). Considering the negative regulatory nature of miRNAs on their
   target genes ([129]32), we integrated these miRNA target genes with the
   down-regulated proteins identified in our proteomic dataset (fold
   change < −1.2 and P < 0.05) (table S1). Cytoscape for further analysis
   ([130]33) of integrated data was used, and we noticed that these 10
   miRNAs might centrally target Gnal, Rem2, Gria1 and Gria2 ([131]Fig.
   6F). Western blots and immunofluorescence data suggested that Gnal,
   Rem2, Gria1 and Gria2 were decreased in the IL neurons upon the
   neuron-specific Hp1bp3 overexpression, neuron-specific Cpeb1 knockout,
   and the Impaired-ext mice ([132]Fig. 6, G to J, and fig. S11, A and B).
   Silencing of Hp1bp3 in neurons effectively restored the levels of Gnal,
   Rem2, Gria1, and Gria2 in neuron-specific Cpeb1 knockout mice
   ([133]Fig. 6J). Considering the critical role of Gnal, Rem2, Gria1, and
   Gria2 in mediating the synaptic transmission and structure
   ([134]34–[135]38), we therefore proposed that neuronal Hp1bp3 promotes
   a complex miRNA network to regulate postsynaptic proteins acting as
   downstream effectors for Cpeb1 in neurons.

Fig. 6. Neuronal Hp1bp3 signal regulates synaptic function via a complicated
miRNA network.

   [136]Fig. 6.
   [137]Open in a new tab

   (A) Volcano plot showing the differentially regulated miRNAs in the
   mPFC from control and OE-Hp1bp3 groups with fold change >1.2 (pink
   dots) or <1.2 (cyan) and P value <0.05. The gray dots indicate the
   unchanged miRNAs (n = 3 per group). (B) Heatmap showing differential
   expression of miRNAs in the mPFC between control and OE-Hp1bp3 mice.
   (C) Quantitative PCR (qPCR) of 19 up-regulated miRNAs from control and
   OE-Hp1bp3 mice was performed in the mPFC between Impaired-ext (Imp) and
   Normal-ext (Nor) groups (*P < 0.05, **P < 0.01, n = 4). (D and E) qPCR
   was performed to measure the relative expression of pre-miRNA (D) and
   pri-miRNA (E) between control and OE-Hp1bp3 mice (*P < 0.05, **P <
   0.01, n = 4). (F) An integrated analysis for the predicted targets of
   these 10 miRNAs. (G and H) The protein levels of Hp1bp3, Gnal, Rem2,
   Gria1, and Gria2 in the mPFC homogenates from control and OE-Hp1bp3
   mice were examined by Western blotting. Representative images are shown
   in (G), and the quantitative analysis is shown in (H) (*P < 0.05, **P <
   0.01, ***P < 0.001, n = 4). GAPDH, glyceraldehyde phosphate
   dehydrogenase. (I) The protein levels of Gnal, Rem2, Gria1, and Gria2
   in the mPFC homogenates from Impaired-ext and Normal-ext mice were
   examined by Western blotting. The representative images are shown in
   right and the quantitative analysis is shown in left (*P < 0.05, **P <
   0.01, n = 4). (J) The protein levels of Gnal, Rem2, Gria1, and Gria2 in
   the mPFC homogenates from Cpeb1 WT+Scr, Cpeb1 cKO+Scr, Cpeb1
   WT+Sh-Hp1bp3, and Cpeb1 cKO+Sh-Hp1bp3 mice were examined by Western
   blotting. The representative images are shown in right and the
   quantitative analysis is shown in left. (*P < 0.05, **P < 0.01, ***P <
   0.001, n = 4).

Disturbance of the microglia-specific translational signal causes microglia
to exhibit an aged-like phenotype and leads to the accumulation of
perisynaptic ECM

   To investigate whether the microglial Cpeb1-Cx3cr1 signal was involved
   in fear extinction, we administered AAV-DIO-(EGFP-U6)-ShCx3cr1 virus
   into the IL of the Cx3cr1-Cre mice (fig. S12, A and B) to specifically
   knock down Cx3cr1 in microglia (DIO-sh-Cx3cr1), which was validated by
   the immunofluorescence and Western blot data ([138]Fig. 7A and fig.
   S12, C and D). In addition, Nissl staining showed that knockout of
   Cx3cr1 did not affect cell density (fig. S12, E and F). Behavioral
   tests showed that microglial Cx3cr1 knockdown did not affect fear
   memory acquisition but elevated the freezing percentage during the
   sessions 1 and 2 of fear extinction ([139]Fig. 7, B and C). To further
   elucidate the specific role of Cx3cr1 in mediating impaired fear
   extinction caused by loss of Cpeb1 in microglia, we administered
   AAV2/6-DIO-CX3CR1-2A-mCherry and AAV2/6-DIO-(EGFP-U6)-ShCpeb1 into the
   IL of Cx3cr1-Cre mice together to overexpress Cx3cr1 in the microglia
   of microglia-specific Cpeb1 knockout mice ([140]Fig. 7D). Western
   blotting and behavioral assays showed that OE-Cx3cr1 not only
   attenuated the down-regulation of Cx3cr1 (fig. S12, G and H) but also
   rescued the impaired fear extinction caused by microglial Cpeb1 loss
   ([141]Fig. 7, E and F). These data indicated that microglial Cpeb1 loss
   hindered fear extinction learning processes by suppressing the levels
   of microglial Cx3cr1.

Fig. 7. Microglial Cpeb1-Cx3cr1 signaling regulates fear extinction and
induces an aged-like microglial phenotype.

   [142]Fig. 7.
   [143]Open in a new tab

   (A) Representative confocal image (right) for virus injection in the IL
   of Cx3cr1-Cre mice. Coronal slices with infected (EGFP) microglia were
   stained with anti-Cx3cr1 (red). Arrowheads indicate virus-infected
   microglia, and arrows show noninfected microglia. (B) Average freezing
   percentage during fear memory acquisition and the first and second
   sessions of fear extinction in DIO-Scr and DIO-ShCx3cr1 groups (*P <
   0.05, **P < 0.01, n = 4 per group). (C) Freezing percentages across 14
   trials in the second extinction session of DIO-Scr and DIO-ShCx3cr1
   groups (n = 4 per group). (D) Representative confocal image for virus
   injection into the IL from Cx3cr1-Cre mice. (E) Average freezing
   percentage during fear memory acquisition and the first and second
   sessions of fear extinction learning in Scr+Con, Scr+OE-Cx3cr1,
   Sh-Cpeb1+Con, and Sh-Cpeb1+OE-Cx3cr1 groups (*P < 0.05, **P < 0.01,
   ***P < 0.001, n = 5 to 7 per group). (F) Freezing percentages across 14
   trials in the second extinction session for Scr+Con, Scr+OE-Cx3cr1,
   Sh-Cpeb1+Con, and Sh-Cpeb1+OE-Cx3cr1 groups (n = 5 to 7 per group). (G)
   Scatter plot showing linear regression of log[2]FC of overlapping genes
   between ADEM gene sets and Cx3cr1-deficient microglia gene sets, with a
   positive correlation. Pink shading shows 95% confidence interval, and
   Pearson’s correlation coefficient and P value are shown. (H to J)
   Microglia from the Impaired-ext group display reduced branching and
   enlarged cell bodies. Confocal images are shown in (H), and Sholl
   analysis (I) and microglial cell size (J) were quantified by ImageJ (*P
   < 0.05, n = 6 cells from four mice per group). (K to M)
   Immunofluorescence on IL sections from Impaired-ext and Normal-ext
   groups using anti-Xaf1 (Red) and anti-Iba1 (Green) antibodies.
   Representative confocal images are shown in (K). Xaf1/Iba1 colocalized
   pixels (L) and normalized intensity of Xaf1 (M) were calculated by
   ImageJ (**P < 0.01, ***P < 0.001, n = 6 cells from four mice per
   group).

   We then aimed to interrogate the underlying mechanisms through which
   microglial Cx3cr1 translation suppression affected fear extinction
   learning. Considering that Cx3cr1-deficient microglia could affect
   microglial homeostasis, we then analyzed transcriptome characteristics
   of isolated brain microglia from 2-month Cx3cr1^+/+, Cx3cr1^+/− and
   Cx3cr1^−/− mice ([144]39). We found that the fold changes of
   overlapping genes in the Cx3cr1-deficient microglia were positively
   correlated with the age-dependent microglia (ADEM) genes ([145]40)
   ([146]Fig. 7G) rather than well-known senescence genes
   ([147]41–[148]45) or senescence-associated secretory phenotype (SASP)
   genes ([149]46, [150]47), which indicates an aged-like but nonsenescent
   phenotype (fig. S13, A and B). Meanwhile, microglia cells in the
   Impaired-ext group exhibited less branching morphology and larger cell
   bodies, indicating an aged-like state ([151]Fig. 7, H to J). The
   aged-like state of microglia in Impaired-ext group was further
   confirmed by the immunostaining of Xaf1 and Tnfsf13b, which were
   significantly up-regulated ADEM genes in Cx3cr1-deficient microglia
   ([152]Fig. 7, K to M, and fig. S13, C and D). However, the level of
   β-galactosidase (β-gal), a marker of cell senescence, did not change in
   microglia in Impaired-ext mice compared to Normal-ext mice, indicating
   that the microglial cells in the Impaired-ext group exhibited an
   aged-like but nonsenescent phenotype (fig. S13, E and F), consistent
   with studies showing that Cx3cr1-deficient microglia display premature
   aging ([153]39).

   To gain insight into the biological characteristics of aged microglia
   with Cx3cr1 deficiency in nonaged brain, we performed pathway
   enrichment analysis of differentially expressed genes between WT and
   Cx3cr1-deficient microglia. Gene Ontology (GO) analysis showed that
   differential genes were mainly related to cell chemotaxis, immune
   response, phagocytosis pathways, and ECM organization ([154]Fig. 8A).
   In addition, gene set enrichment analyses (GSEAs) showed a substantial
   reduction in a set of phagocytosis genes ([155]Fig. 8B). Our
   phagocytosis assay further showed that knocking down Cx3cr1 in primary
   microglia cells inhibited microglial phagocytosis of latex beads
   ([156]Fig. 8C). These findings are consistent with the role of aged
   microglia in dampened phagocytosis ([157]48) and Cx3cr1’s role in
   promoting microglial phagocytosis ([158]49).

Fig. 8. Disturbance of microglial Cpeb1-Cx3cr1 signaling leads to impaired
phagocytic ability in aged microglia and the accumulation of perisynaptic
ECM.

   [159]Fig. 8.
   [160]Open in a new tab

   (A) Functional enrichment analysis of differentially expressed genes
   between Cx3cr1-deficient microglia and WT microglia (fold change >1.2,
   P < 0.05). (B) GSEA of phagocytosis-related genes in Cx3cr1-deficient
   microglia compared with WT microglia. (C) BV2 cells with knocked-down
   Cx3cr1 showed reduced phagocytic ability of fluorescent latex beads.
   Representative confocal images are shown in the top panel of (C), and
   engulfed beads per microglia were calculated in the bottom panel of (C)
   (**P < 0.01, n = 7 microglia cells from two replicates per group). (D
   and E) Aggrecan protein within microglial lysosomes from DIO-Scr and
   DIO-ShCx3cr1 group. Coronal slices from virus infected (EGFP) microglia
   cells were stained with anti-Aggrecan (cyan) and CD68 (red) antibody.
   The representative confocal images and 3D reconstruction are shown in
   (D). The number of engulfed Aggrecan in microglia and the normalized
   intensity of Aggrecan were calculated in (E) (**P < 0.01, ***P < 0.001,
   n = 5 microglia cells from four mice per group). (F) Aggrecan
   colocalization with dendritic spines in the IL labeled with
   AAV9-hSyn-mCherry and AAV9-hSyn-mCherry-shcpeb1 virus from
   M-shCon+N-shCon, M-shCon+N-shCpeb1, M-shCpeb1+N-shCon and
   M-shCpeb1+N-shCpeb1 mice. Representative confocal images of Aggrecan
   colocalization with dendritic spine are showed in the left of (F), and
   the Aggrecan positive dendritic spine was calculated in the right of
   (F) (***P < 0.001, n = 4 neurons from three mice per group).

   We next aimed to investigate the underlying mechanisms by which Cx3cr1
   translation suppression influenced synaptic function. Among the
   enrichment pathways between WT and Cx3cr1-deficient microglia,
   ECM-related pathways have been shown played an important role in
   synaptic and spine plasticity ([161]50) ([162]Fig. 8A). Aggrecan, a
   core component of the ECM structures in perineuronal nets, impaired the
   plasticity of the visual cortex in mice ([163]51). Deposition of
   Aggrecan near synapses resulted in a decrease in the number of spines
   ([164]52). Aggrecan could be degraded by the action of a bacterial
   enzyme, chondroitinase ABC (ChABC) ([165]53), which could improve
   morphological changes in dendritic spines in the visual cortex of adult
   mice ([166]54) and promote the erasure of fear memories ([167]55).
   These data suggest a possible role of ECM remodeling in contributing to
   the impaired fear extinction. Given that aged microglia’s ability to
   engulf was diminished, we investigated whether their capacity to engulf
   Aggrecan was also affected. Using super-resolution imaging and
   three-dimensional (3D) reconstruction, we found that microglia-derived
   Cx3cr1 reduced the amount of engulfed Aggrecan, while the relative
   fluorescence intensity of Aggrecan was increased compared to the
   control group ([168]Fig. 8, D and E). We found that compared to the
   control group, the inhibition of Cpeb1 in microglia significantly
   increased the number of spines fully contacted by Aggrecan, while the
   inhibition of Cpeb1 in neurons did not affect the number of spines in
   contact with Aggrecan ([169]Fig. 8F). These findings indicate that the
   Cx3cr1-translated suppressive microglia exhibit an aged phenotype,
   impairing their capability to engulf the ECM Aggrecan, resulting in the
   increase in the number of Aggrecan-positive spines. In summary, these
   data suggest that the disturbance of microglial Cpeb1-Cx3cr1
   translational signal promotes ECM remodeling and ECM deposition near
   synapses, hindering spine formation.

DISCUSSION

   In this study, we first verified the critical role of protein synthesis
   in fear extinction by using a Tet-On 3G–based genetic approach.
   Subsequently, we demonstrated that cell type–specific translational
   programs regulated by the decay of Cpeb1 in neurons and microglia of
   the IL played an important role in mediating impaired fear extinction.
   Fantastically, the translation activation of Hp1bp3 in neurons and the
   translation suppression of Cx3cr1 in microglia acted as key
   translational downstream events of Cpeb1 loss. Specifically, elevation
   of neuronal Hp1bp3 stimulates a miRNA network that affects postsynaptic
   proteins, hindering spine formation and plasticity, while suppression
   of microglial Cx3cr1 induces the aged-like microglia with impaired
   ability to engulf ECM Aggrecan, promoting perisynaptic ECM deposition
   and reducing dendritic spine density ([170]Fig. 9).

Fig. 9. A working model illustrating Cpeb1 bidirectionally drives neuron- and
microglia-specific translational programs to promote fear extinction.

   [171]Fig. 9.
   [172]Open in a new tab

   Our findings demonstrate that Cpeb1 in the IL regulates fear extinction
   by controlling Hp1bp3 translation activation in neurons and Cx3cr1
   translation suppression in microglia. Neuronal Hp1bp3 translation
   activation increases a network of miRNAs affecting postsynaptic
   proteins, hindering spine formation and plasticity. Microglial Cx3cr1
   translation suppression results in an aged phenotype impairing their
   ability to engulf ECM Aggrecan, promoting perisynaptic ECM deposition,
   and reducing dendritic spine density. Together, Cpeb1 bidirectionally
   drives neuron- and microglia-specific translational program to promote
   fear extinction.

   Evidence suggests that the prelimbic cortex (PL) and anterior cingulate
   cortex (ACC) in the mPFC mediates fear memory expression, while the
   infralimbic cortex (IL) plays a role in fear extinction
   ([173]56–[174]58). Research has demonstrated that inhibiting protein
   synthesis in the prelimbic cortex (PL) disrupts contextual fear memory,
   while similar inhibition in the ACC does not affect contextual fear
   memory ([175]59). In addition, previous studies have highlighted that
   the synthesis of specific proteins within the PL, such as
   N-methyl-d-aspartate receptors (NMDARs) ([176]60), the
   plasticity-associated protein kinase M-ζ (PKM-ζ) ([177]61), and the
   motor protein kinesin family member 3B (KIF3B) ([178]62), is essential
   for fear memory formation. Furthermore, protein synthesis inhibition in
   the lateral ventricles or insular cortex does not impair the
   consolidation or extinction of cued fear memories ([179]16). However,
   inhibition of protein synthesis specifically in the IL disrupts the
   extinction of cued fear memory without affecting its consolidation
   ([180]16). This underscores the critical role of protein synthesis in
   the IL for the successful extinction of cued fear memory. The
   predominant approach involved the use of protein synthesis inhibitors,
   such as anisomycin or rapamycin, administered via subcutaneous
   injections ([181]63, [182]64) or delivered via stereotaxic injection
   into either the lateral ventricle or specific brain regions ([183]18,
   [184]65, [185]66). Anisomycin, which acts by blocking peptidyl
   transferase activity during translation elongation, has been shown to
   result in nearly 30% inhibition of brain protein synthesis by
   stereotaxic injection ([186]67). Subcutaneous anisomycin achieves an
   inhibition rate of more than 90% on brain protein synthesis in the
   first 2 hours and more than 60% in the following 2 hours ([187]64,
   [188]68). Rapamycin inhibits the mechanistic target of rapamycin
   complex 1 (mTORC1) to affect translational initiation and elongation,
   resulting in a 50% decrease in protein synthesis in lymphocytes and
   peripheral skeletal tissue ([189]69, [190]70). However, the
   administration of common drugs via microinjection can induce a range of
   adverse effects. Anisomycin, for instance, has been associated with
   heightened activation of stress signaling pathways ([191]71), induction
   of gene superinduction ([192]72), and facilitation of lysosomal
   degradation ([193]73). Similarly, rapamycin perturbs numerous intricate
   intracellular signaling cascades implicated in cellular growth,
   proliferation, autophagy, and metabolic regulation ([194]74).
   Furthermore, subcutaneous injection of anisomycin, despite high protein
   synthesis inhibition rates, fails to inhibit protein translation in
   certain brain regions and increases adverse reactions, such as skin
   allergies. In addition, conventional protein synthesis inhibitors lack
   temporal control over protein synthesis and the ability to selectivity
   target specific molecularly diverse cell populations, making it
   impossible to accurately regulate protein synthesis within distinct
   cellular subsets. Given the limitations of current approaches for
   protein translation inhibition in memory extinction research, it is
   urgent to propose a more effective and temporally controllable method
   for achieving protein translation inhibition in vivo. Ribosome
   inactivating protein (RIP) ([195]75), that depurinates A4324 on the
   sarcin-ricin loop of the 28S rRNA and blocks translation elongation,
   forms the basis of gePSI and achieves almost complete inhibition of
   neuronal protein synthesis in vitro ([196]20). However, its effects in
   vivo are not clear. Here, we used the Tet-on system, in which the
   introduction of the Dox induces the expression of gePSI in mice. Our
   findings showed that this approach achieved nearly 60% general protein
   inhibition in the brain of live animals and hindered fear extinction,
   while ceasing the administration of Dox reversed fear extinction
   inflexibility by restoring protein synthesis. Moreover, gePSI
   expression in a single neuron in vitro impairs the structural
   plasticity of neuronal synapses ([197]20), which is consistent with our
   results on the disruption of neuronal synapse structure and plasticity
   caused by translational disturbances in neurons. This further proves
   the feasibility of our approach for protein translation inhibition.
   Together, our approach enables the rapid, controllable, and reversible
   achievement of cell type–specific temporal control of protein synthesis
   in a specific brain region, without affecting cellular viability. This
   approach holds significant importance for future investigations into
   the impact of protein synthesis inhibition on learning and memory in
   various cell populations of animals.

   The cellular composition of the mPFC is notably complex, comprising a
   diverse array of neuronal subtypes, glial cells, and intricate synaptic
   networks ([198]76). This complexity underscores the pivotal roles of
   diverse cell types in encoding and regulating many different cognitive
   behaviors ([199]77). For instance, in combination with single-cell RNA
   sequencing, microendoscopic Ca^2+ imaging, and photoinhibition
   approaches, the cell type–specific neural circuits in the mPFC had been
   identified in the goal-directed behavior ([200]77), flexible behavior
   ([201]78), social interaction ([202]79), and choice and reward context
   encoding ([203]80). Optogenetic modulation of specific neuron subtypes
   ([204]81) or cell type–specific modulation of NMDARs ([205]82) in the
   mPFC has been implicated in rapid antidepressant responses. Moreover,
   cocaine administration can induce specific transcriptional adaptations
   in multiple neuron subtypes, with these changes becoming particularly
   pronounced during extended periods of drug withdrawal ([206]83). In
   addition, a cell type–specific transcriptomic analysis suggested the
   oligodendrocytes and inhibitory neurons are most affected in alcohol
   dependence in mPFC ([207]84). Therefore, investigating the cell
   type–specific responses of the mPFC in fear extinction, which rely on a
   complex pattern of brain region and cell type–specific processes
   ([208]85), will benefit the comprehensive understanding of its
   underlying mechanisms and offer promising avenues for therapeutic
   interventions targeting extinction disability disorders.

   CPEB isoforms play distinct roles in regulating synaptic plasticity,
   memory formation, and extinction. Cpeb1, the most studied isoform, is
   essential for extinction but not for memory consolidation, as Cpeb1-gKO
   mice exhibit intact fear memory consolidation but impaired extinction
   of contextual fear memories ([209]24). Cpeb2 is critical for
   hippocampus-dependent memory; Cpeb2-cKO mice display deficits in
   contextual fear memory and spatial memory consolidation in the Morris
   water maze (MWM) tests ([210]86). Cpeb3 appears to enhance memory
   consolidation, with Cpeb3-gKO mice showing improved spatial memory and
   enhanced consolidation of short-term fear memory ([211]86). Cpeb4,
   however, does not seem to affect general learning and memory, as
   Cpeb4-gKO mice exhibit normal performance in MWM and contextual fear
   conditioning, along with unaltered synaptic transmission and spine
   morphology ([212]87). Despite these findings, it remains unclear how
   different CPEB isoforms in specific cell types contribute to memory
   processes. In our study, we observed that reducing Cpeb1 in neurons and
   microglia within the IL region impairs extinction of cued fear memory
   without affecting its consolidation, indicating cell type–specific role
   of Cpeb1 in extinction pathways. Moreover, cell type–specific loss of
   Cpeb1 initiates cell type–specific translation programs, consequently
   playing a crucial role in fear extinction. Although the absence of
   Cpeb1 in both neurons and microglia is implicated in fear extinction,
   the magnitude of their respective contributions varies. Compared to the
   control group, the absence of Cpeb1 in neurons causes mice to display
   increased freezing during the initial phase of extinction learning,
   while its absence in microglia leads to higher freezing levels during
   the second phase of extinction learning rather than the first. In
   addition, the absence of Cpeb1 in neurons leads to mice exhibiting a
   higher freezing tendency compared to the absence of Cpeb1 in microglia
   during the second phase of extinction. These findings indicate that the
   absence of Cpeb1 in neurons may have a more pronounced and direct
   effect on fear extinction inflexibility.

   In our study, we provided previously unknown insights into the
   involvement of Cpeb1-mediated protein translation events across
   distinct cellular populations, further affirming the importance of
   Cpeb1 in fear extinction. It is known that Cpeb1 is a critical
   regulator of RNA translation and orchestrates various important
   physiological functions by modulating cytoplasmic polyadenylation, a
   crucial process for regulating the translation of specific mRNAs
   ([213]25). The precise mechanism by which Cpeb1 modulates translation
   and its specific impact on fear extinction across various cell types
   remains elusive and warrants further investigation. By integrating data
   from proteomics, RNA sequencing, and RNA immunoprecipitation
   sequencing, our study reveals that the absence of Cpeb1 triggers
   distinct translation events in neurons and microglia, respectively,
   thereby impairing fear extinction. Specifically, Cpeb1 binds with the
   unique CPEs and affects the poly(A) tail elongation of Hp1bp3 and
   Cx3cr1, leading to the translation suppression of Hp1bp3 in neurons and
   the translation activation of Cx3cr1 in microglia. In the Impaired-ext
   mice, neuronal Hp1bp3 is elevated while microglial Cx3cr1 is reduced,
   as seen in the Cpeb1 knockout mice. Silencing of neuronal Hp1bp3 or
   supplementing microglial Cx3cr1 effectively rescues fear extinction
   inflexibility in Cpeb1 conditional knockout mice. It may seem somewhat
   paradoxical that Cpeb1 regulates the translation of Hp1bp3 and Cx3cr1
   in neurons and microglia in opposite directions. However, because
   Cpeb1’s activation or inhibition of target gene translation is
   dependent on the number and relative positioning of CPEs within the
   3′UTR, it is reasonable for Cpeb1 to exert a bidirectional regulatory
   effect on Hp1bp3 and Cx3cr1 in different cell types ([214]26).

   Here, we also delve into the contribution of cell type–specific
   translational programs to fear extinction, uncovering that both
   neuronal Hp1bp3 and microglial Cx3cr1 are implicated in regulating
   synaptic function through direct and indirect pathways. Hp1bp3 is a
   multifunctional protein involved in various biological processes,
   including gene expression regulation, cell cycle progression, cell
   development, and miRNA biogenesis. By interacting with heterochromatin
   protein HP1, Hp1bp3 influences chromatin compaction and modulates gene
   silencing ([215]88). Hp1bp3 contributes to cell cycle progression by
   maintaining heterochromatin integrity during G[1]-S phase transition,
   thus regulating the duration of G[1] phase ([216]89). In addition,
   Hp1bp3 binds to nucleosomes and protects linker DNA from degradation by
   micrococcal nuclease, which also has been identified as a mediator of
   cell viability and growth ([217]90). Recently, Hp1bp3 has also been
   found to specifically associate with the microprocessor, promoting
   cotranscriptional miRNA processing ([218]31) and adding a previously
   unknown dimension to its regulatory role in gene expression. In the
   current study, we characterized the miRNA transcriptome and found that
   Hp1bp3 enhanced miRNA biogenesis by facilitating pri-miRNA processing
   rather than transcription, suggesting a widespread and conserved
   mechanism for Hp1bp3 in eukaryotes. Furthermore, we have identified the
   specific downstream molecules (Rem2, Gnal, Gria1, and Gria2) that acted
   as the common targets by the network of miRNAs. Rem2 is a member of the
   small Ras-like guanosine triphosphatase family, identified as an
   important mediator of synaptic development ([219]91). Knocking out Rem2
   in hippocampal neurons reduces dendritic spine density and maturation
   ([220]35). The Gα (olf) subunit of the Gnal-encoded guanine
   nucleotide–binding protein plays a key role in dopamine signaling in
   striatal medium spiny neurons ([221]92). Studies have found that the
   complexity of dendrites in striatal neurons of heterozygous Gnal
   knockout mice is significantly reduced ([222]34). Gria1 and Gria2 are
   two subunits of the AMPA receptor, which play a crucial role in
   synaptic transmission and plasticity ([223]38). It has been reported
   that interfering with AMPAR by peptides to inhibit the expression of
   GluA1 and GluA2 in optic tectal neurons of the Xenopus retinotectal
   system results in decreased dendritic branching and complexity
   ([224]37). Therefore, the down-regulation of these proteins in neurons
   is critical for the synaptic transmission and structure.

   In various settings, microglial cells, as resident brain macrophages,
   exhibit alterations in morphology, phagocytic capacity, and
   transcriptional profile to promptly detect neuronal changes and
   maintain brain homeostasis ([225]93). Cx3cr1 is a chemokine receptor
   expressed on microglial cells in the brain, and disruption of Cx3cr1 is
   accompanied by dysfunction of microglial status and function ([226]39).
   In our study, the transcriptomes of Cx3cr1-deficient microglia and
   fluorescence staining unveiled the aged-like microglia rather than a
   senescent phenotype in impaired fear extinction. Aged and senescent
   microglia are two distinct states of microglial cells that occur with
   advancing age. Aging microglia refer to those that have naturally aged
   over time ([227]94), exhibiting characteristics such as oxidative
   stress, DNA damage, telomere shortening, reduced phagocytic function,
   lipid metabolism disorders, and altered morphology (including fewer
   dendritic branches and larger cell bodies) ([228]94–[229]97).
   Senescence represents a more severe and irreversible stage of aging
   ([230]98), marked by irreversible growth arrest and development of a
   SASP ([231]99). Aged microglia and senescent microglia exhibit
   different characteristics under various stimuli or conditions. For
   example, microglia isolated from newborn mice show an aged phenotype in
   prolonged culture, featuring reduced phagocytic ability, motility, and
   autophagy, along with increased SA-β-gal activity ([232]100).
   Similarly, aged microglia isolated from aging transgenic mice via flow
   cytometry exhibit lipofuscin deposits, reduced process complexity,
   increased granularity, and enhanced expression of proinflammatory and
   anti-inflammatory cytokines ([233]97). Senescent microglia induced by
   chronic tau exposure exhibit cell cycle arrest, DNA damage, lamin-B1
   loss, impaired tau clearance, and formation of a SASP ([234]99).
   Iron-overloaded senescent microglia display decreased branching,
   elevated iron storage and ferritin expression, increased release of
   neurotoxic substances, and reduced capacity for phagocytosis of debris
   and toxic protein aggregates ([235]101). Senescent microglia induced by
   corticosteroid hormones show increased SA-β-gal activity, increased
   expression of tumor suppressor genes, and impaired phagocytic function
   ([236]102). In addition, transcriptomic differences are also observed
   in aged and senescence microglia. Senescent microglia induced by
   replicative stress are enriched in DAM signatures, accompanied by
   increased of SA-β-gal activity, telomere shortening, neuritis, and
   synaptic damage ([237]42). Aged-like microglia induced by three rounds
   of depletion–repopulation stimulation (3×DR) is enriched in ADAM
   landscapes, along with reduced phagocytic capacity, shortened
   telomeres, and hindered myelin development ([238]40). A subset of
   microglia rich in ADAM genes is found in Cx3cr1-deficient mice with
   extinction learning deficits, representing an age-like phenotype
   induced by fear extinction disorder. Moreover, GSEAs showed a
   substantial reduction in a set of phagocytosis genes in
   Cx3cr1-deficient microglia compared to WT microglia, suggesting a
   potential decline in phagocytic function in aged-like microglia
   influenced by fear extinction disorder.

   How does Cpeb1’s downstream effector Cx3cr1 in microglia impair fear
   extinction? Loss of Cx3cr1 in microglia was shown to result in the
   disruptions in the communications of neurons and microglia and
   influence synaptic pruning ([239]30). Previous researches have shown
   that activated microglia inhibit gamma-aminobutyric acid (GABA)
   neuronal activity by engulfing their dendritic spines, ultimately
   leading to the extinction of anxiety-like behaviors induced by
   restraint stress ([240]103). Developing juvenile microglia exhibit an
   immature state reminiscent that contributed to extinction learning
   deficits by elevated spine pruning and reduced spine formation
   ([241]104). Our results would be of much interest to raise that the
   aged-like microglia in nonaged brain disrupt synapse structural
   plasticity by preventing local clearance of the ECM around synapses,
   leading to the impairments in fear extinction. Our findings align with
   previous reports revealing that the ablation of Cx3cr1 in mice resulted
   in the deposition of the ECM in the hippocampal dentate gyrus (DG)
   region ([242]105), the deposition of ECM near synapses resulted in a
   decrease in the number of spines ([243]52), and degradation of ECM by
   the ChABC promoted the extinction of fear memories ([244]55).

   Although our findings suggest that Cpeb1-dependent translational events
   mediating fear extinction is predominantly restricted to the IL cortex,
   the contribution of Cpeb1 and its downstream signaling pathways to fear
   extinction is likely not confined to this region alone. For instance,
   Cpeb1 has been implicated in hippocampal-dependent memory and cortical
   plasticity, suggesting that the processes identified here could extend
   to memory systems beyond the mPFC ([245]106). Overexpression of Cpeb1
   has been shown to rescue deficits in long-term potentiation, a synaptic
   mechanism fundamental to learning and memory across various brain
   regions, including the hippocampus ([246]107). This indicates that
   Cpeb1 plays a crucial role in synaptic plasticity and memory formation,
   and its influence may generalize to other regions beyond the mPFC.
   Hp1bp3 knockdown in the hippocampus results in impaired working memory,
   linked to reduced synaptic plasticity in hippocampal neurons
   ([247]108). Hp1bp3’s epigenetic functions, particularly in chromatin
   remodeling, suggest that it could affect synaptic plasticity and memory
   formation not only in the mPFC but also in regions like the hippocampus
   and neocortex ([248]109). Meanwhile, Cx3cr1 signaling is critical for
   maintaining synaptic structure and plasticity across brain regions.
   Deletion of Cx3cr1 in mice has been shown to disrupt dendritic spine
   pruning, synapse maturation, and connectivity between the hippocampus
   and mPFC ([249]110). These mice also exhibit cognitive deficits,
   including impaired social interaction and repetitive behaviors and
   hippocampus-dependent spatial memory deficits ([250]29), highlighting
   Cx3cr1’s broader role in cognitive function. Our studies further
   indicate that Hp1bp3 overexpression in the IL cortex impairs
   AMPAR-mediated synaptic transmission, dendritic complexity, and spine
   morphology. Similarly, loss of Cx3cr1 in microglia disrupts synaptic
   structural plasticity by preventing local clearance of the ECM around
   synapses. These changes are both indicative of impaired synaptic
   plasticity, which is essential for adaptive learning and behavioral
   flexibility. Thus, the Cpeb1-mediated dysregulation of Hp1bp3 and
   Cx3cr1 likely contributes to deficits not only in fear extinction but
   also in broader cognitive functions, such as hippocampus-dependent
   working memory and behavioral flexibility.

   In summary, we identified cell type–specific translational programs in
   the IL that are essential for fear extinction, underscoring the
   importance of thoroughly investigating the molecular mechanisms of cell
   type–specific translation. These findings have significant implications
   for the development of previously unidentified tools and approaches
   aimed at addressing PTSD and potentially other disorders characterized
   by deficits in extinction.

MATERIALS AND METHODS

Animals

   Adult male C57BL/6 J mice were purchased from the National Resource
   Center of Model mice in Nanjing, China. Genotyping for Cx3cr1-Cre mice
   (obtained from Shanghai Model Organisms, China) was performed via
   multiplex PCR using the primer pair 5′-ACACCAGAGACGGAAATCCATCG-3′ and
   5′-CAACGAGTGATGAGGTTCGCAAG-3′. Cpeb1^flox/flox mice were generated from
   GemPharmatech Co., Ltd. in Jiangsu, China. Genotyping for
   Cpeb1^flox/flox mice was conducted using PCR with primer sequences
   5′-GAGCTCACTGTTTCTTGGTGTCCCT-3′ and 5′-CACAGTTGTTACCAGTCACAAGTGG-3′.
   All mice were housed in a controlled environment, maintaining a 12-hour
   light/dark cycle, with a stable temperature (22° ± 2°C) and humidity
   ranging from 40 to 60%. They were provided with ad libitum access to
   food and water. Ethical approval for all experimental procedures was
   obtained from the Institutional Animal Care and Use Committee of
   Huazhong University of Science and Technology (approval no.
   2020-S2317). The mice, along with their littermates, were bred at the
   Experimental Animal Central of Tongji Medical College, Huazhong
   University of Science and Technology.

Microarray-based mRNA and miRNA expression profiling

   Microarray-based mRNA and miRNA expression profiling were conducted
   following standardized procedures. Tissues were homogenized using a
   TissueRuptor homogenizer (Qiagen, Palo Alto, CA, USA) operating at
   50/60 Hz, in accordance with the manufacturer’s instructions. Total RNA
   extraction was performed, and samples were treated with ribonuclease
   (RNase)–free deoxyribonuclease (Qiagen, Palo Alto, CA, USA). RNA
   concentration was determined using a spectrophotometer (DU-640,
   Beckman, USA), and RNA integrity was assessed via 1% agarose gel
   electrophoresis. miRNA expression analysis was performed using the
   Mouse OneArray v2 (RmiOA.2) chip, while gene expression analysis used
   the OneArray microarray v2 (MOA-002) chip. Data normalization and
   analysis were carried out using the GeneChip Robust Multichip Average
   method, with normalization conducted using OneArray Software (Phalanx
   Biotech Group). Only hybridized chips meeting predefined quality
   control standards were included in the analysis to determine fold
   changes. Statistical significance was defined as P < 0.05.

Mass spectrometry and data analysis

   Proteins were extracted from the mPFC of both the Normal-ext and
   Impaired-ext groups. The protein lysate underwent heat treatment at
   70°C for 1.5 hours to facilitate the release of cross-linked fixed
   proteins. Subsequently, protein precipitation was achieved by adding
   precooled acetone, followed by sequential washing steps with precooled
   acetone, ethanol, and acetone. The resulting protein pellet was
   resuspended in UA buffer (8 M urea in 0.1 M tris-HCl), and
   dithiothreitol (final concentration, 2 mM) was added, followed by
   incubation at 30°C for 1.5 hours. Iodoacetamide (Sigma-Aldrich) was
   then introduced to the sample, which was incubated in darkness for 60
   min. Trypsin was added for overnight digestion, and the reaction was
   terminated with trifluoroacetic acid (Sigma-Aldrich). Following salt
   depletion using C18 spin tips (Thermo Fisher Scientific), the sample
   was loaded onto the Bruker timsTOF Pro mass spectrometer following the
   manufacturer’s protocol. The assay was performed as previously
   described ([251]111).

   Data processing was conducted using PEAKS software. The spectral
   abundance factor was determined by normalizing the number of protein
   spectra to the length of the respective proteins. Subsequently,
   normalization among samples was achieved by dividing the spectral
   abundance factor by the sum of all spectral abundance factors.
   Statistical significance was determined with a threshold of P < 0.05
   and fold change > 1.2. Protein expression hierarchical clustering
   analysis was performed using Biovinci 1.1.5.

Cpeb1 RNA immunoprecipitation sequencing and data analysis

   The mPFC from C57BL6/J mice was isolated, and the tissue in ice-cold
   phosphate-buffered saline (PBS) was gently dissociated using a Dounce
   homogenizer. The cells were collected by centrifuging at 1500 rpm for 5
   min at 4°C, and the supernatant was discarded. The cell pellet was
   resuspended in an equal volume of complete RIP Lysis Buffer (containing
   RIP lysis buffer, protease inhibitor cocktail, and RNase inhibitor).
   The lysate was incubated on ice for 5 min, then transferred into
   nuclease-free microcentrifuge tubes, and stored at −80°C.

   For immunoprecipitation bead preparation, 50 μl of magnetic bead
   suspension was taken and the beads were washed with RIP wash buffer.
   After washing, the beads were resuspended in 100 μl of RIP wash buffer,
   and 5 μg of the Cpeb1 antibody (lot 13274-1-AP, Proteintech, China) was
   added. The mixture at room temperature was rotated for 30 min and
   briefly centrifuged. The tubes were placed on a magnetic separator, and
   the supernatant was discarded. After a final wash with RIP wash buffer,
   the antibody-bound beads were resuspended in RIP immunoprecipitation
   buffer.

   To begin the immunoprecipitation reaction, the RIP lysate was mixed
   with RIP immunoprecipitation buffer (comprising RIP wash buffer, 0.5 M
   EDTA, and RNase inhibitor), adjusting the final volume to 1 ml and
   incubated at 4°C overnight with gentle rotation.

   The following day, the supernatant was discarded, and the beads were
   washed six times with PIP wash buffer, rotating for 5 min at 4°C for
   each wash. Each immunoprecipitate was resuspended in the proteinase K
   buffer (containing RIP wash buffer, 10% SDS, and proteinase K) and
   incubated at 55°C for 30 min with gentle shaking to ensure complete
   protein digestion. The supernatant was transferred to a fresh
   microcentrifuge tube, and 250 μl of RIP wash buffer was added, followed
   by 400 μl of phenol:chloroform:isoamyl alcohol. RNA extraction and
   purification were performed as previous protocol ([252]112).

   CPEB1 immunoprecipitation, RNA extraction and cDNA generation were
   carried out following the protocol of the RIP kit (lot 3778111,
   Millipore, USA). The cDNA library products ranging from 200 to 500 bp
   were enriched, quantified, and ultimately sequenced on a NovaSeq 6000
   sequencer (Illumina) using the PE150 model.

   Raw data analysis was carried out according to the published protocol
   ([253]113). Briefly, STAR software (version 2.5.3a) was used with
   default parameters. RSeQC (version 2.6) was used for reads distribution
   analysis. The ExomePeak software (version 3.8) was used for peak
   calling, with subsequent annotation using BedTools (version 2.25.0).
   DeepTools (version 2.4.1) facilitated peak distribution analysis.
   Differentially binding peaks were identified through a Python script
   using Fisher test. HomerSoftware (version 4.10) was used to determine
   sequence motifs enriched in peak regions. To mitigate potential
   false-positive results, criteria were set for fold change of IP/IgG
   FPKM (fragments per kilobase transcript per million mapping reads) > 2
   and P < 0.05 for each experiment. Scatter plots of Cpeb1-bound genes
   were generated using GraphPad 8. GO analysis was conducted using KOBAS
   software (version: 2.1.1).

Behavioral assays

Fear conditioning and extinction paradigm

   On the first day of fear conditioning, animals underwent habituation in
   the conditioning chamber (context A) for 3 min. Subsequently, they were
   exposed to an auditory fear conditioning paradigm comprising four
   pairings of a 30-s tone (CS, 1 kHz, 80 dB) with a 1-s foot shock (US,
   0.5 mA), with a 20-s intertrial interval. The onset of the US coincided
   with the termination of the CS. Following training, animals were
   returned to their home cages, and the conditioning chambers were
   sanitized with 70% ethanol.

   On the second day, mice were allowed to rest in their cages.
   Concurrently, Con (On) and gePSI (On) groups administered
   Dox-containing water (1 mg/ml) for gene expression manipulation.

   On the third and fourth days, extinction training and recall testing
   were conducted in context B. Animals received 14 CS presentations with
   a 20-s intertrial interval each day. The “No-extinction” group was
   placed in context B for the same duration without any stimuli.
   “Acquisition” referred to the average freezing during the initial three
   trials of extinction training on the third day. “Extinction in session
   1 and session 2” denoted the average freezing across all trials during
   the extinction phase on the third and fourth days, indicative of
   extinction learning and retrieval, respectively ([254]19). Video
   tracking technology was used to monitor animal activity and freezing
   behavior, with freezing defined as the absence of movement for 2 s.

   Behavioral process for fear conditioning training, Tet-on system, and
   fear extinction experiment in [255]Fig. 1B was performed as follows:

   On day 1, mice underwent habituation (Hab) followed by fear
   conditioning (Cond). On day 2, gePSI-expressing mice were treated with
   Dox (gePSI-On) to inhibit protein synthesis, while the control group
   (gePSI-Off) were treated with vehicle. On days 3 and 4, both groups
   underwent two extinction sessions (session 1 and session 2). During
   these sessions, Group gePSI (On) continued receiving Dox, and Group
   gePSI (Off) continued with the vehicle. On day 5, Group gePSI (On) was
   split into two subgroups: gePSI (On) and gePSI (On-Off). The gePSI (On)
   group maintained Dox administration, while the gePSI (On-Off) group
   switched to the vehicle for 24 hours. On days 7 and 8, mice underwent
   another two extinction sessions (session 3 and session 4), with Group
   gePSI (On) continuing Dox and gePSI (On-Off) receiving the vehicle.

Elevated plus maze test

   The elevated plus maze consisted of two closed arms with 15-cm high
   transparent walls and two open arms with a slight threshold along the
   sides. Each mouse was placed in the center of the maze, facing one of
   the open arms. The animal’s movements were recorded for 10 min using
   video capture software. The number of visits and the time spent in the
   open and closed arms were quantified. This experiment was conducted as
   previously explained ([256]112).

Open-field test

   This behavioral paradigm was performed as previously described
   ([257]114). The open-field test involved placing each mouse in a 40 cm
   by 40 cm by 30 cm arena for 10 min. Video capture software (WinAVI
   Video Capture, ZJMedia Digital Technology, China) recorded the mice’s
   movements. The total distance traveled and the time spent in the center
   of the arena were measured. Before each trial, the open-field arena was
   sanitized with 70% ethanol to maintain cleanliness and hygiene.

Principal components analysis

   A comprehensive understanding of the neural mechanisms underlying fear
   extinction heavily depends on distinguishing individual mice that
   exhibit different extinction characteristics, particularly those mice
   that exhibit resilience to the extinction process (Impaired-ext) and
   those more susceptible to it (Normal-ext). To identify these
   differences, we analyzed the mean percentage of freezing across the 14
   trials in session 2 of extinction learning, as well as the average
   freezing percentage during the last three trials in a large sample.

   Following manual pattern identification, we used PCA ([258]19) for
   dimensionality reduction and clustering of the mice’s original test
   data. By integrating manual identification with data-driven
   multivariate analysis, we aim to systematically capture potential
   behavioral variability within the mouse population.

Stereotaxic injection

   Mice were anesthetized with ketamine (100 mg/kg) and dexmedetomidine
   (0.5 mg/kg), The head was then secured in the stereotaxic apparatus
   (RWD Life Science, China). The scalp was sterilized with iodophors,
   bilateral holes were drilled, and 0.3 μl of the virus was microinfused
   into the IL using an automatic microinjection system (Reno, NV, USA).
   The AAV2/8-SYN-Cre-P2A-EGFP, AAV2/8-SYN-P2A-EGFP,
   AAV2/9-SYN-Hp1bp3-3×FLAG-P2A-EGFP, AAV2/9-SYN-3×FLAG-P2A-EGFP, U6-shRNA
   (Hp1bp3)-CMV-mCherry, and U6-shRNA (NC)-CMV-mCherry were purchased from
   Obio Technology (Shanghai, China). The
   AAV/9-SYN-mCherry-5′miR-30a-shRNA (Cpeb1)-3′miR-30a,
   AAV2/6-DIO-Cx3cr1-P2A-mCherry, AAV2/6-CMV-DIO-(EGFP-U6)-shRNA (Cx3cr1),
   and corresponding control virus were purchased from BrainVTA (Wuhan,
   China). In addition,
   AAV9-TRE3G-α-gePSI-P2A-β-gePSI-T2A-EGFP-bGHpolyA-CMV-rtTA-hGHpolyA,
   AAV2/6-CMV-DIO-(EGFP-U6)-shRNA (Cpeb1), AAV/9-SYN-Cre-P2A-mCherry, and
   corresponding control virus were obtained from Brain Case (Shenzhen,
   China). The correct valid sequence for sh-Cpeb1 is GCCGAAGGATGCGCTGCAA,
   for sh-Cx3cr1 is GTTCATGTTCACAAAGAGAAA, and for sh-Hp1bp3 is
   CAATCTTAACTGAGGCCATTA. All viral vectors were targeted bilaterally to
   the IL region in C57BL/6 J, Cpeb1^flox/flox, and Cx3cr1-Cre mice at
   coordinates anterior-posterior (AP): −1.75 mm, medial-lateral (ML):
   ±0.3 mm, dorsal-ventral (DV): −2.8 mm. hU6-shRNA
   (Cpeb1)-Hef1a-mCherry-3×FLAG and hU6-shRNA (NC)-Hef1a-mCherry-3×FLAG
   viruses were targeted bilaterally to the PL region in C57BL/6 J mice at
   coordinates AP: −2.4 mm, ML: ±0.25 mm, DV: −1.8 mm, and 0.3 μl of the
   virus for each site at an infusion rate of 0.03 μl/min using a Hamilton
   microsyringe (Hamilton). For SUnSET in vivo immunoprecipitation, 3
   weeks after the injection of gePSI or control virus, mice underwent a
   second surgical procedure to implant stainless steel guide cannulas
   (RWD, China) into the left lateral ventricle (AP: −0.50 mm, ML: −1.10
   mm, DV: −2.20 mm). A skull screw was inserted, and dental cement (RWD,
   China) was used to securely fix the cannula in place. For SUnSET in
   vivo immunohistochemistry, mice injected with either gePSI or control
   virus were implanted with two guide cannulas (spaced 0.7 mm apart) in
   the IL region to allow for puromycin infusion.

Electrophysiology recording

   Mice were induced into anesthesia, and coronal brain slices were
   prepared in artificial cerebrospinal fluid (ACSF) aerated with 95%
   oxygen and 5% carbon dioxide. The composition of the ACSF solution was
   124 mM NaCl, 3.0 mM KCl, 2.0 mM CaCl[2], 1.2 mM MgSO[4], 1.25 mM
   KH[2]PO[4], 26 mM NaHCO[3], and 11 mM glucose. Brain sections of 300-μm
   thickness were obtained using a vibrating microtome (Leica).
   Subsequently, the slices were incubated in oxygenated ACSF at 30°C for
   a minimum of 30 min before being transferred to a recording chamber
   equipped with a planar multielectrode recording setup (MED64, Alpha Med
   Science, Tokyo, Japan).

   For mEPSC recordings, neurons were maintained at −70 mV in
   voltage-clamp mode. Recordings used an internal solution comprising 140
   mM potassium gluconate, 10 mM Hepes, 0.2 mM EGTA, 2 mM NaCl, 2 mM
   MgATP, and 0.3 mM NaGTP. The external solution contained 10 μM
   bicuculline (GABA[A]R antagonists), 1 μM tetrodotoxin (sodium channel
   blocker) and 50 μM AP-5 (NMDAR antagonists). Events were filtered at 2
   kHz and sampled at 10 kHz (Molecular Devices, Sunnyvale, CA, USA) using
   pClamp 10.2 software. Currents were recorded in 10-s epochs for a
   minimum total duration of 200 s per recording. The electrophysiology
   experiment was performed as described previously ([259]115). Analysis
   of the mEPSC signal was performed in Clampfit 10.0 software.

Paired-pulse stimulation

   We introduced 10 μM CNQX into the perfusate and positioned the
   stimulation electrode in mPFC neurons to record evoked somatic
   currents. The paired-pulse paradigm used a 50-ms interpulse interval,
   with the objective of assessing the presynaptic release probability of
   the observed synapses. We administered five pairs of stimuli with a 5-s
   interval and measured the peak amplitudes of the EPSCs, as described
   previously ([260]115). The paired-pulse ratio was determined by
   calculating the peak amplitude ratios of the first and second
   responses. Signal analysis was conducted using Clampfit 10.0 software.

Golgi staining

   Golgi staining was performed with the FD Rapid Golgi Stain Kit (catalog
   no. PK401, FD Neuro Technologies Inc.) following the manufacturer’s
   protocol. Mice were anesthetized and underwent perfusion with 0.9%
   saline, followed by prompt extraction of the brain. After rinsing with
   double distilled water, the brain was immersed in an impregnation
   solution (A/B solution in equal parts) and left to incubate in darkness
   at room temperature for a duration of 2 weeks, with the solution
   refreshed every 3 days. Subsequently, the brain was transferred to
   solution C and further incubated in darkness at room temperature for 1
   week. Brain sections of 100-μm thickness were obtained using a
   vibrating microtome (Leica VT1000s) and mounted on high-adhesion glass
   slides with solution C. These sections were then dehydrated in
   sequential ethanol solutions for 10 min each and cleared in xylene
   thrice for 10 min each. The staining procedure was carried out in
   accordance with established protocols ([261]32). Images were captured
   using a Coolpix 5000 Nikon camera. Sholl analysis and assessment of
   dendritic spines were performed using Fiji software.

Sholl analysis

   Sholl analysis was conducted to assess the complexity of the dendritic
   trees and microglia using established methods ([262]116, [263]117).

Immunofluorescence

   Mice were anesthetized with ketamine (100 mg/kg) and dexmedetomidine
   (0.5 mg/kg) and perfused with 0.9% saline, followed by 4%
   paraformaldehyde (PFA) for fixation. Then, brains were extracted and
   fixed with 4% PFA for an additional 24 hours. Brain slices (30 μm) were
   then obtained using a vibratome (VT1000S, Leica, Germany). Brain
   sections underwent three PBS washes and were penetrated with 0.1%
   Triton X-100 and 5% bovine serum albumin in PBS for 1 hour at room
   temperature. Subsequently, the sections were incubated with specific
   primary antibodies overnight at 4°C. Following primary antibody
   incubation, the sections were washed three times with PBS and then
   exposed to Alexa Fluor 546– or Alexa Fluor 488–conjugated donkey
   secondary antibodies (Invitrogen, USA) for 1 hour at room temperature.
   4′,6-Diamidino-2-phenylindole (DAPI) staining was performed to
   visualize cell nuclei. Imaging was conducted using a laser confocal
   microscope (LSM800, Zeiss, Germany). Immunofluorescence data
   acquisition was carried out in a blinded manner, with random imaging
   and cell counting performed in multiple fields per sample to quantify
   the number of positively stained cells. Primary antibodies used for
   immunohistochemistry and immunofluorescence are listed in table S2.

3D resconstruction

   The 3D reconstruction of images was performed using Imaris x64 9.5.0
   software (Bitplane Software). Immunofluorescence images were captured
   with a laser confocal microscope (LSM800, Zeiss, Germany), with the
   z-stack z-step size adjusted as necessary to encompass the desired
   range. Subsequently, the acquired 3D images were processed,
   reconstructed, and analyzed using Imaris x64 9.5.0 software.

Western blotting

   Protein extraction from the mPFC was conducted, and the protein
   concentration was quantified using the BCA Assay Reagent (Thermo Fisher
   Scientific, USA). Western blotting was performed as described in
   previous studies ([264]112). The membranes were blocked with 3%
   defatted milk for 1 hour at room temperature, followed by overnight
   incubation with primary antibodies at 4°C. Subsequently, protein bands
   were probed with goat anti-mouse or anti-rabbit antibodies conjugated
   to IRdye 800 (LI-COR Biosciences, USA) at room temperature for 1 hour.
   Visualization of protein bands was achieved using Odyssey software
   (LI-COR Bioscience, USA) after washing with Tris-Buffered Saline with
   Tween 20 (TBST), and analysis was performed using ImageJ software.
   Primary antibodies are shown in table S2.

Surface labeling of translation in vivo

   The SUnSET assay was performed as previously described ([265]11,
   [266]21), In brief, for the SUnSET in vivo immunoprecipitation, mice
   injected with gePSI or control virus were fed with Dox or solvent.
   After 24 hours, 1 μl of 25 μg/μl puromycin (P8833; Sigma-Aldrich) was
   injected into the left lateral ventricle. One hour after puromycin
   injection, mPFC tissue was removed and stored at −80°C, following the
   procedure of the Western blotting. For SUnSET in vivo
   immunofluorescence, mice injected with gePSI or control virus were
   treated with DOX or solvent. Twenty-four hours later, puromycin (0.5 μl
   of 10 μg/μl) was injected into the IL. One hour after puromycin
   injection, the brain was perfused, sliced for staining, and underwent
   the steps outlined for immunofluorescence staining.

Real-time PCR

   Total RNA was extracted from either cultured cells or brain tissues
   using TRIzol Reagent (Life Technologies, China) following the
   manufacturer’s instructions. Reverse transcription PCR (RT-PCR) was
   carried out as previously detailed ([267]118). Briefly, 1 μg of total
   RNA was reverse transcribed into mRNA or miRNA using either the Reverse
   Transcription Kit (FSK-100, Toyobo, Japan) or the miRcute Plus miRNA
   First-Strand cDNA Kit (Tiangen, Beijing, China), respectively. RT-PCR
   was performed using SYBR Green PCR Master Mix (Takara, Japan) on the
   CFX96 Real-Time PCR Detection System (Bio-Rad, CA, USA). Gene
   expression levels were analyzed relative to glyceraldehyde phosphate
   dehydrogenase using the 2^-ΔΔCt method. The primers designed are listed
   in table S3.

Poly(A) tail-length assay

   Total RNAs were extracted from cells treated with different
   interventions. Polyadenylation and reverse transcription were carried
   out using the Poly(A)-Tail Length Assay kit (lot 01272451, Thermo
   Fisher Scientific, USA) as per the manufacturer’s instructions. Poly(A)
   tail PCR was conducted using a universal reverse primer and
   gene-specific forward primers. The poly(A) tail and gene-specific PCR
   products were separated on a 2% agarose gel. In addition, the poly(A)
   tail PCR products were sequenced to determine the actual length of
   poly(A) tails. Gene-specific PCR was performed with a pair of
   gene-specific primers, and the primers designed are listed in table S4.

Dual luciferase assay

   Cpeb1 was inserted into the pcDNA3.1(+) vector (Addgene), while the
   Hp1bp3 or Cx3cr1 3′UTR was inserted into the psiCHECK-2 vector
   (Promega). Mutated CPE sequences in the Hp1bp3 or Cx3cr1 3′UTR were
   inserted into the psiCHECK-2 vector using the ClonExpress Ultra One
   Step Cloning Kit (Vazyme, Nanjing, China, catalog no. C115-02).
   Subsequently, the psiCHECK-2- Hp1bp3 or Cx3cr1 3′UTR, alone with
   pcDNA3.1(+) or pcDNA3.1(+)-Cpeb1, were cotransfected into 293 T cells
   cultured in 24-well plates. Luciferase activity was quantified using
   the Dual Luciferase Reporter Assay Kit (Vazyme, Nanjing, China, catalog
   no. DL101-01) and measured on the BioTek Synergy2.

Cell culture

   Mouse BV2 cells were brought from American Type Culture Collection (MD,
   USA), while mouse HT22 cells were brought from the MilliporeSigma
   (catalog no. SCC129). BV2 cells were cultured in cell culture medium
   (Pricella, China), whereas HT22 cells were cultured in Dulbecco’s
   modified Eagle’s medium (Thermo Fisher Scientific, Carlsbad, USA)
   supplemented with 8% fetal bovine serum (v/v) and 1%
   penicillin/streptomycin. Cell culture was performed as previously
   described ([268]119). Both cells were incubated at 37°C in a 5% CO[2]
   environment. Cells were randomly allocated to different experimental
   groups. Plasmid transfections were performed using Lipofectamine 3000
   (Invitrogen, MA, USA) according to the manufacturer’s instructions.

Phagocytosis assay

   The BV-2 cells were cultured in plates and then randomly assigned to
   two groups. Each group was transfected with plasmids that silenced
   Cx3cr1 gene, along with their respective control groups. In addition,
   these plasmids carried a red fluorescent marker. After 48 hours, the
   yellow-green fluorescence-labeled beads (Sigma-Aldrich, USA, L1030)
   were added for 60 min at 37°C. After that, the cells were rinsed three
   times with PBS to eliminate any beads that were not engulfed and were
   then immobilized using 4% PFA. Then, the cells were treated with 0.1%
   Triton X-100 for 5 min to allow permeabilization, followed by
   incubation with DAPI (Servicebio, Wuhan, China) for 15 min to stain the
   cell nuclei. Last, the phagocytosis of beads by microglia was observed
   using a confocal microscope.

Statistics

   All the statistical analyses were done using GraphPad Prism 8.0
   software. All data are shown as means ± SEM. Unpaired or paired t tests
   (two tailed) were used for individual comparisons, while one-way or
   two-way analyses of variance (ANOVAs) followed by Tukey’s multiple
   comparisons test were used for single-variable comparisons. The
   statistical significance level was defined as P < 0.05. The number of
   biological replicates and the statistical tests used in the article are
   detailed in table S5.

Acknowledgments