Abstract

Background

   Ephrin-B2 (EB2) signaling plays a crucial role in regulating memory and
   synaptic plasticity. Comprehensive identification of cell-type-specific
   transcriptomic changes in EB2 knockout mice is expected to shed light
   on potential mechanisms associated with EB2 signaling in cognitive
   functions.

Results

   Our study captures changes in cell populations in response to EB2
   manipulation and reveals previously uncharacterized cell types
   (CPA6 + inhibitory neurons) in the mPFC. We validated the differential
   transcriptomic activity of Pbx1 and Meis1 in CPA6 + neurons using
   fluorescence in situ hybridization (ISH) in EB2-vGATCre mice. The
   aberrant presence of CPA6 + neurons in the mPFC may correlate with
   cognitive impairments induced by EB2 deletion in vGAT + neurons.
   Analyzing differentially expressed genes (DEGs) in individual cell
   clusters, we identified alterations related to synapse organization and
   development, cognition, amyloid-beta formation, and locomotor behavior.
   Additionally, our DEGs overlapped with human genome-wide association
   study (GWAS) candidate genes related to cognition and anxiety,
   underscoring the relevance of our mouse model to human disease.

Conclusions

   We present a comprehensive atlas of cell-type-specific gene expression
   changes in this synaptic deficiency model and identify novel
   cell-type-specific targets implicated in cognitive deficits. Our
   investigation provides a detailed map of the cell types, genes, and
   pathways altered in this inhibitory synaptic deficiency model.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12915-025-02333-5.

   Keywords: Ephrin-B2, Cognition, Single-nucleus RNA sequencing,
   GABAergic neurons

Background

   GABAergic inhibitory neurons are essential components of neural
   circuits, working together with excitatory neurons and glia to maintain
   the balance necessary for normal brain development and function. They
   constitute approximately 20–30% of the cortical neurons. Dysfunction in
   their synapse can disrupt the delicate balance between excitation and
   inhibition, potentially leading to neurological disorders or cognitive
   impairments [[42]1]. Molecular mechanisms likely underlie the
   development of different inhibitory synapses [[43]2]. These synaptic
   molecules play a crucial role in shaping cellular and synaptic programs
   in inhibitory subpopulations, both during developmental phase and
   adulthood, influencing behavioral processes like learning and memory
   [[44]3]. Genetic manipulation of these synaptic molecules within
   inhibitory neurons could impact a heterogeneous population of neuronal
   cells, and subsequently coordinate an array of fundamental behaviors.
   However, our understanding of how genetic alterations in synaptic
   molecules affect the diverse population of inhibitory neuronal
   population and their interaction with other brain cell types, as well
   as the broader functional connectivity between different brain regions,
   remains limited.

   Ephrin-Eph ligand-receptor synaptic signaling is a crucial molecular
   guidance system in the central nervous system with versatile roles in
   synaptic function and dysfunction [[45]4]. Specifically, Ephrin-B2
   (EB2), which is a transmembrane molecule, can transduce a reverse
   signal into EB2-expressing cells [[46]5]. In excitatory neurons, EB2
   regulates synaptic strength by modulating AMPA receptor trafficking
   [[47]6, [48]7]. While EB2’s importance in excitatory neurons was well
   documented, its function in inhibitory neurons has received less
   attention until recent studies, which detected EB2 expression in nearly
   all inhibitory neurons and suggested its role in transducing reverse
   signals to guide their migration into neocortex during embryonic
   development [[49]8]. Loss of Ephrin-B in the major class of GABAergic
   interneurons (GAD65/67 + interneurons) affected distinct interneuron
   subpopulations in the cortex [[50]9]. Despite these advancements, our
   knowledge of EB2 reverse signaling pathways in GABAergic neurons which
   are critical for shaping higher brain connectivity and other cellular
   functions remains rudimentary. Further investigation is also required
   to determine how EB2 loss affects distinct cell subtypes in the brain
   which is not confined to inhibitory neurons. We hypothesized that EB2
   ablation in GABAergic neurons influences distinct subtypes of cell
   population at transcriptomic level.

   To investigate our hypothesis, we generated mice with Ephrin-B2
   specifically knocked out in vGAT + neurons (vGAT-Cre; Ephrin-B2^loxP,
   EB2-vGATCre mice). Selective restoration of EB2 expression in GABAergic
   neurons of the PFC reversed learning and memory deficits in EB2-vGATCre
   mice. Additionally, these mice exhibited impaired resting-state network
   connectivity between the PFC and the reticular nucleus of the thalamus
   (RNT) region, along with impaired synaptic spatial plasticity in the
   prefrontal cortex (PFC), underscoring the impact of EB2 loss in
   inhibitory neurons. To comprehensively assess changes in cell-type
   diversity following EB2 ablation in vGAT + neurons, we used a
   single-nucleus RNA sequencing to identify novel populations and uncover
   transcriptomic diversity within known populations. Our snRNA-seq data
   enabled us to identify a novel population of transcriptionally and
   topographically distinct neurons following EB2 deletion in
   vGAT + neurons (CPA6 + neurons). Furthermore, we validated the
   differential transcriptomic expression of Pbx1 and Meis1 using
   fluorescence in situ hybridization (ISH), confirming their expression
   changes within EB2-vGATCre mice. Taken together, these findings enhance
   our understanding of the EB2-mediated molecular basis of cell-type
   heterogeneity within the PFC, laying the groundwork for targeted
   investigations into circuit-specific manifestations of neurocognitive
   disorders.

Results

Cognitive deficits in GABAergic neuron-specific EB2 KO mice are rescued by
targeted EB2 over-expression in the PFC

   GABAergic neuron-specific EB2 KO mice were used for this study
   [[51]10]. The specific knockout of EB2 in GABAergic neurons is
   illustrated in Additional file [52]1: Fig. S1A and B. Cells expressing
   vGAT-Cre showed tdTomato fluorescence, marking EB2 KO. In the PFC,
   tdTomato-positive neurons colocalized with GABA and GAD65/67,
   confirming vGAT-Cre specificity in GABAergic neurons (Fig. [53]1A and
   B), with a minor activity of vGATCre in PV neurons (Additional file
   [54]1: Fig. S1C), but not in vGluT2 and CaMK2α neurons (Additional file
   [55]1: Fig. S1D, E).

Fig. 1.

   [56]Fig. 1
   [57]Open in a new tab

   Specific ablation of EB2 in vGAT + neurons induced cognitive deficits.
   A and B In vGATCre mice, the neurons labeled with Ai9 Rosa26 reporter
   tdTomato were expressing GABA and GAD65/67. DAPI indicated the nuclei
   location. Scale bar, 20 μm. C NORT discrimination index in EB2-vGATCre
   mice as compared with EB2-flox mice (left) adolescent, **P = 0.0097,
   Mann–Whitney U test. (right) Adult, **P = 0.0012, Mann–Whitney U test.
   D Left, schematic representation of the tamoxifen treatment protocol.
   Behavioral tests were carried out during postnatal week 4 and week 10.
   Bar plots showing the NORT discrimination index in Cre + T mice
   compared with W/OCre + T and Cre + V mice. Middle, P = 0.0051,
   Kruskal–Wallis ANOVA test, W/OCre + T vs. Cre + V, P > 0.9999,
   W/OCre + T vs. Cre + T, *P = 0.0153, Cre + V vs. Cre + T, *P = 0.0227.
   Right, P = 0.0298, ordinary one-way ANOVA test, W/OCre + T vs. Cre + V,
   P = 0.9987, W/OCre + T vs. Cre + T, *P = 0.0416, Cre + V vs. Cre + T,
   P = 0.0658. E NORT discrimination index in Ephrin-B2.^lacZ/+mice as
   compared with WT mice (*P = 0.0317, Mann–Whitney U test). F Specific
   restoration of Ephrin-B2 by the precise bilateral injections of
   pAAV-EF1a-DIOEfnb22A-EGFP (OE virus) rescues NOR behavior deficit.
   P = 0.0038, ordinary one-way ANOVA test, EB2-vGATCre + Control vector
   vs. EB2-vGATCre + OE vector, **P = 0.0041, EB2-vGATCre + Control vector
   vs. Off target + OE vector, P = 0.7480, EB2-vGATCre + OE vector vs. Off
   target + OE vector, *P = 0.0274. G Mice injected with control vector
   and pAOV-CAMKIIα-GFP-2A-Cre showed no difference in NORT discrimination
   index (P = 0.9623, Mann–Whitney U test)

   To investigate the behavioral consequence of EB2 deletion in
   vGAT + neurons, we conducted a series of tests on EB2-vGATCre mice and
   their littermates (EB2-flox mice). In the novel object recognition
   (NOR) test, EB2-vGATCre mice exhibited pronounced cognitive
   impairments. Unlike EB2-flox mice, neither adolescent nor adult
   EB2-vGATCre mice preferred the novel object compared over the familiar
   one, as indicated by the reduced exploration. The impairment was
   quantified using as a preference index, with EB2-vGATCre mice showing
   significantly lower novelty preference indices for duration and
   frequency compared to EB2-flox mice (Fig. [58]1C). Similar cognitive
   deficits were observed in EB2-flox Cre mice and EB2^lacZ/+ mice
   (Fig. [59]1D, E). Additionally, mice with EB2 ablation in
   vGAT + neurons showed mild deficits in anxiety-like and depression-like
   behaviors (Additional file [60]1: Fig. S2A, B), but no significant
   abnormalities in prepulse inhibition (PPI), or compulsive behavior
   paradigms (Additional file [61]1: Fig. S2C, D).

   Numerous psychiatric and neurological conditions are associated with
   cognitive deficits linked to PFC functions [[62]11]. Inspired by these
   studies, we investigated the involvement of the PFC in the abnormal
   cognitive behaviors observed in EB2 mutants. We bilaterally
   microinjected a Cre-dependent AAV vector (pAAV-EF1a-DIO-Efnb2-2A-EGFP)
   into the mPFC of mice to specifically overexpress EB2 in the vGAT
   Cre-expressing neurons of EB2-vGATCre mice. Control animals received
   viruses expressing only EGFP (pAAV-EF1α-DIO-EGFP). Three weeks after
   virus injection, we validated the overexpression in the injected
   animals and performed NOR behavioral assays. EB2-vGATCre mice with EB2
   overexpression in the PFC showed significant restoration of cognitive
   impairments (Fig. [63]1F). To further confirm whether EB2 contribute to
   cognitive impairment specifically though its absence in inhibitory
   neurons, we use a viral assay to knock out EB2 in excitatory neurons in
   the PFC, followed by cognitive behavioral studies. The results showed
   that knockout of EB2 in CamKIIα + excitatory neurons did not impair
   cognitive function (Fig. [64]1G). Together, these findings demonstrate
   that deletion of EB2 in vGAT + neurons impaired recognition memory.

Impaired paired-pulse ratio (PPR) and altered PFC-thalamic connectivity from
EB2 deletion in PFC inhibitory neurons

   EB2 deletion from vGAT + neurons disturbed local excitatory and
   inhibitory synaptic formation and transmission [[65]8, [66]10]. To
   evaluate functional changes in the PFC neural network, we performed
   microelectrode array recordings (Fig. [67]2A) in mice with specific EB2
   knockout in inhibitory neurons. As shown in Fig. [68]2B and C,
   EB2-vGATCre mice exhibited significantly reduced activity across the
   entire population of cells compared to EB2-flox mice, with a
   progressive decrease from proximal to distal regions relative to the
   stimulation electrode point. In Fig. [69]2D (upper panels), the slope
   of field excitatory postsynaptic potentials (fEPSPs) showed a notable
   alteration in the EB2-vGATCre group of mice. PPR values were calculated
   at graded stimulus intensities of 10–70 μA. A reduction in paired-pulse
   ratio was observed in EB2-vGATCre mice (Fig. [70]2D, lower panels),
   indicating impaired short-term synaptic plasticity due to altered
   presynaptic vesicle release probability.

Fig. 2.

   [71]Fig. 2
   [72]Open in a new tab

   Impaired synaptic plasticity in mice with EB2 ablation in
   vGAT + neurons. A Schematic diagram showing the location of a MED64
   probe placed on the coronal PFC slice (left) and the arrangement of the
   8 × 8 recording array. Light microscopy photograph showing the location
   of the MED64 probe relative to the PFC cortex. The red dot indicates
   the stimulation site. B At various stimulation intensities (20, 40, 70
   μA), the response intensity around the stimulation point (black) is
   shown (each square is 100 μm per side). The bar on the right indicates
   the color changes in response intensity. C Sample traces of evoked
   potentials in both groups of mice. D Significant decrease in fEPSP at
   positions 100 μm above, below, and to the left of the stimulation point
   (the top row). Upper 100 μm: two-way ANOVA, interaction,
   F(8,72) = 1.416, P = 0.2045, row factor, F(8,72) = 8.936, P < 0.0001,
   column factor, F(1,72) = 49.26, P < 0.0001, Sidak’s multiple
   comparisons test, 50 μA, 60 μA, 70 μA, **P < 0.01; left 100 μm: two-way
   ANOVA, interaction, F(8,72) = 1.584, P = 0.1447, row factor,
   F(8,72) = 12.26, P < 0.0001, column factor, F(1,72) = 78.53,
   P < 0.0001, Sidak’s multiple comparisons test, 40 μA, 50 μA, 60 μA,
   **P < 0.01, 70 μA, ***P < 0.001; lower 100 μm: two-way ANOVA,
   interaction, F(8,72) = 2.891, P = 0.0075, row factor, F(8,72) = 8.489,
   P < 0.0001, column factor, F(1,72) = 59.16, P < 0.0001, Sidak’s
   multiple comparisons test, 40 μA, *P < 0.05, 50 μA, ****P < 0.0001, 70
   μA, ***P < 0.001. At positions 100 μm above, below, and to the left of
   the stimulation point, there is a significant decrease in the
   paired-pulse ratio of fEPSP (the bottom row). Upper 100 μm: two-way
   ANOVA, interaction, F(8,72) = 0.6037, P = 0.7689, row factor,
   F(8,72) = 0.7938, P = 0.6098, column factor, F(1,72) = 61.17,
   P < 0.0001, Sidak’s multiple comparisons test, 25 μA *P = 0.0388, 30 μA
   *P = 0.0243, 40 μA, *P = 0.0175, 60 μA, *P = 0.0421, 70 μA,
   *P = 0.0139; left 100 μm: two-way ANOVA, interaction, F(8,72) = 0.4780,
   P = 0.8678, row factor, F(8,72) = 0.3884, P = 0.9234, column factor,
   F(1,72) = 41.39, P < 0.0001, Sidak’s multiple comparisons test, 25 μA,
   *P = 0.0495, 40 μA, *P = 0.0211; lower 100 μm: two-way ANOVA,
   interaction, F(8,72) = 0.9763, P = 0.4618, row factor, F(8,72) = 1.159,
   P = 0.3359, column factor, F(1,72) = 14.1, P = 0.0004, Sidak’s multiple
   comparisons test

   To investigate the impact of EB2 deletion on PFC neural circuitry of
   the at a systemic level, we used functional ultrasound imaging (fUSi),
   which accurately identifies the PFC as a major connectivity hub. Most
   ROI pairs were shown in Fig. [73]3A, and the heatmap presents level
   variations across ROI pairs for both groups of mice (Fig. [74]3B).
   Connectivity between the infralimbic PFC and the RNT was significantly
   impaired in the resting state in EB2-vGATCre mice (Fig. [75]3C). These
   findings suggest that EB2 deletion in inhibitory neurons profoundly
   disrupts local PFC function and connectivity with related brain
   regions, such as the RNT, potentially contributing the cognitive
   changes observed in EB2-vGATCre mice.

Fig. 3.

   [76]Fig. 3
   [77]Open in a new tab

   Functional connectivity changes between EB2-vGATCre and EB2-flox group
   of mice. A Upper panel: head fixation and skull thinning. Lower panel:
   3D registration enabled linear probe positioning (gray bar) through the
   PFC. Right panel: anatomical delineations derived from the Allen Common
   Coordinate Framework overlaid on a representative fUS Doppler image
   show coverage of ROIs. B Heatmap of Pearson correlation coefficient
   between selected ROIs in both groups of mice. C Differences in
   functional connectivity z-values between EB2-vGATCre and EB2-flox group
   of mice for RNT-IA (Mann–Whitney U test, **P = 0.0020). N = 5 per group

Cellular and transcriptional alterations in the PFC of EB2-vGATCre and
EB2-flox mice via snRNA-seq

   Neural activity induces widespread transcriptional changes in neurons,
   resulting in structural and/or functional changes [[78]12]. To
   investigate cell-type-specific changes linked to cognitive
   abnormalities and impaired neural connectivity in the PFC of
   EB2-vGATCre mice, we performed snRNA-seq. A total of 11,308 single
   nuclei were sequenced. After quality filtering (Methods, Additional
   file [79]1: Figs. S3A, S4A–C), 11,090 high-quality nuclei were
   obtained, including 5336 cells from EB2-flox mouse and 5754 from
   EB2-vGATCre mouse. These nuclei were grouped into 17 cell clusters
   (Additional file [80]1: Fig. S4D–H), which were further assigned to 9
   major cell types based on cell-type-specific markers: excitatory
   neurons (Slc17a7^+), inhibitory neurons (Gad2^+), astrocytes (Gja1^+),
   microglia (C1qa^+), oligodendrocytes (Aspa^+), oligodendrocyte
   precursors (Pdgfra^+), smooth muscle cell (Uaca^+), endothelial
   (Flt1^+) and ependymal (Dynlrb2^+) cells (Fig. [81]4A, C, Additional
   file [82]1: Fig. S3B).

Fig. 4.

   [83]Fig. 4
   [84]Open in a new tab

   Widespread transcriptional changes in PFC cell types between EB2-flox
   and EB2-vGATCre mouse. A t-SNE plot showing the broad clustering of PFC
   cells (5336 from EB2-flox mouse PFC and 5754 from EB2-vGAT mouse PFC).
   B t-SNE plot showing the broad clustering of PFC cell (5336 from
   EB2-flox mouse PFC and 5754 from EB2-vGAT mouse PFC). t-SNE plot
   showing the distribution of the merged PFC cells from EB2-flox mouse
   (red) and EB2-vGATCre mouse (green). C t-SNE plot showing the
   expression of well-established marker genes in each broad cell cluster.
   D Bar plot showing the proportion of PFC cell types in EB2-flox and
   EB2-vGAT mice. E Strip chart shows the logFC of all detected genes
   (dots) across all 9 cell types. Genes in colored dots are significantly
   (Padj < 0.05 and |logFC|> 0.25) upregulated or downregulated. Genes in
   gray are not significantly changed. F Representative volcano plots
   showing altered gene expression that are significantly upregulated (red
   dots) or downregulated (blue dots) in the interneuron. G Overlap of
   differentially expressed genes between neurons and glia. H Expression
   changes of selected genes involved in synapse, learning and memory, and
   locomotory behavior. I Enriched pathways related to metabolism,
   cellular stress, synapse, and channel activity per cell type. The
   ancestor terms selected from the Gene Ontology and numbers of their
   associated child terms are listed on the left. On the right, the red
   and green colors represent numbers and percentages of upregulated and
   downregulated enriched pathways

   Comparison of cell composition in the PFC of EB2-vGATCre mouse with
   EB2-flox mouse revealed significant changes in neuronal populations
   (Fig. [85]4B, Additional file [86]2: Table S1). Consistent with the
   reported excitatory/inhibitory (E/I) ratio [[87]13], excitatory neurons
   constituted the largest cell class (54.31%) in the PFC of EB2-flox
   mouse, while inhibitory neurons accounted for 19.45%. Interestingly,
   EB2 deletion in vGAT^+ neurons disrupted the delicate E/I balance,
   resulting in comparable proportions of excitatory (37.81%) and
   inhibitory (34.93%) neurons, which might underlie cognitive deficits
   observed in these mice (Fig. [88]4D). The disruption of E/I homeostasis
   in the GABAergic/glutamatergic system is implicated in the
   pathophysiology of mental disorders [[89]14].

   Comparison of gene expression in the PFC of EB2-vGATCre mouse and
   EB2-flox mouse revealed substantial changes across major cell types.
   Using an adjusted P value cutoff of < 0.05 and |logFC|> 0.25, we
   identified significantly alterations in the expression of 540 genes in
   at least one cell type (Additional file [90]2: Table S2). Excitatory
   and inhibitory neurons showed the most pronounced transcriptional
   changes (292 and 221 DEGs, respectively) (Fig. [91]4E, F and Additional
   file [92]1: Fig. S5A–H). Comparing DEGs across major cell clusters
   revealed that only four genes (Hbb-bs, Cd81, Apoe, Gm28928) out of 540
   DEGs were shared by all neurons and glia, while 413 DEGs were
   cell-type-specific, indicating minimal overlap in expression profiles
   between cell types (Fig. [93]4G and Additional file [94]2: Table S3).

   Cell-type-specific transcriptional changes associated with distinct
   functional phenotypes were identified in this study. Among the 540
   DEGs, many genes involved in cognition, locomotion, and synapse
   development showed altered expression across various cell types
   (Fig. [95]4H). For example, Apoe, which was upregulated in all neurons
   and glia clusters, is a well-known gene related to cognition and
   locomotory behavior. Differentially expressed genes such as Syp,
   Atp1a3, Ndrg4, and Pbx3, involved in synapse development, cognition,
   and locomotion, were upregulated in neuronal populations. In addition,
   Neurexin1 (Nrxn1) and Lsamp identified in glial cells were associated
   with locomotor functions and synaptic activity. A subset of genes,
   including Reln, Apoe, and Gpr88 in inhibitory neurons was implicated in
   both cognition and anxiety, both of which were observed in EB2-vGATCre
   mouse.

   To explore the biological impact of transcriptional changes across cell
   types, we performed gene set enrichment analysis and identified
   significant effects on relevant pathways. Using an adjusted P value
   cutoff of < 0.05, each cluster exhibited significantly affected
   pathways related to synaptic development, neurotransmitter, metabolism,
   etc. (Additional file [96]2: Table S4). Minimal overlap between cell
   clusters, except for Olig and OPC (Additional file [97]1: Fig. S6).
   Based on the hierarchical GO tree structure and previous literature
   [[98]15], we selected a set of GO ancestor terms associated with
   cellular stress, metabolism, synapse, and cation homeostasis,
   reflecting the E/I balance, and a hallmark of behavioral deficit in
   EB2-vGATCre mice (i.e., cognitive anomaly). We then filtered pathways
   linked to these terms (Fig. [99]4I). Pathways related to cellular
   stress, ATP metabolic process, cellular response to DNA damage, and
   proteolysis (e.g., protein catabolic process and protein
   ubiquitination) were upregulated across all cell clusters, indicating
   that EB2 deletion increased metabolism and cellular stress. Conversely,
   pathways involved in synapse signaling (e.g., synaptic signaling,
   chemical synaptic transmission, neurotransmitter transport) and synapse
   development (e.g., axon and dendrite development, cell junction
   organization, and cytoskeleton organization) were significantly
   downregulated, especially in glia and stromal cells. Notably, G
   protein-coupled receptor (GPCR) signaling and cation hemostasis
   pathways were primarily upregulated in neurons but downregulated in
   glia and stromal cells. Additionally, pathways related to amyloid
   processes (e.g., amyloid-beta/amyloid precursor protein formation,
   metabolism, and clearance) were broadly upregulated. Amyloid deposition
   in the brain, a hallmark of Alzheimer’s disease, is strongly linked to
   cognitive deficits. Collectively, these pathways highlight the roles of
   metabolism, cellular stress, synaptic signaling, and development in EB2
   deletion-induced dysfunction.

Molecularly distinct neuronal subtypes in the PFC following EB2 deletion

   Many studies have reported distinct functional subpopulations within
   the same neuron types, classified by morphology, anatomical location,
   electrical properties, histological features, and transcriptomic
   profiles [[100]13]. Consistently, we found that inhibitory and
   excitatory neurons in the PFC comprise distinct cell subtypes with EB2
   deletion in vGAT + neurons causing significant changes in cellular
   composition and transcriptomic profiles.

   We observed a substantial increase in inhibitory neurons in the PFC of
   EB2-vGATCre mouse (Additional file [101]1: Fig. S7). Re-clustering
   these inhibitory neurons revealed 12 transcriptionally distinct
   subtypes (Fig. [102]5A), each identified by unique or combinational
   marker expression. Some subtypes corresponded to well-known inhibitory
   neuron populations, such as Pvalb^+ (Int_2), Sst^+ (Int_5), and Vip^+
   (Int_7) (Fig. [103]5B and Additional file [104]1: Fig. S8A and B),
   whose expression in the PFC was validated using in situ hybridization
   (Allen Brain Atlas; Additional file [105]1: Fig. S8C). Additionally, we
   identified novel inhibitory subpopulations, i.e., Int_8, 10 (Igfbpl1^+,
   with or without Atad2^+), Int_0, 1, 3 (Rgs9^+, with or without
   Mhrt^+/Sh3rf2^+), Int_6 (Cpa6^+), Int_4, 9 (Ndst4^+, with or without
   Gfra1^+), and Int_11 (Atad2^+), which did not belong to the
   aforementioned populations.

Fig. 5.

   [106]Fig. 5
   [107]Open in a new tab

   Inhibitory neuronal subclusters. A t-SNE plot showing that inhibitory
   neurons of PFC can be classified into 12 subtypes based on their
   transcriptome (left), and the distribution of the merged cells from
   EB2-flox (1038 cells, red) and EB2-vGAT (2010 cells, green) PFC cells
   (right). B t-SNE plots highlight marker genes for inhibitory neuronal
   subclusters. C Enrichment map network of GO pathways impacted in Int_6
   based on marker genes. Each node represents a GO term and each edge
   represents the overlap between two GO terms. D Heatmap of the mean
   value of AUCell scores of expression regulation by transcription
   factors, as estimated using SCENIC, per inhibitory neuronal subcluster.
   E t-SNE plots of inhibitory neurons, color-coded for expression of Pbx1
   and Meis1 (up), for AUC of the estimated regulon activity of Pbx1 and
   Meis1 (bottom), corresponding to the degree of expression regulation of
   their target genes. F Strip chart shows logFC of all detected genes
   (dots) across all 12 subclusters. Genes in colored dots are
   significantly (Padj < 0.05 and |logFC|> 0.25) upregulated or
   downregulated. Genes in gray are not significantly changed. G Heatmap
   showing the functional GO pathways impacted in many of subclusters
   based on gene expression changes between EB2-flox and EB2-vGAT mice

   Surprisingly, we discovered that Int_6 was a cell population detected
   exclusively in EB2-vGATCre (Fig. [108]5A, Additional file [109]1: Fig.
   S8D). RNAscope validation confirmed the emergence of int_6 following
   EB2 deletion (Fig. [110]6A and B). Functional enrichment analysis of
   its marker genes revealed that Int_6 is associated with functional
   categories including membrane signaling, cation hemostasis, synaptic
   structure and development, synaptic signaling/neurotransmission, and
   metabolism (Fig. [111]5C). These pathways collectively support
   functions critical for cognition (e.g., learning or memory) and
   locomotory behavior.

Fig. 6.

   [112]Fig. 6
   [113]Open in a new tab

   Validation of gene expression changes in PFC of EB2-vGATCre group of
   mice. A Multi-channel FISH detecting subtypes within a novel inhibitory
   neuronal subpopulation in PFC (CPA6 +, arrow, and CPA6 −, arrowhead
   within vGAT neurons: enlarged view of single cells in box area shown in
   side panel). B Violin plots with boxplots overlaid (right) showing the
   quantification of the RNAscope data (data presents differential
   expression of CPA6 in vGAT + neurons (n = 82 cells from 4 brains of
   EB2-flox mice, n = 442 cells from 4 brains of EB2-vGATCre mice,
   ****P < 0.0001, Mann–Whitney U test)). C and D Ablation of EB2 induced
   the upregulation of the Pbx1 and Meis1 genes in vGAT + neurons
   (Pbx1 + cells; indicated by arrows), scale bar, 50 μm and 10 μm. E
   Enlarged view of single cells in box area shown, scale bar, 10 μm. F
   Violin plots with boxplots overlaid (right) showing the quantification
   of the RNAscope data (data presents median expression of Pbx1 in
   vGAT + neurons (n = 493 cells from 4 brains of EB2-flox mice, n = 540
   cells from 4 brains of EB2-vGATCre mice, *P = 0.0102, Mann–Whitney U
   test) and G Meis1 + vGAT + neurons (n = 493 cells from 4 brains of
   EB2-flox mice, n = 564 cells from 4 brains of EB2-vGATCre mice;
   ****P < 0.0001, Mann–Whitney U test))

   To identify transcription factors underlying differences in expression
   between Int_6 and other subtypes, we applied single-cell regulatory
   network inference and clustering (SCENIC). This analysis identified
   co-expressed transcription factors and their putative target genes,
   highlighting Pbx1, Meis1, Sp4, Sall1, and Nfatc1, with Pbx1 and Meis1
   being prominent candidates (Fig. [114]5D and Additional file [115]1:
   Fig. S9). The expression of Pbx1 and Meis1 increased following EB2
   deletion (Fig. [116]6C–G), and their target genes were highly
   upregulated in Int_6 (Fig. [117]5E). Pbx1 and Meis1 frequently form
   heterodimeric complexes that regulate the transcription of targeting
   genes. These finding suggest that Pbx1 and Meis1 likely function
   cooperatively to orchestrate precise gene expression patterns in Int_6,
   contributing to the abnormal behavior observed in the EB2-vGATCre mice.

   To investigate the contributions of other inhibitory neuron subtypes to
   abnormal behaviors, we compared gene expression between EB2-vGATCre and
   EB2-flox Mice and performed an enrichment analysis on these DEGs. Using
   the aforementioned cutoffs, we identified differentially expressed
   genes, with Int_0, Int_1, Int_2, Int_5, and Int_7 exhibiting the most
   significant transcriptional changes (Fig. [118]5F and Additional file
   [119]2: Table S5). Enrichment analysis revealed that these
   transcriptional changes impacted various functional pathways, including
   synapse-related processes (structure, neurotransmitter, long-term
   synaptic potential, synaptic plasticity, etc.), Rac GTPase, ERBB
   signaling, and amyloid protein across different inhibitory subtypes
   (Fig. [120]5G). These findings shed light on potential pathways through
   which EB2 deletion may contribute to behavioral abnormalities.

   Excitatory neurons were significantly affected by EB2 deletion.
   Consistent with a previous study [[121]16], they accounted for the
   largest cell population in the PFC, but their proportion decreased
   following EB2 deletion. Sub-clustering revealed 14 distinct subtypes of
   excitatory neurons (Additional file [122]1: Fig. S10A and B). Layer
   identity and specific functions of these subtypes were assessed using
   t-SNE overlaid with unique layer marker gene expression (Additional
   file [123]1: Fig. S10C and D). Comparison of gene expression between
   EB2-vGATCre and EB2-flox mice identified 1 (Exc-8) to 138 (Exc-0) DEGs
   across subtypes, with Exc_0, Exc_2, Exc_4, Exc_5, Exc_6, and
   Exc_11—distributed in different layers, showing the most
   transcriptional changes (Additional file [124]1: Fig. S10E and
   Additional file [125]2: Table S6). These findings suggest that EB2
   deletion likely affects distinct cortical layers involved in neuronal
   projection, connections, and related functions, contributing to
   abnormal animal behaviors. Enrichment analysis of DEGs further
   indicated that pathways related to synapse function, nervous system
   development, amyloid protein, and so on might may play roles in the
   pathogenic mechanisms of EB2 deletion (Additional file [126]1: Fig.
   S10F).

EB2 knockout in vGAT + neurons affects genes linked to neuropsychiatric
diseases

   Many neuropsychiatric disorders are characterized by cognitive
   impairment, emotional symptoms, and locomotor abnormalities. The PFC
   plays an important role in cognition and emotion, and the
   pathophysiology of disorders like schizophrenia, Alzheimer’s disease,
   anxiety, and bipolar disorder. Genome-wide association studies (GWAS)
   have identified numerous candidate genes linked to these disorders,
   which show cell-type-specific expression in the PFC [[127]13]. Our
   study demonstrated that EB2 deletion resulted in novel inhibitory
   neuron subtypes, E/I imbalance, and cell-specific transcriptional
   changes in the PFC. Thus, we conclude that EB2 deletion may be closely
   associated with neuropsychiatric disorders.

   To gain preliminary insights into the potential role of EB2 in
   neuropsychiatric disorders, we compared the expression of GWAS
   candidate genes of 12 disorders/phenotypes between EB2-vGATCre mice and
   EB2-flox mice. These disorders/phenotypes include cognition,
   schizophrenia, bipolar disorder, depression, suicide,
   obsessive–compulsive disorder, autism, attention-deficit hyperactivity
   disorder, dementia, anxiety, and Parkinson’s disease. Many GWAS genes
   associated with these disorders showed significant expression changes
   after EB2 deletion, with inhibitory and excitatory neurons accounting
   for the largest portion of genes related to cognition and anxiety
   (Fig. [128]7A and B and Additional file [129]2: Table S7). Some
   candidate genes exhibited selective differential expression in distinct
   cell clusters, while others showed broad changes across neurons and
   glia (Fig. [130]7C: shows some examples). For example, the well-known
   cognition-related gene Apoe was upregulated to varying degrees among
   neurons and glia, while Nrg1 was specifically downregulated in
   excitatory neurons (Fig. [131]7C top). Similarly, the anxiety-related
   gene Elavl2 was downregulated only in inhibitory neurons, whereas Ptprd
   was downregulated in excitatory neurons and most glias (Fig. [132]7C
   down). Given the significant number of differentially expressed GWAS
   genes in inhibitory and excitatory neurons, we analyzed transcriptional
   changes within their subpopulations. Int_0 showed the highest number of
   differential candidate genes across most disorders/phenotypes in
   inhibitory neurons, while Exc_0, Exc_2, and Exc_6 identified the most
   genes in excitatory neurons. These transcriptional changes also
   exhibited cell-specific patterns (Additional file [133]2: Table S7,
   Additional file [134]1: Fig. S11). For instance, Erbb4, a gene critical
   for learning and memory, was upregulated in Int_8 and Exc_6, while Clu,
   an anxiety-related gene, was upregulated in Int_5, Int_7, and Exc_2.
   Finally, phenotype analysis of Int-6 marker genes and DEGs in neuron
   and glia revealed significant enrichment in pathways related to
   cognition and synapse development (Fig. [135]7D).

Fig. 7.

   [136]Fig. 7
   [137]Open in a new tab

   EB2 deletion links to phenotypes of mice and neuropsychiatric diseases.
   A Heatmap showing the number of differentially expressed GWAS candidate
   genes relevant to 12 disorders/phenotypes in the PFC cell types between
   EB2-flox and EB2-vGAT mice. B t-SNE plot indicating the number of
   differentially expressed cognition and anxiety GWAS candidates per
   cluster (shown in distinct colors). C Heatmap showing the
   representative differential expressed GWAS candidates relative to
   cognition (up) and anxiety (down) in PFC cell types. Cognition and
   anxiety are shown as examples. D The phenotypes are found to be
   associated to markers of Int_6 and DEGs of neurons and glia. The left
   panel shows the downregulated DEGs, and the right panel shows the
   upregulated DEGs. The Fisher’s exact test and Benjamini–Hochberg FDR
   correction were used for P values

   The cell-type-specific transcriptional changes in GWAS candidate genes
   suggested that EB2 deletion may lead to abnormal behaviors by altering
   the expression of risk genes, offering a cellular perspective on the
   pathological mechanisms.

Discussion

   In this study, we manipulated endogenous EB2 expression in transgenic
   mice and found that EB2 loss disrupts several behavioral domains,
   including recognition memory and increased anxiety, highlighting its
   indispensable role in the brain. Restoration of EB2 after early brain
   development ameliorated cognition-related impairments. EB2 deletion in
   vGAT + neurons also impaired PPR and disrupted PFC-thalamic functional
   connectivity. To investigate these alterations at the cellular level,
   we conducted a single-cell investigation in EB2 knockout mice and their
   EB2-flox control littermates. Our study revealed cell population
   changes in response to EB2 deficiency in vGAT + neurons and identified
   previously uncharacterized cell types. Differentially expressed gene
   analysis showed that EB2 deficiency primarily affects genes involved in
   synapse organization and development, cognition, amyloid-beta
   formation, and locomotor behavior, overlapping with human GWAS
   candidate genes linked to cognition and anxiety. This underscores the
   relevance of our mice model to human disease. Overall, our
   investigation provides a comprehensive map of cell types, genes, and
   pathways altered in this synaptic deficiency model.

   Previous studies have highlighted the role of Ephrin Bs in regulating
   neuronal migration [[138]17]. Here, we demonstrate for the first time
   that the relative numbers of neuronal subtypes in the PFC change when
   EB2, the most abundantly expressed Ephrin B molecule, is specifically
   knocked out in inhibitory neurons. During early brain development,
   synapses form to guide the migration of newborn neurons to the
   neocortex [[139]18]. Disruption of EB2-mediated signaling significantly
   impairs neuronal migration, affecting brain maturation and population
   dynamics. We confirmed an increase in inhibitory neurons in EB2-vGATCre
   mice, leading to an imbalance in E/I signaling (Additional file [140]1:
   Fig. S10). PPR analysis revealed abnormalities in the PFC, further
   supporting disrupted E/I balance. In PPR, a larger second fEPSP
   indicates paired-pulse facilitation [[141]19], a fundamental form of
   short-term potentiation (STP) that we also observed. STP underpins
   higher cognitive functions, including associative learning, working
   memory, and short-term memory [[142]20].

   In addition to abnormal electrophysiological signals in the local PFC
   region caused by EB2 knockout in vGAT + neurons, ultrasound functional
   imaging revealed disrupted functional connectivity between the PFC and
   the RNT, a region known to be functionally linked to the PFC. The
   midline/paramidline thalamus, which receives projections from the PFC,
   plays a crucial role in working memory, with PFC interneurons
   modulating its function [[143]21]. Thus, EB2 knockout in vGAT neurons
   may impair PFC and RNT connectivity, potentially contributing to the
   cognitive deficits observed in EB2 knockout mice.

   The synaptic balance between excitation and inhibition is a fundamental
   principle of cortical circuits, and disruptions in this balance are
   closely linked to cognitive impairments [[144]22]. Our results
   demonstrate that deleting EB2 in inhibitory neurons during early
   development, but not in adulthood, induces cognitive impairments.
   Evidence suggests that activating the EB2 signaling pathway can restore
   cognitive function [[145]23, [146]24]. Consistent with this, we found
   that targeted viral overexpression of EB2 in inhibitory neurons
   reversed cognitive impairments caused by its loss, while EB2 knockout
   in excitatory neurons does not cause deficits. Since early EB2
   deficiency in inhibitory neurons disrupts synaptic balance and leads to
   cognitive and behavioral deficits, we used single-nucleus sequencing to
   identify affected cell subpopulations and functional pathways
   associated with differentially expressed genes. These findings provide
   insights into the cellular mechanisms linking EB2 deficiency, E/I
   imbalance, and altered cognitive behaviors.

   Our single-nucleus sequencing revealed that a novel inhibitory neuron
   subpopulation (Int_6) exclusively present in the EB2-vGATCre mice,
   representing a significant finding warranting further investigation.
   The emergence of this transcriptionally distinct subtype suggests that
   EB2 deletion in vGAT + neurons profoundly impact the cellular
   composition and transcriptional landscape of the PFC. Int_6 neurons are
   marked by the expression of carboxypeptidase A6 (CPA6), a member of the
   M14 metallocarboxypeptidase family, which is highly expressed in the
   adult mouse olfactory bulb but broadly expressed in the brain and other
   tissues during embryonic development [[147]25]. In EB2-vGATCre mice,
   CPA6 is widely observed in Int_6 neurons within the PFC, suggesting
   potential migration issues for CPA6-expressing neurons. Cognitive
   processing relies on the refinement of limbic circuitry during the
   first 2 weeks of life, with the olfactory bulb influencing limbic
   activity and communication during development [[148]26, [149]27]. The
   reduction of CPA6 in the olfactory bulb (Additional file [150]1: Fig.
   S12) and the presence of CPA + neurons in the adult PFC may contribute
   to the cognitive impairments caused by EB2 deletion in vGAT + neurons.
   Functional enrichment analysis indicated that Int_6 is involved in
   pathways related to membrane signaling, cation homeostasis, synaptic
   structure/development, and neurotransmission, suggesting a role in
   synaptic plasticity, neuronal migration, and E/I balance-processes
   critical for cognition and behavior. Furthermore, transcription factors
   Pbx1 and Meis1 emerged as key regulators of the Int_6 population.
   Evidence linked Pbx1 and Meis1 to mood instability and psychotic
   disorder [[151]28–[152]30], and Meis1 knockout mice exhibited impaired
   cognitive abilities [[153]31]. These findings collectively suggested
   that the synaptic protein EB2 modulates transcriptional networks
   governing Int-6’s unique gene expression patterns that are associated
   with cognitive impairment.

   Apart from the newly identified Int_6 population, DEG analysis revealed
   widespread transcriptional adaptations across subpopulations, primarily
   involving cognition, locomotion, and synaptic function. In excitatory
   neuronal subpopulations, we identified from layer 5/6 excitatory
   neurons in the PFC with significantly differentially expressed genes
   between EB2-vGATCre and EB2-flox mice. Dysfunctional projections from
   the hippocampal CA1 area to the PFC, which are associated with poor
   cognitive performance in mental disorders, preferentially target layer
   5/6 pyramidal neurons and interneurons [[154]32–[155]34]. Thus, the
   DEGs in these excitatory subpopulations may underlie the cognitive
   impairments observed in EB2-vGATCre mice.

   While some pathways are cell-type-specific, common enriched pathways
   include synaptic signaling, neurotransmission, and cell junction
   assembly. These transcriptional alterations across diverse brain cell
   types, combined with disruptions in the E/I neuronal ratio, likely
   underlie the synaptic and circuit-level dysfunctions contributing to
   cognitive and behavioral deficits. Some DEGs were consistently changed
   across cell types, while others were neuron-specific. Notably,
   transcriptional adaptations in neurons affected common signaling
   pathways involved in neuronal development, neurotransmitter transport,
   and neuropsychological behaviors. These findings provide a foundation
   for investigating the molecular mechanisms underlying behavioral
   deficits, including cognitive impairments, and suggest possible changes
   in both neurons and supportive cells.

   Notably, several GWAS candidate genes linked to neuropsychiatric
   disorders, including cognition, schizophrenia, and anxiety, exhibited
   cell-type-specific expression changes upon EB2 deletion. This
   underscores the relevance of the EB2-vGATCre model to human disorders
   and suggests potential cellular mechanisms through which EB2 deletion
   contributes to abnormal behaviors by dysregulating risk gene expression
   across neuronal populations.

Conclusions

   Overall, our study unveiled novel cell types, detailed transcriptional
   remodeling, and links to neuropsychiatric risk genes, paving the way
   for targeted investigations into circuit-specific mechanisms underlying
   EB2-mediated behavioral phenotypes. Further exploration of distinct
   neuronal subtypes like Int_6, their connectivity patterns, and the
   interplay between excitatory and inhibitory populations may provide
   crucial insights into the etiology and pathophysiology of cognitive and
   neuropsychiatric disorders.

Methods

Animals and behavioral assessment

   Mice were maintained in a mixed CD1/B6 genetic background. All
   experiments involving mice were conducted in accordance with National
   Institute of Health Guide for the Care and Use of Laboratory Animals
   and approved by Animal Care and Use Committee in Shanghai Jiao Tong
   University. Mice used in this analysis were previously described
   [[156]10]. They carried loxP-flanked alleles in Ephrin-B2 (EB2^loxP),
   and the Ai9 Rosa26-STOP-tdTomato Cre reporter strain was used to
   monitor Cre activity. Specifically, EB2-flox mice were crossed with
   Vgat-Cre mice [B6.FVB-Tg(Slc32a1-cre)2.1Hzo/FrkJ, provided by Prof.
   Huang Zhili, Fudan University]. Genotyping with WT-EB2, Mutant-EB2, and
   vGAT-Cre primers yielded products of 500 bp, 350 bp, and 400 bp,
   respectively. Genotypes were revealed only after analysis.
   Additionally, mice with loxP-flanked EB2 alleles were crossed with
   tamoxifen-inducible CreER^T2 mice [homozygous CreER^T2
   (B6.129-Gt(ROSA)26Sor tm1(CreER^T2)tyj/J)], allowing EB2 deletion 5–7
   days post-tamoxifen administration (Sigma T5648-5G). Ephrin-B2^lacZ/+
   mice were also used to validate behavioral phenotypes. Mice were housed
   in standard conditions with a standard 12-h dark/light cycle (lights
   were on at AM 7:00 and off at PM 7:00), a room temperature of 24 ± 1
   °C, and ad libitum access to food and water.

Novel object recognition

   A plastic arena (44 cm × 44 cm × 44 cm) and objects with distinct
   shapes was used for examining novel object recognition. Mice were
   habituated to the arena for 5 min per day over 3 days without the test
   objects. Each mouse was initially placed into the box with two
   identical (familiar) objects for 10 min, and then returned to their
   home cage. One hour later, the mouse was reintroduced to the arena, now
   containing one familiar object and one novel object with a different
   shape, for 5 min. Object-related behaviors including the number of
   object contacts and the duration of contacts were recorded and analyzed
   by EthoVision XT 8.5. Novel object recognition task (NORT),
   [DI = (Novel Object Exploration Time/Total Exploration
   Time) − (Familiar Object Exploration Time/Total Exploration
   Time) × 100].

Elevated plus maze

   The elevated plus maze (EPM) apparatus was made of dark gray plastic
   and comprised two open arms (30*7*0.25 cm) opposed to two enclosed arms
   (30*7*15 cm) elevated 60 cm from the floor. Animals were placed in the
   central area facing an enclosed arm (test duration, 5 min). Digitized
   video recordings with EthoVision software (Noldus Information
   Technology) were used for behavioral analysis. The percentages of time
   spent in open arms and open-arm entries were evaluated.

Tailing suspension test

   The mouse is suspended 50 cm above the ground, ensuring it cannot touch
   any surface. The test lasts 6 min, during which movements are recorded,
   and immobility is measured. Immobility is defined as the absence of any
   limb or body movement except those caused by respiration. The first 2
   min serve as an acclimation period, with data collected during the last
   4 min. The percentage of immobility time is calculated.

Prepulse inhibition test

   The prepulse inhibition (PPI) test measures sensorimotor gating in mice
   by assessing their ability to inhibit a startle response when a weaker
   prepulse precedes a stronger startle-inducing stimulus. Mice are
   habituated to a sound-attenuated chamber equipped with a startle
   platform to detect movements. After establishing a baseline startle
   response to 120 dB noise bursts, prepulses (e.g., 75 dB, 80 dB, 85 dB)
   are presented 100 ms before the startle stimuli. Trials include startle
   stimuli alone, prepulse followed by startle, and control (background
   noise). Startle response amplitudes are recorded and PPI is calculated
   using the formula: PPI (%) = (1 − (startle response with
   prepulse/startle response alone)) × 100.

Compulsive checking

   The protocol for assessing compulsive checking behavior in mice
   typically involves the following steps: Mice are habituated to an open
   field arena with several objects for exploration. During testing,
   specific objects or locations are designated as “checkpoints.” The
   mouse is placed in the arena and allowed to explore freely for a set
   period, usually 30–60 min. The number of returns to and interactions
   with the checkpoints is recorded. High frequency and repetitive returns
   to these checkpoints are indicative of compulsive checking behavior.
   This behavior can be further analyzed to evaluate the effects of
   pharmacological interventions or genetic modifications on compulsive
   behavior in mice.

Immunohistochemistry

   Anesthetized mice were transcardially perfused with PBS followed by 4%
   paraformaldehyde (PFA). Brains were fixed overnight with 4% PFA and
   dehydrated in 30% sucrose at 4 °C. Brains were sectioned coronally in
   30 µm slices. Slices between bregma 1.98 to 2.58 mm were collected.
   Sections were washed with PBS, blocked with 10% BSA in 0.1%
   Triton-X-100 (CAS No. 9002–93-1) for 2 h at room temperature (RT), and
   incubated with primary antibodies of anti-rabbit GABA, anti-mouse
   GAD65/67, anti-goat Ephrin-B2, anti-rabbit parvalbumin (1:2000),
   anti-vGLUT2, and anti-goat CamkII at 4 °C overnight. Secondary
   antibodies were used as follows: Alexa 647 donkey anti-mouse IgG
   (1:1000, Jackson lab, Shanghai, China, RRID: AB_2340863) and Alexa 488
   donkey anti-rabbit (1:1000, Jackson lab, RRID: AB_2313584) for 2 h at
   room temperature. Cells were counted in every sixth section. Images
   were captured using Thunder image systems (Leica).

RNA scope

   In situ RNA analysis was conducted using the RNAscope® Multiplex
   Fluorescent Reagent Kit v2 Assay (Advanced Cell Diagnostics, Hayward,
   CA, USA, Cat. No. 323100), following the manufacturer’s guidelines.
   Briefly, paraffin-embedded mouse brains were sectioned into 5 µm slices
   and mounted on Superfrost® plus slides (Thermo Fisher Scientific,
   Waltham, MA, USA, Cat. No. 4951PLUS4). After pretreatment with hydrogen
   peroxide, target retrieval, and protease, the slices were hybridized
   with probes for Pbx1 (Cat. No. 550901-C2), Meis1 (Cat. No. 445231),
   vGAT (Cat. No. 400951), and CAP6 (Cat. No. 311491) using a HybEz
   hybridization system. Following amplification steps, the sections were
   stained with DAPI and mounted using Prolong Gold® Antifade Mountant
   (Thermo Fisher Scientific, Cat. No. [157]P10144). Imaging was performed
   with FV1200 confocal microscopes (Olympus, Tokyo, Japan).

Recording multi-channel field potentials in medial frontal cortical slices

   The method for recording multi-channel field potentials from medial
   frontal cortical slices follows established protocols [[158]35]. After
   incubation, a slice is placed on the MED64 probe ensuring the insular
   cortex is fully covered by the recording chamber on an inverted
   microscope (CKX41; Olympus). Placement follows an anatomical atlas for
   precise localization.

   To stabilize the slice, a fine mesh anchor (Warner Instruments, Harvard
   Apparatus) is used during recording. The slice is continuously perfused
   with oxygenated extracellular solution (95% O2/5% CO2) containing ACSF:
   125 mM NaCl, 2 mM CaCl2, 1.25 mM NaH2PO4, 3 mM KCl, 26 mM NaHCO3, 1 mM
   MgCl2, and 10 mM D‐glucose, pH 7.3, at a flow rate of 2 ml/min,
   maintained at 30 ± 1 °C using a peristaltic pump (Minipuls 3; Gilson).

   After a 20-min recovery period, a single microelectrode from the 64
   available on the probe is selected for stimulation, identified via a
   charge-coupled device camera (DP70; Olympus) linked to the inverted
   microscope. Monopolar, biphasic constant-current pulses (0.2 ms
   duration) are generated by MOBIUS software (Panasonic Alpha-Med
   Sciences) and applied at 0.008 Hz to the deep layer of the insular
   slice. Field excitatory postsynaptic potentials (fEPSPs) evoked at
   other sites are amplified using a 64-channel amplifier, displayed on a
   monitor, and stored for analysis. Stimulation intensities are
   sequentially adjusted from 15 to 70 µA in increments of 5 µA.

fUS headframe attachment surgery

   To conduct fUS recordings on awake mice, we developed a lightweight
   head fixation device compatible with fUS. The design incorporates a
   standard mounting headband with mouse support brackets. Headframe
   placement procedure: Mice were anesthetized with isoflurane gas
   (1–1.5%). During surgery, mice breathed a mixture of air and
   supplemental oxygen, with body temperature maintained at 36.5 °C using
   a feedback-controlled heating pad and monitored via a rectal probe. The
   scalp was carefully incised, using a specialized cranial drill and
   grinder. Care was taken to avoid any abrasion to the skull and the head
   fixation device was securely attached to the skull and surrounding
   muscles (Super-Bond C&B, LOCTTLF 454). Afterward, a protective layer
   was applied to safeguard the skull. Post-surgery, mice were given a
   3-day recovery period with careful monitoring and pain management as
   needed. Starting on day 3, they were habituated to head fixation, with
   session duration gradually increased from 5 min on the first day to 60
   min by day 6 [[159]24]. This acclimation step was critical to reduce
   stress and ensure consistent fUS measurements.

fUS imaging

   fUS hardware and software (Iconeus, Paris, France) were used for
   imaging sessions. A linear ultrasonic probe (15 MHz central frequency)
   with 128 piezo-electric was connected to an ultrafast ultrasound
   scanner (Iconeus One—128 channels). The animal was positioned in a
   stereotaxic frame, and we pre-trained the mice to adapt to a conscious
   state of restraint beforehand, in preparation for acquiring
   resting-state fUS imaging. After animal fixation, the scalp was covered
   with an isotonic coupling gel and the ultrasonic probe was lowered to 1
   mm from the scalp for complete immersion into the gel. The probe’s
   location was determined using Iconeus Studio software to target an
   oblique plane, which encompassed the neuroanatomical regions of
   interest (ROIs) associated with the PFC. Twenty minutes of Power
   Doppler images were captured for each session. These normalized values
   were used to generate bar graphs and conduct statistical analyses
   between the two groups.

Brain tissue dissociation and nucleus isolation

   Two mice from EB2-flox and EB2-vGATCre group respectively were
   anesthetized with isoflurane and rapidly decapitated. Brains were
   extracted and transferred into ice-cold Hibernate A/B27 medium. The PFC
   was removed from slices and isolated intact nucleus suspension. The
   nucleus was isolated and purified as previously described with some
   modifications [[160]36]. Briefly, the frozen tissue was homogenized in
   NLB buffer containing 25 mM sucrose, 10 mM Tris–HCl, 3 mM MgAc2, 0.1%
   Triton X-100 (Sigma-Aldrich), 0.1 mM EDTA, and 0.2 U/μL RNase Inhibitor
   (Takara). Various concentrations of sucrose were used to purify the
   nuclei. Nuclei counts were determined by PI staining. Finally, the
   concentration was adjusted to 1000 nuclei/μL for snRNA-Seq.

Single-nucleus RNA sequencing

   The snRNA-Seq libraries were generated using the 10X Genomics Chromium
   Controller Instrument and Chromium Single Cell 3′ V3 Reagent Kits (10X
   Genomics, Pleasanton, CA). Briefly, nuclei were concentrated to 1000
   nuclei/μL and approximately 10,000 nuclei were loaded into each channel
   to generate single-nuclei gel bead-in-emulsions (GEMs), which results
   into expected mRNA barcoding of ~ 6000 single nuclei for each sample.
   After the reverse transcription (RT) step, GEMs were broken and
   barcoded cDNA was purified and amplified. The amplified barcoded cDNA
   was fragmented, A-tailed, ligated with adaptors, and index PCR
   amplified. The final libraries were quantified using the Qubit High
   Sensitivity DNA assay (Thermo Fisher Scientific) and the size
   distribution of the libraries were determined using a High Sensitivity
   DNA chip on a Bioanalyzer 2200 (Agilent). All libraries were sequenced
   by Novaseq 6000 (Illumina, San Diego, CA) on a 150-bp paired-end run.

Single-nucleus gene expression quantification and determination of cell-type
identity

   Raw data processing were performed by NovelBio Bio-Pharm Technology
   Co., Ltd. with NovelBrain Cloud Analysis Platform. First, we applied
   fastp with default parameter to filter adaptor sequence and remove the
   low-quality reads to achieve the clean data [[161]37]. The
   feature-barcode matrices were then obtained by aligning reads to the
   mouse genome (GRCm38 ensembl: version 92) using CellRanger v3.1.0. We
   performed downsampling analysis among samples based on the mapped
   barcoded reads per nucleus of each sample and obtained the aggregated
   matrix. Nuclei with fewer than 200 or more than 10,000 expressed genes,
   or with more than 20% mitochondria UMI rate, were removed. Mitochondria
   genes were also removed in the expression table.

   For the remaining 11,090 nuclei, the Seurat R package (v.2.3.4)
   [[162]38] was used for cell normalization and regression based on UMI
   counts and mitochondrial rate, generating scaled data. Highly variable
   genes were identified by FindVariableGenes function with default
   parameters. The top 2000 highly variable genes were used to calculate
   principal components (PCs) and the top 10 PCs were selected for t-SNE
   construction. Utilizing graph-based cluster method, unsupervised cell
   cluster was generated based on the top 10 PCs. Marker genes were
   calculated with FindAllMarkers function using the Wilcoxon rank-sum
   test and the following criteria: 1. lnFC > 0.25; 2. P value < 0.05; 3.
   min.pct > 0.1. Clusters were annotated based on well-known marker genes
   of major brain cell types as reported in previous studies [[163]39]. To
   identify subclusters within neurons, inhibitory and excitatory neurons
   were reclustered separately using similar methods and parameters.

Differential gene expression analysis

   Differential gene expression analysis between EB2-flox and EB2-vGATCre
   mouse was performed using the FindMarker function in Seurat. The
   analysis generated logFC (natural log of the FCs), P values, pct
   (percentage of cells where the gene is detected in the case or control
   group), and adjusted P values (based on Bonferroni correction using all
   genes in the dataset). Genes with |logFC|> 0.25 and adjusted P
   value < 0.05 were considered significantly changed. DEGs were shown in
   strip charts and volcano plots, which were generated using ggplot2 R
   package [[164]40]. Overlapping DEGs across cell clusters was displayed
   in Venn diagrams created with jveen [[165]41].

Pathway analysis

   Pathway enrichment analysis was performed to elucidate the biological
   implications of differentially expressed genes. The clusterProfiler
   package [[166]42] in R was used to identify the Gene Ontology (GO)
   terms in the biological process, considering only pathways with an
   adjusted P value < 0.05 as significant. The GO.db package
   ([167]https://web.mit.edu/~r/current/arch/i386_linux26/lib/R/library/GO
   .db/html/00Index.html) was used to explore the ancestor-child
   relationship between representative pathways (ancestor pathways),
   associated with cognition and anxiety [[168]15], and significant
   pathways (child pathways).

SCENIC analysis

   To assess transcription factor regulation strength, we used the
   single-cell regulatory network inference and clustering (pySCENIC,
   v0.9.5) workflow with the 20-thousand motifs database for RcisTarget
   and GRNboost [[169]43].

Association of GWAS genes to cell clusters

   To investigate the relationship between EB2 and neuropsychiatric
   diseases, we downloaded GWAS candidate genes for 12 selected diseases
   from the NHGRI-EBI GWAS catalog (v. 1.0.2) [[170]44] and compared them
   with differentially expressed genes in each cluster. Phenotype analyses
   were performed on Int-6 marker genes and DEGs of neurons and glia using
   model organism Phenotype Enrichment Analysis (modPhEA,
   [171]https://evol.nhri.org.tw/phenome2/) [[172]45].

Statistical analysis

   Statistical analysis was performed using a two-sample Student’s t-test
   or one-way/two-way ANOVA, depending on the experimental design. For
   multiple group comparisons, Bonferroni’s post hoc test was used.
   Results are expressed as the mean ± standard error of the mean (SEM).
   Prior to analysis, data were log-transformed and tested for normality
   using the Shapiro–Wilk test. Nonparametric tests were applied for data
   that did not follow to a Gaussian distribution. Statistical
   significance was defined as P < 0.05 (two-sided). Analyses were
   conducted using GraphPad Prism 8.0.2., and the figures were prepared
   with CorelDraw 2019.

Supplementary Information

   [173]12915_2025_2333_MOESM1_ESM.zip^ (21.7MB, zip)

   Additional file 1: Figures S1–S12. Fig. S1 Immunostaining of EphrinB2
   and GAD65/67 in PFC slices from EB2-flox and EB2-vGATCre mice. Fig. S2
   Behavioral assessments showing mild anxiety- and depression-like
   phenotypes in mutant mice, with no schizophrenia- or OCD-like deficits.
   Fig. S3 Overview of the experimental workflow and transcriptome-based
   classification of PFC cells. Fig. S4 Quality control and initial data
   processing of single-cell RNA-seq from mouse PFC. Fig. S5
   Cell-type-specific transcriptional changes in PFC between EB2-flox and
   EB2-vGATCre mice. Fig. S6 Heatmap of enriched GO pathways across PFC
   cell clusters. Fig. S7 Increased vGAT mRNA expression in PFC of
   EB2-vGATCre mice visualized by RNAscope. Fig. S8 Classification of
   inhibitory neuronal subtypes in PFC based on transcriptomic profiles
   and in situ validation. Fig. S9 SCENIC-based TF motif activity (AUC
   scores) in PFC inhibitory neurons. Fig. S10 Identification and
   characterization of excitatory neuronal subclusters in PFC, including
   layer-specific markers and pathway analysis. Fig. S11 Number of
   differentially expressed GWAS candidate genes in each PFC neuronal
   subcluster. Fig. S12 RNAscope showing reduced CPA6 mRNA in olfactory
   bulb of EB2-vGATCre mice.
   [174]12915_2025_2333_MOESM2_ESM.zip^ (20.3MB, zip)

   Additional file 2: Tables S1–S7. Table S1 Number of cells detected by
   cell clusters. Table S2 Differential gene expression results for major
   cell types between EB2-vGATCre and EB2-flox. Table S3 Matrix of DEGs
   across neurons and glias. Table S4 GO pathways across major cell types.
   Table S5 Differential gene expression data for inhibitory neuronal
   subclusters between EB2-vGATCre and EB2-flox. Table S6 Differential
   gene expression data for excitatory neuronal subclusters between
   EB2-vGATCre and EB2-flox. Table S7 Differential expression of GWAS
   candidate genes.

Acknowledgements