Nr4a1 regulates cell-specific transcriptional programs in inhibitory GABAergic interneurons

Summary

The patterns of synaptic connectivity and physiological properties of diverse neuron types are shaped by distinct gene sets. Our study demonstrates that, in the mouse forebrain, the transcriptional profiles of inhibitory GABAergic interneurons are regulated by Nr4a1, an orphan nuclear receptor whose expression is transiently induced by sensory experiences and is required for normal learning. Nr4a1 exerts contrasting effects on the local axonal wiring of Parvalbumin- and Somatostatin-positive interneurons, which innervate different subcellular domains of their postsynaptic partners. The loss of Nr4a1 activity in these interneurons results in bidirectional, cell-type-specific transcriptional switches across multiple gene families, including those involved in surface adhesion and repulsion. Our findings reveal that combinatorial synaptic organizing codes are surprisingly flexible and highlight a mechanism by which inducible transcription factors can influence neural circuit structure and function.

eTOC

Huang et al. report that the immediate early gene Nr4a1 regulates the connectivity and physiological properties of inhibitory GABAergic interneurons in the mouse brain. Nr4a1 appears to exert these effects by controlling the cell-specific expression of multiple gene families, including those essential for synapse formation and maintenance.

Introduction

The assembly of neural circuits in the developing brain is orchestrated by molecular programs that define the cellular and subcellular patterns of synaptic connectivity through the combinatorial expression of multiple gene families ^1–3. Over the course of the postnatal lifespan, central neurons repetitively change their gene expression profiles in response to sensory cues. These events begin with the transient induction of transcription factors (TFs) encoded by immediate early genes (IEGs), whose mRNAs and proteins are rapidly synthesized in cellular ensembles activated by excitatory synaptic input and calcium influx ^4–6. It is well established that activity-regulated transcription is necessary for synaptic plasticity and memory storage ^4,5,7,8. However, the mechanisms through which IEGs modify diverse neuron classes and their synapses are unclear. It is particularly poorly understood how IEG signaling impacts inhibitory circuits.

In the mammalian forebrain, synaptic inhibition is mediated by >20 subtypes of GABAergic interneurons (INs) with distinct morphologies, synaptic target specificities, and physiological properties. Cortical and hippocampal INs are involved in virtually all cognitive tasks; their abnormalities have been associated with a wide range of neurological disorders in humans, including Schizophrenia, Alzheimer’s disease, Autism Spectrum Disorders and Epilepsy ^9–14. Different IN subtypes play unique physiological roles, reflecting their ability to innervate specific subcellular domains of postsynaptic partners ¹⁰. While this characteristic feature of INs appears to be genetically predetermined, their arborization, local wiring, excitability, and synaptic strength are modulated by spontaneous and/or correlated sensory-driven network activity ^10,15–23. These dynamic changes are thought to be critical for various aspects of brain development and function, ranging from the refinement of immature circuits during early postnatal life to high order sensory processing and learning ^{10,16,18,21,24}.

The possibility that IEGs are pivotal for tuning the synaptic inhibition has not been appreciated until recently, and current insights into the molecular bases and physiological implications of IEG signaling in GABAergic INs are still limited. Cortical INs express all classical IEG TFs, such as Fos, Fosb, Npas4, Egr1–4, Nr4a1–3 and Jun ^25,26, but the functions of these TFs and the nature of their effectors in different IN subtypes are largely unknown. The only exception is Npas4, whose involvement in regulating GABAergic synapse numbers via distinct mechanisms in glutamatergic neurons and INs has been recently demonstrated in several seminal reports ^26–28. Furthermore, the current understanding of how IEGs are engaged in INs in response to external sensory cues is largely based on observations in the visual system ^25,26. Therefore, it is unclear if these TFs act synergistically or in unrelated pathways, whether their effector genes are transcribed ubiquitously or in an IN subtype-specific manner, whether their cellular expression patterns vary depending on sensory stimuli or the internal state of the brain, and how they affect GABAergic inhibition in a given brain region and, ultimately, behavior.

Here, we set out to address these questions by studying Nr4a1 in mouse models using genomics, behavior, imaging and electrophysiology. Our work reveals a previously unappreciated relationship between IEGs and cell-specific transcription of genes essential for neuronal wiring and function. We predict that the surprising flexibility of synaptic organizing codes described herein reflects a general mechanism by which inducible TFs regulate the architectures of synaptic networks in the mammalian central nervous system.

Results

Comparative analysis of activity-dependent transcription in PV⁺ and Sst⁺ INs.

Among cortical and hippocampal INs, the two most abundant populations express Parvalbumin (PV) or Somatostatin (Sst). The majority of PV⁺ INs are fast spiking basket cells that innervate the somas of glutamatergic projection neurons (PNs), whereas Sst⁺ Martinotti cells provide GABAergic inputs onto distal parts of PN dendrites ¹⁰. Considering that PV⁺ and Sst⁺ cells are involved in memory coding ^18,29,30, we sought TFs that could be recruited in these INs during associative learning. This was achieved using RiboTag, a technique for immuno-isolation of translating mRNAs in complexes with ribosomes containing a genetically targeted HA epitope-tagged protein, Rpl22 ³¹. We introduced RiboTag into each IN population by crossing the Cre recombinase-dependent Rpl22-HA mouse allele with well characterized Pvalb^Cre and Sst^Cre drivers ^32,33, extracted cell-specific pools of mRNAs from the cortices and hippocampi of young adults at postnatal day (p) 60, and analyzed these mRNAs by deep sequencing (RNA-seq) (Figure 1A). Stringent quality controls confirmed that mRNA libraries from Pvalb^Cre/RiboTag and Sst^Cre/RiboTag mice had appropriate enrichment of markers of IN lineages and negligible contamination with markers of other cell classes (Figures S1A to S1D). We therefore surveyed acute changes in IN transcriptomes that were elicited by contextual fear conditioning (CFC), a neurobehavioral paradigm for the acquisition of stable associative memory from an adverse environment where animals received series of foot shocks ³⁴ (Figures 1B and 1C). In parallel experiments, Pvalb^Cre/RiboTag and Sst^Cre/RiboTag mice received single doses of the GABA receptor antagonist, pentylenetetrazol (PTZ) ³⁵, to identify genes whose mRNA levels were altered by artificial widespread network excitation as a frame of reference (Figures 1D and 1E). We selected all known and putative TFs from sets of significantly up- and down-regulated genes with a false discovery rate (FDR) of <0.05, ranked them by fold-change (FC) of mRNA levels, and compared expression across datasets. These analyses led us to investigate Nr4a1, an orphan nuclear hormone receptor encoded by a stress-sensitive IEG ^36–38 whose role in GABAergic INs had remained entirely unknown. Nr4a1 was strongly induced in PV⁺ (FC=7.3; FDR=2.2⁻¹¹) and Sst⁺ (FC=5.4; FDR=4.2⁻²⁶) cells of fear-conditioned mice (30 minutes post-CFC), and was the only TF activated in both groups of INs under two experimental settings. By contrast, CFC failed to trigger a similar increase in mRNA levels of several other classical IEGs, including Npas4 whose transcription is regulated in cortical INs by visual experience ^25,26 (Figures 1F, 1G, S1E and S1F).

Figure 1. Comparative analysis of activity-dependent transcription in PV⁺ and Sst⁺ INs.

(A) Schematics of interneuron (IN) wiring with principal neurons (PNs) and experimental workflow for deep sequencing of translating mRNAs isolated with genetically targeted RiboTag.

(B and C) Mice were examined after contextual fear conditioning (CFC).

(B) Retrieval of short-term memory, measured as percentage of freezing under the baseline settings (BL) or in the same context (RET). n = 7 to 8 mice per group. Graph represents mean values (circles), standard errors (boxes), standard deviations (whiskers), and medians (horizontal lines). *** p <0.01 (defined by t-test).

(C) Volcano plot of differentially expressed mRNAs in cortical PV⁺ and Sst⁺ cells 30 minutes following CFC (FC = fold change). n = 3 p60 mice/group.

(D and E) Mice were examined after widespread pharmacological induction of neuronal activity with PTZ (1 dose, i.p. injection, 50 μg/gm).

(D) Confocal images of Fos immunofluorescence in cortical columns.

(E) Volcano plot of differentially expressed mRNAs 30 minutes following PTZ delivery. n = 3 to 4 p60 mice/group.

(F) Venn diagrams of significantly up- and down-regulated genes for indicated experimental settings.

(G) Heat maps of mRNA levels of TFs. Pairwise comparisons are ranked by fold change in each IN population (arrows). Asterisks mark TFs whose expression was significantly altered in both populations.

See also Figure S1 and Data S1.

Spatiotemporal dynamics of Nr4a1 signaling in PV⁺ and Sst⁺ INs.

To establish the physiological role of Nr4a1 in inhibitory circuits, we focused on the hippocampal area CA1. In this brain region, PV⁺ and Sst⁺ INs participate in the acquisition and retrieval of episodic memories, and their presynaptic terminals are clearly segregated in separate layers containing PN somas and dendritic fields ^30,39–41.

As the first step, we elucidated the spatiotemporal dynamics of Nr4a1 signaling using quantitative real time PCR (qPCR), immunofluorescent microscopy, and genetically encoded reporters. These experiments were performed with untrained and fear conditioned animals in which INs carried RiboTag or, alternatively, were permanently labeled with tdTomato introduced with the Cre-dependent Ai9 allele ⁴² (Figure 2A). In agreement with our initial RNA-seq screens, qPCR analyses of transcripts isolated with RiboTag from cortical and hippocampal PV⁺ and Sst⁺ cells at p60 revealed that both IN populations had a pronounced increase in Nr4a1 mRNA levels 30 minutes after CFC. This effect was fully reversed to baseline by 3 hours, indicating that sharp upregulation of Nr4a1 mRNA synthesis only occurs within a narrow window following novel experience (Figure 2B). We further evaluated the cellular expression patterns of Nr4a1 by confocal imaging of the CA1 in brain sections stained with the antibody whose specificity was validated in Nr4a1 knockouts (Figure S3D). In animals that were maintained in home cages under the standard housing conditions, Nr4a1 immunoreactivity could only be detected in a few cells scattered across the CA1sp and adjustment subfields. Consistent with mRNA profiling, both the total numbers of cells and the numbers of Nr4a1-positive INs identified with the tdTomato reporter were sharply and transiently increased 30 minutes after CFC declining to baseline within 3 hours. The fractions of Nr4a1-positive PV⁺ and Sst⁺ cells were ~6 and ~3-fold larger, respectively, in fear conditioned mice compared to control mice (~20% of the total pool in each case) (Figures 2C to 2F).

Figure 2. Temporal dynamics of Nr4a1 signaling in inhibitory circuits.

(A to F) Both Nr4a1 mRNA and protein are induced in INs during associative learning in a rapid and reversible manner.

(A) Experimental workflow. Early and late phases of gene expression (ER and LR) were surveyed at indicated time points after CFC in p60 mice whose INs carried RiboTag or Ai9 tdTomato reporter (TdT).

(B) Quantitative real time PCR measurements of Nr4a1 mRNA levels. IN-specific transcripts were isolated as shown in Figure 1A. Left: input material, n = 4 p60 mice/group; middle: PV⁺ cells, n = 3/group; right: Sst⁺ cells, n = 3/group.

(C and D) Overall cellular expression patterns of Nr4a1 protein in area CA1 of the hippocampus. Panels show immunofluorescent images of coronal sections co-stained with the antibody against Nr4a1 and the nuclear marker, DAPI (C) and quantifications (D, n = 3 mice/group).

(E and F) CFC transiently increases the fractions of Nr4a1-positive INs, as identified with tdTomato in Pvalb^Cre/Ai9 and Sst^Cre/Ai9 mice. Panels show high magnification immunofluorescent images of the CA1 (E, INs expressing Nr4a1 above background are marked by arrows) and quantifications (F, n = 3 mice/group).

(G) Schematics of NrRAM (top) and Cre-dependent viral expression of NrRAM-DIO:GFP reporter in Pvalb^Cre/Ai9 and Sst^Cre/Ai9 mice (bottom).

(H and I) Nr4a1 activates a sustained transcriptional response. AAV NrRAM-DIO:GFP was delivered into CA1 at p60. The fractions of PV⁺ and Sst⁺ cells expressing GFP were measured in untrained and trained animals (24 hours post-CFC). Typical fluorescent images (H, GFP-positive INs are marked by arrows) and quantifications are shown (I, n = 3 mice/group).

In B, D, F and I, graphs represent mean values (circles), standard errors (boxes), standard deviations (whiskers), and medians (horizontal lines). * p <0.05; *** p <0.01 (t-test and/or ANOVA).

See also Figure S2.

To test if native Nr4a1 is capable of driving transcription in INs, we designed a fluorescent reporter, NrRAM:GFP, by adapting a recently described strategy for monitoring TF activities of Fos and Npas4 ⁴³. NrRAM was comprised of 4 short Nr4a1 enhancer repeats with a downstream minimal promoter and reliably sensed pharmacologically augmented synaptic excitation or the presence of exogenous Nr4a1 in primary neuronal cultures (Figures S2A and S2B). We introduced NrRAM:GFP into PV⁺ and Sst⁺ cells of adult Pvalb^Cre/Ai9 and Sst^Cre/Ai9 mice in vivo with a Cre-dependent adeno-associated virus (AAV) and then imaged labeled INs 7 days later in brain slices from untrained and fear conditioned animals (24 hours post-training). This protocol was chosen to account for a delay in GFP mRNA and protein synthesis following a brief induction of NrRAM by Nr4a1 (Figures 2G and S2C). The fractions of GFP⁺ INs were significantly increased in the CA1 after CFC, albeit the effect was not as dramatic as acute CFC-dependent induction of Nr4a1 itself (1.6- and 1.4-fold) with 19% to 28% of cells exhibiting detectable GFP fluorescence in untrained animals (Figures 2H, 2I and S2D). The higher “background” expression of NrRAM:GFP as compared to native Nr4a1 is not surprising since the two proteins have drastically different lifespans. Unlike Nr4a1 and other IEGs, the GFP driven by NrRAM does not degrade shortly after synthesis and may therefore remain detectable in neurons with relatively recent history of activity elicited by unrelated stimuli. Taken together, these results suggest that Nr4a1 is induced in INs by sensory stimuli in a transient manner, but its effects on expression levels of downstream genes can be sustained.

Genetic ablation of Nr4a1 in PV⁺ and Sst⁺ cells interferes with normal memory coding.

Next, we generated two conditional knockout (cKO) mouse lines that lacked Nr4a1 in PV⁺ or Sst⁺ cells. This was achieved by crossing the Pvalb^Cre and Sst^Cre drivers with an allele containing loxP sites flanking exons 2 and 4 of the Nr4a1 gene on chromosome 15 ⁴⁴ (Pvalb^Cre/Nr4a1 ^loxP/loxP and Sst^Cre/Nr4a1 ^loxP/loxP; Figure 3A and S3A). We confirmed that the Nr4a1 ^loxP/loxP allele works as expected by genotyping and sequencing genomic DNA extracted from tail tissues, PCR/qPCR analysis of DNA recombination and Nr4a1 mRNA levels in unstimulated and stimulated cortical neurons from Nr4a1 ^loxP/loxP mice that were broadly infected in vitro with lentiviruses (LVs) encoding Cre or inactive ΔCre, and immunofluorescent microscopy. Since INs are sparce and expression of Nr4a1 across cell types is strongly regulated by sensory experience (Figures 2B to 2F), the latter was done using brain sections from fear conditioned mice with a broader cKO of Nr4a1 in excitatory telencephalic neurons of the Emx1 lineage to validate the antibody and to unambiguously confirm the ablation of Nr4a1 protein in the presence of Cre in vivo (Figures S3B to S3D). Loss of Nr4a1 immunoreactivity in the CA1 of Emx1 cKOs indicated that Nr4a1 is induced in both INs and subsets of glutamatergic PNs (Figure S3D). As demonstrated below, we also confirmed selective Cre-dependent loss of mRNA fragments encoded by Nr4a1 exons 2, 3 and 4 in PV⁺ and Sst⁺ cells of new mouse models by RNA-seq (Figure S6A).

Figure 3. Behavior of PV+ and Sst⁺ Nr4a1 conditional knockout (cKO) mice.

(A) Annotations of genotypes.

(B and C) Analysis of associative learning and memory.

(B) Experimental workflow. Short- and long-term forms of fear memory (STM and LTM) were monitored 2 and 24 hours, after CFC, respectively.

(C) Quantifications of STM and LTM, expressed as percentages of freezing in the same context. PV⁺ (WT), n = 16 mice; PV⁺ (cKO), n = 16; Sst⁺ (WT), n = 17; Sst⁺ (cKO), n = 18.

(D and E) Analysis of spatial learning and memory.

(D) Schematics of the Barnes maze and experimental workflow.

(E) Quantifications of time spent in the target (T) and other (O) quadrants of the Probe test of memory retrieval. PV⁺ (WT), n = 17; PV⁺ (cKO), n = 17; Sst⁺ (WT), n = 22; Sst⁺ (cKO), n = 23.

(F and G) Behavior in open fields.

(F) Typical tracks of individual animals during 30 minute recording sessions.

(G) Quantifications of time spent in central zones and total travel distances for trials of indicated durations. 10 minutes: PV⁺ (WT), n = 27; PV⁺ (cKO), n = 29; Sst⁺ (WT), n = 19; Sst⁺ (cKO), n = 18. 30 minutes: PV⁺ (WT), n = 35; PV⁺ (cKO), n = 34; Sst⁺ (WT), n = 19; Sst⁺ (cKO), n = 18.

All graphs represent mean values (circles), standard errors (boxes), standard deviations (whiskers), and medians (horizontal lines). * p < 0.05; ** p < 0.02; *** p <0.01 (defined by t-test and ANOVA).

See also Figure S3.

The homozygous offspring of both IN-specific cKO lines were born at the expected Mendelian ratio, had regular lifespans, and did not display apparent developmental defects or behavioral abnormalities in the standard laboratory environment. However, quantifications of contextual freezing 2 and 24 hours after CFC demonstrated that Nr4a1-deficient mutants had aberrant short- and long-term associative memory (STM and LTM). Compared to their p60 Nr4a1 wildtype (WT) littermates, Pvalb^Cre/Nr4a1^loxP/loxP cKOs exhibited prolonged freezing in the same context, though this change was only statistically significant for STM. In contrast, the freezing of Sst^Cre/Nr4a1^loxP/loxP mice was significantly reduced at both timepoints, suggesting that STM and LTM were partially impaired (Figures 3B and 3C). These phenotypes were not attributed to abnormal locomotion, elevated anxiety or widespread disruption of forebrain circuit function, as evidenced by measurements of spatial memory performance in the Barnes maze as well as travel distances and zone preference of cKOs in open fields (Figures 3D to 3G). Furthermore, confocal imaging of brain sections from adult cKOs carrying the Ai9 tdTomato reporter allele showed no detectable changes in the densities, positioning and gross morphologies of labeled PV⁺ and Sst⁺ cells throughout the hippocampal pathway indicating that their migration, differentiation and survival were preserved (Figures S3E to S3H).

Nr4a1-deficeint INs have abnormal physiological properties.

To explore whether Nr4a1 signaling in INs is required for appropriate GABAergic inhibition of principal projection neurons, we injected the area CA1 of Pvalb^Cre/Nr4a1^loxP/loxP and Sst^Cre/Nr4a1^loxP/loxP cKOs with a Cre-dependent AAV encoding Channelrhodopsin2 (ChR2). Subsequently, we sampled evoked and spontaneous inhibitory postsynaptic currents (eIPSCs and sIPSCs) from PNs in acute hippocampal slices at p21 (Figure 4A). Whole-cell recordings of synchronous eIPSCs elicited by optical stimulation of INs demonstrated that inhibitory synaptic input to CA1 PNs from Nr4a1-deficient PV⁺ cells was significantly weaker, whereas input from Nr4a1-deficient Sst⁺ cells was stronger than in corresponding Nr4a1 WT littermates (Figures 4B and 4C). Likewise, the PNs of Pvalb^Cre/Nr4a1^loxP/loxP and Sst^Cre/Nr4a1^loxP/loxP cKOs had reduced and increased frequencies of spontaneous IPSCs, respectively, while the sizes and kinetics of isolated sIPSCs were unchanged (Figures 4D to 4G). As the functional maturation of PV⁺ cells continues beyond p21, we extended these analyses to adult mice (Figure S4A). PNs of p60 Pvalb^Cre/Nr4a1^loxP/loxP cKOs had lower rates of spontaneous inhibition, suggesting that the physiological consequences of disrupted Nr4a1 expression are not restricted to early postnatal development (Figure S4B). We also observed a slight but significant decrease of sIPSC amplitudes in adults (Figure S4B), appearing to be presynaptic in nature, as the sIPSCs of mature neurons represent a mixture of truly quantal events mediated by GABA release from single neurotransmitter vesicles and occasional multivesicular events elicited by action potentials. Indeed, recordings of miniature (m) currents in the presence of the voltage-gated sodium channel blocker, Tetrodotoxin, showed that only the mIPSCs frequencies, not amplitudes, were lowered in p60 Pvalb^Cre/Nr4a1^loxP/loxP cKO mice (Figure S4C). Notably, the synaptic abnormalities observed in both mouse models were not generic, as whole-cell recordings from INs identified in juvenile cKOs and their littermate controls by Cre-dependent expression of tdTomato showed no apparent differences in the frequencies and amplitudes of spontaneous excitatory synaptic currents mediated by release of glutamate from PNs (Figures S4D to S4H). However, quantifications of action potential firing upon membrane depolarization in the current clamp mode revealed that both Nr4a1-deficient PV⁺ and Sst⁺ cells had higher intrinsic excitability (Figures 4H to 4J).

Figure 4. Physiological properties of Nr4a1-deficient INs.

(A to G) Evoked (e) and spontaneous (s) inhibitory postsynaptic currents (IPSCs) were monitored from CA1 PNs in acute hippocampal slices from juvenile Nr4a1 WT and IN-specific cKO mice. eIPSCs were elicited by optogenetic stimulation of INs.

(A) Schematics of Cre-dependent expression of Channelrhodopsin2 (ChR2)-YFP from an AAV, annotations of genotypes, and whole-cell voltage-clamp recording configurations.

(B and C) Sample traces of eIPSCs triggered by 1 ms pulses of blue light (arrows, 450 nm) and quantifications of eIPSC amplitudes. Panel B: PV⁺ (WT), n = 5 mice/24 neurons; PV⁺ (cKO), n = 5/32. Panel C: Sst⁺ (WT), n = 3/20; Sst⁺ (cKO), n = 3/21.

(D and E) Sample traces of sIPSCs and quantifications of event frequencies. Panel D: PV⁺ (WT), n = 6 mice/25 neurons; PV⁺ (cKO), n = 6/21. Panel E: Sst⁺ (WT), n = 3/22; Sst⁺ (cKO), n = 3/25.

(F and G) Examples of quantal sIPSCs and cumulative probability histograms of sIPSC amplitudes. Panel F: PV⁺ (WT), n = 6 mice/25 neurons; PV⁺ (cKO), n = 6/21. Panel G: Sst⁺ (WT), n = 3/22; Sst⁺ (cKO), n = 3/25.

(H) Schematics of current-clamp recordings from INs of Nr4a1 WT and cKO mice carrying the Ai9 tdTomato (tdT) reporter.

(I and J) Sample traces of action potentials (APs) elicited by current injection (2x threshold) and quantifications of AP numbers. Panel I: PV⁺ (WT), n = 4 mice/14 neurons; PV⁺ (cKO), n = 4/13. Panel J: Sst⁺ (WT), n = 3/26; Sst⁺ (cKO), n = 3/26.

In B, C, D, E, I and J, graphs represent mean values (circles), standard errors (boxes), standard deviations (whiskers), and medians (horizontal lines). * p < 0.05; *** p <0.01 (defined by t-test). cKO values were normalized to WT for each littermate pair.

See also Figure S4.

Nr4a1 regulates the local connectivity of IN axons.

Our electrophysiological studies revealed that cKO of Nr4a1 in two major IN populations leads to similar shifts in their intrinsic membrane properties. Intriguingly, it changes the strengths of synchronous and spontaneous GABAergic inhibition of PNs in opposite directions. Since the frequencies of spontaneous and miniature synaptic currents generally correlate with synapse numbers, a plausible interpretation of these phenotypes is that, at a structural level, disruption of Nr4a1 signaling in INs alters the morphologies of their axons, target selection, and/or local wiring with postsynaptic partners. To test these predictions, we performed dual Cre-dependent labeling of CA1 INs in vivo with tdTomato and a virally expressed fluorescent fusion protein localized to presynaptic neurotransmitter vesicles, Synaptophysin (SyP)-Venus (Figure 5A). This strategy allowed us to examine the distribution of IN processes and their presynaptic terminals across the CA1 stratum oriens (so) to pyramidal cell layer (sp) to stratum radiatum (sr) to stratum lacunosum moleculare (slm) axes by confocal imaging of brain sections. At p60, axons of PV⁺ and Sst⁺ cells in cKO mice projected to the correct zones and their SyP-positive boutons were spatially segregated similarly to normal counterparts (Figure 5B). Nonetheless, the numbers of these boutons were altered in opposite directions, mirroring the consequences of Nr4a1 loss on amplitudes of optogenetically triggered eIPSCs and rates of sIPSCs/mIPSCs sampled from PNs. Nr4a1-deficient PV⁺ cells formed ~2 times less terminals on PN somas in the CA1sp, whereas the innervation of PN dendritic fields by Nr4a1-deficient Sst⁺ cells was more robust (Figures 5C to 5F). Interestingly, the upregulation of GABAegic synapse numbers was only pronounced in the CA1sr but not in the CA1slm of Sst^Cre/Nr4a1^loxP/loxP mice. These domain-specific changes were confirmed by imaging of brain sections stained with antibodies against the native presynaptic marker of PV⁺ basket cells, Syt2 ⁴⁵, and pan-inhibitory presynaptic marker, GAD67 (Figures 5C to 5F and S5A to S5E). Importantly, the structural abnormalities of inhibitory presynaptic networks in mutant mice were not associated with global delay in IN development or overgrowth. Nr4a1 cKO did not alter the frequencies of spontaneous glutamatergic excitation of INs, as shown above, or the morphologies of dendritic trees of PV⁺ cells visualized by viral Cre-dependent expression of membrane-bound GFP (Figures S3G, S3H, and S4D to S4H). We were unable to quantify dendritic arborization of Sst⁺ cells because their dendrites intermingled with axons.

Figure 5. Nr4a1 exerts opposite effects on local wiring of PV⁺ and Sst⁺ cells with PNs.

Inhibitory synaptic networks in the CA1 of p60 Nr4a1 WT and IN-specific cKO mice were analyzed by confocal imaging.

(A) Schematics of dual Cre-dependent labeling of PV⁺ and Sst⁺ cells with genetically encoded tdTomato and presynaptic marker, Synaptophysin (SyP)-Venus in mice of indicated genotypes.

(B) Nr4a1-deficient INs innervate appropriate receptive fields. Representative low magnification images of labeled projections and nerve terminals are shown. CA1so = stratum oriens; CA1sp = pyramidal cell layer; CA1sr = stratum radiatum. Note preserved segregation of terminals along the CA1so-CA1sp-CA1sr axes in both cKO lines.

(C and D) Ablation of Nr4a1 in PV⁺ cells diminishes their innervation of PN somas. Panels show images of terminals labeled in the CA1sp of Nr4a1 WT and cKO mice with SyP-Venus or antibody against Syt2 (C) and quantifications of terminal numbers (D).

(E and F) Ablation of Nr4a1 in Sst⁺ cells promotes their innervation of PN dendrites. Panels show images of terminals labeled in the CA1sr of Nr4a1 WT and cKO mice with SyP-Venus or antibody against GAD67 (E) and quantifications of terminal numbers (F).

In the top row of panel E, PN dendrites were stained with MAP2. In the bottom row, IN projections were labeled with the Ai9 tdTomato reporter.

Quantifications are from 3 to 4 mice/genotype. Graphs represent mean values (circles), standard errors (boxes), standard deviations (whiskers), and medians (horizontal lines). * p < 0.05; *** p <0.01 (defined by t-test). cKO values were normalized to WT for each littermate pair.

See also Figure S5.

To extend these observations, we selectively overexpressed the full-length Nr4a1 cDNA in PV⁺ and Sst⁺ cells on the WT background from a Cre-dependent AAV. We found that cell-autonomous Nr4a1 gain-of-function in INs altered the innervation of PN somas in CA1sp and dendrites in CA1sr in directions that were opposite to phenotypes cKO mice (Figures S5F to S5J). Collectively, these experiments with loss- and gain-of-function mouse models support the hypothesis that Nr4a1 is involved in transcriptional control of inhibitory circuit structure and function.

Loss of Nr4a1 in INs leads to cell-specific transcriptional switches.

Neuronal connectivity is organized by secreted and transmembrane proteins that guide axons to receptive fields and mediate surface adhesion or repulsion ¹. These proteins are often expressed in a cell-type-specific manner, thereby generating combinatorial codes that dictate the wiring selectivity of diverse neuron classes and/or their synapse numbers in local connectomes ^2,3,46. To determine how Nr4a1 regulates the presynaptic networks of INs at a molecular level, we performed in vivo screens for downstream effectors of this TF in PV⁺ and Sst⁺ cells. For this purpose, we generated cKO mouse lines carrying the RiboTag allele (Pvalb^Cre/Nr4a1^loxP/loxP/RiboTag and Sst^Cre/Nr4a1^loxP/loxP/RiboTag) and surveyed transcriptional profiles of Nr4a1-deficient INs at p60 by RNA-seq. Considering that pronounced changes in the expression of Nr4a1 effector genes may occur with a substantial and variable delay after each sensory stimulus we performed these experiments without fear conditioning. We reasoned that this strategy is advantageous as even subtle effects of repetitive induction of Nr4a1, or lack thereof, by natural cues will likely accumulate in INs of behaving mice over time. We were also concerned that stress response to foot shocks may engage compensatory mechanisms, thereby complicating the interpretation of RNA-seq data. Hence, we only used CFC to ensure the expected loss of Nr4a1 mRNA fragments encoded by loxP-flanked exons 2–4 in both cKO lines (Figure S6A).

Analysis of RNA-seq reads with the cutoff criteria of absolute FC of expression level >2 yielded 171 and 269 putative Nr4a1 effectors in PV⁺ and Sst⁺ cells, respectively. Approximately halves of these genes were upregulated in each mutant mouse line, indicating that Nr4a1 may act on different substrates as a transcriptional activator and repressor (Figure 6A). Remarkably, ablation of Nr4a1 produced partial switches in lineage-specific programs by inducing genes that are repressed and, vice versa, by repressing genes that are transcribed in a particular IN lineage in the normal brain (Figure 6B). For example, of 83 genes whose mRNA levels were >2-fold lower in Nr4a1-deficient PV⁺ cells, nearly 80% had enriched mRNAs in their WT counterparts, as compared to WT Sst⁺ cells. Consistent with these findings, more than 96% of all Nr4a1 effectors were lineage-specific (Figure 6C). Bioinformatic searches showed that many of these genes contained Nr4a1 consensus DNA binding sites ⁴⁷ and encoded proteins with axonal and synaptic functions (Figures 6D, 6E, and S6B).

Figure 6. Transcriptional profiling of Nr4a1-deficient INs.

(A to C) Genome-wide analysis of transcriptional programs in PV⁺ and Sst⁺ INs from Nr4a1 WT and cKO mice. Translating mRNAs were isolated at p60 with genetically targeted RiboTag and analyzed by RNA seq, as shown in Figure 1A (n = 3 mice/group).

(A) Volcano plots of RNA-seq data from mice of indicated genotypes (FC = fold change).

(B) Heat maps show comparisons of mRNA levels of genes that were significantly up- or down-regulated in each population of Nr4a1-deficient INs (top rows, FDR<0.05, −1<log2 FC>1), with their expression in other Nr4a1-deficient population (middle rows) and relative expression in Nr4a1 WT mice (bottom rows). Note transcriptional switches in cKO mice (compare top and bottom rows in each panel).

(C) Venn diagrams for data shown in panel (B).

(D and E) Pathway analysis of genes that are differentially expressed in Nr4a1-deficient PV⁺ (D) and Sst⁺ cells (E).

See also Figure S6 and Data S1.

Nr4a1 regulates the combinatorial expression of gene families involved in surface adhesion.

To address the possibility that Nr4a1 controls synaptic organizing codes, we compared the mRNA levels of genes from families essential for axonal and/or synaptic differentiation in p60 WT and cKO mice. Extending recent studies in the developing cerebral cortex ², we found that PV⁺ and Sst⁺ cells in the adult brain expressed different combinations of Neurexins (Nrxn), Neuroligins (Nlgn), Cerebelins (Cbln), Leucine-rich glioma-inactivated (Lgi), classical Cadherins (Cdh), Protocadherins (Pcdh), Semaphorins (Sema), Ephrin receptors (Epha), Complement 1q (C1q), Contactins (Cntn), Contactin associated proteins (Cntnap), Fibroblast growth factors (Fgf), Immunoglobulin superfamily proteins (Igsf), Neurexophilins (Nxph) and Latrophilins (Adgrl). These unique patterns were disrupted in the absence of Nr4a1 due to pronounced switches in lineage-specific transcription of Sema, Epha, Pcdh and C1q while other genes were either unaffected or affected to a lesser extent. Yet, the Nr4a1-deficient INs had preserved combinatorial expression of Cbln and Lgi whose members influence the subcellular specificity of synaptic targeting ² (Figures 7A, S6C, and S6D). This is consistent with our findings that neither Nr4a1 cKOs nor constitutive viral overexpression of the exogenous Nr4a1 interfere with projection of GABAergic axons to appropriate receptive fields (Figures 5B and S5).

Figure 7. Nr4a1 regulates molecular programs essential for IN wiring.

(A) Nr4a1-deficient INs have cell-specific switches in transcription of genes involved in synaptic differentiation. Examples of affected (left panel) and unaffected (right panel) genes from families encoding surface adhesion/repulsion molecules are shown. Annotated heatmaps demonstrate relative expression levels in normal INs (first columns, plotted as Sst/PV ratios) and expression in each population of Nr4a1-deficient INs (second and third columns).

(B to F) Gains-of-function of developmentally repressed Epha3 and Sema3c mimic the effects of Nr4a1 cKO in PV⁺ cells.

(B) Schematics of Cre-dependent viral expression of Epha3 or Sema3c cDNAs in Pvalb^Cre/Ai9 mice.

(C) Confocal images of tdTomato-positive INs and their Syt2-immunoreactive somatic terminals in the CA1.

(D) Quantifications of terminal numbers in CA1sp. n = 3 mice/experimental condition.

(E and F) Reduced GABAergic inhibition of PNs by PV⁺ cells overexpressing Epha3 or Sema3c cDNAs. INs were selectively activated with ChR2-YFP, as shown in Figure 4A.

(E) Sample traces of light-induced eIPSCs recorded from PNs.

(F) Quantifications of eIPSC amplitudes. Control, n = 3 mice/9 neurons; Epha3, n = 3/13; Sema3c, n = 3/9.

In D and F, graphs represent mean values (circles), standard errors (boxes), standard deviations (whiskers), and medians (horizontal lines). * p < 0.05; *** p <0.01 (defined by t-test). Values were normalized to control for each littermate pair.

See also Figures S6 and S7.

Nr4a1-dependent changes in somatic innervation of PNs by PV⁺ INs involve Sema3c and Epha3.

Transcriptional switches detected in Nr4a1 mutant mouse lines by RNA-seq are reconcilable with effects of Nr4a1 loss- and gain-of-function on local IN wiring because the members of Sema, Epha, Pcdh and C1q gene families can mediate attraction and repulsion in various contexts ^1,48–50. To start exploring mechanistic links between Nr4a1 and its effectors, we focused on Sema3c and Epha3. Both of these genes had enriched mRNAs in WT Sst⁺ cells, both contained conserved Nr4a1 consensus DNA binding motifs in promoter regions upstream transcription start sites, and both were strongly and selectively de-repressed in PV⁺ cells lacking Nr4a1 (Figures 7A, S6E, and S6F). We therefore predicted that Nr4a1 promotes the somatic innervation of PNs by PV⁺ basket cells by restricting the transcription of Sema3c and Epha3, which act as local repulsive cues. To test this model experimentally, we introduced recombinant Sema3c and Epha3 cDNAs into CA1 PV⁺ cells of neonatal Pvalb^Cre/Ai9 reporter mice in vivo with Cre-dependent AAVs in the Nr4a1 WT background (Figure 7B). Confocal imaging of tdTomato and Syt2-immunoreactive presynaptic terminals in young adults showed no changes in the survival of INs overexpressing individual constructs, their distribution, and axonal targeting to CA1sp. However, the numbers of inhibitory synapses on PN somas were reduced in both cases, as revealed by quantitative measurements of Syt2⁺ puncta density in CA1sp and whole-cell recordings of synchronous IPSCs elicited in PNs by optogenetic stimulation of PV⁺ cells (Figures 7C to 7F). These changes were not unspecific as targeted overexpression of Sema3c and Epha3 cDNAs in Sst⁺ cells did not alter GABAergic terminal densities in CA1sr, in contrast to overexpression of Nr4a1 (compare Figures S5I, S5J and S7A to S7C). We then developed a CRISPR-based strategy for Cre-inducible knockdowns (KD) of Sema3c and Epha3 to investigate the relationship between Nr4a1 and its two downstream effectors in PV⁺ cells using a complementary loss-of-function approach. To accomplish this task, we designed AAV vectors driving tandem guide RNAs under U6 promoters and validated the potency and specificity of KDs through qPCR measurements of mRNA levels in cultured cortical neurons (Figures S7D and S7E). These vectors were then introduced into the CA1 of neonatal Pvalb^Cre/Nr4a1^loxP/loxP mice together with an AAV expressing a Cre-dependent form of Cas9 (Figure S7F). Confocal imaging of Syt2 immunofluorescence in young adults showed that KDs of Sema3c and Epha3 upregulated the somatic innervation of PNs by Nr4a1-deficient INs, though only the former was statistically significant (Figures S7G and S7H). Taken together, these results indicate that Sema3c and Epha3 gains-of-function mimic the phenotypes of Nr4a1 cKO in PV⁺ cells, while their CRISPR-mediated KDs partially rescue these phenotypes.

Discussion

In summary, we investigated transcriptional programs that regulate synaptic inhibition in the mouse forebrain. Our results demonstrate that Nr4a1, a nuclear receptor encoded by the immediate early gene, is induced in GABAergic INs during associative learning and is required for normal memory storage. At the cellular level, Nr4a1 can promote or suppress the local axonal connectivity of INs which provide GABAergic inputs onto distinct subcellular compartments of their postsynaptic partners. At the molecular level, Nr4a1 instructs axonal wiring, at least in part, through cell-specific effectors mediating surface adhesion and repulsion.

A large body of work shows that the assembly of synaptic contacts in the developing nervous system is coordinated by numerous proteins localized in axons and dendrites. Knockouts of individual players in whole animals often result in only minor phenotypes ¹. For example, the bulk of synaptogenesis is preserved in mice lacking all isoforms of Neurexins ⁵¹, even though asymmetric Neurexin-Neuroligin interactions are sufficient for the formation of artificial synapses in simple heterologous systems ⁵². It is therefore plausible that the repertoires of molecular players essential for the maintenance and structural plasticity of established circuits are equally complex.

In agreement with recent studies of wiring specificity of excitatory and inhibitory neurons in the cerebral cortex and hippocampus ^2,3, we find that INs in these forebrain areas of adult mice employ non-overlapping sets of synaptic organizers. Nr4a1 signaling resembles a partial reprogramming of this code, achieved via on- and off-switches in several gene families. Hence, the unique combinatorial expression patterns of genes that define a neuron’s synaptic architecture are not determined solely during development and maintained across the lifespan in an intrinsic manner. Rather, these patterns can be re-arranged by inducible TFs whose transcription is driven by sensory cues. Since IEGs are induced in activated neuronal ensembles throughout the brain ^3,4,53, similar transcriptional switches may mediate structural and functional changes in other circuits and cell types.

Our study has limitations and raises several questions. First, all genetic manipulations used herein were prolonged. We chose this strategy to ensure no important phenotypes are overlooked, as Nr4a1’s role in inhibitory circuits had not been previously investigated. While Nr4a1 is rapidly and reversibly induced by sensory stimuli (Figure 2), the gene sets whose expression levels are altered in Nr4a1-deficient INs are not known to be similarly influenced by neuronal activity. Therefore, we refrain from asserting that Nr4a1 is solely involved in the acute induction of experience-dependent plasticity. Nevertheless, it is noteworthy that Nr4a1 mRNA becomes detectable in the forebrain only during the second postnatal week (see https://developingmouse.brain-map.org). The temporal dynamics of Nr4a1-dependent transcriptional switches and resulting changes in IN wiring, intrinsic excitability and synaptic strengths should be further elucidated. However, it should be noted that even contemporary methods for drug-inducible gene knockouts or expression of exogenous cDNAs offer suboptimal temporal control. Second, the associative memory defects of Nr4a1 cKO mice (Figure 3) may arise from the loss of TF function in multiple brain areas, owing to DNA recombination patterns in Pvalb^Cre and Sst^Cre drivers. For example, our observations are reminiscent of recent studies demonstrating that fear conditioning induces differential plasticity in dendrites and somas of principal neurons in the amygdala through compartment-specific inhibition ⁵⁴. Additionally, our observations do not exclude the possibility that cognitive abnormalities of cKO mice are attributed to a history of aberrant transcription in INs or other cell types that express PV and Sst outside of the forebrain over weeks. Third, our results establish a framework for better understanding how Nr4a1 signaling influences the complex inhibitory circuits at cellular and molecular levels. PV⁺ and Sst⁺ INs have multiple subtypes with distinct molecular signatures, synaptic partners and subcellular specificities of target innervation ^10,55. This diversity might explain the effects of Nr4a1 loss, or lack thereof, on synaptic networks of Sst⁺ cells in the CA1sr and CA1slm (Figures 5 and S5). It is intriguing to predict that transcription of some Nr4a1 effector genes identified in our RNA-seq screens is switched only in particular IN subtypes. Furthermore, many of these genes lack Nr4a1 binding motifs, suggesting that their induction or repression is mediated by Nr4a1 indirectly, via intermediate chromatin-binding proteins.

Although our study eventually focused on understanding how Nr4a1 signaling impacts inhibitory synapses, it is noteworthy that Nr4a1 does not solely control “synaptic” genes. Some putative Nr4a1 effectors appear to act in unrelated pathways, including metabolic pathways whose de-regulation may contribute to observed shifts in IN excitability (Figures 4 and 6). Viewed in a wider scope, our unbiased molecular profiling of Nr4a1-deficient INs indicates that IEG signaling networks and their downstream effectors are likely multifactorial even within the same brain regions, neuronal types, and developmental stages. As both dysfunctions of inhibitory circuits and abnormalities in experience-dependent transcription are associated with a broad spectrum of neurological disorders in humans ^4,5,13, these results may have translational implications.

It is also important to emphasize that Nr4a1 expression is not restricted to INs. Recent reports have shown that perturbations of Nr4a1 activity in forebrain glutamatergic neurons influence the morphogenesis of dendritic spines, long-term potentiation of excitatory synapses, and behavior. However, these conclusions are based on in vitro experiments with dissociated cultures and/or broad RNA interference without rescue controls or overexpression of dominant-negative constructs whose specificity is unclear ^{36–38,56,57}. Thus, the role of Nr4a1 in regulating the morphologies, wiring and function of diverse cellular building blocks of the mammalian central nervous system, the underlying molecular mechanisms, the generalizability of these mechanisms, and their functional relevance need to be investigated in the future. Lastly, our results suggest that Nr4a1 and its enhancers could be leveraged for development of new genetic tools for temporally restricted labeling of activated neural ensembles to complement the available tools that are based on Fos and Npas4 ^34,43,58. Such efforts will benefit from systematic comparative analyses of cellular patterns of induction of Nr4a1 and other classical IEGs in response to various types of sensory cues across the brain.

STAR★METHODS

RESOURCE AVAILABILITY

Lead Contact

All requests should be directed to the Lead Contact, Dr. Anton Maximov (amaximov@scripps.edu).

Materials Availability

All expression vectors reported in this study are available upon request.

Data and Code Availability

The deep sequencing datasets are enclosed in Supplemental materials and are available at Mendeley Data. All other raw datasets are available upon request. This study did not generate new code.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

Pvalb^Cre, Sst^IRES-Cre (which we call Sst^Cre for simplicity), Emx1^IRES-Cre, Nr4a1^fl/fl, RiboTag and Ai9 alleles used in this study have been described previously ^{31–33,42,44,59}. These alleles were crossed to established conditional lines and resulting offspring was analyzed according to protocols approved by the IACUC committee. All animals used herein were a mix of C57BL/6 and 129/SV. Males and females were analyzed together, unless indicated.

METHOD DETAILS

Expression vectors

Adeno-associated viruses (AAVs) were produced with well-characterized shuttle vectors containing the EF1α, Ubc, Syn and U6 promoters and the DIO cassette for inducibility with Cre ^20,60,61. For targeted overexpression of Epha3 and Sema3C cDNAs, the 1.26 kb EF1α promoter was replaced with the 212 bp EF1α-core promoter to maintain the overall sizes of inserts within the packaging limit (~4.4kb). This smaller promoter was amplified by PCR from the pJEP308 plasmid (Addgene #113685). The coding sequences of full length Epha3, Sema3c and Nr4a1 were amplified by PCR from the custom-made mouse hippocampal cDNA library with the following primers:

Epha3

Forward: 5′- GCTAGCATGGATTGTCACCTCTCCATCCT -3′

Reverse: 5′- TTAATTAATTACACTGGAACTGGACCATTCTTAGATTG -3′

Sema3c

Forward: 5′- GCTAGCATGGCATTCCGGACAATTTGC -3′

Reverse: 5′- TTAATTAATTATGACTCTGGCAACTGATTCCTCCT -3′

Nr4a1

Forward: 5′- CGGATCCATGCCCTGTATTCAAGC -3′

Reverse: 5′- GGAATTCGAAAGACAATGTGTCCAT -3′

The Epha3 and Sema3c coding sequences were inserted downstream of EF1α-core promoter in the 3′–5′ orientation using PacI and NheI restriction sites and were flanked by two pairs of loxP sites (DIO). The Nr4a1 coding sequence was inserted in the same manner into AAV-EF1α-DIO-GFP vector using BamHI and EcoRI sites.

The NrRAM reporter was constructed according to previously described strategy for reporters of Fos and Npas4 ⁴³. This reporter contained 4 tandem Nr4a1 enhancer repeats of the following sequence with underlined nucleotides representing the binding motif:

5′- CTAGAAGTTTGTTAAAAGGTCAGA -3′

The Fos minimal promoter was amplified by PCR from the pAAV-CRAM-tdTomato plasmid (Addgene #84468). The enhancer modules were inserted upstream of this promoter and the entire cassette was then used to replace the EF1α promoter in the EF1α:DIO-GFP AAV shuttle vector.

The sgRNAs for Epha3 and Sema3c were designed in CRISPick. The following oligonucleotide sequences were used to generate AAV shuttle vectors for tandem viral expression of sgRNAs under the control of two U6 promoters (one human and one mouse):

Epha3α

Forward: 5′- ACCGAAGCAGTTCATGAGTTCGCGA -3′

Reverse: 5′- AAACTCGCGAACTCATGAACTGCTTC -3′

Epha3β

Forward: 5′- TTTGGAATGCGATCCCCGAGATCCAT -3′

Reverse: 5′- AACATGGATCTCGGGGATCGCATTC -3′

Epha3χ

Forward: 5′- ACCGATGGAACCAACAGCCGCAAAT -3′

Reverse: 5′- AAACATTTGCGGCTGTTGGTTCCATC -3′

Epha3δ

Forward: 5′- TTTGGCTTTGAGATCGATGCTGTTAA -3′

Reverse: 5′- AACTTAACAGCATCGATCTCAAAGC -3′

Sema3cα

Forward: 5′- ACCGGATTCACACTTAGAGTCGATC -3′

Reverse: 5′- AAACGATCGACTCTAAGTGTGAATCC -3′

Sema3cβ

Forward: 5′- TTTGGGCATGGCTGTGGAAATTTCGT -3′

Reverse: 5′- AACACGAAATTTCCACAGCCATGCC -3′

Sema3cχ

Forward: 5′- ACCGGTCCCACCAGCAGTTAGACTA -3′

Reverse: 5′- AAACTAGTCTAACTGCTGGTGGGACC -3′

Sema3cδ

Forward: 5′- TTTGGATGATTGCAAGAATATGCCCT -3′

Reverse: 5′- AACAGGGCATATTCTTGCAATCATC -3′

The AAV and LV shuttle vectors for expression of mGFP, SyP-Venus, ChR2-EYFP, Cre and ΔCre have been described previously ^20,62,63.

AAV production and injection

AAVs were generated in house in HEK293T cells, purified by Heparin-based affinity chromatography, tittered by real-time quantitative PCR, and injected into brains of mice carrying the Cre drivers at titers of 2×10¹² GC/ml as we have previously described ^20,60,61. For synaptic tracing, reconstruction of neuronal morphologies, optogenetic stimulation, overexpression of cDNAs and CRISPR knockdowns, AAVs were delivered to neonates. Pups were anesthetized on ice, bilaterally injected into the cerebral lateral ventricles with 1 μl of viral stocks using a glass micropipette (1 mm O.D., 0.5 mm I.D., 10 μm tip diameter), warmed up for 5 minutes on a heat pad, and returned to home cages until experiments. To monitor the transcriptional activity of Nr4a1 with NrRAM:GFP, the reporter was delivered to young adults and imaged 7 days later. Mice were anesthetized with 1.5%–2% isoflurane in O₂; stereotaxic injections were performed bilaterally into the dorsal CA1 with the following coordinates (relative to bregma) and volumes: AP −1.82 mm, ML ± 1.4 mm, DV −1.2 mm. Viruses were infused for 5 minutes at a rate of 100 nl per minute. To prevent artifacts associated with variable AAV transduction efficiencies, all groups of mice were injected for each experiment side-by-side with the same viral stock; the AAV titers were the same throughout the study. We also confirmed the efficacies of synaptic labeling through simultaneous imaging of reporters encoded by AAVs and native inhibitory presynaptic markers, such as Syt2 or GAD67.

LV production and infection

Recombinant lentiviruses (LVs) were produced by co-transfection of HEK293T cells with corresponding shuttle vectors, and pVSVg, pGag-Pol and pRev plasmids that encode the elements essential for packaging of viral particles. Transfections were performed using FuGENE HD reagent (Roche). Conditioned medium with secreted viruses was collected 48 hours later and centrifuged at 5,000g for 5 minutes to remove cellular debris. Neurons were infected with 200 μl of viral supernatant per each well of a 24-well plate at 4 days in vitro (DIV). This protocol enabled reliable infection of ~95% neurons in culture.

Neuronal cultures

Cortices of p0 pups were dissociated with papain and seeded onto 24-well plates coated with poly-D-Lysine (Millipore). Cultures were maintained for 1 hour in MEM (Invitrogen) supplemented with fetal bovine serum (Invitrogen), and glucose (Sigma), followed by incubation in the serum-free Neurobasal-A (Invitrogen) supplemented with B27 (Invitrogen) and Glutamax (Invitrogen). Cultures were kept at 37°C in humidified incubators with 5% CO2 until use.

Validation of Nr4a1^loxP/loxP allele, NrRAM reporter, and CRISPR knockdown efficiency in vitro

Neurons from Nr4a1^loxP/loxP neonates were infected at DIV4 with LVs expressing Cre or catalytically inactive ΔCre under the control of the Ubiquitin (Ubc) promoter and assayed 3 days later. Genotyping was performed by PCR with the following primers:

Forward: 5′- TGACACCCTCACACGGACAA-3′

Reverse: 5′- CCAGTACATAGAGGATGCTTGTT-3′

Cre-dependent ablation of Nr4a1 was also confirmed by qPCR of reverse transcribed mRNAs with the following primers:

Fos

Forward: 5′- GGGACAGCCTTTCCTACTACC-3′

Reverse: 5′- AGATCTGCGCAAAAGTCCTG-3′

Nr4a1

Forward: 5′- TTGAGTTCGGCAAGCCTACC-3′

Reverse: 5′- GTGTACCCGTCCATGAAGGTG-3′

Gapdh

Forward: 5′- TCAACGGGAAGCCCATCA-3′

Reverse: 5′- CTCGTGGTTCACACCCATCA-3′

To validate the NrRAM reporter, neurons from C57BL/6 neonates were infected at 4DIV with LV Ubc-Cre, AAVDJ NrRAM:DIO-GFP and AAVDJ EF1α-DIO:Nr4a1. At 14DIV, cultures were stimulated overnight with Bicuculline (50 μM, Sigma) and 4-aminopyridine (250 μM, Tocris). Cultures were then rinsed briefly with HBSS (Invitrogen) and reporter expression was assessed by fluorescent imaging.

A similar strategy was used to drive Cre-dependent expression of Cas9 from AAVDJ CMV-DIO:saCas9 vector. To assess the efficiencies of CRISPR-mediated knockdowns of Epha3 and Sema3c, cultures with activated Cas9 were then infected at 7DIV with AAVDJs encoding 2xU6-Syn:BFP (control vector), 2xU6:gEpha3α/β-Syn:BFP, 2xU6:gEpha3χ/δ-Syn-BFP, 2xU6:gSema3cα/β-Syn:BFP or 2xU6:gSema3cχ/δ-Syn:BFP. Epha3 and Sema3c mRNA levels were tested 7 days later by qPCR with the following primers:

Epha3

Forward: 5′- TGGAGTTACGGGATTGTTCTC-3′

Reverse: 5′- CCCTCATCCACAGCCTTAAT-3′

Sema3c

Forward: 5′- GGCCAGCATCAACAATCAAAG-3′

Reverse: 5′- TCCACTCCCACAGACATACA-3′

Drug injection

Pentylenetetrazol (PTZ, Sigma) was dissolved in saline prior to each experiment and administered to mice by intraperitoneal injections through a 29 g needle at a dose of 50 mg/kg body weight. Control vehicle solution contained saline alone.

Isolation of mRNA with RiboTag

Cortices of p60 Pvalb^Cre/RiboTag and Sst^Cre/RiboTag mice were dissected in ice-cold PBS and homogenized in the buffer containing 100 mM KCl, 50 mM Tris-HCl pH 7.4, 12 mM MgCl2, 100 μg/ml cycloheximide (Sigma), 1 mg/ml heparin (Sigma), 1x complete mini, EDTA-free protease inhibitor cocktail (Roche), 200 units/ml RNasin© plus inhibitor (Promega) and 1 mM DTT (Sigma) (500 μl ~3% w/v). The lysate was centrifuged at 2,000g for 10 minutes at 4°C. Igepal-CA380 was then added to the supernatant at the final concentration of 1%. The lysate was briefly incubated on ice and then centrifuged at 12,000g for 10 minutes at 4°C. 25 μl of the supernatant was collected as input sample. 30 μl/ml of anti-HA coupled magnetic beads (Pierce) were then added to the remaining supernatant. Incubation was performed on a rotator for 3 hours at 4°C. The beads were washed four times in the high salt buffer containing 350 mM KCl, 1% Igepal-CA380, 50 mM Tris-HCl, pH7.4, 12 mM MgCl2, 100 μg/ml Cycloheximide (Sigma) and 1 mM DTT (Sigma). Ribosomes were eluted in 350 μl of RLT plus buffer (Qiagen). RNA purification was performed using RNeasy Plus Mini kit (Qiagen) following the manufacturers’ instructions. For qPCR, 20 ng of total RNA was reverse transcribed using High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). qPCR quality controls were performed with the following primers:

Fos

Forward: 5′- GGGACAGCCTTTCCTACTACC-3′

Reverse: 5′- AGATCTGCGCAAAAGTCCTG-3′

Nr4a1

Forward: 5′- TTGAGTTCGGCAAGCCTACC-3′

Reverse: 5′- GTGTACCCGTCCATGAAGGTG-3′

Gapdh

Forward: 5′- TCAACGGGAAGCCCATCA-3′

Reverse: 5′- CTCGTGGTTCACACCCATCA-3′

Syt2

Forward: 5′- AGAACCTGGGCAAATTGCAGT-3′

Reverse: 5′- CCTAACTCCTGGTATGGCACC-3′

Pvalb

Forward: 5′- CATTGAGGAGGATGAGCTG-3′

Reverse: 5′- AGTGGAGAATTCTTCAACCC-3′

Lhx6

Forward: 5′- CTACTTCAGCCGATTTGGAACC-3′

Reverse: 5′- GCAAAGCACTTTCTCCTCAACG-3′

Deep sequencing

RNA-seq was performed at the TSRI Next Generation Sequencing Core on the Illumina HiSeq platform. The libraries were generated, barcoded, and sequenced according to the manufacturers’ recommendations. The reads were trimmed from adapter sequences using cutadapt 1.18 with Python 3.6.3. The trimmed reads were mapped to the reference genome using the STAR aligner 2.5.2a and gene abundance was estimated with python 2.7.11, and HTSeq 0.11.0. Differential expression analysis was performed in the R package DESeq2. DESeq2 first adjusts read counts based on a normalization factor that accounts for sample size. This was followed by dispersion estimates based on a negative binomial model which accounts for genes with very few counts. Finally, the Wald test was performed to test for statistical significance. Genes with adjusted p-values (padj) of <0.05 were identified as differentially expressed. To remove genes with high fold changes due to low expression, a minimum normalized expression level (basemean) filter of 50 was added. The filtered data were used to generate volcano plots and heatmaps.

Bioinformatics

Pathway analyses were performed using Gene Ontology, Metascape, and STRING. The STRING networks were generated with the confidence score of 0.7. Putative Nr4a1 DNA binding motifs proximal to transcription start sites (TSS) were identified in HOMER v4.11 with the function findMotifs.pl and the criteria ‘-start -2000 -end 100 -len 8,10 -p 4’. For comparative analysis of cell-specific genetic programs in normal and Nr4a1-deficient neurons, dot plots were generated using the “ggplot2” R package. In these plots, different symbols were used to mark cell populations and/or enrichment. Symbol sizes corresponded to log2 of average expression level (log2(basemean)) of each gene. Differentially expressed genes were presented with bold outlines (padj<0.05). Up- and down-regulated genes were color coded based on log2 of fold changes (log2 FC). Previously published datasets from the same IN populations at earlier developmental stages ² were used as internal controls.

Immunofluorescent and reporter imaging

For immunocytochemistry, cultured neurons attached to the glass coverslips were rinsed once in PBS, fixed for 15 minutes on ice in 4% paraformaldehyde (PFA), 4% sucrose in PBS, and permeabilized for 5 minutes at room temperature in 0.2% Triton X-100 (Roche). After permeabilization, neurons were incubated for 1 hour in the blocking solution containing 5% BSA (Sigma, fraction V) in PBS, followed by overnight incubation with primary and corresponding fluorescently labeled secondary antibodies diluted in the same blocking solution. Samples were washed three times in PBS (10 minutes per wash) after each antibody incubation. The coverslips were then mounted on glass slides with PVA-DABCO (Sigma).

For immunohistochemistry, mice were anesthetized with isoflurane and perfused transcardially with 25 ml of ice-cold PBS followed by 25 ml of 4% PFA in PBS using a peristaltic pump. Brains were removed, incubated overnight in 4% PFA, and sliced in ice-cold PBS using a vibratome. 90 μm thick coronal sections were briefly boiled in the citrate solution containing 10 mM sodium citrate and 0.05% Tween20 (pH 6.0) for antigen retrieval, and subsequently incubated for 5 minutes in the same solution at room temperature. Sections were then washed three times in PBS (5 minutes per wash), blocked for 3 hours in the solution containing 4% BSA, 3% donkey serum and 0.5% Triton, and incubated overnight with primary antibodies diluted in the same blocking solution. Subsequently, sections were washed three times in PBS, incubated with corresponding fluorescently labeled secondary antibodies, washed, mounted on glass slides, and embedded in PVA-DABCO (Sigma).

The following antibodies were used at indicated dilutions:

Chicken anti-GFP (1:1,000, Aves), Mouse anti-Syt2 (1:20, ZIRC), mouse anti-Nr4a1 (1:50, Santa Cruz), Rabbit anti-VGAT (1:1,000, Synaptic Systems), chicken anti-MAP2 (1:10,000, Abcam), mouse anti-GAD67 (1:1,000, Millipore) and Alexa 488, 546, 647 secondary antibodies (1:500, Invitrogen).

3D images were collected under the Nikon C2 or A1 confocal microscopes with 0.3–0.5 μm Z-intervals using 20x, 40x and 60x objectives. All images were subjected to uniform digital filtering to reduce non-specific background.

Image data analysis

SyP-Venus or Syt2-positive presynaptic terminals of PV⁺ cells were automatically marked with Volocity software. The CA1 pyramidal cell layer was visualized by DAPI staining and manually assigned as a region of interest (ROI). Data were collected as density (number of terminals per unit of ROI volume). Threshold values were established for individual channels and applied equally across datasets. A similar strategy was used to quantify the overall numbers of terminals of Sst⁺ cells in CA1sr followed by analysis of dendritic innervation in Imaris (Bitplane). The dendrites of pyramidal cells were visualized by immunolabeling for MAP2 and then reconstructed with the “Create surface” tool to automatically calculate areas and create masks. SyP-Venus-positive boutons were detected with the “spot” tool within these masks. Data were collected as density (number of contacts per unit of dendritic surface). GAD67- and tdTomato-positive boutons were detected with the “spot” tool. Colocalization was analyzed with the “Colocalize spots” tool (Imaris XT extension). Threshold values were calculated individually for each brain using the “above automatic threshold” tool. Individual values were averaged and applied to all images. For analysis of dendrite morphologies, 3D images of single neurons labeled with mGFP were reconstructed from serial stacks and analyzed in Neurolucida (MBF Bioscience). The numbers of PV⁺ and Sst⁺ cells within different hippocampal regions of Pvalb^Cre/Ai9 and Sst^Cre/Ai9 mice were quantified manually in NIS elements (Nikon). For analysis of Nr4a1-RAM induction, cells expressing GFP and tdTomato were detected with the “Spot” tool in Imaris with a spot diameter of 18 μm and 2 additional filters (“Intensity Max Ch=tdTomato” above automatic threshold, “Intensity Median Ch=GFP” above automatic threshold). For analysis of native Nr4a1 induction, ROIs were drawn in NIS Elements (Nikon) using DAPI as a nuclear marker of PV⁺ or Sst⁺ cells labeled with tdTomato. Background ROIs were assigned in CA1sr distant from CA1sp, and their intensity values were subtracted from the values of Nr4a1 intensity. The threshold for Nr4a1-immunoreactive cells was generated based on sample background and uniformly applied to all quantifications from all samples.

Electrophysiology

Mice were anesthetized with isoflurane. Brains were removed and placed into ice-cold oxygenated buffer (95%O2/5%CO2) containing 228 mM sucrose, 2.5 mM KCl, 0.5 mM CaCl₂, 7 mM MgCl₂, 26 mM NaHCO₃, 1 mM NaH₂PO₄, and 11 mM glucose. Transverse, 300 μm thick slices were sectioned with a vibratome and initially stored at 32°C in oxygenated aCSF containing 119 mM NaCl, 2.5 mM KCl, 1 mM NaH₂PO₄, 26 mM NaHCO₃, 1.3 mM MgCl₂, 2.5 mM CaCl₂, 11 mM glucose (pH 7.4, 292 mOsm), and then allowed to recover for 1 hour in oxygenated ACSF at 24 °C prior to recording. Synaptic currents were monitored in whole-cell voltage-clamp mode using Multiclamp 700B amplifier (Molecular Devices, Inc.). Recordings were performed at room temperature. The pipette solution contained 135 mM CsMeSO₄, 8 mM CsCl, 0.25 mM EGTA, 10 mM HEPES, 2 mM MgATP, 0.3 mM Na₂GTP, 5 mM QX-314, and 7 mM Na₂phosphocreatine (pH 7.4, 302 mOsm). Synchronous release was triggered by 1 ms optogenetic stimulation through the Lambda DG-4 illumination system. The frequency, duration, and magnitude of stimuli was controlled by Lambda DG4 (Sutter Instrument, Inc.) and pClamp10 through TTL. Traces were analyzed offline with pClamp10 (Molecular Devices, Inc.) and OriginPro (Origin Lab) software packages.

Behavior

Open field tests (OFT).

Locomotor activity was measured in polycarbonate cages (42 × 22 × 20 cm) placed into frames (25.5 × 47 cm) mounted with two levels of photocell beams (2 and 7 cm) above the bottom of the cage (San Diego Instruments, San Diego, CA). These two sets of beams allow for the recording of both horizontal (locomotion) and vertical (rearing) behavior. A thin layer of bedding material was applied to the bottom of the cage. Mice were tested for 10 and 30 minutes.

Contextual fear conditioning (CFC).

Mice were allowed to explore the fear-conditioning boxes (Context A, Med Associates SD) for 3 minutes and were then subjected to four bursts of foot shocks (0.55 mA, 1 minute inter-shock intervals). Memory tests consisted of one 3 min. exposure to the training box. Freezing was determined in 0.75 seconds bouts and expressed as percent time in the context.

Spatial memory was examined in the circular Barnes maze with 20 holes and escape box. Distinct spatial cues were located around the maze and kept constant throughout the study. Mice were subjected to four sequential daily training sessions followed by the Probe Test of memory retrieval. In this test, the escape tunnel was removed, and mice were allowed to freely explore the maze for 3 min. The time spent in each quadrant was determined and the percent time spent in the target quadrant (the one originally containing the escape box) was compared to the average percent time in the other three quadrants.

QUANTIFICATIONS AND STATISTICAL ANALYSIS

All quantifications and statistical analyses were performed in OriginPro (Origin Lab). Means, medians, standard deviations (S.D.) and standard errors (S.E) are shown in all graphs. The p values were determined with Student’s t-test (for two groups) and analysis of variance (for multiple groups). Welch’s correction was applied to Student’s t-test when two groups have unequal sample size. All manual quantifications were performed in a blind manner, by investigators who were unaware of genotypes and experimental conditions.

Key Resources Table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Chicken polyclonal anti-GFP	AVES	Cat#GFP1010, RRID: AB_2307313
Mouse monoclonal anti-Syt2	ZIRC	CAT#Znp-1, RRID: AB_10013783
Mouse monoclonal anti-Nr4a1	Santa Cruz	Cat#sc-166166, RRID: AB_2153745
Rabbit polyclonal anti-VGAT	Synaptic Systems	Cat#131003, RRID: AB_887869
Chicken polyclonal anti-Map2	Abcam	Cat# ab5392, RRID: AB_2138153
Mouse monoclonal anti-Gad67	Millipore	Cat# MAB5406, RRID: AB_2278725
Bacterial and Virus Strains
AAVDJ EF1α-DIO:mGFP	Maximov lab	PMID: 25277456
AAVDJ EF1α-DIO:Synaptophysin-Venus	Maximov lab	PMID: 25277456
AAVDJ EF1α-DIO:ChR2-EYFP (H134R)	Deisseroth Lab	N/A
AAVDJ EF1α-DIO:Nr4a1	This manuscript	N/A
AAVDJ nrRAM-DIO:GFP	This manuscript	N/A
AAVDJ EF1αcore-DIO:GFP	This manuscript	N/A
AAVDJ EF1αcore-DIO:Epha3	This manuscript	N/A
AAVDJ EF1αcore-DIO:Sema3C	This manuscript	N/A
AAVDJ CMV-DIO:saCas9	This manuscript	N/A
AAVDJ 2xU6:gEpha3α/β-Syn:BFP	This manuscript	N/A
AAVDJ 2xU6:gEpha3χ/δ-Syn:BFP	This manuscript	N/A
AAVDJ 2xU6:gSema3cα/β-Syn:BFP	This manuscript	N/A
AAVDJ 2xU6:gSema3cχ/δ-Syn:BFP	This manuscript	N/A
AAVDJ 2xU6-Syn:BFP	This manuscript	N/A
LV-Ubc:Cre	Südhof lab	PMID: 30808661
LV-Ubc:ΔCre	Südhof lab	PMID: 30808661
Chemicals, Peptides, and Recombinant Proteins
APV	Tocris	Cat#01061
CNQX	Tocris	Cat#01045
Picrotoxin	Tocris	Cat#1128
Tetrodotoxin	Tocris	Cat#4368289
DTT	Sigma	Cat#646563
cOmplete™ Protease Inhibitor Cocktail	Roche	Cat# 4693159001
RNAsin	Promega	Cat#2115
Cycloheximide	Sigma	Cat#C7698
Heparin	Sigma	Cat#H3393
Anti-HA Magnetic Beads	Pierce	Cat#88836
Poly-D-Lysine	Millipore	Cat#A003E
Kainic acid monohydrate	Sigma	Cat#K0250
Experimental Models: Cell Lines
HEK293T	ATCC	N/A
Primary neuronal cultures	This paper	N/A
Experimental Models: Organisms/Strains
SstIRES-Cre mouse allele	JAX	PMID: 21943598
PvalbCre mouse allele	JAX	PMID: 15836427
Emx1IRES-Cre mouse allele	JAX	PMID: 12151506
Ai9 reporter mouse allele	JAX	PMID: 20023653
Nr4a1fl/fl mouse allele	Chambon Lab	PMID: 23334790
Ribotag mouse allele	JAX	PMID: 19666516
Software and Algorithms
pClamp10	Mol. Devices	N/A
Imaris 9	Bitplane	N/A
Volocity	Quorum	N/A
OriginPro	Origin Lab	N/A
Nikon Elements	Nikon	N/A
Neurolucida	MBF Bioscience	N/A
FIJI	ImageJ	N/A
Salmon	Github	PMID: 28263959
Python 2.7.11 and 3.6.3	Python	N/A
HOMER 4.11	Benner Lab	PMID: 20513432
cutadapt 1.18	Cutadapt	DOI: https://doi.org/10.14806/ej.17.1.200
STAR aligner 2.5.2a	STAR	PMID: 23104886
HTSeq 0.11.0	HTSeq	PMID: 25260700
R 3.6.1	CRAN	R Core team, 2019
DESeq2 1.20.0	Bioconductor	PMID: 25516281
Metascape	Metascape	PMID: 30944313

Highlights.

Nr4a1 is an immediate early gene whose expression is induced by sensory cues

Nr4a1 regulates cell-specific transcriptional programs in GABAergic interneurons (INs)

Disruption of Nr4a1 signaling in INs affects associative learning

Nr4a1-deficient INs exhibit abnormal connectivity and physiological properties