Procognitive restoration of PV neuron plasticity in neurodevelopmental disorders

The hippocampus has a crucial role in memory formation, storage and retrieval to facilitate the experience-dependent calibration of motivated and defensive behaviours². During the early postnatal period, experience refines the hippocampal circuitry to influence cognitive trajectory²⁰. Disruptions in hippocampal circuitry result in maladaptive neural circuit functions, cognitive impairments and seizures that characterize different NDDs, including autism spectrum disorder (ASD). The identification of ultrahigh-confidence genetic risk factors for NDDs underscores the need to instantiate how experience and genetic risk interact to impair experience-dependent mechanisms that support hippocampal-dependent cognition.

PV INs in the dentate gyrus (DG)–CA3/CA2 circuit of the hippocampus influence encoding, storage, retrieval and routing of memories through activity-dependent regulation of CA3/CA2 PN perisomatic inhibition^{4,5,6,7,9,10,11}. Feedforward inhibition of CA3/CA2 PNs mediated by PV INs dictates their spiking, synchronization of PN activity to form ensembles and the generation of network oscillations to regulate intrahippocampal and inter-regional communication in hippocampal–cortical–subcortical networks^11,21,22. To exert these effects on circuitry and network properties, PV INs cell-autonomously coordinate experience-dependent changes in their intrinsic properties, structural reorganization of axonal arborizations and perisomatic synapses in CA3/CA2, regulation of feedforward inhibition of CA3/CA2 and synaptic plasticity^{9,10,12,13,14,15}. Here, these properties are collectively referred to as experience-dependent PV IN plasticity. Loss of PV IN functions in NDDs, including ASD, bipolar disorder, schizophrenia and epilepsies, may arise from genetic risk factors that impair experience-dependent refinement of inhibition mediated by PV INs during the early postnatal period^{16,17,18,23,24}.

Transcription factors and epigenetic regulators co-ordinate changes in gene expression that underlie synaptic physiology, synaptic and structural plasticity and input–output connectivity to mediate experience-dependent inhibitory and excitatory neuron plasticity. We know a substantial amount about the molecular mechanisms that regulate cortical PV IN identity and experience-dependent plasticity^{12,17,25,26,27}, but less about developmental regulators of hippocampal PV IN properties^13,28,29. Moreover, evidence for transcription factors and epigenetic regulators that control experience-dependent PV IN plasticity in the adult hippocampus is scarce.

PV INs in CA3/CA2 retain experience-dependent plasticity in adulthood. Specifically, learning^6,10 and social experience⁹ increase mossy fibre excitatory synaptic inputs onto PV INs to trigger experience-dependent PV IN plasticity. This process subsequently leads to increased feedforward inhibition in DG–CA3/CA2 circuits⁹. Enhanced mossy fibre excitatory drive onto PV INs increases intrinsic excitability, inhibitory synapses onto CA3/CA2 and perisomatic inhibition of CA3/CA2 PNs to support spatial and social memory^9,10. At a network level, DG recruitment of PV-IN-mediated inhibition of CA3/CA2 PNs promotes stable and context-specific neuronal ensembles in hippocampal–cortical networks and network oscillations during memory consolidation^10,11. Here we build on this previous work to screen for cell-autonomous regulators of XPGs in the adult CA3/CA2 circuit. We report a suite of candidate XPGs, including transcription factors and epigenetic regulators, that are encoded by ultrahigh-confidence risk genes for NDDs and epilepsies. We test the hypothesis that restoration of experience-dependent PV IN plasticity in the adult hippocampus in a widely used NDD risk mouse model is sufficient to reverse developmental deficits in circuitry, ensemble specificity, sharp-wave ripple (SWR) properties, seizures and cognition.

To identify XPGs in CA3/CA2 PV INs, we designed an in vivo functional screen in adult mice to isolate translated mRNAs from CA3/CA2 PV INs that receive increased mossy fibre inputs, a trigger for experience-dependent PV IN plasticity (Fig. 1a). To this end, we used lentiviruses to express short hairpin RNA (shRNA) to downregulate Ablim3 (shAblim3) or a non-target shRNA as a control (shNT). ABLIM3 is a molecular brake of connectivity between mossy fibre filopodia and PV INs, and its downregulation in DG mossy fibres^9,10,11 reproduces the effects of experience on mossy fibre filopodial inputs onto PV INs^9,10,11. We used PV^cre;Rpl22^HA/HA mice³⁰ (PV is also known as Pvalb), in which expression of the haemagglutinin (HA) epitope-tagged ribosomal protein L22 (RPL22) is restricted to PV INs¹⁰. After optimization of the signal-to-noise ratio for mRNA isolation from this extremely sparse population of PV INs³¹ (Extended Data Fig. 1a,b), we compared mRNAs biochemically isolated from genetically tagged PV INs in the naive state versus an experience-induced activated state in CA3/CA2. Using this approach, we identified differentially expressed candidate XPGs associated with the PV IN activated state (Fig. 1b,c and Supplementary Table 1). Significantly upregulated genes included 82 category 1 and S1 ASD genes reported in the SFARI database (https://gene.sfari.org/)³². Other genes identified included 4 genes causally implicated in schizophrenia through exome sequencing (data from the Schizophrenia Exome Sequencing Meta-analysis (SCHEMA) project)³³ and 12 bipolar disorder candidate genes, including AKAP11 (P < 0.01, out of a total of 64 bipolar disorder candidate genes that scored P < 0.05 according to Bipolar Exome (BipEx) data; https://bipex.broadinstitute.org/)³⁴. A curated and updated annotated XPG list is available online (https://xpg-atlas.onrender.com/). By contrast, we identified only three category 1, syndromic ASD genes in the list of significantly downregulated XPGs (Supplementary Table 1). Therefore, we focused on upregulated XPGs in this study. A third of the identified ASD genes encode histone methyltransferase, histone demethylases, chromatin remodelling factors and transcription factors, the functions of which in PV INs are poorly defined (Supplementary Table 1). Gene set enrichment analysis of all the genes upregulated in the experience-induced PV IN activated state suggested that these XPGs have a role in synaptic physiology, structural remodelling and presynaptic terminal specialization (Extended Data Fig. 1c). Half of the ASD-linked upregulated XPGs are implicated in epilepsies, consistent with the high incidence of seizures in individuals with ASD (https://gene.sfari.org/) (Fig. 1c). Moreover, individuals with ASD or schizophrenia often have loss-of-function mutations in these genes, and analyses of data from BipEx suggest that the bipolar disorder risk candidate XPGs have damaging missense or protein-truncating mutations. Because the transcripts of these XPGs are upregulated or poised for translation in PV INs in response to increased mossy fibre inputs, our findings indicate that loss of PV IN experience-dependent plasticity is a convergent mechanism for different NDDs, including ASD, bipolar disorder, schizophrenia and epilepsies.

The upregulation of select candidate ASD-linked XPGs (Meis2, Bcl11a and Tbr1; https://gene.sfari.org/) and the SCHEMA gene Herc1 (ref. ³³) in activated CA3/CA2 PV INs was validated by qPCR (Extended Data Fig. 1d). Therefore, we asked whether inducing the expression of these XPGs in PV INs via S5E2 enhancer-driven AAV vectors³⁵ promotes the formation of PV IN–CA3/CA2 synapses in adult mice. Virus-mediated expression of Tbr1, Bcl11a, Meis2 or Herc1 in S5E2-targeted PV INs increased PV IN synapses in CA3/CA2 PNs (Fig. 1d and Extended Data Fig. 1f–h). Meis2 is expressed at negligibly low levels in a subpopulation of hippocampal PV INs in the hippocampus³⁶, where its function is unclear. MEIS2 is a homeodomain-containing transcription factor that acts as a cellular-context-dependent coactivator to regulate the developmental specification of GABAergic projection neurons^19,37 and maintains intrinsic excitability and terminal arborization in sensory neurons³⁸. On the basis of these observations, we prioritized Meis2 as a candidate XPG to restore experience-dependent PV IN plasticity in NDDs. As the S5E2 enhancer preferentially, but not exclusively, targets PV INs in CA3/CA2 (Extended Data Fig. 1e), we asked whether Meis2 overexpression in CA2 PNs may also contribute to the increase in PV IN synapses onto CA2 PNs. Using a CA2-specific Amigo2^cre mouse line in combination with an AAV-DIO-Meis2-P2A-mCherry virus, we found that Meis2 overexpression in CA2 PNs does not increase PV IN synapses onto mCherry⁺ CA2 PNs (Fig. 1e–g and Extended Data Fig. 1i). Many NDDs, including ASD, are characterized by a range of developmental deficits, intellectual disability, seizures and impaired social cognition. Therefore, we selected Cntnap2 knockout (Cntnap2^−/−) mice to test our hypothesis. The goal was not to model a specific NDD but to use an NDD risk mouse model, like Cntnap2^−/− mice, which exhibits a range of developmental deficits, including altered neuronal migration and excitability, loss of PV INs, cognitive impairments and seizures by around 6 months of age^39,40,41,42.

We tested whether CA3/CA2 PV INs in adult (2 months old) Cntnap2^−/− mice retain competence to increase their perisomatic synapses in response to social experience. Cntnap2^−/− mice had fewer PV IN perisomatic synapses (as indicated by synaptotagmin 2 (SYT2⁺) staining) than wild-type littermates (Cntnap2^+/+) (Fig. 2a,b). Moreover, exposure to a social stimulus following habituation to a context resulted in an increase in PV IN perisomatic synapses in Cntnap2^+/+ mice but not in Cntnap2^−/− littermates (Fig. 2a,b). Next, we tested whether Meis2 expression levels in CA3/CA2 PV INs in adult Cntnap2^−/− mice and Cntnap2^+/+ littermates change in response to social experience (Fig. 2c). Multiplex fluorescent in situ hybridization for Meis2, PV and RGS14 (Fig. 2d,e) and immunohistochemistry for corresponding protein levels (Fig. 2f,g) in hippocampal sections obtained from mice habituated to a context (social-experience naive) and mice exposed to a social stimulus in the habituated context revealed a social-experience-dependent increase in Meis2 transcripts and protein (Meis2 and MEIS2 intensity per PV IN and proportion of PV INs that express Meis2 and MEIS2) in Cntnap2^+/+ mice but not in Cntnap2^−/− mice (Fig. 2c–g).

AAV-S5E2-Meis2-P2A-nlsdTomato (S5E2-Meis2) virus-mediated expression of Meis2 in CA3/CA2 PV INs of adult Cntnap2^−/− mice (hereafter referred to as S5E2-targeted PV INs) or genetically restricted expression of Meis2 in adult CA3/CA2 PV INs using AAV-DIO-Meis2-P2A-mCherry in combination with adult PV^cre;Cntnap2^−/− mice (hereafter referred to as PV-Cre-targeted PV INs) was sufficient to restore PV IN perisomatic synapses in CA2, as quantified using SYT2⁺ and SYT2⁺gephyrin⁺ immunohistochemistry, respectively⁸ (Fig. 2h,i). We next determined whether anatomical restoration of PV IN synapses in CA2 of adult Cntnap2^−/− mice was accompanied by changes in PV IN properties, physiology and synaptic transmission in the DG–CA2 circuit. We targeted expression of Meis2 or control dTomato in PV INs along the stratum lucidum mossy fibre pathway adjacent to the CA2/CA3a PN layer of the dorsal hippocampus in adult Cntnap2^−/− mice and in Cntnap2^+/+ littermates using AAV-S5E2-Meis2-P2A-nlsdTomato virus or AAV-S5E2-dTom-P2A-nlsdTomato control virus, respectively. Ex vivo whole-cell patch-clamp recordings of CA2 PNs revealed that Cntnap2^−/− mice exhibited increased spontaneous excitatory postsynaptic currents (sEPSCs) in CA2 PNs compared with Cntnap2^+/+ mice. Moreover, Meis2 expression in S5E2-targeted PV INs reduced the sEPSC frequency and amplitude in Cntnap2^−/− and Cntnap2^+/+ mice (Fig. 3a,b). Spontaneous inhibitory postsynaptic current (sIPSC) frequency in CA2 PNs from Cntnap2^−/− mice was reduced compared with Cntnap2^+/+ littermates, and Meis2 expression in S5E2-targeted PV INs reversed this reduction in Cntnap2^−/− mice (Fig. 3c). To assess action-potential-independent spontaneous vesicle release, we recorded miniature excitatory postsynaptic currents (mEPSCs) and miniature inhibitory postsynaptic current (mIPSCs) from CA2 PNs. Cntnap2^−/− mice exhibited a higher mEPSC frequency than Cntnap2^+/+ littermates, and Meis2 expression in S5E2-targeted PV INs reversed this increase (Fig. 3d). There was no difference in the frequency or amplitude of mIPSCs from CA2 PNs between Cntnap2^−/− and Cntnap2^+/+ mice. However, Meis2 expression in S5E2-targeted PV INs increased the mIPSC frequency and amplitude in CA2 PNs of Cntnap2^−/− mice (Fig. 3e). We assessed sEPSCs and mEPSCs in S5E2-targeted PV INs and found that Meis2 expression in these neurons reversed the increase in frequency and amplitude of sEPSCs in Cntnap2^−/− mice and increased mEPSC amplitude in both Cntnap2^−/− and Cntnap2^+/+ mice (Fig. 3f–h).

Analyses of mossy-fibre-driven evoked current onto CA2 PNs revealed that Cntnap2^−/− mice exhibited higher-amplitude optically evoked EPSCs and lower-amplitude IPSCs across the stimulus train than Cntnap2^+/+ mice (Fig. 3i–k). Meis2 expression in S5E2-targeted PV INs attenuated evoked EPSCs and increased evoked amplitude IPSCs to levels comparable to those observed in Cntnap2^+/+ mice (Fig. 3k). Compared with all other groups, Cntnap2^−/− mice exhibited an overall increased excitation to inhibition ratio (Fig. 3k). We did not observe differences in vesicle release probability across the stimulation train, which indicated that stimulation frequency does not induce short-term plasticity (Fig. 3k). These results suggest that Cntnap2^−/− mice have an increased frequency of excitatory transmission and reduced inhibition onto CA2 PNs and that Meis2 expression in S5E2-targeted PV INs reverses these developmental phenotypes. The reduction in excitatory synaptic transmission in CA2 may arise in response to sustained increases in MEIS2-dependent inhibition of PNs.

Using a mossy fibre electrical-stimulation protocol, we independently corroborated the finding that increased Meis2 expression in S5E2-targeted PV INs enhances feedforward inhibition on CA2 PNs in Cntnap2^+/+ and Cntnap2^−/− mice (Extended Data Fig. 2a–c). Treatment of brain slices with gabazine to block GABA_A receptor-mediated feedforward inhibition and to unmask excitatory synaptic input abrogated significant differences in the EPSC amplitude between Cntnap2^−/− mice and other groups. This result provides support for our observations that increased Meis2 expression in S5E2-targeted PV INs mediates feedforward inhibitory control of excitation on CA2 PNs (Extended Data Fig. 2d,e). Analyses of the Schaffer collateral CA3–CA2 pathway using an electrical-stimulation protocol with and without gabazine treatment also revealed a role for increased Meis2 expression in S5E2-targeted PV INs in restoring feedforward inhibition on CA2 PNs and mediating feedforward inhibitory control of excitation on CA2 PNs in Cntnap2^−/− mice (Extended Data Fig. 2f–j).

Next, we asked whether Meis2 expression in S5E2-targeted PV INs influences mossy-fibre-driven, inhibitory-dependent synaptic plasticity (that is, inhibitory long-term depression (iLTD)), which is thought to facilitate routing of information⁴³. To induce inhibitory-dependent plasticity, we stimulated the mossy fibre pathway by electrical theta burst stimulation (TBS) and recorded from CA2 PNs⁴³. TBS along the mossy fibre pathway induced iLTD of the IPSC and long-term potentiation (LTP) of the EPSC on CA2 PNs (Fig. 3l,m and Extended Data Fig. 2k–m). Cntnap2^−/− mice exhibited reduced occurrence of iLTD compared with Cntnap2^+/+ littermates, and Meis2 expression in S5E2-targeted PV INs restored the occurrence of iLTD in both Cntnap2^−/− and Cntnap2^+/+ mice (Fig. 3m). Vesicle release probability of IPSCs was reduced from baseline (5 min before TBS) to after iLTD (between 35 and 40 min after TBS) in all other groups except Cntnap2^−/− mice. There were no changes in vesicle release probability in the EPSC across all groups (Extended Data Fig. 2m). This result suggests that mossy-fibre-driven iLTD and subsequent LTP of the EPSC on CA2 PNs is a presynaptic mechanism in inhibitory synapses.

Current injections in S5E2-targeted PV INs revealed reduced excitability in Cntnap2^−/− mice, and Meis2 expression in these neurons restored excitability and reduced rheobase in both Cntnap2^−/− and Cntnap2^+/+ mice (Fig. 3n–p). There were no differences in the resting membrane potential in any of the groups. However, S5E2-targeted PV INs from Cntnap2^−/− mice exhibited a more depolarized threshold. Notably, boosting Meis2 did not affect the threshold in Cntnap2^−/− or in Cntnap2^+/+ mice, which suggests that the depolarized threshold arises from a Meis2-independent compensatory mechanism (Fig. 3p). Whole-cell recordings of S5E2-targeted PV INs before and after bath application of cyanquixaline (an AMPA and kainite receptor competitive antagonist) and d-2-amino-5-phosphonopentanoic acid (AP5; a selective NMDA receptor antagonist) revealed that excitatory synaptic transmission does not contribute to Meis2-dependent restoration of PV IN excitability in Cntnap2^−/− mice (Extended Data Fig. 2n–q).

The S5E2 enhancer also targets non-PV INs, albeit to a substantially lesser extent (Extended Data Fig. 1e). Therefore, we conducted essential control experiments to address the possibility that S5E2 targeting of Meis2 expression in putative CA2 PNs (or non-PV INs) may recruit PV-IN-mediated inhibition onto CA2 PNs. We analysed optically evoked EPSCs and IPSCs in S5E2-targeted PV INs and CA2 PNs after AAV-S5E2-ChR2-mCherry injections into CA3/CA2 of adult Cntnap2^+/+ mice. We recorded IPSC but no EPSC responses in PV INs or in CA2 PNs. This result suggests that S5E2-enhancer-driven expression of channelrhodopsin does not occur in CA2 PNs (Extended Data Fig. 3a,b). We also assessed the contribution of S5E2-enhancer-driven expression of Meis2 in non-PV neurons (Extended Data Fig. 3c). For current-clamp recordings, we targeted dTomato⁺GFP^– (non-PV) neurons located in the CA2 PN layer and in the adjacent stratum lucidum layer. PNs were distinguished from INs by differences in their electrophysiological properties; for example, PN action potential adaptation and asynchronous firing rate (Extended Data Fig. 3c). Of the neurons patched, none exhibited characteristics of PNs. There were no significant differences in action potential spiking or intrinsic properties across groups, apart from input resistance in non-PV INs. Specifically, non-PV INs located in the CA2 PN layer or stratum lucidum of Cntnap2^−/− mice exhibited reduced input resistance compared with Cntnap2^+/+ littermates, and Meis2 overexpression in non-PV INs significantly increased input resistance in both groups (Extended Data Fig. 3d,e).

To determine whether another XPG, Tbr1 (which encodes T-brain-1, a T-box transcription factor), and Meis2 have overlapping or distinct roles in regulating PV IN physiology, we performed recordings from S5E2-targeted PV INs and CA2 PNs after transducing AAV-S5E2-Tbr1-P2A-nlsdTomato in CA3/CA2 of Cntnap2^+/+ mice (Extended Data Fig. 4). Tbr1 expression in S5E2-targeted PV INs resulted in reduced sEPSC frequency and amplitude, reduced mEPSC frequency and increased mIPSC frequency on CA2 PNs (Extended Data Fig. 4a–c). Mossy-fibre-driven evoked current onto CA2 PNs revealed an increase in IPSC amplitude; however, there were significant differences in the excitation to inhibition ratio compared with mice transduced with control virus (Extended Data Fig. 4d,e). Tbr1 expression in S5E2-targeted PV INs did not result in any differences in intrinsic excitability or passive membrane properties from S5E2-targeted PV INs (Extended Data Fig. 4f,g).

Together, these observations demonstrate that Meis2 expression in S5E2-targeted CA3/CA2 PV INs of adult Cntnap2^−/− mice reverses developmental deficits in PV IN intrinsic excitability, input–output synaptic connectivity, excitatory and inhibitory synaptic transmission and inhibitory-dependent synaptic plasticity. Our results of Meis2 and Tbr1 in wild-type mice also suggest that these two XPGs differentially regulate PV IN excitability.

To independently establish that Meis2 expression in adult CA3/CA2 PV INs restores feedforward inhibition on CA2 PNs in Cntnap2^–/– mice, we performed a genetic–temporal strategy. Whole-cell recordings of CA2 PNs after Meis2 expression in PV-Cre-targeted CA3/CA2 PV INs (AAV-DIO-Meis2-P2A-mCherry or AAV-DIO-mCherry injections into CA3/CA2 of adult PV^cre;Cntnap2^+/+ or PV^cre;Cntnap2^−/− mice) revealed significantly reduced frequency and amplitude of sEPSCs in both genotypes (Extended Data Fig. 5a,b). Meis2 expression in PV-Cre-targeted PV INs reversed the reduction in sIPSC frequency in Cntnap2^−/− mice (Extended Data Fig. 5c). CA2 PNs of Cntnap2^−/− mice exhibited higher mEPSC frequency than Cntnap2^+/+ littermates, and Meis2 expression in PV-Cre-targeted PV INs reversed this increase (Extended Data Fig. 5d). Meis2 expression in PV-Cre-targeted PV INs restored mIPSC frequency and amplitude in CA2 PNs of Cntnap2^−/− mice (Extended Data Fig. 5e). Meis2 expression in PV-Cre-targeted PV INs reversed the increase in sEPSC frequency in these neurons in Cntnap2^−/− mice (Extended Data Fig. 5f–h).

We next assessed mossy-fibre-driven evoked currents onto CA2 PNs after PV-Cre-targeted Meis2 expression in CA3/CA2 PV INs in Cntnap2^+/+ and Cntnap2^−/− littermates (Extended Data Fig. 5i). Cntnap2^−/− mice exhibited higher amplitude optically evoked EPSCs and lower amplitude IPSCs across the stimulus train than Cntnap2^+/+ mice. Meis2 expression in PV-Cre-targeted PV INs attenuated evoked EPSCs and increased evoked amplitude IPSCs to restore the excitation to inhibition ratio in Cntnap2^−/− mice (Extended Data Fig. 5j–l). We did not observe differences in vesicle release probability across the stimulation train, which indicated that stimulation frequency does not induce short-term plasticity (Extended Data Fig. 5k).

The alignment of results obtained using genetic–temporal (PV-Cre + AAV-DIO) and S5E2-enhancer-targeted PV INs strategies, taken together with extensive control experiments, establish that Meis2 expression in adult CA3/CA2 PV INs in Cntnap2^–/– mice restores feedforward inhibition on CA2 PNs.

We next asked whether Meis2-dependent restoration of experience-dependent PV IN plasticity in CA3/CA2 PV INs of adult Cntnap2^−/− mice rescues spatial- and social-cognition impairments. After Meis2 expression in S5E2-targeted PV INs, adult (2–3 months old) Cntnap2^+/+ and Cntnap2^−/− mice were tested in the novel object location task (Fig. 4a,b). Consistent with the role that the DG–CA3/CA2 circuitry has in encoding configural relationships between objects and their locations in contexts, Cntnap2^+/+ but not Cntnap2^−/− mice spent more time exploring the displaced object than the non-displaced object in the habituated context. Meis2 expression in S5E2-targeted PV INs in Cntnap2^−/− mice rescued cognitive behavioural impairments in this task, and performance matched the behaviour of Cntnap2^+/+ littermates (Fig. 4c). In the social-recognition memory paradigm, which is sensitive to feedforward inhibition in the DG–CA2 circuit⁹, Cntnap2^−/− mice spent an equivalent time exploring the empty cup and social stimulus in trial 2 and did not discriminate between the novel social stimulus and the familiar social stimulus in trial 3 (Fig. 4d–f). Meis2 expression in S5E2-targeted PV INs in Cntnap2^−/− mice reversed these impairments in social recognition and social discrimination (Fig. 4f). Analyses of this same cohort of mice in assays for anxiety-like behaviour (open field), novel object recognition and homecage group social behaviour using social leap estimates animal poses (SLEAP)⁴⁴ did not detect differences between these groups (groups shown in Extended Data Figs. 6a–d and 7a–d). Independently, Meis2 expression in PV-Cre-targeted CA3/CA2 PV INs (AAV-DIO-Meis2-P2A-mCherry or AAV-DIO-mCherry injections into CA3/CA2 of adult PV^cre;Cntnap2^+/+ or PV^cre;Cntnap2^−/− mice) rescued object-location memory and social-recognition memory deficits (Fig. 4g,h). No differences in anxiety-like behaviour or novel object recognition were observed between these groups of PV^cre;Cntnap2^+/+ and Cntnap2^−/− mice (Extended Data Fig. 6e–g).

Increased MEIS2 levels in CA3/CA2 PV INs may rescue social recognition in Cntnap2^−/− mice by promoting the specificity of neuronal ensembles that encode social stimuli. To test this hypothesis, we used a dual light and calcium-dependent neuronal activity tagging system, soma-targeted Cal-Light⁴⁵, to tag active neurons (with induction of GFP) during exploration of the social stimulus in a habituated context. We then quantified reactivation of the social stimulus-associated ensemble after re-exposure to the same social stimulus. After habituation to the context, epoch-specific light illumination induced GFP expression only in neurons that were active during social behaviour (Extended Data Fig. 8a–d). Elevated reactivation (based on the percentage of FOS⁺GFP⁺/GFP⁺ cells) indicated increased specificity of the neuronal ensemble (Fig. 4i and Extended Data Fig. 8e). In both Cntnap2^+/+ and Cntnap2^−/− mice, we detected equivalent numbers of active cells in recognition and retrieval phases. However, Meis2 expression in S5E2-targeted CA3/CA2 PV INs of Cntnap2^−/− mice led to significantly enhanced ensemble reactivation of the social-stimulus-associated ensemble compared with mice infected with control virus (Fig. 4j,k and Extended Data Fig. 8f,g). We did not detect PV INs tagged with GFP, presumably owing to technical limitations or PV IN calcium dynamics (Extended Data Fig. 8h).

In vivo recordings^21,46, calcium and voltage imaging^47,48, chemogenetic manipulations⁴⁹ and modelling⁵⁰ strongly suggest that PV-IN-mediated perisomatic inhibition has a role in SWRs, a neural substrate for memory consolidation^{3,51,52,53,54}. Local field potential (LFP) recordings in CA1 for 2 h before and after social interaction with a sex-matched juvenile revealed that 2-month-old Cntnap2^−/− mice exhibit reduced SWR rate, peak amplitude, ripple power and duration (Extended Data Fig. 9a–i). Meis2 expression in S5E2-targeted CA3/CA2 PV INs restored ripple rate, peak amplitude, power and duration during non-rapid eye movement (NREM) sleep in Cntnap2^−/− mice after social interaction (Extended Data Fig. 9a–i and Supplementary Table 2). Notably, ripple power was restored to levels comparable with the baseline period before social interaction (Extended Data Fig. 9h and Supplementary Table 2). Taken together, these observations demonstrate that Meis2 expression in CA3/CA2 PV INs of adult Cntnap2^−/− mice reverses developmental deficits in ensemble and SWR properties and hippocampal-dependent cognition.

Cntnap2^−/− mice can develop seizures after 6 months of age^39,41,42. After 4–5 months of Meis2 expression in S5E2-targeted CA3/CA2 PV INs of 2-month-old Cntnap2^+/+ and Cntnap2^−/− mice, electrocorticography (ECoG) electrodes were implanted, and continuous ECoG recordings were performed for 2 weeks after recovery from surgery (Fig. 5a). Seizure frequency and duration were quantified in all experimental groups. Overall, 10 out of 14 Cntnap2^−/− mice treated with a control AAV (AAV-S5E2-dTom) displayed spontaneous electrographic seizures, with a mean frequency of 1.83 (± 5.48 s.d., ± 1.46 s.e.m.) per day. By contrast, 3 out of 12 Cntnap2^−/− mice treated with Meis2 AAV (AAV-S5E2-Meis2-nlsdTom) displayed spontaneous seizures, with a mean frequency of 0.15 (± 0.34 s.d., ± 0.1 s.e.m.) per day (Fig. 5b–e). This result shows that S5E2-targeted Meis2 expression in CA3/CA2 PV INs significantly reduces seizure frequency in Cntnap2^−/− mice and fewer mice experienced spontaneous seizures (Fig. 5d). Significant differences in seizure frequency remained even after we excluded a Cntnap2^−/− mouse in the control group that experienced more than 20 seizures per day (Extended Data Fig. 10a–f). In Cntnap2^+/+ mice, none treated with the control virus experienced seizures (0 of 3 mice). However, one Cntnap2^+/+ mouse treated with Meis2 virus had two seizures during the 2-week recording window (1 out of 6 mice) (Extended Data Fig. 10a,b,d). Seizure duration was similar across all groups, and equivalent numbers of PV INs expressed S5E2-targeted Meis2 (Extended Data Fig. 10d,g). We did not detect an increase in gliosis in these mice (Extended Data Fig. 10h). These findings demonstrate that Meis2 expression in S5E2-targeted CA3/CA2 PV INs reduces seizure phenotypes in Cntnap2^−/− mice.

Impaired experience-dependent refinement of PV inhibitory circuits contributes to the development of NDDs, including ASD^{16,17,18,23,24}. The identification of shared risk genes for different NDDs, including ASD, bipolar disorder and schizophrenia, suggest overlapping genetic architectures^32,33,34,55. But how these different genes converge on biological mechanisms that govern PV IN functions essential to cognition is poorly understood. Experience-dependent PV IN plasticity is a candidate biological mechanism for NDDs as it underlies the capacity of PV INs to coordinate changes in their intrinsic properties, synaptic physiology, input–output connectivity and synaptic plasticity in response to different inputs. However, the identities of cell-autonomous regulators of experience-dependent PV IN plasticity in adult hippocampus have remained elusive. We designed an input-specific activity sensitive screen for XPGs in CA3/CA2 PV INs of adult mice that does not rely on the imposition of artificial patterns of activity onto PV INs but mimics physiological experience-dependent changes in mossy fibre inputs onto PV INs that trigger a change in PV IN cell-state properties. We discovered a suite of XPGs whose expression in PV INs is increased in response to increased mossy fibre excitatory drive. A significant proportion of upregulated XPGs in CA3/CA2 PV INs in our unbiased screen exhibit loss-of-function mutations in ASD, schizophrenia, bipolar disorder and epilepsy risk genes. Based on the directionality of expression change in XPGs in our screen and loss-of-function mutations in these NDD risk genes, we propose that impaired experience-dependent PV IN plasticity is a convergent mechanism for NDDs, including ASD, bipolar disorder and schizophrenia.

Using both genetic–temporal and viral strategies, we demonstrated that Meis2-dependent restoration of experience-dependent PV IN plasticity in CA3/CA2 in adult Cntnap2^–/– mice is sufficient to reverse developmental deficits in excitation–inhibition balance, cognitive impairments and seizure phenotypes. MEIS2-dependent suppression of seizures may reflect a coordinated reduction in excitatory and increased inhibitory synaptic transmission in the DG–CA2 circuit⁵⁶ and, potentially, in the DG–CA3 circuit. XPGs are viable candidates for restoring impaired experience-dependent PV IN plasticity and PV IN homeostatic plasticity in different NDDs that exhibit deficits in these mechanisms^9,57,58.

MEIS2 regulation of CA3/CA2 PV IN physiology, synaptic connectivity and plasticity reflect an amalgam of previously identified roles for MEIS2 in mediating GABAergic projection neuron identity, sensory neuron physiology and terminal specialization^19,37,38. We infer that MEIS2 may function as a permissive coactivator that acts in concert with other transcription factors and chromatin remodellers to instruct different facets of experience-dependent PV IN plasticity (Extended Data Fig. 11b). The observation that Meis2 expression is not detected in the medial ganglionic eminence^8,19, the embryonic source for hippocampal PV INs, suggests an evolutionarily parsimonious repurposing of Meis2 based on function rather than PV IN identity specification to regulate experience-dependent programs in PV INs in the adult brain.

Our study provides a mechanistic framework to investigate how PV INs dynamically change their properties, physiology and connectivity to define characteristics of the cell state in response to experience to influence encoding of contextual, social and spatial information in DG–CA2/CA3. Induction of different XPGs in CA3/CA2 PV INs may enable DG to flexibly switch between two modes of CA3/CA2 PN activity regulation—that is, global suppression of all PNs through provision of PV-IN-mediated blanket inhibition⁷ or selective inhibition of distinct PNs⁵⁹ to dictate ensemble dynamics, remapping and network oscillations. First, XPG-encoding transcription factors may recruit different gene modules to modify PV IN spatial tuning, timing of inhibition, strength, number and distribution of perisomatic synapses to dictate recruitment of PNs into ensembles and network oscillations. Upregulation of XPGs encoding ion channel subunits such as Grin2a may increase PV IN excitability⁶⁰, whereas the GluA2 subunit Gria2 may calibrate PV IN spatial selectivity in response to increased mossy fibre inputs¹⁵. XPGs encoding axon guidance cues may sculpt the distribution of PV IN–PN synapses across the CA2/CA3 region⁶¹. Second, different XPGs may be upregulated by experience in distinct subsets of PV INs (for example, vertical and horizontal basket cells or axo-axonic cells)³⁶ and through differences in output connectivity, regulate spiking of distinct subsets of PNs. Third, signalling mechanisms of XPG induction may differ. For instance, Tbr1 and Meis2 are expressed at low levels in distinct hippocampal PV subtypes³⁶, and we showed here that MEIS2 and TBR1 differentially regulate experience-dependent PV IN plasticity. Future studies will dissect how the Meis2 targets identified here contribute to experience-dependent PV IN plasticity. We speculate that such efforts may illuminate how XPG-dependent combinatorial codes that specify experience-specific ‘PV IN cell states’ support distinct computations. Conversely, loss of distinct XPG-associated codes may underlie circuit and network alterations that distinguish ASDs, bipolar disorder and schizophrenia.

In conclusion, we identified experience-dependent PV IN plasticity as a convergent biological mechanism for cognitive impairments and seizures in NDDs and demonstrated that restoring experience-dependent PV IN plasticity in adulthood offers an opportunity for therapeutic intervention in adulthood to alleviate cognitive impairments and reduce seizures in NDDs. As the principal cell–PV IN–principal cell feedforward inhibition motif is found in circuits that support sensory processing, memory, cognitive flexibility, decision-making and attention, targeting experience-dependent PV IN plasticity may affect different domains of cognition in NDDs.

Mice

All mice were group-housed and experiments were conducted in accordance with procedures approved by the Institutional Animal Care and Use Committees at the Massachusetts General Hospital and the Tufts University School of Medicine and NIH guidelines. All mice were housed in a 12-h (7:00 to 19:00) light–dark colony room at 22–24 °C, 30–50% humidity and with ad libitum access to food and water. Cntnap2^−/−, Pvalb^cre, Amigo2^cre and Rpl22^HA mice were obtained from the Jackson Laboratory (strain 017482, 017320, 030215 and 011029, respectively).

Viruses and virus constructs

AAV-S5E2-dTom-nlsdTom plasmid (plasmid 135630), AAV5-CamKIIa-hChR2(H134R)-eYFP (26969), AAV-S5E2-ChR2-mCherry (135634) and AAV PHP.eB virus were purchased from Addgene. AAV-S5E2-XPGs-nlsdTom viruses was generated by subcloning mouse Meis2 (NM_001346036.1), Bcl11a (NM_016707.3), Tbr1 (NM_009322.3) or dSACas9/VP64 into AAV-S5E2-dTom-nlsdTom plasmid. AAV-U6-sagRNA#1#2-Syn1-P2A-mCherry and AAV-Herc1gRNA#1-CAGGAAAAAGCCTGGTCTTCA-#2-AAAACAAATTCATGTGTATGT-Syn1-P2A-mCherry were generated by VectorBuilder. AAV-EF1a-DIO-mCherry virus was purchased from UNC. AAV-EF1a-DIO-Meis2-mCherry virus was generated by VectorBuilder. The Cal-Light plasmids pAAV-pCMV-Myc-TM-KA2-CaM-NES-TEV-N-AsLOV2-TEVseq-tTA, pAAV-hSYN-M13-TEV-C-P2A-tdTomato and pAAV-TRE-eGFP were gifts from the Hyun Laboratory. AAV.PHP.eB viruses were produced by Boston Children’s Hospital Viral Core. Note, because of differences in constructs, S5E2-dTom-P2A-nlsdTom (control) results in dTomato distributed throughout the neuron, whereas S5E2-Meis2-P2A-nlsdTom is localized to the nucleus.

Immunohistochemistry

Mice were injected with viruses 2 weeks before perfusion. Mice were anaesthetized with ketamine and xylazine (10 mg ml^–1 and 1.6 mg ml^–1, respectively, intraperitoneally) and transcardially perfused with 4% paraformaldehyde (PFA), and their brains were removed and incubated in 4% PFA at 4 °C overnight. Brains were placed in 30% sucrose–PBS for 2 days and then embedded in medium (OCT, Fisher HealthCare). Cryosections (35 μm) were obtained (Leica) and stored in PBS (0.01% sodium azide) at 4 °C. For immunostaining, floating sections were permeabilized, blocked in blocking solution for 2 h (PBS containing 0.3 % Triton X-100 and 10% normal donkey serum) and then incubated with primary antibodies (PBS containing 10% normal donkey serum and 0.1% Triton X-100) at 4 °C overnight. Sections were then washed with PBS 3 times, 10 min each, then incubated with secondary antibodies in PBS for 2 h at room temperature. Sections were then washed with PBS 3 times, 10 min each, mounted on glass slides and coverslipped with DAPI Fluoromount-G (SouthernBiotech).

Image analysis

For PV puncta and SYT2 puncta, images were obtained from three sections per mouse hippocampus blind to treatment and genotype. A Leica SP8 confocal laser microscope and LAS software were used to capture images in the stratum lucidum at high-resolution (2,048 × 2,048 pixels). Single confocal-plane images were captured in the CA2 and CA3ab subfields using a ×63 oil objective plus ×43 digital zoom. For quantification of sample sizes of PV⁺ puncta and SYT2⁺ puncta, densities were averaged from 18 images per mouse for CA3 and 12 images per mouse for CA2. Puncta were analysed using the StarDist 2D plugin and particle analysis tools in Fiji ImageJ. Threshold values were held constant across images. For Cal-Light, dCA3a cells were manually counted. To reduce the contribution of background fluorescence, a threshold based on maximizing Yen entropy⁶² was applied to GFP images during quantification.

Antibodies

The following antibodies were used in the study: PV (rabbit, Swant PV25, 1:5,000, RRID: AB_10000344; goat, Swant PVG213, RRID: AB_2721207, 1:1,000); RGS14 (mouse, NeuroMab 75-170, RRID: AB_2179931, 1:500; rabbit, Proteintech 16258-1-AP, RRID: AB_2179918, 1:500); SYT2 (mouse, Abcam AB154035-1001, RRID: AB_2916272, 1:250); RFP (rabbit, Rockland 600-401-370, RRID: AB_2209751, 1:1,000; goat, Sicgen AB1140-100, RRID: AB_2877097, 1:500); FOS (guinea pig, Synaptic Systems 226-004, RRID: AB_2619946, 1:3,000); GFP (chicken, Invitrogen A10262, RRID: AB_2534023, 1:500); gephyrin (rabbit, Synaptic Systems 147-008, RRID: AB_2619834, 1:500); MEIS2 (rabbit, Protein Tech 11550-1-AP, RRID: AB_2143028, 1:400); BCL11A (mouse, Abcam ab19487, RRID: AB_444947, 1:500); TBR1 (rabbit, Abcam ab31940, RRID: AB_2200219, 1:200); GFAP (Chicken, Millipore AB5541, RRID:AB_177521, 1:2,000); IBA1 (rabbit, FujiFilm 019-19741, RRID: AB_839504, 1:500); and SST (mouse, Santa Cruz G-10, RRID: AB_831726, 1:500). The following fluorescent-label-coupled secondary antibodies from Jackson Immuno Research were used at 1:500 dilution: Alexa-Fluor-488-conjugated donkey anti-rabbit IgG (711-545-152); Cy3-conjugated donkey anti-rabbit-IgG (711-165-152); Alexa-Fluor-488-conjugated donkey anti-mouse-IgG (715-545-151); Cy3-conjugated donkey anti-mouse-IgG (715-165-151); Alexa-Fluor-488-conjugated donkey anti-chicken-IgG (703-545-155); Cy3-conjugated donkey anti-goat-IgG (705-165-147); 647-conjugated donkey anti-guinea pig-IgG (706-605-1481); and Cy3-conjugated donkey anti-guinea-pig-IgG (706-165-148).

RNA in situ hybridization and immunofluorescence imaging

RNAscope in situ hybridization was performed following the RNAscope Multiplex Fluorescent Reagent kit v.2 protocol (323100, Advanced Cell Diagnostics). In brief, mice were perfused with PBS followed by 4% PFA. Brain slices (10 µm) were mounted and dried at –20 °C for 2 h. The slides were washed in PBS, then dehydrated in a gradient of ethanol (50%, 70%, 100% and 100%) for 5 min each and air-dried for 5 min. RNAscope hydrogen peroxide was applied to tissues and incubated at room temperature for 10 min, followed by treatment with RNAscope 1× Target Retrieval Reagent. Tissues were incubated for 10 min at 95 °C followed by washes in ddH₂O and 100% ethanol. Tissues were dried at room temperature for 3 min. RNAscope Protease III was applied and incubated for 30 min at 40 °C, then washed with water. The target probes were prepared (MmRgs14, 416651; Mm-Pvalb-C2, 421931-C3; and Mm-Meis2-C3, 436371-C3). Target probes were applied and incubated at 40 °C for 2 h, after which tissues were washed with 1× wash buffer and placed in 5× SSC buffer overnight. Tissues were washed with wash buffer twice followed by incubation with AMP1, AMP2 and AMP3 (30 min for AMP1 and AMP2, 15 min for AMP3, at 40 °C). Tissues were washed with wash buffer twice in between each amplification step, then incubated with HRP specific to each channel (15 min, 40 °C). Tissues were washed twice, after which the TSA fluorophore specific to each channel or probe (Opal 520, 570 and 690, Akoya Biosciences) was added and incubated for 30 min at 40 °C. Tissues were washed twice and incubated with RNAscope HRP Blocker for 15 min at 40 °C. Fluorescent images were captured using a SP8 Leica confocal microscope. All Pvalb⁺ cells were outlined and Meis2 intensity was quantified.

RNA sequencing

Tissue collection and RNA isolation

Hippocampal CA2/CA3 regions were collected following injections of lentivirus expressing shAblim3 or shNT into the DG of 2-month-old PV^cre;Rpl22HA^f/f mice for 2 weeks. Brain tissues were snap-frozen and pooled from 6 mice (male and female) for every sample. Samples were immunoprecipitated with anti-HA magnetic beads (Pierce, 88836) for 3 h at 4 °C^30,63. After elution using a RNeasy Plus Micro kit (Qiagen), purified ribosome-associated RNAs were stored at –80 °C. RNA quality was assessed using a TapeStation (Agilent), and RNA amounts were quantified using a Qubit 4.0 Fluorometer (Life Technologies). Only RNA samples with a RIN value above 8.0 were used for library preparation and sequencing.

Library preparation, Illumina sequencing and bioinformatics

NGS libraries were constructed from total RNA using a Clontech SMARTer v.4 kit (Takara), followed by sequencing on an Illumina HiSeq 2500 instrument, which resulted in 20–30 million 50 bp reads per sample. STAR aligner⁶⁴ was used to map sequencing reads to the transcriptome in the mouse mm9 (GRCm37) reference genome. Read counts for individual genes were produced using the unstranded count function in HTSeq (v.0.6.0)⁶⁵, followed by the estimation of expression values and detection of differentially expressed transcripts using EdgeR⁶⁶. Only the genes with count per million reads of 1 for one or more samples⁶⁷ were included for subsequent analyses. DEGs were defined by at least 1.5-fold change with FDR < 0.05.

RT–qPCR analysis

Hippocampal DG and CA2/CA3 regions were collected and snap-frozen⁶⁸. In brief, cDNA samples were reverse transcribed from RNA collected from PV^cre;Rpl22^HA/HA mice injected with either shNT or shAblim3 into DG. Total RNA was quantified using a NanoDrop spectrophotometer (Thermo Scientific), and then equal amounts of RNA were used for reverse transcription (SuperScript IV First-strand synthesis system, Invitrogen). RT–qPCR was carried out using SYBR green (Bio-Rad) and primers (Primer bank)⁶⁹ with the following sequences: Ablim3-F 5′-GGTCCGTGTCCACAACAAC-3′; Ablim3-R 5′-GTCCCGGCAGCTATCACAG-3′; Bok-F 5′-CCACAGACAAGGAGCTGGT-3′; Bok-R 5′-TAGCCAAGGTCTTGCGTACA-3′; Dsp-F 5′-CGGACATTCATGCGAGATAC-3′; Dsp-R 5′-GCCTTGAACTGGGAACACTC-3′; Bcl11a-F 5′-TGGTATCCCTTCAGGACTAGGT-3′; Bcl11a-R 5′-TCCAAGTGATGTCTCGGTGGT-3′; Meis2-F 5′-CAGGGTGGTCCAATGGGAATG-3′; Meis2-R 5′-GGGGGTCCATGTCTTAACTGAG-3′; Tbr1-F 5′-CAAGGGAGCATCAAACAACA-3′; Tbr1-R 5′-GTCCTCTGTGCCATCCTCAT-3′; Herc1-F 5′-GAAGATGTGGATGCAGCAGA-3′; Herc1-R 5′-GGTCTGTCCGGTGAAGGATA-3′; Gapdh-F 5′-GCTTGTCATCAACGGGAAG-3′; Gapdh-R 5′-TTGTCATATTTCTCGTGGTTCA-3′. Primers for the following MEIS2 downstream candidates were used: Cdh2-F 5′-AGCGCAGTCTTACCGAAGG-3′; Cdh2-R 5′-TCGCTGCTTTCATACTGAACTTT-3′; Chd7-F 5′-GGAGAACCCTGAGTTTGCTAGCGCAGTCTTACCGAAGG-3′; Chd7-R 5′-CCCTGAAGTAGAGGCGACAG-3′; Kirrel3-F 5′-TCGGAGAATGAATGAAGGTCAAG-3′; Kirrel3-R 5′-GATGGCACACAGCAGAGTCA-3′; Reln-F 5′-CTGTGTCATACGCCAAGAACA-3′; Reln-R 5′-GGGGAGGTACAGGATGTGGAT-3′; Syt2-F 5′- AGAACCTGGGCAAATTGCAGT-3′; Syt2-R 5′-CCTAACTCCTGGTATGGCACC-3′; Tanc1-F 5′-CGAGCGCCACACACATAAC-3′; Tanc1-R 5′-AGGCAGTGCTGCTTGTGAG-3′; Kcnab1-F 5′-TTCCGCACTGTCGCTATCATC-3′; Kcnab1-R 5′-AGCATGAAACTCTGAGTCCTGA-3′; Pvalb-F 5′-CATTGAGGAGGATGAGCTG-3′; and Pvalb-R 5′-AGTGGAGAATTCTTCAACCC-3′.

Gene set enrichment analysis

Gene set enrichment analysis was performed using the package from a joint project of UC San Diego and the Broad Institute (https://www.gsea-msigdb.org/gsea/index.jsp), using FDR < 0.05 and default settings. Mouse MSigDB (v.2024.1.Mm) was used for pathway (Gene Ontology terms) analysis^70,71.

Stereotactic viral injection

Mice were administered carprofen (5 mg kg^–1, subcutaneously) before surgery and were then anaesthetized with ketamine and xylazine (10 mg ml^–1 and 1.6 mg ml^–1, intraperitoneally) or with 1.75% isoflurane (Baxter Healthcare Med Delivery Systems). Mice were placed in a stereotaxic frame, and a small hole was drilled at each injection site (Foredom K.1070 High Speed Rotary Micromotor kit). Bilateral injections were performed using Hamilton microsyringes (Hamilton, Neuros Syringe 7001) or Nanoject digital microinjectors that were slowly lowered into target sites and remained in place for 8 min before viral infusion at a rate of 50 nl min^–1. The following coordinates relative to bregma were used: dorsal DG: –1.8 mm (anterior–posterior (AP)), ±1.35 mm (medial–lateral (ML)) and –2.25 mm (dorsal–ventral (DV)); dorsal CA2/CA3: –1.8 mm (AP), ±2.45 mm (ML) and –2.35 mm (DV). Recombinant AAVs (titre of 1 × 10¹³) were injected for a total volume of 100 nl per injection site. After 10 min of infusion, the microsyringes were slowly withdrawn and the skin above the incision was sutured with coated vicryl sutures. For Cal-Light tagging experiments⁴⁵, viruses were mixed (1:1:1 ratio) and injected in a total volume of 300 nl. One week later, animals were injected with AAV-S5E2 control virus or AAV-S5E2-Meis2 virus and implanted with an optical fibre probe above CA2/3 and given 2 weeks to recover before behavioural testing. The following coordinates relative to bregma were used: –1.80 (AP), ±2.28 (ML) and –2.20 (DV) for virus injection; –1.80 (AP), 2.28 (ML) and –1.80 (DV) for optical fibre probe implantation. After surgery, mice were placed in a clean empty cage on top of a heating pad with ambient temperature to 36 °C until full recovery from anaesthesia. Mice were monitored and received a daily injection of carprofen (5 mg kg^–1, intraperitoneally) for 3 days after surgery⁹.

Electrode implantation

Mice were allowed to recover and 2 weeks later underwent a second surgery of tetrode implantation. Mice were placed in a stereotaxic frame after anaesthesia (1.75% isoflurane) to implant three bone screws and a 16-channel, custom-built, twisted wire electrode bundle. A four-wire electrode bundle was created by twisting together four 75-µm diameter nichrome wires (California Fine Wire). The bundle was cut at an angle that spanned 0.5 mm. Four different sites were chosen for twisted wire implantation: the tips of the twisted wire were placed in the dorsal hippocampus at the following locations: DG: –1.8 mm (AP), +1.35 mm (ML) and –2.25 mm (DV); CA1, –1.8 mm (AP), +1.35 mm (ML) and –2.18 mm (DV); CA3, –1.8 mm (AP), +2.45 mm (ML) and –2.35 mm (DV); and CA2, –1.8 mm (AP), +2.45 mm (ML) and –2.31 mm (DV). The twisted electrodes were each attached to a pin in a Mill-Max connector. After confirming that the recording electrode array extended through CA1, the wires and connectors were secured in place with dental cement. The Omnetics connector of the recording electrode and the Mill-Max connector of the electrode bundle were anchored to the skull along with bone screws using dental cement mixture (C&B Metabond, Parkel, and TEETs Denture Material, Cooralite Dental Mfg).

Recording setup

The implanted mice were attached to the headstage (RHD 32 ch, C3314, Intan Technologies) through a custom-made Omnetics to Mill-Max adaptor (Mill-Max model 852-10-100-10-001000). Behaviour was recorded using a monochromatic camera (30 frames per s (fps), Flea3 USB3, FLIR). Electrophysiological signals were sampled at 30 kHz and recorded using an Open Ephys Acquisition system via an ultrathin SPI interface cable connected to the headstage board. Electrophysiological recordings were synchronized with recorded video using a TTL trigger pulse and by recording camera frame strobes. The implanted mice (10 weeks old) were allowed to explore their home cage for 2 h (pre-social), followed by 5 min of social interaction in the home cage with a novel, sex-matched juvenile mouse. After the social encounter, another 2 h of home-cage exploration and recording were conducted. Mouse movement was analysed on the basis of markerless tracking of the snout using DeepLabCut⁷². A snout speed of less than 1.5 cm s^–1 in the home cage was considered an immobility period.

Extracting NREM periods

LFP data processing and ripple analysis were conducted using a custom written Python script. Initially, the LFP data from the selected CA1 channel was scaled to microvolts and downsampled to 1,000 Hz. The NREM sleep periods were identified in the downsampled CA1 LFP by first computing a spectrogram (10-s window, 1-s step, 0–300 Hz). Features, including the first principal component (PC1, derived from the z-transformed 0–300 Hz spectral power) and theta dominance (5–10 Hz power/2–16 Hz power) were then extracted. NREM states were defined as periods in which PC1 exceeded its 75th percentile for at least 5 s, accompanied by a low theta dominance ratio⁷³.

Ripple analysis

Ripple events were subsequently detected in these NREM epochs from the downsampled CA1 LFP. The LFP was bandpass-filtered (100–250 Hz, third-order Butterworth) and z-scored. To calculate the power envelope of this ripple-band signal, the z-scored filtered signal was rectified and then filtering again, in the 1–20 Hz band, using third-order Butterworth^74,75. Ripple events were marked when this power envelope exceeded 2 Z, reached a peak greater than 5 Z, had a duration between 20 and 150 ms and were merged if the inter-event interval was less than 20 ms^76,77. An artefact rejection was performed when the z-scored envelope of the DG channel concurrently exceeded the high threshold, thereby coinciding with the detected ripple event in the CA1 channel. For each valid ripple event, the following properties were extracted: the event start time (the point at which the power envelope went above 2 Z); the end time; the peak power time (the point at which the power envelope fell back below 2 Z); and the peak LFP time (the point between the start and end times at which the highest amplitude ripple-band cycle occurs). These provided precise timestamps for the boundaries of the event and its peak activity points. The duration of each ripple was calculated in milliseconds, which represented its total temporal extent. The peak power quantified the maximum deviation of the power envelope of the ripple from the baseline, normalized by its median and median absolute deviation. The peak LFP amplitude was determined as the maximum absolute value the detected boundaries of each ripple. For visualization of each ripple, time–frequency power was calculated by applying a continuous complex Morlet wavelet transform to the raw LFP signal (0–300 Hz) surrounding the event using PyWavelets⁷⁸. Power was estimated for frequencies from 100 to 250 Hz, and the resulting spectrograms were generated on a decibel (dB) scale.

Behavioural procedures

Two weeks after viral injections, mice were handled for 3 days before behavioural experiments to habituate them to human handling and transportation from the vivarium to the behavioural testing rooms. The behavioural assays were performed in the following order: open-field (OF, day 1), visual cue habituation (day 2), novel object location (NOL) followed by novel object recognition (NOR, day 3), and social recognition and discrimination (day 4). All the behavioural assays were performed in the same chambers (40 × 40 cm, MazeEngineers). Videos were recorded and exported from Freezeframe (Actimetrics) and analysed using EthoVision XT 15 (Noldus). Centre point tracking was used to record movement, and nose point tracking was used to evaluate object and social interaction. An interaction was registered when the test mouse’s nose position to object or stimulus mouse was within 1 cm⁹.

Behavioural paradigm for PV IN synapse, SYT2 and puncta analysis

Mice were handled for 7 days before behavioural experiments to habituate them to human handling, transportation from the vivarium to the behavioural testing rooms and the context for 1 h. On day 8, mice were either exposed to context only or with one stimulus mouse in one pencil cup for 10 min. Mice were returned to their home cage and subsequently perfused 90 min later.

Behavioural paradigm for Meis2 and MEIS2 intensity analysis

Mice were handled for 3 days before behavioural experiments to habituate them to human handling, transportation from the vivarium to the behavioural testing rooms and the context for 1 h. On day 4, mice were either exposed to context only or with one stimulus mouse in one pencil cup for 10 min. Mice were returned to their home cage and subsequently perfused 10 min later.

OF paradigm

Mice were transported into a holding room and habituated for 1 h before testing. The total distance travelled and the time spent in the centre of the arena were quantified over 30 min⁹.

NOL and NOR

Two identical objects (2 × 4 × 6 cm) were placed in the OF chamber along one side (5 cm distance from the wall). Mice were placed in the opposite side of the object and allowed to explore freely for 5 min and then returned to their home cage for 2 h. Then one object was moved to the opposite side, and mice were placed in the middle of objects and allowed to explore for 5 min. Mice were returned to home cage for 10 min. Next, one object was replaced with a novel object (4 × 4 × 6 cm, 5 cm distance from the wall), and mice were placed in the middle of the objects and allowed to explore freely for 5 min. The time spent exploring the objects (noise point within 2 cm) was quantified⁹.

Social recognition and social discrimination

Stimulus mice (matched for strain, age and sex) were habituated to being placed in a pencil wire cup in the OF chamber before the task day for 3 days (15 min per day). The task consisted of three trials with 5 min of inter-trial intervals: habituation (empty cup versus empty cup); recognition (empty cup versus stimulus); and discrimination (novel versus familiar). Mice were placed in the centre of the chamber for 10 min. The locations of the familiar and novel stimuli were counterbalanced across trials. The time spent exploring the stimulus mouse was quantified (nose point within 1 cm)⁹.

Tagging socially active neurons with soma-targeted Cal-Light

Behavioural paradigm and tagging protocol

Before testing, animals were placed in holding cages for a minimum of 1 h, using the same cage each day. Subject mice underwent a 3-day habituation protocol, during which they explored an OF chamber with two empty cups placed in opposite corners while attached to lightless fibre optic cables on the final 2 days. One day after habituation, subject mice were exposed to a novel stimulus mouse (stimulus 1) placed under one of the cups and allowed to explore the chamber for 10 min. When subjects approached within 5 cm of the stimulus mouse, blue light was delivered through fibre optic cables to mediate Cal-Light tagging of the engram. Immediately after the session, subjects were returned to their holding cages. Four hours after initial exposure, subjects were reintroduced to the chamber for a second 10-min session, with a familiar mouse under the same cup as before. Following this, subjects were returned to holding cages and subsequently perfused 90 min later. For social versus contextual comparison of Cal-Light activity, no mouse was placed under the cup in the first second 10-min session. For light versus no light comparisons, mice underwent one 10-min social session with a novel mouse, with or without blue light delivery though fibre optic cables, and were perfused 5.5 h later.

Optical fibres and laser delivery

Optical fibres were constructed using a 200 μm core, 0.37 numerical aperture multimode fibre from Thorlabs. Fibres were inserted through a 230 μm core zirconia ferrule 701 (Precision Fiber Products) and glued in place before thoroughly polishing for a smooth connection between the fibre and optical cable from the laser. For Cal-Light activation, a 100 mW 475 nm blue laser diode was used (OEM Laser Systems). When the subject animal was within a predefined zone, the light was delivered at a frequency of 1 kHz and an intensity of 5–7 mW using an external arbitrary waveform generator (Agilent) attached to the signal port of the laser.

Ex vivo electrophysiology

Mice were 2–3 months old before viral transduction (see above for stereotactic coordinates). At 2–3 weeks after viral infusion, mice were anaesthetized with ketamine and xylazine (10 mg ml^–1 and 1.6 mg ml^–1, respectively, intraperitoneally) then transcardially perfused with ice-cold (4 °C) choline chloride-based artificial cerebrospinal fluid (ACSF) composed of (in mM) the following factors: 92 choline chloride, 2.5 KCl, 1.25 NaH₂PO₄, 30 NaHCO₃, 20 HEPES, 25 glucose and 10 MgSO₄·7H₂O. Mouse brains were rapidly extracted following decapitation. Coronal slices (300-µm thick) containing the dorsal hippocampus were cut in ice-cold (4 °C) choline chloride ACSF using a Leica VT1000 vibratome (Leica Biosystems) and transferred to warm (33 °C) normal ACSF for 30 min. Normal ACSF contained (in mM) the following factors: 124 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 24 NaHCO₃, 5 HEPES, 12.5 glucose, 2 MgSO₄·7H₂O and 2 CaCl₂·2H₂O. All ACSF solutions were adjusted to pH 7.4 and mOsm of 305, and were continuously saturated with carbogen (95% O₂ and 5% CO₂). Slices were allowed to cool to room temperature (20–22 °C) for 1 h before recordings.

Whole-cell patch-clamp recordings were amplified, low-pass filtered at 1.8 kHz with a 4-pole Bessel filter and digitized (Muliclamp 700B, Digidata 1550B, Molecular Devices). Slices were placed in a polytetrafluoroethylene submersion chamber and continually perfused with normal ACSF (>2 ml min^–1). Neurons were visually identified by infrared differential interference contrast imaging combined with epifluorescence using LED illumination (pE-300 white, CoolLED). PNs in CA2 and CA3ab were distinguished by their anatomical location and distinct electrophysiological properties. Borosilicate patch pipettes had a resistance of 4–5 MΩ and filled with an internal solution containing (in mM) the following factors: 120 CsMeS, 4 MgCl₂, 1 EGTA, 10 HEPES, 5 QX-314, 0.4 Na₃GTP, 4 MgATP, 10 phosphocreatine and 2.6 biocytin, at pH 7.3, 290 mOsm. For current-clamp recordings, patch pipettes were filled with 130 mM potassium gluconate in place of CsMeS and QX-314 was excluded. Once the GΩ seal was obtained, neurons were held in voltage-clamp configuration at –70 mV and the input resistance, resting membrane potential and capacitance were measured. Series resistance (<30 MΩ) was monitored throughout recordings, and recordings were discarded if the series resistance changed by >20% from baseline.

Optically evoked EPSCs and IPSCs were evoked with 1 ms 473 light pulses delivered above the mossy fibre pathway–the hilus of the DG. Current responses were recorded at 1.5× threshold, defined as the minimum stimulation intensity required to produce a consistent current response beyond baseline noise. Isolation of EPSCs was done by voltage clamp at –70 mV and IPSCs at 0 mV. Optical 10-pulse stimulation trains were evoked 5 times at an interval of 20 s between trains. The inter-event interval between pulse stimulation was 100 ms.

Electrically evoked EPSCs and IPSCs were evoked with 0.2 ms stimulation from a bipolar tungsten electrode in a patch pipette with a resistance of 1–2 MΩ and filled with ACSF. The stimulation pipette was placed on the mossy fibre pathway–the hilus of the DG or the stratum oriens between CA2 and CA3a to elicit Schaffer collateral stimulation. Pharmacological validations were conducted to ensure selective mossy fibre stimulation or selective Schaffer collateral stimulation using DCG-IV (1 µM) or tetrodotoxin (TTX) and 4AP (1 µm and 100 µm, respectively).

LTP of the EPSCs and long-term depression of the IPSCs (iLTD) was induced by electrically evoked TBS, which consisted of bursts of 4 pulses at 100 Hz, with and inter-burst interval of 200 ms, and repeated 4 times at 10 s intervals. A bipolar tungsten electrode in a patch pipette was placed on the mossy fibre pathway. Stimulation was set to 1.5× threshold of the resulting EPSC. The recording consisted of a 10 min baseline of the optically evoked EPSC or IPSC, TBS, then 40 min of the optically evoked postsynaptic current. Plasticity was calculated on the basis of the per cent change from the last 5 min of the baseline current amplitude to the last 5 min of the 40 min recording.

A current-clamp configuration was used to record intrinsic membrane properties. To assess action potential spiking activity, the neuron was clamped at –70 mV and 10 pA ascending step currents were delivered at 500 ms durations. Bursting neurons were identified by their asynchronous rapid firing within the first 50 ms of the current step followed by failure to fire and were excluded from analyses. In some experiments, intrinsic membrane properties were recorded independent of excitatory synapses by bath application of cyanquixaline (CNQX, 10 µM), an AMPA and kainite receptor competitive antagonist, and d-2-amino-5-phosphonopentanoic acid (AP5, 50 µM), a selective NMDA receptor antagonist.

sEPSCs and sIPSCs were recorded by voltage clamp at –70 mV for sEPSCs and 0 mV for sIPSCs. Pharmacological validations were conducted to ensure no sIPSCs were visible during sEPSC recordings and vice versa (data not shown). No inward events were detected after bath application of CNQX (10 µM) while voltage clamped at –70 mV. Likewise, no events were detected after bath application of gabazine (10 µM), a GABA_A receptor antagonist, while voltage clamped at 0 mV.

mEPSCs and mIPSCs were recorded after bath application of TTX (1 µM). Isolation of mEPSCs was done by voltage clamp at –70 mV and mIPSCs at 0 mV. Autodetection parameters for the inclusion of spontaneous and miniature events were determined by calculating the minimum threshold: root mean square 2 × 1.5. Data acquisition was performed using Clampex and analysed with Clampfit (Molecular Devices v.11) and EasyElectrophysiology (v.2.8.0) software.

ECoG surgical implant

All animal surgeries and subsequent experiments were conducted in accordance with Tufts University’s Institutional Animal Care and Use Committee guidelines and animal use protocols. In brief, 6–7-month-old Cntnap2^−/− and Cntnap2^+/+ mice that had previously undergone CA2/CA3 AAV injection were anaesthetized with isoflurane (3% induction, 1.5% for maintenance, 2 l min^–1 O₂ flow rate) and given systemic analgesic (buprenorphine, 0.1 mg kg^–1, subcutaneously) as well as local analgesia near the incision site (bupivacaine, 4 mg kg^–1, subcutaneously). Mice were then placed in a stereotactic frame, and the dorsal side of the skull was shaved and disinfected with iodine and ethanol. A scalp incision was made, and the skull surface cleaned and dried. Four burr holes were drilled into the skull without puncturing the dura mater. The drill bit tip diameter was 0.7 mm (item no. 19007-07, Fine Science Tools). Stereotaxic coordinates of the four drill holes are as follows: anterior burr holes were −0.6 mm (bregma, AP axis) and 2 mm left (ECoG) and right (ground) of the midline (ML); posterior burr holes were – 2.6 mm (bregma, AP) and 2.5 mm left (reference) and right (ECoG) to ML. Four 0.25 cm stainless steel screw electrodes with attached silver wires (part no. 8403, Pinnacle Technologies) were gently screwed into the burr holes and fixed with dental cement and superglue to increase implant stability for chronic ECoG recordings. The silver wires of the electrode screws were then soldered to the headmount (part no. 8402, Pinnacle Technologies). The headmount–electrode construct was then secured to the skull with dental cement. After surgery, animals were allowed to recover for at least 7 days before chronic ECoG recordings started and were given analgesia (buprenorphine, 0.1 mg kg^–1, subcutaneously) as needed for the first 3 days after surgery.

Chronic ECoG recording

After recovering from surgery, a preamplifier (100× gain, 1 Hz high-pass filter, part no. 8213, Pinnacle Technologies) was plugged into the implanted headmount and preamplifiers were attached to a commutator in the ECoG recording system (Pinnacle Technologies). Mice were housed in a round, acrylic ECoG recording chamber with access to food and water and kept on a standard light–dark cycle. ECoG data were recorded using LabChart Pro software (AD instruments) with a 1 kHz sampling rate. ECoG data were collected continuously 24 h a day for at least 14 days. After recording, ECoG signals were filtered (100 Hz low-pass filter) and recordings were manually reviewed manual by an experienced, blinded investigator. When seizures were identified, their duration was quantified, and their time of incidence was noted.

SLEAP

Mice were pair-housed for over 2 weeks before recording. On the recording day, the cage lid was removed, and the cage was placed inside a recording chamber overnight with access to food and water. Behaviour was continuously recorded for 8 h under infrared lighting, with a 1 h segment, captured 1 h after the cage was placed in the chamber, selected for analysis. We used SLEAP⁴⁴ to track and estimate the poses of freely behaving mice in their home cages. To analyse social interactions between mice, a multi-instance, bottom-up model with a U-Net architecture was trained on 2,751 instances across 917 frames, with 306 validation instances from 102 frames. Six body parts were labelled as nodes: nose, right ear, left ear, head, body centre and tail base. Our average distance between the predicted and ground truth nodes was 4.3 pixels. Training frames were randomly selected from videos containing different mice, with slight variations in contrast between subjects and background. Recordings were captured continuously during a 1-h session, and the frame rate was 23.97 fps. A batch size of 6 was used for training, with data augmentation through rotation (360°).

For inference, identity tracking across frames was performed using a flow tracker, with centroid similarity and the Hungarian matching algorithm applied over an elapsed frame window of 4 frames. Instances were manually inspected for identity switches and corrected before data export. Data were analysed using a Python script modified from SLEAP analysis examples (https://sleap.ai/notebooks/Analysis_examples.html). These were used to analyse social interactions based on the distance between nodes. First, missing nodes were linearly interpolated, and predictions were smoothed using a Savitzky–Golay filter to reduce jitter. We then measured nose-to-nose, nose-to-centre and nose-to-rear distances. Interactions were defined using the following pixel thresholds: 61 pixels (nose-to-nose), 100 pixels (nose-to-centre) and 70 pixels (nose-to-rear). Thresholds were verified against manually scored data to ensure accuracy.

Sex as a biological variable

Male and female mice were included in most of the experiments in this study. Attempts were made to balance sex in each experimental group. However, we acknowledge that the ratio of males and females are not consistent across every group, and that group sizes may be underpowered for statistical comparisons of sex.

Statistics, rigour and reproducibility

All experimenters were blind to treatment conditions throughout data collection, scoring and analyses. Animals were assigned to experimental groups based on genotype. Representative images and traces were selected from three or more independent variables in each experiment (mice). Statistical analyses were conducted using Prism (v.10; GraphPad) and the minimum sample size was determined on the basis of previous experience of experimental paradigms, existing literature and power analyses. Significance was defined as P < 0.05 and exact P values are provided whenever possible. Appropriate nonparametric tests were used when datasets did not meet parametric assumptions. Grubbs’ or ROUT tests were used to identify outliers with α or Q = 0.05 (Fig. 4k). Detailed statistical analyses are provided in Supplementary Table 3.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

1. Yuste, R., Cossart, R. & Yaksi, E. Neuronal ensembles: building blocks of neural circuits. Neuron 112, 875–892 (2024).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

2. Lisman, J. et al. Viewpoints: how the hippocampus contributes to memory, navigation and cognition. Nat. Neurosci. 20, 1434–1447 (2017).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

3. Oliva, A., Fernandez-Ruiz, A., Leroy, F. & Siegelbaum, S. A. Hippocampal CA2 sharp-wave ripples reactivate and promote social memory. Nature 587, 264–269 (2020).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

4. Szabadics, J. & Soltesz, I. Functional specificity of mossy fiber innervation of GABAergic cells in the hippocampus. J. Neurosci. 29, 4239–4251 (2009).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Pouille, F. & Scanziani, M. Enforcement of temporal fidelity in pyramidal cells by somatic feed-forward inhibition. Science 293, 1159–1163 (2001).

   Article  CAS  PubMed  Google Scholar 

6. Ruediger, S. et al. Learning-related feedforward inhibitory connectivity growth required for memory precision. Nature 473, 514–518 (2011).

   Article  ADS  CAS  PubMed  Google Scholar 

7. Neubrandt, M., Olah, V. J., Brunner, J. & Szabadics, J. Feedforward inhibition is randomly wired from individual granule cells onto CA3 pyramidal cells. Hippocampus 27, 1034–1039 (2017).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

8. Pelkey, K. A. et al. Hippocampal GABAergic inhibitory interneurons. Physiol. Rev. 97, 1619–1747 (2017).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

9. Shih, Y. T., Alipio, J. B. & Sahay, A. An inhibitory circuit-based enhancer of DYRK1A function reverses Dyrk1a-associated impairment in social recognition. Neuron 111, 3084–3101 (2023).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

10. Guo, N. et al. Dentate granule cell recruitment of feedforward inhibition governs engram maintenance and remote memory generalization. Nat. Med. 24, 438–449 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

11. Twarkowski, H., Steininger, V., Kim, M. J. & Sahay, A. A dentate gyrus–CA3 inhibitory circuit promotes evolution of hippocampal-cortical ensembles during memory consolidation. eLife https://doi.org/10.7554/eLife.70586 (2022).

12. Okur, Z. et al. Control of neuronal excitation–inhibition balance by BMP–SMAD1 signalling. Nature 629, 402–409 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

13. Yap, E. L. et al. Bidirectional perisomatic inhibitory plasticity of a Fos neuronal network. Nature 590, 115–121 (2021).

    Article  ADS  CAS  PubMed  Google Scholar 

14. Favuzzi, E. et al. Activity-dependent gating of parvalbumin interneuron function by the perineuronal net protein brevican. Neuron 95, 639–655 (2017).

    Article  CAS  PubMed  Google Scholar 

15. Hong, I. et al. Calcium-permeable AMPA receptors govern PV neuron feature selectivity. Nature 635, 398–405 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

16. Reh, R. K. et al. Critical period regulation across multiple timescales. Proc. Natl Acad. Sci. USA 117, 23242–23251 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

17. Exposito-Alonso, D. & Rico, B. Mechanisms underlying circuit dysfunction in neurodevelopmental disorders. Annu. Rev. Genet. 56, 391–422 (2022).

    Article  CAS  PubMed  Google Scholar 

18. Contractor, A., Ethell, I. M. & Portera-Cailliau, C. Cortical interneurons in autism. Nat. Neurosci. 24, 1648–1659 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

19. Su, Z. et al. Dlx1/2-dependent expression of Meis2 promotes neuronal fate determination in the mammalian striatum. Development 149, dev200035 https://doi.org/10.1242/dev.200035 (2022).

20. Donato, F. et al. The ontogeny of hippocampus-dependent memories. J. Neurosci. 41, 920–926 (2021).

    Article  CAS  PubMed  Google Scholar 

21. Klausberger, T. et al. Brain-state- and cell-type-specific firing of hippocampal interneurons in vivo. Nature 421, 844–848 (2003).

    Article  ADS  CAS  PubMed  Google Scholar 

22. Csicsvari, J., Hirase, H., Mamiya, A. & Buzsaki, G. Ensemble patterns of hippocampal CA3–CA1 neurons during sharp wave-associated population events. Neuron 28, 585–594 (2000).

    Article  CAS  PubMed  Google Scholar 

23. Sohal, V. S. Neurobiology of schizophrenia. Curr. Opin. Neurobiol. 84, 102820 (2024).

    Article  CAS  PubMed  Google Scholar 

24. Dudok, B., Klein, P. M. & Soltesz, I. Toward understanding the diverse roles of perisomatic interneurons in epilepsy. Epilepsy Curr. 22, 54–60 (2022).

    Article  PubMed  Google Scholar 

25. Favuzzi, E. et al. Distinct molecular programs regulate synapse specificity in cortical inhibitory circuits. Science 363, 413–417 (2019).

    Article  ADS  CAS  PubMed  Google Scholar 

26. Allaway, K. C. et al. Genetic and epigenetic coordination of cortical interneuron development. Nature 597, 693–697 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

27. Dehorter, N. et al. Tuning of fast-spiking interneuron properties by an activity-dependent transcriptional switch. Science 349, 1216–1220 (2015).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

28. Calfa, G., Li, W., Rutherford, J. M. & Pozzo-Miller, L. Excitation/inhibition imbalance and impaired synaptic inhibition in hippocampal area CA3 of Mecp2 knockout mice. Hippocampus 25, 159–168 (2015).

    Article  CAS  PubMed  Google Scholar 

29. Bartley, A. F. et al. Interneuron transcriptional dysregulation causes frequency-dependent alterations in the balance of inhibition and excitation in hippocampus. J. Neurosci. 35, 15276–15290 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

30. Sanz, E. et al. RiboTag analysis of actively translated mRNAs in Sertoli and Leydig cells in vivo. PLoS ONE 8, e66179 (2013).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

31. Bjerke, I. E. et al. Densities and numbers of calbindin and parvalbumin positive neurons across the rat and mouse brain. iScience 24, 101906 (2021).

    Article  ADS  CAS  PubMed  Google Scholar 

32. Satterstrom, F. K. et al. Large-scale exome sequencing study implicates both developmental and functional changes in the neurobiology of autism. Cell 180, 568–584 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

33. Singh, T. et al. Rare coding variants in ten genes confer substantial risk for schizophrenia. Nature 604, 509–516 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

34. Palmer, D. S. et al. Exome sequencing in bipolar disorder identifies AKAP11 as a risk gene shared with schizophrenia. Nat. Genet. 54, 541–547 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

35. Vormstein-Schneider, D. et al. Viral manipulation of functionally distinct interneurons in mice, non-human primates and humans. Nat. Neurosci. 23, 1629–1636 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

36. Que, L., Lukacsovich, D., Luo, W. & Foldy, C. Transcriptional and morphological profiling of parvalbumin interneuron subpopulations in the mouse hippocampus. Nat. Commun. 12, 108 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

37. Dvoretskova, E. et al. Spatial enhancer activation influences inhibitory neuron identity during mouse embryonic development. Nat. Neurosci. 27, 862–872 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

38. Desiderio, S. et al. Touch receptor end-organ innervation and function require sensory neuron expression of the transcription factor Meis2. eLife https://doi.org/10.7554/eLife.89287 (2024).

39. Penagarikano, O. et al. Absence of CNTNAP2 leads to epilepsy, neuronal migration abnormalities, and core autism-related deficits. Cell 147, 235–246 (2011).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

40. Jurgensen, S. & Castillo, P. E. Selective dysregulation of hippocampal inhibition in the mouse lacking autism candidate gene CNTNAP2. J. Neurosci. 35, 14681–14687 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

41. Thomas, A. M., Schwartz, M. D., Saxe, M. D. & Kilduff, T. S. Cntnap2 knockout rats and mice exhibit epileptiform activity and abnormal sleep–wake physiology. Sleep https://doi.org/10.1093/sleep/zsw026 (2017).

42. Paterno, R. et al. Hippocampal gamma and sharp-wave ripple oscillations are altered in a Cntnap2 mouse model of autism spectrum disorder. Cell Rep. 37, 109970 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

43. Nasrallah, K. et al. Routing hippocampal information flow through parvalbumin interneuron plasticity in area CA2. Cell Rep. 27, 86–98 (2019).

    Article  CAS  PubMed  Google Scholar 

44. Pereira, T. D. et al. SLEAP: a deep learning system for multi-animal pose tracking. Nat. Methods 19, 486–495 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

45. Hyun, J. H. et al. Tagging active neurons by soma-targeted Cal-Light. Nat. Commun. 13, 7692 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

46. Varga, C. et al. Functional fission of parvalbumin interneuron classes during fast network events. eLife https://doi.org/10.7554/eLife.04006 (2014).

47. Vancura, B., Geiller, T., Grosmark, A., Zhao, V. & Losonczy, A. Inhibitory control of sharp-wave ripple duration during learning in hippocampal recurrent networks. Nat. Neurosci. 26, 788–797 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

48. Huang, Y. C. et al. Dynamic assemblies of parvalbumin interneurons in brain oscillations. Neuron 112, 2600–2613 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

49. Pochinok, I., Stober, T. M., Triesch, J., Chini, M. & Hanganu-Opatz, I. L. A developmental increase of inhibition promotes the emergence of hippocampal ripples. Nat. Commun. 15, 738 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

50. Schlingloff, D., Kali, S., Freund, T. F., Hajos, N. & Gulyas, A. I. Mechanisms of sharp wave initiation and ripple generation. J. Neurosci. 34, 11385–11398 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

51. Girardeau, G., Benchenane, K., Wiener, S. I., Buzsaki, G. & Zugaro, M. B. Selective suppression of hippocampal ripples impairs spatial memory. Nat. Neurosci. 12, 1222–1223 (2009).

    Article  CAS  PubMed  Google Scholar 

52. Joo, H. R. & Frank, L. M. The hippocampal sharp wave-ripple in memory retrieval for immediate use and consolidation. Nat. Rev. Neurosci. 19, 744–757 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

53. Fernandez-Ruiz, A. et al. Long-duration hippocampal sharp wave ripples improve memory. Science 364, 1082–1086 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

54. Norman, Y. et al. Hippocampal sharp-wave ripples linked to visual episodic recollection in humans. Science https://doi.org/10.1126/science.aax1030 (2019).

55. Grotzinger, A. D. et al. Mapping the genetic landscape across 14 psychiatric disorders. Nature 649, 406–415 (2026).

    Article  ADS  CAS  PubMed  Google Scholar 

56. Whitebirch, A. C. et al. Enhanced excitability of the hippocampal CA2 region and its contribution to seizure activity in a mouse model of temporal lobe epilepsy. Neuron 110, 3121–3138 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

57. Monday, H. R. et al. Physiological and molecular impairment of PV circuit homeostasis in mouse models of autism. Preprint at bioRxiv https://doi.org/10.1101/2025.01.08.632056 (2025).

58. Kaneko, K. et al. Developmentally regulated impairment of parvalbumin interneuron synaptic transmission in an experimental model of Dravet syndrome. Cell Rep. 38, 110580 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

59. Jung, K. et al. An adaptive behavioral control motif mediated by cortical axo-axonic inhibition. Nat. Neurosci. 26, 1379–1393 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

60. Camp, C. R. et al. Loss of Grin2a causes a transient delay in the electrophysiological maturation of hippocampal parvalbumin interneurons. Commun. Biol. 6, 952 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

61. Huang, M. et al. Nr4a1 regulates cell-specific transcriptional programs in inhibitory GABAergic interneurons. Neuron 112, 2031–2044 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

62. Yen, J.-C. et al. A new criterion for automatic multilevel thresholding. IEEE Trans. Image Process. 4, 370–378 (1995).

    Article  ADS  CAS  PubMed  Google Scholar 

63. Furlanis, E., Traunmuller, L., Fucile, G. & Scheiffele, P. Landscape of ribosome-engaged transcript isoforms reveals extensive neuronal-cell-class-specific alternative splicing programs. Nat. Neurosci. 22, 1709–1717 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

64. Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013).

    Article  CAS  PubMed  Google Scholar 

65. Anders, S., Pyl, P. T. & Huber, W. HTSeq—a Python framework to work with high-throughput sequencing data. Bioinformatics 31, 166–169 (2015).

    Article  CAS  PubMed  Google Scholar 

66. Robinson, M. D., McCarthy, D. J. & Smyth, G. K. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26, 139–140 (2010).

    Article  CAS  PubMed  Google Scholar 

67. Anders, S. et al. Count-based differential expression analysis of RNA sequencing data using R and Bioconductor. Nat. Protoc. 8, 1765–1786 (2013).

    Article  PubMed  Google Scholar 

68. Hagihara, H. et al. Dissection of hippocampal dentate gyrus from adult mouse. JoVE https://doi.org/10.3791/1543-v (2009).

69. Wang, X., Spandidos, A., Wang, H. & Seed, B. PrimerBank: a PCR primer database for quantitative gene expression analysis, 2012 update. Nucleic Acids Res. 40, D1144–1149 (2012).

    Article  CAS  PubMed  Google Scholar 

70. Mootha, V. K. et al. PGC-1α-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat. Genet. 34, 267–273 (2003).

    Article  CAS  PubMed  Google Scholar 

71. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

72. Mathis, A. et al. DeepLabCut: markerless pose estimation of user-defined body parts with deep learning. Nat. Neurosci. 21, 1281–1289 (2018).

    Article  CAS  PubMed  Google Scholar 

73. Yang, W. et al. Selection of experience for memory by hippocampal sharp wave ripples. Science 383, 1478–1483 (2024).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

74. Tingley, D. & Buzsáki, G. Routing of hippocampal ripples to subcortical structures via the lateral septum. Neuron 105, 138–149 (2020).

    Article  CAS  PubMed  Google Scholar 

75. Zutshi, I. & Buzsáki, G. Hippocampal sharp-wave ripples and their spike assembly content are regulated by the medial entorhinal cortex. Curr. Biol. 33, 3648–3659 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

76. Abbaspoor, S. & Hoffman, K. L. Circuit dynamics of superficial and deep CA1 pyramidal cells and inhibitory cells in freely moving macaques. Cell Rep. https://doi.org/10.1016/j.celrep.2024.114519 (2024).

77. Oliva, A., Fernández-Ruiz, A., Buzsáki, G. & Berényi, A. Role of hippocampal CA2 region in triggering sharp-wave ripples. Neuron 91, 1342–1355 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

78. Lee, G., Gommers, R., Waselewski, F., Wohlfahrt, K. & O’Leary, A. PyWavelets: a Python package for wavelet analysis. J. Open Source Softw. 4, 1237 (2019).

    Article  ADS  Google Scholar 

79. Mohapatra, A. N., Ahmed, O. & Sahay, A. Scripts for Ripples analysis of “Pro-cognitive restoration of experience-dependent parvalbumin inhibitory neuron plasticity in neurodevelopmental disorders”. Zenodo https://doi.org/10.5281/zenodo.18496478 (2026).

Download references

Author notes

1. These authors contributed equally: Yu-Tzu Shih, Jason Bondoc Alipio

Authors and Affiliations

1. Center for Regenerative Medicine, Massachusetts General Hospital, Boston, MA, USA

   Yu-Tzu Shih, Jason Bondoc Alipio, Nathaniel Green, Alok Nath Mohapatra, Travis D. Goode, Muthu Panchanatham, Devesh Pathak & Amar Sahay

2. Harvard Stem Cell Institute, Cambridge, MA, USA

   Yu-Tzu Shih, Jason Bondoc Alipio, Nathaniel Green, Alok Nath Mohapatra, Travis D. Goode, Muthu Panchanatham, Devesh Pathak & Amar Sahay

3. Department of Psychiatry, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA

   Yu-Tzu Shih, Jason Bondoc Alipio, Nathaniel Green, Alok Nath Mohapatra, Travis D. Goode, Muthu Panchanatham, Devesh Pathak & Amar Sahay

4. BROAD Institute of MIT and Harvard, Cambridge, MA, USA

   Yu-Tzu Shih, Jason Bondoc Alipio, Nathaniel Green, Alok Nath Mohapatra, Travis D. Goode, Muthu Panchanatham & Amar Sahay

5. Department of Neuroscience, Tufts University School of Medicine, Boston, MA, USA

   Zin-Juan Klaft & Chris Dulla

6. Department of Molecular Biology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA

   Lai Ping Wong & Ruslan Sadreyev

7. Department of Brain Sciences, Daegu Gyeongbuk Institute of Science and Technology, Daegu, South Korea

   Jung Ho Hyun

8. Department of Psychology, University of Michigan, Ann Arbor, MI, USA

   Omar Ahmed

9. Neuroscience Graduate Program, University of Michigan, Ann Arbor, MI, USA

   Omar Ahmed

10. Department of Biomedical Engineering, University of Michigan, Ann Arbor, MI, USA

    Omar Ahmed

Contributions

A.S. conceived and administered the project. A.S., Y.-T.S. and J.B.A. co-developed the project. Y.S. performed the XPG screen. L.P.W. and R.S. performed bioinformatics. Y.-T.S. performed molecular, cellular and behavioural experiments. J.B.A. performed electrophysiological experiments. Y.-T.S., N.G. and T.D.G. performed Cal-Light experiment. N.G., D.P. and J.B.A. performed SLEAP. N.G. and J.B.A. performed LFP recordings. A.N.M. and O.A. performed SWR analyses. Z.-J.K. and C.D. designed and performed ECoG recordings and seizure analyses. M.P. built the XPG portal. J.H.H. provided Cal-Light viruses. A.S. and C.D. provided funding. A.S., Y.-T.S. and J.B.A. wrote the manuscript with input from co-authors.

Corresponding author

Correspondence to Amar Sahay.

Competing interests

A.S., Y.-T.S. and J.B.A. are named co-inventors on patent filing related to this work. All other authors declare no competing interests.

Peer review information

Nature thanks Nathalie Dehorter, Rebecca Piskorowski and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions