Dlx1 and Dlx2 Promote Interneuron GABA Synthesis, Synaptogenesis, and Dendritogenesis

Abstract

The postnatal functions of the Dlx1&2 transcription factors in cortical interneurons (CINs) are unknown. Here, using conditional Dlx1, Dlx2, and Dlx1&2 knockouts (CKOs), we defined their roles in specific CINs. The CKOs had dendritic, synaptic, and survival defects, affecting even PV⁺ CINs. We provide evidence that DLX2 directly drives Gad1, Gad2, and Vgat expression, and show that mutants had reduced mIPSC amplitude. In addition, the mutants formed fewer GABAergic synapses on excitatory neurons and had reduced mIPSC frequency. Furthermore, Dlx1/2 CKO had hypoplastic dendrites, fewer excitatory synapses, and reduced excitatory input. We provide evidence that some of these phenotypes were due to reduced expression of GRIN2B (a subunit of the NMDA receptor), a high confidence Autism gene. Thus, Dlx1&2 coordinate key components of CIN postnatal development by promoting their excitability, inhibitory output, and survival.

Introduction

The Dlx homeodomain transcription factor (TF)-encoding genes are remarkable because their CNS expression is restricted to developing and mature GABAergic and dopaminergic forebrain neurons (Stuhmer, Anderson, et al. 2002; Stuhmer, Puelles, et al. 2002; Cobos, Broccoli, et al. 2005), and is excluded from glutamatergic neurons and glia (Petryniak et al. 2007). There are 6 mammalian Dlx genes (Dlx1-Dlx6), each organized in bigene pairs (Dlx1/2; Dlx3/4; Dlx5/6) co-regulated by multiple enhancers (Ghanem et al. 2007). During forebrain development only Dlx1,2,5&6 are expressed; Dlx2 expression begins in the neuroepithelium at ~E9.5, shortly followed by Dlx1; at E10.5 Dlx5&6 expression is present in secondary progenitors in the subventricular zone (SVZ), which also co-express Dlx1&2 (Eisenstat et al. 1999). As neurons are generated, they express different combinations and levels of the Dlx genes. For instance, parvalbumin⁺ (PV) cortical interneurons (CINs) express Dlx5&6, and no detectable Dlx1 (Cobos, Calcagnotto, et al. 2005; Cho et al. 2015).

The primordia of the basal ganglia are the ganglionic eminences (lateral, medial and caudal: LGE, MGE, CGE), septum and preoptic area (Flames et al. 2007), which generate the local GABAergic projection neurons and glia, and interneurons for the cortex, hippocampus, olfactory bulb, amygdala, and basal ganglia nuclei (Marin et al. 2001). The MGE and CGE generate the majority of pallial interneurons (Wonders and Anderson 2005; Rudy et al. 2011). The Dlx genes are required for development of telencephalic GABAergic neurons, including the striatum, globus pallidus, central nucleus of the amygdala and pallial interneurons (Anderson, Eisenstat, et al. 1997; Long et al. 2007; Wang et al. 2011). As the Dlx1&2 mutation leads to loss of Dlx1,2,5&6 expression, it creates an operational Dlx null expression state, resulting in regional fate changes (dorsal LGE takes on pallial properties), and cell fate changes due to disruption of TF expression (Long, Cobos, et al. 2009; Long, Swan, et al. 2009; Wang et al. 2013). For instance, Dlx1&2 mutants fail to maintain Zfhx1b (SIP1) SVZ expression; as a result MGE-derived interneurons fail to turn off Nkx2-1 expression and take on properties of subpallial interneurons (McKinsey et al. 2013).

Mice lacking a single Dlx gene have more specific neuronal phenotypes. For instance, adult Dlx1 constitutive and conditional mutants have pallial interneuronal apoptosis, affecting all subtypes except PV⁺ (Cobos, Calcagnotto, et al. 2005; Seybold et al. 2012). Even Dlx heterozygote mice have CIN phenotypes, as Dlx5/6^+/− mutants have abnormal function of PV⁺ interneurons (Cho et al. 2015).

Currently, Dlx1/2 postnatal functions have not been described, since the publications report phenotypes of constitutive or prenatal conditional mutants. Here, we address postnatal CIN subtype-specific functions of Dlx1&2 using conditional alleles of Dlx1 (Seybold et al. 2012), Dlx2 (described herein), and Dlx1&2 (Silbereis et al. 2014), by deleting them in calretinin (CR)⁺, PV⁺, or in all CINs. A summary of Dlx1 and/or Dlx2 CIN deletion phenotypes presented in this paper, and from previous studies, can be found in Supplementary Table 1.

We discovered that Dlx1/2 conditional knockouts (CKO) had greater CIN postnatal loss than Dlx1 or Dlx2 single mutants (constitutive and conditional), affecting even PV⁺ CINs. We provide evidence that DLX2 directly drives Gad1, Gad2, and Vgat expression, and show that the Dlx1/2 CKOs had reduced mIPSC amplitude. Furthermore, mutants had fewer GABAergic synapses on excitatory neurons and reduced mIPSC frequency.

Finally, Dlx1/2 CKO CINs had hypoplastic dendrites, reduced excitatory synaptic structures, and reduced excitatory input. The mutants had reduced expression of GRIN2B, a high confidence Autism gene encoding the NR2B subunit of the NMDA receptor. Restoring GRIN2B expression in Dlx1/2 mutant CINs rescued the hypoplastic dendrite phenotype. Thus, Dlx1&2 coordinate key components of CIN postnatal development by promoting their excitability, inhibitory functions, and survival.

Materials and Methods

Animals

A floxed Dlx2 allele (Dlx2^f) was generated by Siu-Pok Yee’s team at the Gene Targeting and Transgenic Facility of the University of Connecticut Health Center. LoxP sites were inserted so that Cre excision removes exons 2 and 3, including the entire homeodomain, similar to the constitutive null allele (Supplementary Fig. 1).

The following mouse strains and their genotyping protocols have been previously reported: DlxI12b-Cre (I12b-Cre) (Potter et al. 2009), Dlx1-CreER (Taniguchi et al. 2011), Calretinin-IRES-Cre (CR-Cre) (Taniguchi et al. 2011), Parvalbumin-Cre (PV-Cre) (Hippenmeyer et al. 2005), Nkx2.1-Cre (Xu et al. 2008), Dlx1^f/+ conditional mutants (Seybold et al. 2012), Dlx1/2^f/+ conditional mutants (Silbereis et al. 2014), Dlx2^−/− constitutive mutants (Qiu et al. 1995), Dlx1/2^−/− constitutive mutants (Qiu et al. 1997), and the Cre-dependent reporter lines CAG-GFP (Kawamoto et al. 2000) and Ai14 (which expresses tdTomato upon Cre-mediated recombination (Madisen et al. 2010)).

Initial transgenic analysis of the putative enhancer Gad2En1 was made, at Lawrence Berkeley labs, by having this candidate enhancer drive LacZ expression. We found expression in the GEs, (X-gal assay) consistent with normal Gad2 expression at E11.5 (data not shown, but available on request; data is recorded as mm1196 at the VISTA enhancer browser (Visel et al. 2013)). Following those promising results we made Gad2En1-CreER^T2-GFP mice by pronuclear injection at the Gladstone Transgenic Gene-Targeting Core using the C57BL/6 strain. The 1.7 kb Gad2 upstream enhancer region was amplified from C57BL/6 mouse genomic DNA and subcloned into a Hsp68-CreERT2-IRES-GFP vector (Visel et al. 2013). Founders were screened by PCR using specific primers against the 3′ end of the enhancer region (Fwd: 5′-CTC CAG AGG TCT CCA CGG CGC-3′) and the 5′ end of CreER^T2 (Rev: 5′-TTA TTC AAC TTG CAC CAT GCC GCC-3′).

We observed a partial penetrance for the Dlx1/2^f/−; I12b-Cre CKO phenotype. For instance, if the background was C57Bl/6, incomplete I12b-Cre-mediated recombination was observed. For this reason, we used a CD1 background. None-the-less, roughly 1/4 of the mutants in CD1 background did not show a clear reduction in P30 cortical interneurons (2/6 animals) or neonatal reduction of Gad1, Gad2, and Vgat by in situ RNA hybridization (1/4 animals) or by qRT-PCR (1/4 animals; the hypomorphic mutant had Dlx1 and Dlx2 RNA levels that were 2-fold higher than fully penetrant mutants). Hence, I12b-Cre recombination can be incomplete. Hypomorphic mutants were not included in our quantifications from the histological and the qRT-PCR analysis.

All strains were maintained on a CD1 background. P1 wild type (WT) hosts were CD-1 (Charles River Laboratories). All animals were housed in a vivarium with a 12 h light, 12 h dark cycle. Postnatal animals eventually used for experiments were kept with their littermates (litter size ranging from 5 to 16 animals). For timed pregnancies, noon on the day of the vaginal plug was counted as embryonic day 0.5. All animal care and procedures were performed according to the University of California at San Francisco Laboratory Animal Research Center guidelines.

Histology

Mice were euthanized by CO₂ inhalation followed by cervical dislocation. Isolated brains (E15.5) that were used for in situ hybridization (ISH) were fixed overnight in 4% paraformaldehyde (PFA), transferred to 15% sucrose overnight followed by 30% sucrose overnight for cryoprotection, and then embedded and frozen in OCT for cryosectioning. Cryostat sections were cut coronally at 20 um thickness. For E18.5 and P0 experiments, animals were anesthetized on ice while postnatal (P8, P29, P30) animals were anesthetized with intraperitoneal avertin (0.015 ml/g of a 2.5% solution) injection. Animals were perfused transcardially with PBS and then with 4% PFA, followed by brain isolation, 4–5 h fixation in 4% PFA, cryoprotection with 30% sucrose in PBS, and cut frozen (coronally or sagittal) on a freezing sliding microtome at 40 μm (100 μm in dendrite experiments) for immunohistochemistry or ISH. For GABA immunohistochemistry on Gad2En1-GFP mice at P30, mice were perfused with 4% PFA/0.1% glutaraldehyde in 0.14 M phosphate buffer, with further processing as described above.

All primary and secondary antibodies were diluted in PBS containing 10% Normal Serum, 0.25% Triton X-100 and 2% BSA. The following primary antibodies were used: Rabbit anti-DsRed (1:500, Clontech), chicken anti-GFP (1:2000, Aves), rabbit anti-GFP (Life Technologies, 1:1000), rat anti-GFP (Nacalai Tesque, 1:500), mouse anti-parvalbumin (1:3000, Swant Swiss Abs), rat anti-somatostatin (1:200, Chemicon), goat anti-somatostatin (1:100, Santa Cruz), mouse anti-calretinin (1:1000, Millipore), rabbit anti VIP (1:300, Immunostar), rabbit anti-GABA (Sigma, 1:500), rabbit anti-DLX2 (1:300, (McKinsey et al. 2013)), rabbit anti-GRIN2B (1:1000, Sigma), mouse anti-VGLUT1 (1:200, Synaptic Systems), rabbit anti-Vglut1 (1:500, Synaptic Systems), mouse anti- PSD95 (1:200, NeuroMab), rabbit anti-PSD95 (1:200, Cell Signaling), mouse anti-gephyrin (1:200, Synaptic Systems), rabbit anti-synaptophysin (1:200, Invitrogen). The secondary antibodies for immunofluorescence were Alexa Fluor-conjugated and purchased from Invitrogen. For indirect immunohistochemistry, sections were incubated with biotinylated secondary antibodies (Jackson), diluted 1:300, and processed by the ABC histochemical method (Vector), black reaction was obtained as described before (Adams 1981). For synapse immunohistochemistry (except for GRIN2B), sections were pre-treated with pepsin to enhance the staining as described before (Corteen et al. 2011). Immunofluorescence specimens were counterstained with 1% DAPI to assist in the delineation of cortical layers. For light microscopy preparations, sections analyzed were stained with Cresyl Violet to delineate the layers of the cortex (not shown for a better visualization). For ISH a rostrocaudal series of at least ten sections were examined. ISH on sections were performed using digoxigenin-labeled riboprobes as described (Stanco et al. 2014).

Electrophysiology

Whole-cell electrophysiological recordings were made in acute cortical slices made from WT and mutant mice. Mice were aged between P10–12 and P30–35. Mice were deeply anesthetized with a lethal dose of a ketamine/xylazine cocktail, brains were removed into ice-cold sucrose solution containing (in mM): 150 sucrose, 50 NaCl, 25 NaHCO₃, 10 dextrose, 2.5 KCl, 1 NaH₂PO₄, 0.5 CaCl₂, 7 MgCl₂. About 300 μm coronal sections were made on a vibratome. Slices were incubated in oxygenated artificial cerebrospinal fluid (ACSF) containing (in mM): 124 NaCl, 3 KCl, 1.25 NaH₂PO₄, 2 MgSO₄, 26 NaHCO₃, 10 dextrose, 2 CaCl₂. For recording, slices were submerged and continuously perfused (2–3 ml/min) with room temperature ACSF in a recording chamber mounted to a microscope (Olympus BX50WI; Center Valley, PA) equipped for differential interference contrast (DIC) and epifluorescence. Whole-cell recordings were made from visually identified pyramidal neurons under DIC optics, or GFP⁺ neurons under epifluorescence. Borosilicate patch pipettes (3–5 gigaohm resistance) were filled with internal solution appropriate for each experiment. For spontaneous and miniature EPSCs, the solution contained (in mM): 135 CsMeSO₄, 8 NaCl 5 QX-314, 10 Hepes, 4 ATP-Mg, 0.3 GTP-Na₂, and 0.3 EGTA. For evoked EPSCs, CsMeSO4 was reduced to 125 mM and replaced with 10 mM BAPTA to eliminate activity-dependent NMDA receptor subunit switching. For IPSCs, the solution contained (in mM): 140 CsCl, 1 MgCl₂, 10 Hepes, 11 EGTA, 2 ATP-Mg, 0.5 GTP-Na. Formation of gigaohm seals, and subsequent recordings were made using a Multiclamp 700B amplifier and pClamp8 software (Molecular Devices; Sunnyvale, CA). After obtaining whole-cell configuration, spontaneous postsynaptic currents (sEPSCs or sIPSCs) were recorded for a period of 5 min, followed by a 5 min epoch in which slices were perfused with ACSF containing tetrodotoxin (TTX; 0.1 μM; Alomone Labs, Jerusalem) and either gabazine (100 μM; Sigma, St. Louis) to pharmacologically isolate mEPSCs or kynurenic acid (1 mM; Sigma) to isolate mIPSCs. mEPSCs were then recorded for a period of 5 min. For evoked EPSCs, neurons were held at −70 mV and stimulated with a bipolar electrode at 0.05 Hz while kynurenic acid was washed onto the slice. AMPA receptor currents were recorded until their amplitude was indistinguishable from noise. Cells were then held at +40 mV to unblock NMDA receptors. Stimuli frequency was varied in an attempt to eliminate rundown. Ifenprodil (Sigma) was used at 3 micromolar. Only recordings in which input resistance and holding current were steady across the entire recording period were used for analysis. Analysis was performed using Mini Analysis v.6 (Synaptosoft; Fort Lee, NJ), Excel 2008 (Microsoft; Redmond, WA), and Matlab 2011 (MathWorks; Natick, MA), with experimenter blinded to genotype.

MGE Transplantation and In Vivo Rescue Assay

A detailed protocol for the MGE transplantation assay has been previously described (Vogt, Wu, et al. 2015). First, E13.5 MGEs from individual Dlx1/2^f/; CAG-GFP; I12b-Cre control and mutant embryos were dissected in ice-cold HBSS. Next, cells were mechanically dissociated by repeated pipetting (10–15 times) through a 1000 μl plastic pipette tip in DMEM media that contained 10% fetal bovine serum. The cells were then pelleted by centrifugation (3 min, 700 × G), and resuspended in 2–3 μl of DMEM, put on ice, and then remaining media was removed before loaded into the injection needle. For injections, a glass micropipette of 50 μm diameter (with a beveled tip) was preloaded with sterile mineral oil and cells were front-loaded into the tip of the needle using a plunger connected to a hydraulic drive (Narishige) that was mounted to a stereotaxic frame. Wild type (WT) P1 pups were anesthetized on ice for 1–2 min before being placed into a molded surface (modeling clay) for injections. Each pup received 3–5 injections of cells (~70 nl per site) in one hemisphere. These sites were about 1 mm apart along the rostral to caudal axis; cells were injected into layers V–VI of the neocortex. After injections, pups were put back with the mother to recover after they began to move around on their own. Mice were sacrificed 28 days after transplantation and transcardially perfused with PBS followed by 4% PFA.

For the rescue assay, we modified a previously reported protocol that transduced MGE cells before transplantation (Vogt et al. 2014; Vogt, Cho, et al. 2015; Vogt, Wu, et al. 2015), in order to dual transfect MGE cells with 2 expression vectors (one expressing Cre and a gene of interest and the other a Cre-dependent reporter). Briefly, E13.5 MGE cells were collected from either Dlx1/2^f/+ or Dlx1/2^f/− embryos in HBSS. Cells were dissociated in the same manner as above, except that they were dissociated in DMEM with 10% FBS that was preconditioned in a tissue culture incubator at 37 °C and with 5% CO₂ to achieve a physiological pH. The cells were then transfected with a combination of a Cre-expressing plasmid (empty or GRIN2B) and a Cre-dependent reporter (CAG-Flex-GFP) using Lipofectamine2000 (Invitrogen). Cells were transfected for 30 min at 37 °C then centrifuged as before. The remaining procedures followed the above protocol to transplant the cells.

Primary Cell Culture

MGE tissue was dissected from E13.5 Dlx1/2^f/; CAG-GFP; CR-Cre embryos and mechanically dissociated with a P1000 pipette tip. A total of 200 000 cells were seeded into tissue culture slides precoated with poly-L-lysine (10 mg/ml, Sigma) and then laminin (5 mg/ml, Sigma), and grown in vitro with media containing DMEM-H21 with 10% fetal bovine serum for 2 h. After the cells recovered, DMEM-H21 media was replaced by Neurobasal medium containing B27 supplement, 25% glucose, and glutamax and cultures were grown for 14 days in vitro. Cultures were fixed with 4% PFA for 10 min and processed for immunocytochemistry. Briefly, they were washed in PBS, quenched 2 times for 15 min with 2 mg/ml sodium borohydrate solution, blocked in PBS containing 10% Normal Serum, 0.1% Triton X-100 and 2% BSA, incubated in primary antibody overnight (4 °C), washed in PBS, incubated in secondary antibody for 1–2 h (room temperature), washed in PBS, and mounted.

DNA Expression Vector Generation

Luciferase reporter constructs (pGL4.23-Enhancer-luciferase) were generated from the PGL4.23 vector (Promega). The enhancers were PCR-amplified from mouse genomic DNA and ligated upstream of the minimal promoter with the following primers and introduced restriction sites: Grin2b Enhancer3: (5′ GAGACTCGAGGAAAGTTGTAATGGTGTGG, 3′ GAGACCCGGGTCCTACTTAACTTCCGAGTC), with XhoI and SmaI underlined; Gad2 Enhancer: (5′ TGTCACCTCGAGACTTTGGAACTTCCTGTGCTAACT, 3′ TTACGCATCCCGGGGTGGAGACCTCTGGAGTTGC), with XhoI and SmaI underlined. The pCAGGS-DLX2 and pCAGGS-GFP expression vectors have been described previously (Maira et al. 2010).

The CMV-empty-IRES-Cre vector was made by excising the T2a sequence from a previously reported vector that contains a GFP-T2a-Cre cassette (Vogt et al. 2014) with the enzymes BsrGI and EcoRI. Next, an IRES fragment was PCR-amplified from the pEGFP-IRES2 vector (Clontech), with primers (Forward: 5′ GAGATGTACAACCGGGATCCGCCCCTCT 3′, Reverse: 5′ GAGAGAATTCTGTGGCCATATTATCATCG 3′) that introduced 5′ BsrGI and 3′ EcoRI sites. This product was then ligated into the aforementioned vector at the same restriction sites. This vector was used as the empty control in experiments. Next, human GRIN2B was PCR-amplified from a cDNA (R&D Systems, RDC1221) with primers (Forward: 5′ GAGATCTAGAATGAAGCCCAGAGCGGAG 3′, Reverse: 5′ GAGATGTACATCAGACATCAGACTCAAT 3′) that introduced 5′ XbaI and 3′ BsrGI restriction sites. GRIN2B was then ligated into the corresponding sites in the CMV-empty-IRES-Cre vector. The Cre-dependent expression vector, CAG-Flex-GFP, has been previously described (Vogt, Wu, et al. 2015).

Human DLX2 was PCR-amplified from a pCAAGS/ES-DLX2 expression vector (Maira et al. 2010) with primers (5′ GAGAGAATTCATGACTGGAGTCTTTGAC, 3′ GCAACCGGTTCATTAGAAAATCGTCC) with introduced EcoRI and AgeI restriction sites underlined. Next, DLX2 was ligated into the EcoRI and XmaI restriction sites within the multiple cloning site (MCS) of a CMV-DlxI12b-Bg-Cre-T2a-empty vector (CMV-empty) that has been previously described (Vogt, Cho, et al. 2015). These latter 2 plasmids are lentiviral DNA vectors and the DlxI12b-BG enhancer/minimal promoter only directs expression if virus is made and used. Since these vectors were used for transfection, the CMV overrides the enhancer/minimal promoter and cassettes are ubiquitously expressed. All vectors were verified by restriction digest and sequencing.

DLX2 and H3K27Ac Chromatin Immunoprecipitation (ChIP)

Transcription factor ChIP was performed similar to a published method (McKenna et al. 2011) with few modifications. Briefly, E13.5 or E16.5 ganglionic eminences were dissected, fixed in 1.5% formaldehyde for 20 min and neutralized with glycine. Fixed chromatin was lysed and sheared into 200–1000 basepair (bp) fragments using a bioruptor (Diagenode). Immunoprecipitation (IP) reactions of E13.5 and E16.5 were performed using an anti-Dlx2 antibody raised in rabbit against purified MBP-Dlx2 fusion protein (MBP-Dlx2, amino acids 1-154, tagged by Maltose Binding Protein). Two negative controls were performed: 1) 30× molar fold MBP-Dlx2 blocking peptide was included in the IP reactions; 2) DLX2 ChIP-seq was performed using E16.5 Dlx2−/− constitutive mutant; called peaks were not present in either of these negative controls. Precipitated fractions were purified using Dynabeads (Invitrogen).

The H3K27Ac histone ChIP was done on native E13.5 ganglionic eminences as previously described (Magklara et al. 2011). A hypotonic buffer was used to extract nuclei, which were digested with micrococcal nuclease (MNase, Sigma) to yield a population of mono- to tetra-nucleosomes. This native chromatin was imunoprecipitated overnight using a specific H3K27Ac antibody (05-1334, Millipore) and purified using Dynabeads (Invitrogen).

Libraries preparation and sequencing were performed as described (Sandberg et al. 2016). Briefly, DLX2 and H3K27ac reads were aligned to the mouse genome (mm9) and regions of significant enrichment were identified based on comparison to genomic background and to input (no antibody) and IgG control. After sequencing, reads were filtered to remove low quality bases and adapter sequences using the fastx toolkit. Remaining reads were aligned to mouse genome (mm9) using BWA and peaks were identified via MACs (version 2). For regions of interest, binding of DLX2 and presence of H3K27ac were identified. For the figures, data from the DLX2 E13.5 and E16.5 ChIP-seq experiments and the H3K27ac E13.5 ChIP-seq experiment are shown along with called peaks. The aligned reads are plotted on the same scale (0–150 for DLX2 ChIP and 0–5 for H3K27ac) across all figures.

Quantitative RT-PCR

RNA was harvested from cortical tissue of P2 mice using Mini RNeasy kit (Qiagen). The RNA samples were DNAse treated (Turbo DNA free kit, Ambion), 2 μg RNA from each sample were used to generate cDNA (SuperScript III First Strand Synthesis System, Thermofisher) and minus Reverse Transcriptase reactions were included as negative controls. Each cDNA sample was used in quantitative PCR performed using an ABI 7900HT instrument. Data were analyzed using the ΔΔCT method. Primers used were Gad1 (5′ ACTCCTGTGACAGAGCCGA3′, 5′ TACGGTTCAAGGGTCCCCC 3′), Gad2 (5′ CACCGTGTATGGGGCTTTTG 3′, 5′ TCCACTTGTGTTTCCGGGAC 3′), vGAT (5′ GGGTCACGACAAACCCAAGA 3′, 5′ GAGGAACAACCCCAGGTAGC 3′), and GAPDH (5′ GGCGCGCGTCATCAG 3′, 5′ TGACCAGGCGCCCAATAC 3′). As control for the i12b-Cre deletion of Dlx1/2, primers against Dlx2 (5′ ACACTCCCGTAGCTCCTTCA 3′, 5′ AAATGAGGTCATCCGCAAAG 3′) and Dlx1 (5′ TGGAATCCGAACTCCTCATC 3′, 5′ TGCTGCATAGCTTCTTGGTG 3′) were included.

Luciferase Enhancer-reporter Assay

P19 cells were seeded in 24-well culture plates at density of ~100 000 cells/well in DMEM supplemented with 10% FBS. Cells were transfected using Xtreme Gene HP (Roche) at 16 h with 200 ng of pGL4.23-Enhancer-luciferase construct DNA, 200 ng of pCAGGS-Dlx2 or pCAGGS-GFP, and 0.5 ng Renilla luciferase vector (Promega) as an internal control. Cells were harvested 30 h later and assessed for luciferase activity according to the Dual-Luciferase Reporter Assay System protocol (Promega).

Image Acquisition and Analysis

All the image acquisition was performed in the somatosensory cortex. Fluorescent and brightfield images were taken using a Coolsnap camera (Photometrics) mounted on a Nikon Eclipse 80i microscope using NIS Elements acquisition software (Nikon). Confocal images were taken using a Yokagawa CSU22 confocal scanner mounted on a Nikon Ti-E Microscope with an ×100 objective at 1024 × 1024 pixels of resolution from the Nikon Imaging Center at UCSF. Brightness and contrast were adjusted and images merged using Photoshop or ImageJ software. Neurolucida software was used to draw and analyze the dendritic length of CR cells in the Dlx1/2^f; I12b-Cre. ImageJ software was used to image processing, draw, and analyze the dendritic arbor on grafted cells. For synapse counting (GRIN2B, presynaptic, and postsynaptic boutons), confocal image stacks from all cortical layers (0.4 μm step size) were processed with ImageJ software. In brief, we applied background subtraction and smooth filter. We then ran a threshold to convert the pictures into “masks” and colocalized the channels with the Image Calculator plugin. We measured the length of the dendrites and counted the number of particles or colocalizations on the dendrites in each of the focal planes. The staining and imaging of controls and mutants were done in parallel.

Cell Counting

For assessing cell density in the neocortex on fluorescence sections 10× images at postnatal or 20× images at embryonic stages were taken of the somatosensory cortex, encompassing all neocortical layers, from both hemispheres for each replicate (Bregma level in postnatal brains around −0.7 mm and around 3 mm from the most rostral section in perinatal brains). Images were opened with ImageJ or Canvas software to delimit and measure the region of interest (ROI). All cells in the ROI were counted, and divided by the ROI area to determine cell density. For lamination counts, we used DAPI to subdivide neocortical layers. For the quantification of cell density by bins the cortex was analyzed with a grid of 10 equal horizontal bins. Bin 1 corresponds to the MZ, and bin 10 corresponds to the VZ for prenatal or WM for postnatal mice. Cell density on DAB sections was measured with Neurolucida software.

Statistics

All statistical analyses were done with SPSS15 software. The statistical significance of single comparison on discrete data was performed using the nonparametric Chi-square’s test. For continuous data the statistical significance of single comparisons was performed using 2-tailed t-test with Welch’s correction when required (non-equal variances) or Mann–Whitney nonparametric test when data did not fit to a normal distribution (assessed by Shapiro–Wilk normality test). For multiple comparison we used ANOVA with a Tukey HSD post hoc to determine the significance between groups after checking our data fitted to a normal distribution (assessed by Shapiro–Wilk normality test) and the variance was equal (determined by Levene’s test). n refers to the number of mice analyzed, unless otherwise stated in the Figure Legends.

Results

Reduction of Adult Cortical Interneurons in the Dlx1/2; I12b-Cre Conditional Mutant

Dlx1/2 are expressed in GE progenitors, newly generated and migrating subpallial neurons, and postnatally in subsets of GABAergic neurons (Eisenstat et al. 1999; Cobos, Broccoli, et al. 2005; Cobos et al. 2006). While Dlx1/2 are required in GE progenitors for the fate, differentiation, migration and survival (Anderson, Eisenstat, et al. 1997; Anderson, Qiu, et al. 1997; Cobos et al. 2007; Long et al. 2007; Long, Cobos, et al. 2009; Long, Swan, et al. 2009), postnatal Dlx1/2 functions are unknown. Here, we generated 3 types of Dlx1/2 conditional knockouts (CKOs) by deletion: 1) in the GE SVZ using the I12b-Cre allele (~E10.5); (Dlx1/2; I12b-Cre CKO) (see Material and Methods for the percentage of penetrance); 2) in CR⁺ CINs, using the Calretinin-Cre allele (beginning around P0); (Dlx1/2; CR-Cre CKO); 3) in PV⁺ CINs (beginning around P10); (Dlx1/2; PV-Cre CKO). We used the Cre reporter (CAG-GFP) to follow the fate of Cre⁺ cells. A typical cross was: Dlx1/2^+/−; Cre allele (male) × Dlx1/2^f/f; CAG-GFP (female), producing Dlx1/2^f/−; Cre; CAG-GFP CKO mutants and Dlx1/2^f/+; Cre; CAG-GFP controls.

We used I12b-Cre to test whether loss of Dlx1/2 in the SVZ phenocopied the constitutive Dlx1/2^−/− KO (Supplementary Table 1). No change in GE radial migration (not shown) nor tangential migration (Fig. 1a–d) was detected at E15.5. As expected Dlx1 and Dlx2 RNA expression was maintained in the VZ, reduced in the SVZ (Supplementary Fig. 2a,b), and was absent in the cortex (Supplementary Fig. 2c–c’). Unlike the constitutive null mutant, which loses Dlx5 expression (Anderson, Eisenstat, et al. 1997; Long, Cobos, et al. 2009), the I12b-Cre CKO showed no major change in Dlx5 expression (Supplementary Fig. 3b–b’).

Figure 1.

Loss of Dlx1/2 function leads to a decrease of CINs only at late postnatal stages. (a, b, e, f) Coronal sections through the telencephalon of control (a, e) and Dlx1/2^f; GFP; I12b-Cre mutant (b, f) E15.5 embryos (a, b) or P30 (e, f) mice showing the distribution of GFP-expressing cells after GFP staining. (a) Numbers identify bins for quantification. (c, d, g, h) Quantification of GFP⁺ cells in cortex in E15.5 embryos (c, d) and P30 (g, h) by (c) (control: 2004.2 ± 151.8 cells/mm², n = 4; mutant: 1775.7 ± 128.5 cells/mm², n = 4; P = 0.2) (g) (control: 370.5 ± 17.01 cells/mm², n = 4; mutant: 269.9 ± 9.04 cells/mm², n = 4; P = 0.0014) cell density and (d, h) distribution along the cortex by bins (d) (n = 4, a total of 2141 counted cells, P = 0.4) or layers (h) (control; layer2/3:494.3 ± 45.2; layer4:401.6 ± 58.4; layer5/6:370.02 ± 26.3 cells/mm², n = 4; mutant; layer2/3:329.03 ± 23.3; layer4:232.2 ± 21.3; layer5/6:305.3 ± 13.3 cells/mm², n = 4; P layer2/3 = 0.01, layer4 = 0.03, layer5/6 = 0.07) of control (black bars) and Dlx1/2 mutant (white bars) mice. *P < 0.05, **P < 0.01 (t-test). Scale bar (in a) a–b, 100 μm and (in e) e–f, 200 μm. (i–l, m–p) Coronal sections through the cortex of P30 control (i–l) and Dlx1/2 mutant (m–p) mice showing the distribution of positive cells after DAB immunohistochemistry for PV (i, m), SOM (j, n), CR (k, o) and VIP (l, p). (q) Quantification of the density of PV (control: 248.06 ± 25.4 cells/mm2, n = 4; mutant: 132.2 ± 19.6 cells/mm2, n = 4; P = 0.011); SOM (control: 202,4 ± 3.7 cells/mm², n = 3; mutant: 121.7 ± 11.2 cells/mm², n = 3; P = 0.02); CR (control: 52.6 ± 1.4 cells/mm², n = 4; mutant: 35.4 ± 4.3 cells/mm², n = 4; P = 0.0096) and VIP (cells control: 50.6 ± 2.9 cells/mm², n = 3; mutant: 23.03 ± 1.8 cells/mm², n = 4; P = 0.0003) cells in the cortex. *P < 0.05, **P < 0.01, ***P < 0.001 (t-test). I–VI, cortical layers from I to VI. Histograms show average ± SEM. Scale bar (in m) i–l, m–p 200 μm.

On the other hand, VZ deletion of Dlx1/2 in the MGE using Nkx2-1-Cre phenocopied the Dlx1/2 constitutive mutant, with loss of Dlx5 expression in the MGE (Supplementary Fig. 3a–a’; asterisk in MGE), ectopic migration to the caudoventral subpallium (not shown), and ~4-fold reduced tangential migration to the cortex (Supplementary Fig. 3f–h). Thus, while Dlx1/2 have critical VZ functions including driving Dlx5 expression and the generation of migrating neurons, Dlx1/2 do not have these functions in SVZ and MZ.

We next analyzed CIN development in the Dlx1/2; I12b-Cre CKO. In P0 cortex, the density and laminar distribution of GFP⁺ cells was normal (Supplementary Fig. 3c–e), whereas at P14 and P30 there was a ~30% reduction of GFP⁺ cell density, preferentially in superficial layers (Fig.1e–h and Supplementary Fig. 2d–g), that was associated with ~3.5-fold increased apoptosis at P8 (normal at P0, P14) (Supplementary Fig. 2h–j). Unlike Dlx1^−/− constitutive mutants, in which PV⁺ CINs were not reduced, the Dlx1/2; I12b-Cre CKO had fewer PV⁺ CINs (~50% of control numbers), in addition to reductions in SOM⁺, CR⁺ and VIP⁺ CINs (Fig. 1i–q).

PV⁺ CINs are Derived from Cells that Express Dlx1 and Dlx2

Previously we reported that Dlx1 expression was not detectable in PV⁺ CINs at P30 (Cobos, Calcagnotto, et al. 2005), raising the question of why PV⁺ CINs were reduced in the Dlx1/2; I12b-Cre CKO. We tested whether Dlx2 is expressed in PV⁺ CINs; immunofluorescence showed that DLX2 and PV were co-expressed in P30 cortical and hippocampal INs (Fig. 2f–i”). Thus, to analyze Dlx2’s postnatal function we generated a Dlx2 floxed allele (Dlx2^f) (Materials and Methods), as Dlx2^−/− constitutive mutants die at P0 (Qiu et al. 1995). Dlx2^f/−; I12b-Cre CKOs had normal numbers of PV, SOM and VIP CINs or tdTomato⁺ cells at P30 (Supplementary Fig. 4a–d). Thus, loss of Dlx2 or Dlx1 alone does not account for the reduced numbers of PV CINs in the Dlx1/2; I12b-Cre CKO.

Figure 2.

Dlx1 is expressed prenatally, but not neonatally, in cells that become PV⁺CINs; adult PV⁺ interneurons maintain Dlx2 expression. (a) Schematic diagram of the experimental design. Dlx1Cre-ER; AI14 pregnant females received a single injection of tamoxifen at E12 or P4. Embryos were allowed to be born and analyzed at P30. (b–e) Coronal sections through the somatosensory cortex of Dlx1Cre-ER; AI14 P30 mice showing the distribution after the immunohistochemistry of PV⁺ (green) and tdTomato⁺ (red) cells after E12 (b, c) or P4 (d, e) tamoxifen injection. (c’, c”, e’, e”) Magnification of the areas boxed in c and e, respectively. Scale bar (in d) b–c, d–e 200 μm and (in e”) c’, c”, e’, e” 50 μm. (f–i) Coronal sections through the cortex (f–g) and the hippocampus (h–i) P30 WT mice showing immunohistochemistry expression for PV⁺ (green) and DLX2⁺ (red) cells. (g’, g”, h, i) Confocal images of somatosensory cortex (g’–g”) and hippocampal pyramidal layer (h, i) showing immunohistochemistry expression for PV⁺ (green) and DLX2⁺ (red) cells. (j) Quantification of the DLX2/PV colocalization. Colocalized (arrowhead), non-colocalized (open arrowhead) cells are noted in the panels. Scale bar (in f) f, g 100 μm, (in g”) g’, g” 20 μm and (in h) h, i 20 μm.

We next assessed the timing of when Dlx1&2 are required to establish the proper PV CINs numbers. Thus, we generated Dlx1/2; PV-Cre CKO mice, deleting Dlx1/2 between P10–P20 (Taniguchi et al. 2011). PV numbers were normal at P30 and P60 showing that Dlx1/2 function is needed prior to ~P10 for PV CIN survival (Supplementary Fig. 4e–h).

We hypothesized that Dlx1 and Dlx2 have redundant functions in PV CINs development. As Dlx2 is expressed in these neurons (Fig. 2f–i’), we tested whether Dlx1 was expressed at earlier stages of their development. We carried out fate mapping experiments using a Dlx1-CreER allele, in which tamoxifen induced tdTomato expression. We gave tamoxifen at either E12.5 or P4 (Fig. 2a), analyzing PV and tdTomato co-expression at P30. E12.5 tamoxifen administration led to PV/tdTomato co-expression (Fig. 2b–c”), whereas none was detected from the P4 treatment (Fig. 2d–e”). Thus, Dlx1 is expressed prenatally in SVZ cells and/or immature neurons that become PV CINs, whereas at P4, Dlx1 expression is absent from neurons that will later be PV CINs. Therefore Dlx1&2 are required prenatally for the survival of PV interneurons.

Dlx1/2 Neonatal Function is not Required for CR⁺ CIN Survival

Prenatal deletion of Dlx1/2 caused a robust reduction in the numbers of all CIN subtypes (Fig. 1i–q), whereas postnatal deletion beginning around P10 in cells that express PV did not lead to cell loss (Supplementary Fig. 4e–h). To pinpoint when Dlx1/2 function is required for CIN survival, we used a CR-Cre allele (Taniguchi et al. 2011), whose expression begins around P0 (Supplementary Fig. 4n–p’). We generated Dlx1/2; CR-Cre CKO mutants and assessed GFP and GFP/CR expression at P30. We did not detect an alteration in CR⁺ CIN or GFP⁺ cell number (Supplementary Fig. 4i–m). Thus, Dlx1/2 functions are required prenatally (in progenitors and/or immature neurons), but not postnatally, for CR⁺ (from ~P1) and PV⁺ (from ~P10) CIN survival.

Synaptic Inhibition Frequency and Amplitude are Reduced in the Cortex of Dlx1/2; I12b-Cre CKO

To measure the physiological effects of CIN loss in Dlx1/2; I12b-Cre CKOs we studied synaptic inhibition on layer 2/3 pyramidal neurons of somatosensory cortex at P10–12 and P30. P30 miniature inhibitory postsynaptic currents (mIPSCs) showed a significant reduction in frequency and amplitude (Fig. 3a–c). Cumulative probability plots confirmed reductions in mIPSC frequency and amplitude (Fig. 3d). Analysis of mIPSC kinetics to assess defects on GABA receptors or synapse maturation revealed significant increases in both rise time and decay suggesting the synapses are immature (Supplementary Fig. 5d). This constellation of changes is seen during synaptic maturation, and could indicate delayed synaptic development. Several mechanisms could account for such differences in mIPSC amplitude and kinetics, including differences in the number and subunit composition of synaptic receptors (Tia et al. 1996), as well as structural changes to the synapse (Cathala et al. 2005). Spontaneous IPSCs recordings showed similar trends but did not reach statistical significance (Fig. 3b). At P10–12 no phenotypes were observed (Supplementary Fig. 5a–c).

Figure 3.

Synaptic transmission is altered in the cortex of adult mice after Dlx1/2 gene deletion. (a) Scheme of Dlx1/2 deletion in cortical interneurons and recording location (pipette) in the cortex. (b) Whole-cell recordings at P30 reveal that sIPSC frequency (control: 23.3 ± 3.4 Hz, n = 12; mutant: 18.5 ± 2.5 Hz, n = 17; P = 0.2) and amplitude (control: 32.9 ± 13.2 pA, n = 11; mutant: 26.2 ± 2.4 pA, n = 17; P = 0.1) are unchanged between WT and mutant mice. (c) Synaptic inhibition is reduced, with significant reductions in mIPSC frequency (control: 15.6 ± 1.9 Hz, n = 12; mutant: 10.2 ± 1.2 Hz, n = 15; P = 0.02) and amplitude (control: 21.3 ± 1.8 pA, n = 11; mutant: 13.5 ± 1.4 pA, n = 15; P = 0.002). Analysis of mIPSC kinetics show significant increases in both rise time (control: 1.59 ± 0.08 ms, n = 11; mutant: 2.00 ± 0.11 ms, n = 15; P = 0.01) and decay tau (control: 12.11 ± 0.84 ms, n = 10; mutant: 15.17 ± 0.88 ms, n = 15; P = 0.02). *P < 0.05 **P < 0.01, (t-test), as well as in the probability distributions (d) of inter-event intervals and amplitudes ***P < 0.001, (Kolmogorov–Smirnov test). (e) Scheme of Dlx1/2 deletion in Calretinin interneurons and recording location (pipette) in the cortex. Measures of excitatory synaptic transmission in cortical layer 2 on CR⁺ interneurons at P30, (f) show a significantly decreased sEPSC frequency (control: 13.1 ± 1.4 Hz, n = 7; mutant: 7.01 ± 0.7 Hz, n = 4; P = 0.01), but normal sEPSC amplitude (control: 12.5 ± 1.0 pA, n = 7; mutant: 13.2 ± 0.8 pA, n = 4; P = 0.6). (g) mEPSC frequency was significantly reduced (control: 10.5 ± 1.3 Hz, n = 5; mutant: 4.4 ± 0.8 Hz, n = 4; P = 0.009) but not the mEPSC amplitude (control: 15.6 ± 0.8 pA, n = 5; mutant: 13.9 ± 1.6 pA, n = 4; P = 0.3). Analysis of mEPSC kinetics show no significant differences in both rise time (control: 1.46 ± 0.12 ms, n = 5; mutant: 1.72 ± 0.13 ms, n = 4; P > 0.05), and decay tau (control: 5.72 ± 1.0 ms, n = 5; mutant: 5.46 ± 0.70 ms, n = 4; P > 0.05). *P < 0.05, **P < 0.01 (t-test), an effect also revealed by analysis of inter-event interval (h); ***P < 0.001, (Kolmogorov–Smirnov test). Histograms show average ± SEM. n, number of cells.

The reduced mIPSC frequency likely reflected the reduced CIN numbers (and thus GABAergic synapses). mIPSC amplitude is related of the amount of GABA released into the synaptic cleft by the presynaptic neuron and the number of receptors present in the postsynaptic density. Thus, mIPSC amplitude can be altered by both presynaptic (e.g., vesicle GABA content) and postsynaptic (e.g., receptor number) mechanisms. The changes we report could result from direct control by Dlx1/2, or from activity-dependent regulation of synaptic transmission. To explore potential direct, presynaptic (interneuron-based) mechanisms, we investigated the effect of the Dlx1/2 mutation on expression of genes that control GABA synthesis and transport into synaptic vesicles.

Decreased Expression of Gad1, Gad2, and Vgat in Dlx1/2; I12b-Cre CKO

Reduced mIPSC amplitude (Fig. 3c) could be caused by decreased GABA in the synaptic cleft. We explored that possibility using ISH to analyze Gad1 (Gad67), Gad2

(Gad65) and Vgat (SLC32a1) expression. These genes encode the enzymes that synthesize GABA, and the GABA Vesicular Transporter (Juge et al. 2009). We analyzed their expression at P0 and P8 and found a reduction in the number of neocortical cells that express Gad1, Gad2, and Vgat (Fig. 4a–g, and not shown). We then used qRT-PCR for further quantification, and found a ~2-fold reduction in RNA levels encoded by these 3 genes (Fig. 4h).

Figure 4.

Decreased expression of Gad1, Gad2, and Vgat in the Dlx1/2;I12bCre CKO. (a–f) Coronal sections through the cortex of control (a–c) and Dlx1/2 mutant (d–f) mice at P0 showing the expression of Gad2 (a, d), Gad1 (b, e) and Vgat (c, f) mRNA. (g), Quantification of the expression of Gad2 (control: 283.5 ± 23.4 cells/mm², n = 3; mutant: 141.2 ± 6.5 cells/mm², n = 3; P = 0.004), Gad1 (control: 253.2 ± 4.7 cells/mm², n = 3; mutant: 189.3 ± 5.6 cells/mm², n = 3; P = 0.0009) and Vgat (control: 233.2 ± 14.3 cells/mm², n = 3; mutant: 156.6 ± 21.1 cells/mm², n = 3; P = 0.04) by cell density in the cortex from controls (black bars) and Dlx1/2 mutants (white bars) mice. (h) Quantitative RT-PCR measurement RNA levels of Gad1 (control: 102.4 ± 15.1%, n = 3; mutant: 54.9 ± 6.01%, n = 3; P = 0.04), Gad2 (control: 112.8 ± 17.8%, n = 3; mutant: 57.5 ± 8%, n = 3; P = 0.04), Vgat (control: 101.7 ± 14.3%, n = 3; mutant: 54.6 ± 2.8%, n = 3; P = 0.07), Dlx2 (control: 100.1 ± 4.4%, n = 3; mutant: 6.04 ± 0.4%, n = 3; P = 2.9e−005) and Dlx1 (control: 84.3 ± 5.9%, n = 3; mutant: 1.6 ± 0.2%, n = 3; P = 0.0001) on RNA from P2 cortical tissue from control (closed symbols) and Dlx1/2 mutant (open symbols) mice. Data are expressed as the percentage of GAPDH control RNA. *P < 0.05, **P < 0.01 and ***P < 0.001 (t-test). Histograms and scatter plots show average ± SEM. Scale bar (in d) a–f, 200 μm.

Next to test whether Dlx1/2 directly regulate Gad1, Gad2, and Vgat, we used DLX2 ChIP-Seq on E13.5 and E16.5 ganglionic eminences (GEs). We found robust DLX2 ChIP-Seq peaks at several loci in the regions of these genes (Fig. 5a–c). DLX2 binding was detected in the Gad2/Myo3a region at 2 sites within the Myo3a locus (Fig. 5a). The Gad1/Myo3b region had multiple DLX2 peaks within the Myo3b locus (Fig. 5b). The Vgat (Slc32a1)/Arhgap40 region had one large and multiple small DLX2 peaks between the genes (Fig. 5c).

Figure 5.

Evidence for Direct Control of Gad1, Gad2, and Vgat by DLX2. (a–c) Diagram showing DLX2 binding peaks in E13.5 and E16.5 GE in the (a) Gad2/Myo3a locus (peaks #1 and 2), (b) in the Gad1/Myo3b locus, and (c) the Vgat (Slc32a1)/Arhgap40 locus. H3K27Ac ChIP-Seq peaks show candidate active regulatory elements in the E13.5 GE; these peaks are generally in the same position as the DLX2 peaks. The orange box in (a) shows putative Gad2 enhancer 1. Arrows shows the direction of transcription from the promoter. Thick lines underneath the peaks are the computationally “called” peaks. (d) Schematic of the luciferase reporter transcriptional assay in P19 cells. (e) Luciferase assay data, presented as fold change in activation with DLX2 over activation with GFP, and normalized to activation of the empty pGL4.23 control (pGL4.23: 1.0 ± 0.08, n = 3; Gad2 En1: 7.7 ± 0.54, n = 3, P = 0.0053). Histograms show average ± SEM. **P < 0.01, (t-test). (f–h”) Immunohistochemistry from the Gad2En1-GFP transgenic mouse. (f) GFP and DAPI labeling of an E12.5 telencephalic coronal section. (g) GFP and DAPI labeling of P30 neocortex (coronal section). (h–h”) Colocalization (arrows) of GFP (h, h’) and GABA (h, h”) in P30 neocortex. Scale bar in f, 250 μm, g, 250 μm, h, 50 μm. Cx, Cortex; LGE, Lateral Ganglionic Eminence; MGE, Medial Ganglionic Eminence.

To gain evidence of whether these peaks represent candidate active regulatory elements, we performed ChIP-Seq for a histone variant that is associated with active enhancers in vivo in the developing brain (H3K27Ac; (Creyghton et al. 2010; Rada-Iglesias et al. 2011; Nord et al. 2013)). H3K27Ac peaks were present at the DLX2 peak loci (Fig. 5a–c).

To investigate the function of these candidate DLX2-bound regulatory elements we used a cell-culture transcription assay and transgenic mice (Visel et al. 2013). We concentrated on region #1 near Gad2 because it had a DLX2 peak at E13.5 and E16.5, as well as an E13.5 H3K27Ac peak (Fig. 5a). First we tested whether DLX2 could modulate the activity of the candidate Gad2 enhancer (Gad2 En1) using a luciferase reporter assay in P19 cells. Co-transfection of a DLX2 expression vector (pCAGGS-Dlx2) with the enhancer-reporter construct (pGL4.23-Gad2En1-luciferase) resulted in an ~8-fold increase in luciferase activity compared with the pGL4.23 empty luciferase control (Fig. 5d,e).

Next, we assessed Gad2 En1 activity in vivo by testing whether it could drive GFP expression in an E12.5 stable transgenic mouse (Gad2En1-CreER^T2-GFP, hereafter referred to as Gad2En1-GFP). GFP expression was present in the GEs, preoptic area and in cells tangentially migrating to the cortex (probably subpallially-derived interneurons) (Fig. 5f). At P30 we found GFP expression in the forebrain was restricted to GABAergic domains (Fig. 5g). Within the somatosensory cortex, 67.6% ± 4.6% of the GFP⁺ cells co-localized with GABA, and 36.2% ± 2.6% % of GABA cells were labeled by GFP (Fig. 5h–h”); of the 60 brightly labeled GFP cells with a clear morphology, none had a pyramidal shape. Thus, this candidate Gad2 enhancer is active in both developing CINs and a subset of mature CINs. Together, these data show that: 1) DLX2 directly binds to a candidate Gad2 enhancer in GE cells (at E13.5 and E16.5); 2) this candidate Gad2 enhancer’s activity patterns closely match expression of Gad2 prenatally and at P30; and 3) DLX2 activates this enhancer in tissue culture transcription assays, providing evidence that DLX2 promotes inhibitory interneuron function through direct activation of Gad2 expression.

Dlx1/2 Control Inhibitory Synapse Numbers

The reduced mIPSC frequency in projection neurons (Fig. 3c) could be caused by a decreased number of interneurons, decreased synapses, or both. To test the second possibility, we quantified the density of inhibitory terminals on projection neurons in the somatosensory cortex of P30 Dlx1/2; CR-Cre CKOs (Fig. 6a), a mutant with a normal number of CINs (Supplementary Fig. 4i–m). CR⁺ cells inhibit CINs and projection neurons. To identify presynaptic punctae on CR⁺ interneurons we performed GFP and Synaptophysin immunohistochemistry. To identify the neuronal postsynaptic structures that were apposed to CIN presynaptic punctae, we performed immunohistochemistry with Gephryin (Fig. 6b–c). Confocal fluorescent microscopic analysis showed a ~36% reduction in inhibitory synaptic punctae from CR⁺ onto dendrites (these could be dendrites of CINs and/or excitatory neurons) (Fig. 6d). A reduction of cortical inhibitory synapse numbers was also found in P30 Dlx1 sensitized CKO (Dlx1^f/−; Dlx2^+/−; CR-Cre) CR⁺ interneurons (Supplementary Fig. 6a–d). Together, these data show that postnatal Dlx1/2 function promotes inhibitory synapses.

Figure 6.

Loss of Dlx1/2 in CR⁺ CINs reduces synaptic boutons. (a, e) Schematic drawings of Dlx1/2 deletion in CR⁺ (green) CINs. The circles indicate the synapses were analyzed onto the axon (a) or the dendrites (e) of neurons in the somatosensory cortex. (b, c) Single confocal images of Dlx1/2^f; GFP; CR-Cre control (b), and mutant (c), showing the colocalization of Synaptophysin⁺ boutons and Gephryn⁺ clusters onto CR axon (arrowheads) at P30. (b’, c’) Binary images after ImageJ software processing on confocal images (b, c) for quantification. (d) Quantification of Synaptophysin⁺ boutons and Gephryn⁺ clusters colocalization onto the CR axon (control: 0.22 ± 0.02 puncta/μm, n = 17; mutant: 0.15 ± 0.009 puncta/μm, n = 19; average ± SEM, P = 0.006). *P < 0.05 (t-test). (f, g) Single confocal images of Dlx1/2^f; GFP; CR-Cre control (f) and mutant (g) showing the colocalization of VGlut1⁺ boutons and PSD95⁺ clusters onto CR dendrite (arrowheads) at P30. (f’, g’) ImageJ software was used to process confocal images (f, g) for quantification. (h) Quantification of VGlut1⁺ boutons and PSD95⁺ clusters colocalization onto the CR dendrite (control: 0.3 ± 0.05 puncta/μm number, n = 15; mutant: 0.2 ± 0.02 puncta/μm, n = 17; P = 0.04). *P < 0.05 (U of Mann–Whitney). Scale bar (in f) b, c, f, g 5 μm. n, number of cells.

Synaptic Excitation is Reduced in CR⁺ CINs in Dlx1/2; CR-Cre CKO

Next we studied excitatory inputs onto CINs by recording excitatory postsynaptic currents (EPSCs). To negate potential homeostatic changes to neural circuitry caused by the loss of CINs, we performed the recordings in the Dlx1/2; CR-Cre CKOs in which survival of CR⁺ CINs was not affected. We analyzed the excitatory currents on CR⁺ interneurons in layer 2/3 of the somatosensory cortex at P10 and P30. At P10 no defects were detected. Spontaneous and miniature EPSCs (sEPSCs and mEPSCs), including frequency, amplitude and decay tau recorded in CR⁺ interneurons were normal (Supplementary Fig. 5e–g).

However, by P30 the mutant CR⁺ interneurons exhibited a deficit in receiving excitatory neurotransmission, which was corroborated by cumulative probability plots (Fig. 3e–h). Other characteristics were normal, including mEPSC amplitude and rise time (Fig. 3g). Similar changes were seen in recordings of sEPSCs (Fig. 3e–f). Additionally, the less severe Dlx1 sensitized CKO (Dlx1^f/−; Dlx2^+/−; CR-Cre) exhibited a similar deficit in excitatory synaptic transmission at P30 (Supplementary Fig. 7).

Dlx1/2 CKO CINs Have Less Elaborate Dendrites

The reduction in sEPSC and mEPSC frequencies onto the CR⁺ cells (Fig. 3e–g and Supplementary Fig. 7c,d) could result from several mechanisms, including reduced numbers of functional excitatory synaptic contacts (addressed in the next section) and defects in dendrite morphology. Indeed, bipolar CR⁺ CINs had reduced dendritic length in Dlx1/2; I12b-Cre CKO (Fig. 7a–c). We did not detect a change in dendritic length of bipolar CR⁺ CINs in the Dlx1/2; CR-Cre CKO (data not shown). Thus, prenatal but not postnatal Dlx1/2 loss of function reduces dendritic length of bipolar CR⁺ CINs.

Figure 7.

Loss of Dlx1/2 function (I12b-Cre) shortens dendrite length of CR⁺ and PV⁺ CINs. (a, b) Immunohistochemistry for CR on coronal sections through the cortices of control (a) and Dlx1/2 mutant (b) P30 mice. (c) Quantification of dendrite morphology of control (black bars) and Dlx1/2 mutants (white bars) (control: 140.2 ± 14.2 μm, n = 21; mutant: 49.9 ± 5.3 μm, n = 28; P = 3.6e−008) ***P < 0.001 (U of Mann–Whitney). Scale bar (in a) a, b 100 μm. (d) Schematic diagram of the experimental design. MGE from E13.5 Dlx1/2^f; GFP; I12b-Cre controls and mutants were dissected, dissociated and transplanted into the cortices of P1 WT mice. Host mice were analyzed 28 DPT (days post-transplant). (e, f) Coronal sections trough the cortex of transplanted mice showing the dendritic morphology of a Dlx1/2^f; GFP; I12b-Cre control (e) and mutant (f) PV⁺ cells after staining of GFP (green) and PV (red) and highlighted by imageJ software. (g, h) Quantification of dendrite length (g) (control: 1420.5 ± 124.7 μm, n = 11; mutant: 896.1 ± 103.1 μm, n = 12; P = 0.004) and dendrite number (h) (control: 15.6 ± 1.06 dendrite number, n = 11; mutant: 13.9 ± 1.7 dendrite number, n = 12; P = 0.2) in controls (black bars) and Dlx1/2 mutant (white bars) after 28 DAT. **P < 0.01 (t-test). Histograms show average ± SEM. n, number of cells. Scale bar (in e) e, f 100 μm.

To further explore the role of prenatal loss of Dlx1/2 on dendrite length, in particular whether this phenotype was cell autonomous and whether it could affect other CIN subtypes, we studied the dendrite morphology of PV⁺ CINs using the MGE transplantation method (Vogt, Wu, et al. 2015). E13.5 MGE cells from either control (Dlx1/2^f/+; I12b-Cre; CAG-GFP) or mutant (Dlx1/2^f/−; I12b-Cre; CAG-GFP) embryos were transplanted into the P1 neocortex of a WT host, and analyzed 28 days post-transplant (DPT) (Fig. 7d–f). We measured 2 dendrite morphology parameters in PV⁺ CINs: total dendrite length and the number of dendrites. The Dlx1/2^f/−; I12b-Cre CINs had shorter dendrites compared to controls (Fig. 7g), whereas the total number of dendrites did not change (Fig. 7h). Together, these results show that Dlx1&2 promote dendrite outgrowth of CR⁺ and PV⁺ CINs, and that this Dlx1/2 function is required prenatally in at least the CR⁺ interneurons.

Dlx1/2 Control Excitatory Synapse Numbers

We next investigated whether postnatal Dlx1/2 expression regulated the number of excitatory synapses on CINs using Dlx1/2; CR-Cre CKO mutants at P30 in somatosensory cortex. We analyzed the number of excitatory terminals (VGLUT1⁺ presynaptic structures) apposed to dendritic postsynaptic zones (PSD95⁺) of control and mutant GFP-expressing CR⁺ CINs (Fig. 6e–g). Confocal fluorescent microscopic analysis showed a ~40% reduction in excitatory synaptic punctae onto CR⁺ dendrites (Fig. 6h). A similar reduction was found in P30 Dlx1 sensitized CKO (Dlx1^f/−; Dlx2^+/−; CR-Cre) CR⁺ interneurons (Supplementary Fig. 6e–h). These results are concordant with the reduction in mEPSC frequency onto interneurons (Fig. 3g–h and Supplementary Fig. 7c,d).

Reduced GRIN2B⁺ Punctae on Dlx1/2 CKO Dendrites

Towards establishing molecular mechanisms through which Dlx1/2 postnatally regulate excitatory synapse number (Fig. 6h) and the changes in excitatory transmission (Fig. 3f–h), we investigated the function and expression of receptors involved in glutamatergic neural transmission. Experimental conditions of mEPSC recordings isolated AMPA receptors. The normal mEPSC kinetics and amplitude provide evidence that synaptic AMPA receptors were normal in these Dlx1/2 mutants.

Next, we explored Dlx1/2 regulation of NMDA receptors. There are 7 NMDA receptor genes encoding GRIN1 (an obligatory subunit for all NMDA receptors), GRIN2A-D, and GRIN3A-B (Paoletti et al. 2013). In the developing cortex GRIN2A and GRIN2B are the major GRIN2 subunits expressed by CINs (Matta et al. 2013). GRIN2B is expressed at high levels throughout CIN development, while high expression and synaptic receptor contribution of GRIN2A start during the second postnatal week (Matta et al. 2013). GRIN2B plays a key role in synaptogenesis and plasticity (Akashi et al. 2009; Espinosa et al. 2009; Paoletti et al. 2013). In fact, loss of GRIN2B in hippocampal interneurons reduces glutamate synapse numbers onto interneurons (Kelsch et al. 2014). Because we did not detect differences in Grin2a expression at P8 by ISH (data not shown), and we found reduced excitatory synapses on CR⁺ cells by P15 in the Dlx1 sensitized CKO (Dlx1^f/−; Dlx2^+/−; CR-Cre) (Supplementary Fig. 6i), we focused on expression of GRIN2B.

We explored GRIN2B expression by immunohistochemistry and confocal fluorescent microscopy in Dlx1/2 CKO mutant CINs (CR-Cre and I12b-Cre). We found reduced GRIN2B expression on P8 and P30 sections of somatosensory cortex (Fig. 8a–d). CR-Cre (P30) and I12b-Cre (P8 and P30) mutants had reduced GRIN2B⁺ punctae on CIN dendrites (Fig. 8e).

Figure 8.

Grin2b rescues the dendritic phenotype of Dlx1/2 CKO CINs. (a–d) Single confocal images of Dlx1/2^f; GFP; CR-Cre (a, b) and Dlx1/2^f; GFP; I12b-Cre (c, d) control (a, c) and mutant (b, d) mice at P30 showing the colocalization of GRIN2B (magenta) boutons onto GFP⁺ dendrite (arrowheads) at P30. (e) Quantification of GRIN2B⁺ boutons density on control (black and dark gray bars) and mutant (white and light gray bars) Dlx1/2^f; GFP; CR-Cre (control: 0.67 ± 0.04 puncta/μm number, n = 23; mutant: 0.42 ± 0.03 puncta/μm, n = 24; P = 1.8e−005) or Dlx1/2^f; GFP; I12b-Cre (control: 0.64 ± 0.06 puncta/μm number, n = 23; mutant: 0.37 ± 0.03 puncta/μm, n = 29; P = 2.4e−004) P30 mice and P8 Dlx1/2^f; GFP; I12b-Cre (control: 0.76 ± 0.06 puncta/μm number, n = 25; mutant: 0.44 ± 0.03 puncta/μm, n = 29; P = 2.0e−005). ***P < 0.001 (t-test). Histograms show average ± SEM. n, number of cells. Scale bar (in b) a–d 10 μm. (f) Schematic diagram of the experimental assay. MGE from E13.5 Dlx1/2^f control and mutants were dissected and dissociated. Cells, transfected with a mixture of CMV-Cre⁺Flex-GFP or CMV-hGRIN2B-T2a-Cre⁺Flex-GFP, were transplanted into cortices of P1 WT mice. Mice were analyzed 28 DAT (days after transplant). (g–i) Images and drawings of PV cells transfected with CMV-Cre⁺Flex-GFP (g, h) or CMV-hGRIN2B-T2a-Cre⁺Flex-GFP (i) from Dlx1/2^f controls (g) or mutants (h, i). (j) Quantification of dendrite length of controls (black bar), mutants (white bar), and human GRIN2B rescued mutants (gray bar) 28 DAT (control: 2824.6 ± 237.3 μm, n = 10; mutant: 1550.1 ± 223.1 μm, n = 10; rescued: 2533.8 ± 163.2 P = 0.0005). ***P < 0.001, **P < 0.01 (ANOVA test, Tukey’s B post hoc). Histograms show average ± SEM. Scale bar (in g) G–I 100 μm. n, number of cells. (k) Diagram showing GE ChIP peaks in the Grin2b gene locus at E13.5 after H3K27Ac ChIP and at E13.5 and E16.5 after DLX2 ChIP. Three DLX2 peaks are identifiable at both ages (#1-3). (l) Luciferase reporter assay in P19 cells testing DLX2 activation of region #3 (orange box in k). Data are presented as fold change in activation with DLX2 over activation with GFP, and normalized to activation of the empty pGL4.23 control (pGL4.23: 1.0 ± 0.08, n = 3, same as control in Figure 5; Grin2b En3: 2.5 ± 0.20, n = 3, P = 0.01). Histograms show average ± SEM. *P < 0.05 (t-test).

To investigate whether reduced GRIN2B affected the function of the remaining synapses at P8–10, we made whole-cell electrophysiological recordings of evoked EPSCs while pharmacologically isolating NMDA currents. NMDA current kinetics and amplitude were unchanged between Dlx1/2; CR-Cre CKO and control cells (Supplementary Fig. 8a,b), providing indirect evidence that the evoked synapses did not have changes in NMDA receptor numbers or subunit composition.

To make a more direct measurement of GRIN2B function at these synapses, we applied the GluN2B-specific blocker ifenprodil (3 micromolar) to the cells (P8–10). Stable measurement of ifenprodil’s effect requires long-term (~45 min) synaptic stimulation (Gray et al. 2011). Unfortunately, we found that the excitatory synapses on CR⁺ CINs at P8–10 exhibited a stimulation-dependent amplitude rundown, confounding the use of ifenprodil for this analysis (Supplementary Fig. 8c). Nonetheless, the histological data showing reduced GRIN2B at P8 (Fig. 8e) provides support for the hypothesis that reduced NMDA receptors in Dlx1/2 CKOs alters early steps of CIN maturation, including dendrite morphogenesis and synaptogenesis (Paoletti et al. 2013).

GRIN2B has been implicated in remodeling actin cytoskeleton (Akashi et al. 2009; Paoletti et al. 2013). Thus, we tested whether the reduced dendrite length of the mutant CINs could be rescued by increasing GRIN2B expression by transfecting a GRIN2B expression vector into E13.5 Dlx1/2^f/− MGE cells. The transfection mixture included a vector that ubiquitously expressed Cre and a Cre-dependent reporter vector (Flex-GFP). MGE cells were transfected and transplanted into the cortex of neonatal WT mice and allowed to mature in vivo for 30 days. Then we assessed the dendritic arbor of PV⁺;GFP⁺ double-labeled CINs. As before, we found that PV⁺Dlx1/2 CKO had reduced dendrite length (Figs 7g and 8g,h,j). This phenotype was largely rescued in CINs transfected with GRIN2B (Fig. 8i,j), supporting a model that Dlx1/2 promote interneuron dendrite complexity, at least in part, through GRIN2B expression.

The decrease in GRIN2B at P8 and P30 suggests Dlx1/2 could regulate Grin2b expression. Thus, we tested whether DLX2 directly binds the Grin2b locus by performing DLX2 ChIP-Seq on E13.5 and E16.5 GEs. We found 3 intragenic DLX2 peaks in the Grin2b locus (Fig. 8k). To gain further evidence for whether these peaks represent candidate active regulatory elements, we performed ChIP-Seq against a histone variant that is associated with active enhancers in vivo in the developing brain (H3K27Ac; (Creyghton et al. 2010; Rada-Iglesias et al. 2011; Nord et al. 2013)). Each of the DLX2 peaks was associated with an H3K27Ac peak; the large H3K27Ac peak at the 5′ end of the gene includes the promoter region.

To test whether DLX2 could modulate the activity of the candidate Grin2b regulatory elements we co-transfected a DLX2 expression vector (pCAGGS-Dlx2) with a pGL4.23-Grin2bEn3-luciferase reporter construct. DLX2 activation of the enhancer resulted in a ~2.5-fold increase in luciferase activity compared with the pGL4.23 empty luciferase control (Fig. 8l). Overall, the data provide evidence that Dlx1/2 may directly promote Grin2b expression, which in turn controls dendrite arborization.

Discussion

Dlx regulation of Cortical Interneuron Development

Dlx1,2,5&6 have redundant and unique functions in the generation, maturation and function of different subtypes of CINs (Supplementary Table 1). Dlx1 is required for the survival of SOM⁺, CR⁺ and VIP⁺ CINs. Dlx1 is not expressed in mature PV⁺ interneurons; nor is it required for their survival (Cobos, Calcagnotto, et al. 2005). Here we used tamoxifen-regulated fate mapping to demonstrate that prenatally Dlx1 is expressed in SVZ and/or immature neurons that later become PV⁺ CINs, whereas neonatally Dlx1 is not expressed in cells that become PV⁺ CINs (Fig. 2). We provided evidence that Dlx2 is expressed in PV⁺ interneurons. Thus, Dlx1/2, Dlx5 and Dlx5/6 mutants effect PV⁺ interneuron development and function (Cho et al. 2015).

Dlx1/2 Mutants Provide Insights into How Genetic Defects in Interneuron Development Disrupt Pallial Function

We propose that Dlx genes are at the top of a regulatory hub governing interneuron development and function. Prenatal loss of Dlx1/2 expression leads to early postnatal CIN death that includes PV interneurons (Fig. 1), reduced excitatory drive on the surviving interneurons (Fig. 6) and reduced inhibitory tone (Fig. 3), in part due to reduced expression of Gad1, Gad2, and Vgat (Fig. 4). Postnatal loss of Dlx1/2 expression leads to reduced excitatory drive (Fig. 3) and reduced GABA synapses (Fig. 6), without interneuron loss (Supplementary Fig. 4). Thus, reduced Dlx function, either prenatally or postnatally, has qualitatively similar ramifications on reducing interneuron functions. Interneuron loss at one month of age in Dlx1 mutants is associated with a delayed onset of inducible seizures at 2 months of age (Cobos, Calcagnotto, et al. 2005) and finally spontaneous seizures and epilepsy at 3 months of age (Howard, Rubenstein and Baraban, unpublished results). More subtle phenotypes are present in Dlx5/6^+/− mice that have normal interneuron numbers; they exhibit deficits in medial prefrontal cortex function and related behaviors, which can be corrected with diazepam (Cho et al. 2015).

Insights into how interneuron defects in Dlx mutants disrupt cortical (A1 and V1) and hippocampal neural systems have been made. Dlx1 mutants have reduced excitatory amplitudes (Howard et al. 2014), consistent with the idea that pallial development with reduced interneuron-mediated inhibition results in a compensatory downscaling of excitatory synapses. In addition, possibly as a result of homeostatic downscaling, the Dlx1 mutant hippocampus exhibited a poised hyperexcitable state, with a marked increase in LTP induced in vitro (Howard et al. 2014). Of note, many of the Dlx1 mutant phenotypes were rescued by interneuron replacement using MGE transplantation (Howard et al. 2014).

Temporal Dependence of Dlx1/2 Regulation of Dlx5&6 Expression and GABAergic Neuronal Production and Migration

Dlx1/2 ^−/− constitutive mutants, and Nkx2-1-Cre induced VZ deletion of Dlx1/2, resulted in lack of Dlx5 expression (Anderson, Qiu, et al. 1997) (Supplementary Fig. 3; Supplementary Table 1). In both cases, CINs failed to migrate to the cortex, and instead coalesced as subpallial ectopia (Anderson, Eisenstat, et al. 1997; Cobos et al. 2007; Long et al. 2007; McKinsey et al. 2013) (Supplementary Fig. 3). Thus, Dlx1/2 function in the VZ, and perhaps in the early SVZ, are required to induce Dlx5/6; the lack of expression of all 4 Dlx genes leads to a profound deficits in the development of telencephalic GABAergic neurons(Anderson, Eisenstat, et al. 1997; Cobos et al. 2007; Long et al. 2007; McKinsey et al. 2013).

On the other hand, deletion of Dlx1/2 in the late SVZ and immature neurons does not lead to loss of Dlx5/6 expression (Supplementary Fig. 3). Thus, Dlx1/2 expression in the VZ and/or early SVZ induces Dlx5/6, whose subsequent expression is maintained independent of Dlx1/2. As a result, loss of Dlx1/2 in the late SVZ (I12b-Cre) and in immature neurons (CR-Cre, PV-Cre) does not affect CIN generation, migration and integration, allowing us to investigate postnatal Dlx functions of these types of Dlx1/2 conditional mutants.

Dlx Dosage and Timing Determine Interneuron Survival

Dlx1 function is required for pallial interneuron survival (Cobos, Calcagnotto, et al. 2005). Here we expanded the information about the relationship of Dlx dosage, and the timing of when Dlx expression is lost, with respect to the magnitude, cell type specificity and timing of interneuron loss (Supplementary Table 1). The major findings are that prenatal loss of both Dlx1/2 using I12b-Cre (beginning in the late SVZ) increased interneuron death around P7, affecting all subtypes, whereas loss of Dlx1 leads to interneuron death around P30. Deletion of Dlx1 prenatally in immature SOM⁺ neurons also led to interneuron death (Supplementary Table 1, data not shown). However, deletion of Dlx2 prenatally (I12b-Cre) or Dlx1/2 genes postnatally (CR-Cre and PV-Cre) did not reduce interneuron numbers. Thus, only prenatal loss of Dlx1/2 reduced interneuron survival. Despite the synaptic, dendritic and physiological defects of Dlx1/2; CR-Cre mutants (Figs 3, 6, and 7; Supplementary Table 1), interneuron survival was not affected.

Dlx1&2 are Required for the GABAergic Phenotype: Gad1, Gad2, and Vgat Expression

Dlx function is required in forebrain progenitors and neurons for the GABAergic phenotype. In Dlx1/2^−/− constitutive mutants, the dorsal LGE fate undergoes a partial fate switch to ventral pallial properties, with reduced Gad1 and Vgat expression, and ectopic expression of Vglut1 (Long et al. 2007). Dlx1/2; Mash1 (Ascl1) and Dlx1/2; Gsx2 compound mutants exhibit an almost complete loss of LGE properties (Long et al. 2007; Wang et al. 2013). The MGE is less sensitive than the LGE to loss of Dlx1/2, perhaps due to maintenance of Nkx2-1 expression (Long, Swan, et al. 2009). Loss of Dlx1/2 beginning in the SVZ (I12b-Cre CKO) leads to maintenance of Dlx5 expression; the mutant interneurons migrate and integrate in the pallium, yet express less Gad1, Gad2, and Vgat (Fig. 4). Therefore, reduced Dlx dosage in GE progenitors and in maturing cortical interneurons reduces Gad1, Gad2, and Vgat expression.

We demonstrated using ChIP-Seq that DLX2 binds to a region upstream of Gad2 in the E13.5 and E16.5 GEs (Fig. 5). Transgenic assays show that this candidate enhancer has E12.5 and P30 activity consistent with Gad2 expression. In vitro transcription assays showed that the enhancer is activated by DLX2 (Fig. 5). Another recent study showed that DLX1/2 can also directly activate Gad1 and Gad2 expression through binding to TAAT/ATTA motifs located within their promoter regions (Le et al. 2017). Thus, there is strong evidence that Dlx1/2 promote the GABAergic phenotype of immature and mature forebrain neurons, in part by direct regulation of Gad2 expression through either an enhancer or the promoter.

Dlx1/2 Function Promotes CIN Dendrite Growth, and Inhibitory and Excitatory Synapses: Potential Role of GRIN2B

All previous Dlx publications assessed phenotypes in mice after the Dlx genes were deleted prenatally. Here, we assessed Dlx1/2 postnatal functions primarily in CR⁺ CINs using CR-Cre, which deleted Dlx1/2 in the early postnatal period. CR⁺Dlx1/2 mutant CINs survived, but formed fewer inhibitory synapses onto pyramidal neurons and received fewer excitatory synapses and less synaptic excitation. In contrast, only prenatal deletion of Dlx1/2 using I12b-Cre resulted in dendritic defects.

We made an inroad into understanding the molecular basis of the Dlx1/2 CKO dendritic defects and excitatory defects by finding fewer GRIN2B⁺ punctae in vivo (Fig. 8a–e) in mutant CIN (I12b-Cre and CR-Cre). Restoring GRIN2B in the mutant CINs rescued their dendrite phenotype (Fig. 8f–j). ChIP-Seq experiments revealed that DLX2 binds within the Grin2b locus in E13.5 and E16.5 GE (Fig. 8k), and in vitro transcription assays showed DLX2 can activate a candidate Grin2b enhancer (Fig. 8l), providing evidence that Grin2b is a direct DLX2 target. Importantly, note that a reduction in GRIN2B⁺ punctae was not sufficient to affect dendrite length in Dlx1/2; CR-Cre CKOs (Fig. 8a,b,e and data not shown), demonstrating that additional molecular defects also contribute to the dendritic phenotype.

While our data support that Dlx1/2 cell autonomously control many of these cellular phenotypes, non-autonomous mechanisms may contribute to the phenotypes, particularly in the CKOs with reduced interneuron numbers. Changes in network activity can feedback on activity-regulated processes, such as synapse formation and maintenance (Turrigiano and Nelson 2004; Cobos, Calcagnotto, et al. 2005; Howard et al. 2014). Furthermore, there is evidence that activity regulates Dlx expression and morphology of cortical interneurons (De Marco Garcia et al. 2011). Therefore, alterations in neuronal activity in the Dlx1/2 CKOs could modify Dlx5 and Dlx6 expression in CINs, further modifying the phenotypes. Note, however, that CR-specific elimination of Dlx1/2 is less likely to induce network hyperexcitability (as CR cells inhibit other CINs), thus supporting the hypothesis that the phenotypes observed in this mutant are largely due to cell autonomous functions of the gene products in the Dlx1/2 transcription pathway.

Grin2b is important in postnatal development (Sheng et al. 1994; Flint et al. 1997). Its high calcium permeability makes it crucial in sensing activity and initiating activity-dependent processes (Lynch et al. 1983; Bliss and Collingridge 1993). It also plays a role in cell survival/cell death (Cull-Candy et al. 2001; Endele et al. 2010). Thus, we hypothesize that Grin2b reduction is linked to Dlx1/2 CKO CIN defects including the reductions in EPSCs, excitatory synapses and dendrite complexity.

Dlx1/2 Mutants Provide Insights into How Genetic Defects in Interneuron Development Contribute to Neuropsychiatric Disorders

The evidence that Grin2b is a target of DLX2 is important because of this receptor’s role in development of excitatory signaling and synaptic plasticity, and because GRIN2B mutations are linked to epileptic encephalopathy, mental retardation, schizophrenia and autism (Endele et al. 2010; Pan et al. 2015). Furthermore, interneuron dysfunction may underlie epilepsy, autism, schizophrenia and Alzheimer’s (Marín 2012; Verret et al. 2012; Gonzalez-Burgos et al. 2015; Hardingham and Do 2016). One model posits that schizophrenia is due to reduced excitation onto interneurons, through defects in the NMDA receptor (Fazzari et al. 2010; del Pino et al. 2013; Hardingham and Do 2016). While Dlx mutations have not been proven to cause neuropsychiatric disorders, we suggest that Dlx alleles could sensitize to disease (Hamilton et al. 2005; Poitras et al. 2010), which is supported by the observation that multiple Dlx-regulated genes are implicated in human brain disorders, including Arx, Gad1, Grin2b and Zfhx1b (OMIM: Online Mendelian Inheritance in Man). Ongoing studies are aimed at elucidating the other DLX2 genomic targets that regulate synapse and dendrite development and function.

Author Contributions

Experiments were largely conducted by RP and AS; particular experiments were done and or supervised by: SL (ChIP-Seq), MAH and SCB (physiology), knockout generation (GBP, DHR), Gad2 enhancer transgenic (ANR and JDP), transcription assays (ANR), ChIP-seq data analysis (ASN and AV), initial data on synapse defects (IC), (DV) molecular cloning and MGE transplantation, NM helped in the PV-Cre characterization. (CES) generated the SOM-Cre;Dlx1/2 CKOs. The project was managed by JLRR who co-wrote the paper with RP, ANR and the other authors.

Funding

This work was supported by the research grants to JLRR from: Nina Ireland endowment to the UCSF Department of Psychiatry, and the National Institute of Mental Health (R37 MH049428); to RP from Spanish Ministry of Science and Innovation (ME/BMED-2010-0787); to AS from the National Institute of Health (T32 MH089920); to MAH from the Dravet Syndrome Foundation (Postdoctoral Research Fellowship); to SCB from the National Institute of Neurological Disorders and Stroke (R37 NS071785); to CES from the National Institute of Health (2R01DC002260-21); to AV from the National Institute of Health (R01HG003988, U54HG006997, and U01DE024427).