TrkB Signaling Influences Gene Expression in Cortistatin-Expressing Interneurons

Abstract Brain-derived neurotrophic factor (BDNF) signals through its cognate receptor tropomyosin receptor kinase B (TrkB) to promote the function of several classes of inhibitory interneurons. We previously reported that loss of BDNF–TrkB signaling in cortistatin (Cort)-expressing interneurons leads to behavioral hyperactivity and spontaneous seizures in mice. We performed bulk RNA sequencing (RNA-seq) from the cortex of mice with disruption of BDNF–TrkB signaling in cortistatin interneurons, and identified differential expression of genes important for excitatory neuron function. Using translating ribosome affinity purification and RNA-seq, we define a molecular profile for Cort-expressing inhibitory neurons and subsequently compare the translatome of normal and TrkB-depleted Cort neurons, revealing alterations in calcium signaling and axon development. Several of the genes enriched in Cort neurons and differentially expressed in TrkB-depleted neurons are also implicated in autism and epilepsy. Our findings highlight TrkB-dependent molecular pathways as critical for the maturation of inhibitory interneurons and support the hypothesis that loss of BDNF signaling in Cort interneurons leads to altered excitatory/inhibitory balance.


Introduction
Signaling of brain-derived neurotrophic factor (BDNF) via its transmembrane receptor tropomyosin receptor kinase B (TrkB) plays a significant role in the maturation and function of inhibitory neurons in the cortex and hippocampus (Yamada et al., 2002;Alcántara et al., 2006). Although inhibitory GABAergic interneurons represent only 10 -15% of neurons in the rodent cortex (Meyer et al., 2011), they are highly heterogeneous, differing in morphology, firing patterns, response to neuromodulators, and molecular profiles (Tremblay et al., 2016). At least 26 different types of GABAergic interneurons have been identified in the hippocampus (Somogyi et al., 2004), and perhaps more in the cerebral cortex (Myers et al., 2007;Habib et al., 2017). Differences in firing properties, connectivity patterns, and molecular expression profiles are hypothesized to contribute to nonoverlapping functions of the respective classes.
BDNF-TrkB signaling plays an important role in the development of several classes of inhibitory interneurons. For example, BDNF regulates the differentiation and morphology of hippocampal interneurons (Marty et al., 1996), and BDNF deletion leads to reduction in several neuropeptide transcripts that define GABAergic populations, including somatostatin (SST), neuropeptide Y (NPY), substance P, and cortistatin (Cort) in the cortex (Glorioso et al., 2006;Martinowich et al., 2011). BDNF decreases the excitability of parvalbumin (Pvalb) interneurons in the dentate gyrus (Nieto-Gonzalez and Jensen, 2013), and accelerates their maturation in the visual cortex (Huang et al., 1999). While BDNF is expressed primarily in excitatory pyramidal neurons, but not in inhibitory interneurons, its receptor TrkB is widely expressed in both excitatory and inhibitory neurons (Cellerino et al., 1996;Gorba and Wahle, 1999;Swanwick et al., 2004). Levels of TrkB expression across different interneuron classes have not been explicitly quantified, but we previously reported that ϳ50% of Cort-expressing interneurons express TrkB in the cortex (Hill et al., 2019).
Cortistatin is a secreted neuropeptide that is expressed in a distinct set of interneurons. This population partially overlaps with both Pvalb-and SST-expressing inhibitory interneurons, but its expression is seen prominently in the cerebral cortex and hippocampus (de Lecea et al., 1997). Cortistatin is similar in structure to SST and can bind all fived cloned somatostatin receptors (Veber et al., 1979;Csaba and Dournaud, 2001). However, Cort possesses some notably distinct functions, including its ability to induce slow-wave sleep activity (de Lecea et al., 1996) and regulated synaptic integration by augmenting the hyperpolarization-activated current I H (Schweitzer et al., 2003). Cortistatin is expressed earlier than most inhibitory neuron markers in the brain, peaking at 2 weeks of age in rodents (de Lecea et al., 1997), which closely parallels the pattern of BDNF expression during neurodevelopment (Katoh-Semba et al., 1997). Reductions in BDNF signaling are associated with decreased expression of Cort transcripts (Martinowich et al., 2011;Guilloux et al., 2012), and conversely, the administration of cortistatin increases BDNF expression (Souza-Moreira et al., 2013). We previously demonstrated that TrkB expression in Cort interneurons is required to suppress cortical hyperexcitability. Specifically, mice in which TrkB is depleted in Cort interneurons develop spontaneous seizures and die ϳ1 month after birth. Before developing seizures, these mice sleep for significantly less time and display hyperlocomotion (Hill et al., 2019). While this study established that TrkB signaling in Cort interneurons is critical to maintain appropriate levels of cortical excitability, the molecular mechanisms mediating Cort interneuron dysfunction downstream of TrkB signaling remain known.
To better understand the molecular mechanisms by which BDNF-TrkB signaling influences Cort interneuron development and function, we investigated the impact of TrkB deletion in these cells on the Cort interneuron transcriptome as well as gene expression in the surrounding cellular milieu. Translating ribosome affinity purification (TRAP) has been used to identify molecular profiles for many cell types in the mouse brain, including Cort interneurons (Doyle et al., 2008). Here, we used TRAP to assess how TrkB deletion impacts the molecular profile of these cells, and bulk RNA-sequencing (RNA-seq) to assess how this perturbation affects the surrounding milieu. Using this strategy, we identified several differentially regulated genes, including those encoding molecules important for calcium signaling as well as molecules that influence inhibitory/excitatory balance. Identification of the TrkB-dependent gene pathways that support Cort interneuron function contributes to our understanding of cortical hyperexcitability, which is important because changes in cortical excitability have been implicated in several brain disorders, including epilepsy and autism (Wang et al., 2013;van Diessen et al., 2015).
For bulk homogenate RNA-seq experiments, the groups were postnatal day 21 (P21) Cort Cre or TrkB flox/flox (control group contained both genotypes) and Cort Cre ; TrkB flox/flox (experimental group). As seizure onset begins at P21 (Hill et al., 2019) All mice were housed in a temperature-controlled environment with a 12 h light/dark cycle and ad libitum access to standard laboratory chow and water. Mice were group housed based on genotype. All experimental animal procedures were approved by the SoBran Biosciences Institutional Animal Care and Use Committee. Male and female mice were included and analyzed for all experiments.

RNA extraction and quantitative PCR
Mice were cervically dislocated, and cortices were flash frozen in isopentane. For bulk homogenate experiments in P21 control and Cort Cre ;TrkB flox/flox mice, RNA was extracted using Life Technologies TRIzol (Thermo Fisher Scientific), purified using RNeasy minicolumns (Qiagen), and quantified using a Nanodrop spectrophotometer (Agilent Technologies). RNA concentrations were normalized and reversed transcribed using Life Technologies Superscript III (Thermo Fisher Scientific). Quantitative PCR (qPCR) was performed using a Realplex Thermocycler (Eppendorf) with Life Technologies GEMM mastermix (Thermo Fisher Scientific) and 40 ng of synthesized cDNA. Individual mRNA levels were normalized for each well to Gapdh mRNA levels. For validation of genes differentially expressed in control and experimental Ribotag samples, cDNA was synthesized using the Ovation RNA Amplification System V2 Kit (described below), and qPCR was performed as above. TaqMan probes were commercially available from Thermo Fisher Scientific (Gad1 Mm00725661_s1, Cort Mm00432631_m1, Gfap Mm01253033_m1, Wt1 Mm01337048_m1; Cxcr4 Mm01292123_m1; Calb1 Mm00486647_m1; Lgals1 Mm00839408_g1; Trpc6 Mm01176083_m1; Syt6 Mm04932997_m1; Gng4 Mm00772342_m1; Ttc9b Mm01176446_m1; S100a10 Mm00501458_g1; Nxph1 Mm01165166_m1; and Syt2 Mm00436864_m1; Gsn Mm00456679_m1) or as described in the study by Martinowich et al. (2011). Statistical analysis was conducted using GraphPad Prism (GraphPad Software). Comparisons between two groups were performed using unpaired Student's t test. Data are presented as the mean Ϯ SEM and statistical significance was set at ‫ء‬p Ͻ 0.05, ‫‪p‬ءء‬ Ͻ 0.01, ‫‪p‬ءءء‬ Ͻ 0.001, ‫‪p‬ءءءء‬ Ͻ 0.0001.

RNAscope single-molecule fluorescent in situ hybridization
Control and Cort Cre ;TrkB flox/flox P21 mice were cervically dislocated and the brains were removed from the skull, flash frozen in isopentane, and stored at Ϫ80°C.
Brain tissue was equilibrated to Ϫ20°C in a cryostat (Leica), and serial sections of cortex were collected at 16 m. Sections were stored at Ϫ80°C until completion of the RNAScope assay. We performed single-molecule fluorescent in situ hybridization using the RNAscope Fluorescent Multiplex Kit version 2 [catalog #323100, Advanced Cell Diagnostics (ACD)] according to the study by Colliva et al. (2018). Briefly, tissue sections were fixed with a 10% neutral buffered formalin solution (catalog #HT501128, Sigma-Aldrich) for 20 min at room temperature and pretreated with protease for 20 min. Sections were incubated with commercially available Wt1 (catalog #432711, ACD) and Cre (catalog #312281-C2, ACD) probes. Probes were fluorescently labeled with orange (excitation, 550 nm), green (excitation, 488 nm), or far red (excitation, 647) fluorophores using the Amp 4 Alt B-FL. Confocal images were acquired in z-series at 63ϫ magnification using a Zeiss 700LSM confocal microscope. Images were blinded, and transcript colocalization was quantified using custom MATLAB functions. Briefly, cell nuclei were isolated from the DAPI channel using the cellsegm toolbox (Gaussian smoothening, adaptive thresholding, and splitting of oversized segmented nuclei; Hodneland et al., 2013). Once centers and boundaries of individual cells were isolated, an intensity threshold was set for transcript detection, and watershed segmentation was used to split detected pixel clusters in each channel into identified transcripts. Custom MATLAB functions were then used to determine the size of each detected transcript (regionprops3 function in Image Processing toolbox). Each transcript was then assigned to a nucleus based on its position in three dimensions. Transcripts with centers outside the boundaries of a nucleus were excluded from further analysis. A cell was considered to be positive for a gene if more than two transcripts were present.

Bulk cortex RNA-Seq
Cortices of control (n ϭ 5) and Cort Cre ;TrkB flox/flox (n ϭ 5) mice were collected and flash frozen in isopentane. RNA was extracted from one hemisphere of each animal using Life Technologies TRIzol (Thermo Fisher Scientific), purified with RNeasy minicolumns (Qiagen), and quantified using Nanodrop. The Nextera XT DNA Library Preparation Kit was used to generate sequencing libraries according to manufacturer instructions. Samples were sequenced on the HiSeq2000 (Illumina).

Ribotag and RNA-Seq of Cort interneurons
Cortices of Cort Cre ; Rpl22 HA mice (n ϭ 3) were collected and flash frozen in isopentane. For each sample (n ϭ 3 Input, n ϭ 3 IP), one hemisphere of the cortex from each animal was homogenized according to previously described protocols (Sanz et al., 2013). An aliquot of homogenate was flash frozen and reserved for "Input" samples. Ribosome-mRNA complexes ("IP" samples) were affinity purified using a mouse monoclonal HA antibody (MMS-101R, Covance; RRID:AB_2565334) and Pierce A/G magnetic beads (catalog #88803, Thermo Fisher Scientific). RNA from Input and IP samples was purified using RNeasy microcolumns (Qiagen) and quantified using the Invitrogen Ribogreen RNA Assay Kit (catalog #R11490, Thermo Fisher Scientific). Sequencing libraries were prepared using the SMARTer Stranded RNA-Seq Kit (Clontech) and sequenced on the HiSeq2000 (Illumina).

Ribotag and RNA-Seq of Cort interneurons following disruption of BDNF-TrkB signaling
Cortices of control (n ϭ 6) or Cort Cre ;TrkB flox/flox ; Rpl22 HA (n ϭ 6) mice were collected and flash frozen in isopentane. One hemisphere of the cortex from each animal was homogenized according to previously described protocols (Sanz et al., 2013). Sixty-five microliters of total homogenate was flash frozen and reserved for Input samples. Ribosome-mRNA complexes (IP samples) were affinity purified using a mouse monoclonal HA antibody (MMS-101R, Covance) and PierceA/G magnetic beads (88803, Thermo Fisher Scientific). RNA from Input and IP samples was purified using RNeasy microcolumns (Qiagen) and quantified using the Invitrogen Ribogreen RNA Assay Kit (R11490 Thermo Fisher Scientific). The Ovation RNA Amplification System V2 Kit (7102, NuGEN) was used to amplify cDNA from 10 ng of RNA according to manufacturer instructions. cDNA was used for qPCR validation for Cort enrichment in IP versus Input samples. Sequencing libraries were generated with the Ovation SoLo RNA-seq System Mouse (0502-32, NuGEN) according to manufacturer instructions from 10 ng of RNA. Library concentration was quantified using the KAPA Library Quantification Kit (KR0405, KAPA Biosystems). Libraries were sequenced using the MiSeq Reagent Kit v3 (MS-102-3001, Illumina) and NuGEN Custom SoLo primer.

RNA-Seq data processing and analyses
RNA-seq reads from all experiments were aligned and quantified using a common processing pipeline. Reads were aligned to the mm10 genome using the HISAT2 splice-aware aligner (Kim et al., 2015), and alignments overlapping genes were counted using featureCounts version 1.5.0-p3 (Liao et al., 2014) relative to Gencode version M11 (118,925 transcripts across 48,709 genes; March 2016). Differential expression analyses were performed on gene counts using the voom approach (Law et al., 2014) in the limma R/Bioconductor package (Ritchie et al., 2015) using weighted trimmed means normalization factors using the statistical models described below (Table 1). For each analysis, multiple testing correction was performed using the Benjamini-Hochberg approach to control for the false discovery rate (FDR; Kasen et al., 1990). Gene set enrichment analyses were performed on marginally significant genes using the subset of genes with known Entrez gene IDs against a background of all expressed genes using the clusterProfiler R Bioconductor package, which uses the hypergeometric test (Yu et al., 2012).
Cross-species enrichment analyses of the human SFARI (Banerjee-Basu and Packer, 2010) and Harmonizome (Rouillard et al., 2016) gene sets were performed with Fisher exact tests (which is identical to the above hypergeometric test on these 2 ϫ 2 enrichment tables) on the subsets of homologous and expressed genes in each mouse dataset. For SFARI analyses, we considered the sets of (1) all genes in the mouse model database, (2) all genes in the human gene database (N ϭ 1079 genes), (3) only genes that were syndromic or had gene scores of 1 or 2 (N ϭ 235 genes, which correspond to high-confidence genes), and (4) only genes that had gene scores of 1 or 2, ignoring syndromic genes (N ϭ 91 genes). All RNA-seq analysis code is available on GitHub: https://github.com/Lieber-Institute/cst_trap_seq. Raw RNA-seq reads are available at BioProject Accession PRJNA602667.

Bulk cortex analysis for genotype effects
We used paired end read alignment and gene counting for these 10 samples (5 per genotype group). We analyzed 21,717 genes with reads per kilobase per million counted/ assigned (RPKM normalizing to total number of gene counts, not mapped reads) Ͼ0.1. We performed differential expression analysis with limma voom using genotype as the main outcome of interest, further adjusting for the gene assignment rate (measured by featureCounts), the chrM mapping rate, and one surrogate variable.

Input versus IP analysis
We used paired end read alignment and gene counting for these six samples (three input and three IP). We analyzed 21,776 genes with RPKM Ͼ 0.1. We performed differential expression analysis with limma voom using fraction (IP vs Input) as the main outcome of interest, further adjusting for the gene assignment rate (measured by featureCounts) and also using the duplicateCorrelation function in limma to treat each mouse as a random intercept by using linear mixed-effects modeling.

IP analysis for genotype effects
We used single end read alignment and gene counting for these 12 samples (6 per genotype). We analyzed 21,187 genes with RPKM Ͼ 0.1. We performed differential expression analysis with limma voom using genotype as the main outcome of interest, further adjusting for the gene assignment rate (measured by featureCounts).

Translatome profiling delineates a comprehensive molecular identity for Cort-expressing interneurons in the cortex
Given the critical role of Cort neurons in maintaining cortical excitatory/inhibitory balance, we sought to better understand the molecular profile of Cort neurons using TRAP followed by RNA-seq. We first crossed mice expressing Cre recombinase under control of the cortistatin promoter (Cort Cre ) to mice expressing a Cre-dependent HA peptide tag on the RPL22 ribosomal subunit (Rpl22 HA ; Fig. 2A) to allow for HA tagging of ribosomes selectively in Cort neurons. Tagged ribosomes were immunoprecipitated (IP) from cortical homogenate tissue (Input) using an anti-HA antibody. Ribosome-associated RNA was isolated from IP samples and total RNA was isolated from Input samples.
To discover the potential functional significance of the mRNAs enriched and depleted in Cort-expressing interneurons in the cortex, we performed GO analysis on the subset of 848 Bonferroni-significant genes differentially expressed in Cort IP compared with Input with Entrez gene IDs, stratified by directionality (440 more highly expressed in IP, 408 more highly expressed in Input; Fig.  2D, Extended Data Fig. 2-3). Genes enriched in IP fractions are involved in cellular component category terms such as axon part and neuron projection terminus and in biological process category terms such as synapse organization and cerebral cortex tangential migration. Genes enriched in Input fractions are involved in cellular component category terms such as myelin sheath and biological processes category terms such as gliogenesis. Deenrichment of myelin and gliogenesis pathways would be expected in the neuronal IP fractions, and hence further validate our approach.

Loss of BDNF-TrkB signaling in Cort interneurons impacts genes critical for structural and functional plasticity
To better understand signaling pathways and cellular functions modulated by BDNF-TrkB signaling in Cort cells, we performed TRAP followed by RNA-seq in TrkBdepleted Cort interneurons. We intercrossed Cort Cre ; Rpl22 HA mice to mice expressing a floxed TrkB allele (TrkB flox/flox ) to allow for HA tagging of ribosomes in control Cort interneurons (Cort Cre ; Rpl22 HA ) or TrkB-depleted Cort interneurons (Cort Cre ;TrkB flox/flox ;Rpl22 HA ; referred to hereafter as Cort Cre ; TrkB flox/flox ; Fig. 3A). For control and experimental animals (n ϭ 6 each), tagged ribosomes were selectively immunoprecipitated (IP) from cortical homogenate tissue (Input) using an anti-HA antibody. Ribosome-associated RNA was isolated from IP samples and total RNA was isolated from Input samples.
To explore the functional significance of the mRNAs enriched and depleted in Cort interneurons with disrupted BDNF-TrkB signaling, we performed GO analysis on the subset of 161 Entrez genes more highly expressed in TrkB-depleted Cort neurons and the 269 Entrez genes more highly expressed in control Cort neurons (Fig. 3D, Extended Data Fig. 3-2). Terms in both the molecular function and biological processes categories showed that cortistatin interneurons with disrupted BDNF-TrkB signaling were depleted for calcium ion binding, cellular calcium ion homeostasis, and calcium ion transport (FDR Ͻ 0.05). Ion channel complex and axon part, two cellular component category terms, were both depleted and enriched in Cort Cre ;TrkB flox/flox interneurons, which indicates that disrupting BDNF-TrkB signaling modulates important cellular responses. For the biological process terms, positive regulation of neuron projection development and axon development were both enriched and depleted in those interneurons. Terms associated with both enrichment and depletion in Cort interneurons include different genes, which suggests that these cells may undergo gene-specific changes that support their ability to respond to different cellular signaling pathways. Together, these results support the hypothesis that BDNF-TrkB signaling regulates structural and functional plasticity in Cort interneurons to maintain excitatory/inhibitory balance.

Genes important for Cort neuron identity and function overlap with those identified in autism spectrum disorder
Finally, we explored the potential clinical relevance of deficits in Cort neuron function using predefined genes sets from autism-sequencing studies [using Simons Foundation Autism Research Initiative (SFARI)] and disease ontologies (using Harmonizome). We found no significant enrichment in our bulk RNA-seq data with those identified in both autism spectrum disorder (ASD) and animal models relevant for ASD by the SFARI (Banerjee-Basu and Packer, 2010). However, we found enrichment for ASD genes among those genes highly expressed in Cort interneurons (Extended Data Fig. 3-3). For example, of the 239 genes in the "Mouse models" SFARI database expressed in our data, 27 (11.0%) were differentially expressed in Cort interneurons compared with total cortex, constituting a 6.3-fold enrichment (p ϭ 9.25 ϫ 10 Ϫ13 ). Similarly, of the 937 genes in the "Human gene" SFARI database (which contains genes with rare variations associated with ASD from sequencing studies) with homologs expressed in our data, 93 were differentially expressed (9.9%, 4.2-fold enrichment; p ϭ 9.03 ϫ 10 Ϫ25 ). These enrichments were preserved in the more stringent subset of ASD genes, either with [odds ratio (OR) ϭ 5.71, p ϭ 8.98 ϫ 10 Ϫ14 ] or without (OR ϭ 6.55, p ϭ 8.06 ϫ 10 Ϫ8 ) syndromic genes (see Materials and Methods). We further found significant enrichment for the overlap of genes differentially expressed in TrkB-depleted Cort cells compared with control Cort cells with those identified in both human and animal models of ASD as identified by SFARI (Extended Data Fig. 3-3). Here, of the 237 genes in the Mouse models SFARI database expressed in our data, 26 (11.0%) were differentially expressed in Cort Cre ; TrkB flox/flox mice compared with control, constituting a sixfold enrichment (p ϭ 6.85 ϫ 10 Ϫ12 ). Similarly, of the 917 genes in the Human gene SFARI database with homologs expressed in our data, 66 were differentially expressed (7.2%, 2.8-fold enrichment, p ϭ 3.06 ϫ 10 Ϫ11 ), which were preserved in the smaller subset of more stringent ASD genes with (OR ϭ 2.83, p ϭ 0.002) or without (OR ϭ 2.68, p ϭ 0.02) syndromic genes (see Materials and Methods).
In addition to the overlap of ASD-relevant genes with those important for Cort identity and function, there was significant enrichment of many gene sets related to psychiatric disorders (at both the diseases and endophenotype levels) in the Harmonizome database (Rouillard et al., 2016) with those sets of genes preferentially expressed in Cort neurons and those dysregulated following TrkB depletion (Extended Data Fig. 3-4). In addition to enrichment for psychiatric disorders, we further found enrichment of epilepsy-related genes among TrkB-depleted and control Cort neurons (35 genes, p ϭ 1.93 ϫ 10 Ϫ12 ; Extended Data Fig. 3-4). Together, these results further implicate Cort neurons in several debilitating human brain disorders (Xu et al., 2014).

TrkB signaling in Cort interneurons regulates gene pathways that modulate cortical excitability
To better understand how disrupting BDNF-TrkB signaling in Cort cells impairs cortical function, we performed bulk RNA-seq on cortical tissue derived from Cort Cre ; TrkB flox/flox and control mice and identified significant differential expression of genes important for excitatory neuron function. Pathway analysis of these differentially expressed genes revealed functions associated with glutamatergic synapses and synaptic membranes (Fig. 1C). For example, we observed altered expression of neuronal pentraxin II (Nptx2; log2FC ϭ 1.26, p ϭ 3.60 ϫ 10 Ϫ5 ), which encodes a synaptic protein implicated in excitatory synapse formation and neural plasticity (Gu et al., 2013) that is bidirectionally regulated by BDNF in hippocampal neurons both in vitro and in vivo (Mariga et al., 2015). Our dataset also shows increases in cAMP-responsive element Binding Protein 3 Like 1, which is necessary and sufficient to activate Nptx2 transcription after BDNF treatment (Mariga et al., 2015). The protein encoded by Nptx2 is also directly implicated in BDNF-mediated modulation of glutamatergic synapses, where it facilitates targeting and stabilization of AMPA receptors on excitatory synapses (Chang et al., 2010;Martin and Finsterwald, 2011;Pelkey et al., 2015). Npy is another differentially expressed gene (log2FC ϭ 0.75, p ϭ 2.24 ϫ 10 Ϫ5 ) that influences cortical excitability by reducing excitatory transmission onto neurons in the lateral habenula (Cheon et al., 2019) and inhibiting glutamatergic synaptic transmission in the hippocampus (Xapelli et al., 2008). NPY expression can slow the spread of seizures and has neuroprotective effects against excitotoxicity via increased BDNF signaling (Richichi et al., 2004;Xapelli et al., 2008).
Bdnf transcripts are paradoxically upregulated when comparing Cort Cre ;TrkB flox/flox to controls in the bulk RNAseq dataset (log2FC ϭ 0.95, p ϭ 3.55 ϫ 10 Ϫ5 ). Because TrkB receptors were selectively depleted from Cort interneurons, which do not synthesize BDNF (Gorba and Wahle, 1999;Swanwick et al., 2004), Bdnf increases likely result from upregulation in cortical excitatory neurons. Bdnf expression may be induced in excitatory neurons following the loss of TrkB in Cort interneurons for several reasons. First, in Cort Cre ;TrkB flox/flox mice, impaired Cort interneuron function may facilitate disinhibition of excitatory neurons leading to increased cortical excitability and subsequent activity-induced Bdnf expression (Lu, 2003). Alternatively, increased Bdnf expression may be a compensatory mechanism attempting to counterbalance TrkB depletion in cortistatin cells. BDNF levels increase following seizures Isackson et al., 1991;Mudò et al., 1996), and increases in BDNF can subsequently contribute to hyperexcitability and seizure propagation (Kokaia et al., 1995;Scharfman, 1997;Binder et al., 1999;Croll et al., 1999). Therefore, initiation and progres- sive worsening of seizures seen in Cort Cre ;TrkB flox/flox mice could be exacerbated by increases in Bdnf. It should be noted that at the time of brain extraction (P21), mild seizures may have already have begun and could be influencing gene expression. In summary, depletion of TrkB receptors from Cort inhibitory interneurons may disrupt inhibitory signaling, leading to disinhibition of cortical excitatory neurons and disruption of network activity. This imbalance may push the cortex toward elevated excitation and increased expression of activity-regulated genes such as Bdnf, Nptx2, and Npy.

Cortistatin neurons are enriched in genes relevant to ASD
Translatome profiling in Cort neurons showed enrichment of neuron-relevant genes such as Syt2, a synaptic vesicle membrane protein (Bornschein and Schmidt, 2018) and Nxph1, a protein important for dendrite-axon adhesion (Born et al., 2014). We also observed expected depletion of genes such as Mal, which is implicated in myelination (Schaeren-Wiemers et al., 1995a), and Apoe, which is synthesized in astrocytes (Holtzman et al., 2012). These data expand on a similar translatome profiling experiment previously performed by Doyle et al. (2008) using different mouse models and methodology. In that study, investigators used a mouse in which the EGFP-L10a ribosomal fusion protein is expressed under control of the Cort promoter in a bacterial artificial chromosome, and gene expression data were obtained using a microarray approach combined with TRAP. Here, we used a mouse that expresses Cre from the endogenous Cort promoter, and gene expression data were obtained using a Ribotag/ RNA-seq approach. Reassuringly, there is significant overlap between the Doyle microarray dataset and our RNA-seq analysis (Extended Data Fig. 2-1). Xu et al. (2014) showed candidate autism genes from human genetics studies are enriched in Cort cells, supporting the notion that cortical interneurons play a significant role in the etiology of ASD. Epilepsy, a common neurologic disorder characterized by recurrent seizures, is highly comorbid with ASD (Viscidi et al., 2013), and it has been proposed that these disorders may have overlapping genetic risk that points to shared underlying molecular and cellular mechanisms. Of note, interneuron dysfunction has been identified as a potential shared cellular mechanism in mouse models of both disorders (Jacob, 2016). Our results further demonstrate enrichment of genes associated with epilepsy and ASD in Cort neurons and highlight differential expression of several ASD and epilepsy genes in Cort neurons following the disruption of TrkB signaling. Our findings support the overlapping developmental origins of the two illnesses and highlight BDNF-TrkB signaling as potentially relevant to their etiology.

Genes associated with calcium signaling and axonal development are disrupted following TrkB depletion in cortistatin interneurons
To identify putative molecular mechanisms that contribute to Cort interneuron dysfunction in Cort Cre ;TrkB flox/flox mice, we compared the translatomes of intact Cort interneurons and Cort interneurons depleted of TrkB receptors. TrkB-depleted Cort neurons show dysregulation of genes associated with calcium ion homeostasis (Calb1, calcium binding protein; Schmidt, 2012) or calcium-dependent functions (Syt6, calcium dependent exocytosis; Fukuda et al., 2003), as well as genes associated with axon development (Robo1, axon guidance; Andrews et al., 2006) and cell-cell or cell-matrix interactions (Lgals1, plasma membrane adhesion molecule; Camby et al., 2006).
During development, cortical interneurons are generated in the ventral subcortical telencephalon and travel long distances to reach their final destination in cortical circuits, both tangentially from their birthplace in the ganglionic eminences and radially to their correct laminar position (Cooper, 2013). Chemokine signaling is important for the transition from tangential to radial migration, and the expression of chemokine receptors is directly affected by BDNF-TrkB signaling in the central nervous system, as well as in disease states such as cancer (Azoulay et al., 2018). We found that expression of Cxcr4, a chemokine receptor, is reduced in TrkB-depleted cortistatin interneurons by a factor of 4, which supports previous work showing modulation of CXCR4 expression and receptor internalization by BDNF-TrkB signaling (Ahmed et al., 2008). Degradation of this protein has been identified as a permissive signal for interneurons to leave tangential migratory streams (Sánchez-Alcañiz et al., 2011). Deletion of the gene leads to defects in cortical layer positioning (Li et al., 2008;Wang et al., 2011) and mutations result in premature accumulation of interneurons in the cortex. Although laminar distribution of Cort cells does not appear to be significantly altered by loss of TrkB (Hill et al., 2019), premature entry into the cortex may result in incorrect integration into the circuitry or improper axonal projections that cannot be inferred by laminar position. This explanation is further supported by altered expression of genes associated with axonogenesis, axon guidance, neuron projection terminus, and cell-matrix interactions (Fig. 3). The fact that CXCR4 is normally expressed in axons and functions to define their trajectory (Lieberam et al., 2005;Vilz et al., 2005;Miyasaka et al., 2007) provides additional strength to this hypothesis. In addition to Cxcr4, calcium signaling is important for stimulating (Behar et al., 1999) and halting (Bortone and Polleux, 2009) neuronal migration to the cortex, and Cort Cre ;TrkB flox/flox mice show decreased expression of genes in calciumrelated GO categories compared with control mice (Fig.  3), such as Calb1. Importantly, exogenous application of BDNF induces the elevation of intracellular calcium (Berninger et al., 1993;Marsh and Palfrey, 1996), and endogenous BDNF signaling elicits calcium responses at synapses (Lang et al., 2007). Additional work would be necessary to tease out the effects of interneuron migration, migratory stream maintenance, and correct development of projections during embryonic development in these mutant mice. An important future direction will be to evaluate the morphology of Cort neurons following the disruption of BDNF-TrkB signaling.
In summary, we provide evidence that the loss of BDNF-TrkB signaling in Cort interneurons leads to alterations in calcium signaling and axon development in these cells, which may contribute to altered excitatory/inhibitory balance in the cortex. Several of the genes enriched in Cort neurons and differentially expressed in TrkBdepleted neurons are implicated in both ASD and epilepsy. These data shed light on the role of BDNF-TrkB signaling in the function of Cort-expressing interneurons and provide the rationale for further functional studies of these interneurons.