In order to identify Otx2 binding sites in the genome of postnatal mouse cortex, we attempted ChIP-seq analysis. Previous studies have indicated that cortical Otx2 proteins are provided by extracellular sources and transported mainly into PV cells throughout the cerebral cortex (Sugiyama et al., 2008; Spatazza et al., 2013). Consistent with these studies, Otx2 proteins in juvenile (P28) mice were enriched mostly in PV cells in the cingulate, primary motor and somatosensory areas as well as in the primary visual cortex (V1) (Figure 1A). Some Otx2 protein was distributed outside GAD67-positive interneurons, suggesting the existence of Otx2-positive pyramidal neurons (Figure 1A); however, we expected that the Otx2 ChIP signals would represent binding events mainly in interneurons. Because of the low abundance of cortical Otx2 and the similar populations of Otx2-positive cells across the cortex, ChIP-seq analysis was performed using a large part of the P28 cortex, in which neuronal circuits are more plastic during postnatal development than in adulthood. To increase reliability, we used three independent antibodies (see Section Materials and Methods), two of which (CH and Operon2) can be used in immunohistochemistry studies to detect nuclear Otx2 within cortical PV cells, and the other (R&D) having been used in previous ChIP-seq studies in different systems (Samuel et al., 2014; Hoch et al., 2015). The quality of each ChIP was confirmed by ChIP-qPCR analysis to detect enrichment of the Otx2 binding sites identified in our preliminary ChIP-seq trial (Figure S1).

In ChIP-seq analysis, peak calling in comparison to input was carried out using MACS software (Figure 1B, see Section Materials and Methods) and 13,596, 8,845, and 4,332 peaks were obtained for samples using CH, Operon2, and R&D antibodies, respectively. We focused on the peaks that overlapped between ChIP samples from at least two antibodies so that relatively weak but important peaks could be included while excluding false positives (Figure S2A). Under this consensus peak set, 5,490 peaks were identified as Otx2 binding sites in P28 cortex. We searched for peaks containing the consensus binding sequences of Otx2 and found that 1,273 (23.2%) possessed the major target sequence 5′-TAATCC-3′. In addition, two related motifs, 5′-TAAGCC-3′ and 5′-TAACCC-3′ (Chatelain et al., 2006), were found in 1,369 and 1,262 peaks, respectively. In total, 2,253 (41%) of our common peak set contained Otx2 consensus and/or related motifs, confirming the reliability of the data. As expected from Otx2 homeoprotein being a transcription factor, there were prominent peak signals at specific gene promoter regions over the mouse genomic map (Figure 1B). Analyses of peak positions relative to genes demonstrated that the majority of peaks were within genes (genic: 52%) and in the 5 kb region upstream (34%; Figure 1C). Notably, when peak locations were compared to transcription start sites (TSS), nearly half of the peaks were positioned within 5 kb downstream of a TSS, suggesting that most “genic” signals were located near TSS (Figure 1D). Examining the peaks both within gene and in the 5 kb upstream/downstream vicinity, a total of 5,199 genes were annotated to have Otx2 binding sites.

To characterize Otx2-bound genes, we investigated their ontology using DAVID (the Database for Annotation, Visualization and Integrated Discovery; Huang et al., 2009a,b). In the biological process category, the top six enriched terms were related to transcriptional regulation and chromatin modification, suggesting that Otx2 may regulate transcriptional cascades in postnatal cortex (Figure 2A). In addition, there were categories related to nervous system and forebrain development including migration and axon guidance, as well as particular functions such as rhythmic process and transport. Gene ontology (GO) of cellular component categories further confirmed enrichment of neuronal components such as axon, synapse, neuron projection, and myelin sheath (Figure 2B). Of note, the term “mitochondrion” was also enriched, probably reflecting the functional importance of energy metabolism in PV cells and the potential role of Otx2 in its regulation. This notion was strengthened by the fact that the enriched KEGG pathway contained a number of neurodegenerative diseases such as Alzheimer's, Huntington's, and Parkinson's diseases, whose common components are mitochondrial factors (Figure 2C). In addition, enrichment of the FoxO signaling pathway, implicated in the oxidative stress response, further suggested that Otx2 targets may involve factors for the maintenance of cellular integrity. Correspondingly, functional disease ontology analysis showed that genes related to “Alzheimer's disease” and “Schizophrenia” were enriched (Table 1), and that those related to schizophrenia (Hess et al., 2016; Wu et al., 2016b) indeed possessed Otx2-binding sites around the TSS (Table S1). Together, ChIP-seq data in P28 cortex revealed potential Otx2 targets that regulate transcription, neuronal development and mitochondrial function and are implicated in cellular integrity and disease.

Distinct genomic Otx2 binding sites have recently been reported in different tissues and developmental stages (Samuel et al., 2014; Yang et al., 2014; Hoch et al., 2015). It prompted us to compare Otx2 peaks with other systems to identify unique Otx2 binding sites in postnatal cortex. Raw data from previous studies were used for our peak calling to minimize analytical differences between laboratories. The same antibody (R&D) was also used for ChIP experiments in all previous studies and ours, to further reduce differences in experimental conditions between groups. We found that 191 peaks were shared among all four systems including P28 cortex, embryonic subpallium, neural retina and RPE (Figures S2B,C) despite the distinct functions and genomic targets of Otx2 depending on cellular context (Samuel et al., 2014). Conversely, P28 cortex also had unique Otx2 binding sites (2,683 peaks) compared to embryonic subpallium, neural retina and RPE.

Next, we examined genes bound by Otx2 in postnatal cortex but not in embryonic subpallium which includes progenitors of cortical interneurons at the ganglionic eminence. Because the published embryonic data contain only a few hundred common peaks among triplicate datasets (Hoch et al., 2015), we used “replicate-3 (replica3)” data of them, which produced a large number of peaks (29,533) in our peak calling, allowing us to find as many common peaks between embryo and P28 cortex as possible. As a result, over 2,000 overlapping peaks were detected between P28 cortex and embryo (Figure 3A). Next, we focused on genes carrying Otx2 binding sites in the 5 kb upstream and downstream from the TSS to include the peaks that did not overlap exactly between embryo and P28 cortex but existed in the promoter region of same gene. Comparison of P28 cortex (4,722 genes in total) with embryonic subpallium (6,940 genes in total) revealed that 2,664 genes appeared in both P28 and embryo, suggesting overlapping targets of Otx2 during development (Figure 3B). Common genes (2,664) or P28-specific genes (2,058) were classified based on GO terms of biological processes (Tables S2, S3) and subjected to gene clustering analysis (Figures 3C,D). We found that terms related to “development of nervous system and forebrain” appeared only in common genes (Figure 3C, Table S2). In fact, genes encoding transcription factors involved in the development of the forebrain (such as Emx1/2, Otx1/2, and Pax6/7) or interneurons (Dlx2, Foxg1, and Etv1) were bound by Otx2 in embryonic and P28 cortex (Figure S2D and data not shown). These results suggest that Otx2 may directly regulate these genes by binding to their promoter both in embryonic progenitors and in postnatal interneurons. Categories such as “transcription” and “mRNA processing” enriched in both common and P28-specific genes, indicating that these functions were maintained in embryonic and postnatal development via common genes (Figure 3C) and that postnatally, P28-specific genes were added in the same categories (Figure 3D). In contrast, terms implicated in “transport,” “protein catabolic process,” “signal transduction,” and “phosphorylation” were unique to P28-specific Otx2 targets (Figure 3D, Table S3).

Gene clusters containing “vesicle-mediated transport,” “potassium/calcium ion transport,” and “autophagy/mitophagy” were especially enriched in P28 cortex (Figure 3D). Examination of the genes related to “ion transport” with Otx2-bound promoter showed that 21 out of 28 had P28-specific binding (Table 2, Figure S2E). Interestingly, part of “ion transporters” tends to be developmentally upregulated between P10 and P25 of fast-spiking interneurons in the somatosensory cortex (Okaty et al., 2009). In particular, potassium channels such as Kcnc1 (K_V3.1), Kcna1/6 (K_V1.1/1.6), and Kcnk1 contribute to the maturation and fast-spiking properties of PV cells (Rudy and McBain, 2001; Okaty et al., 2009). Moreover, some glutamate and GABA receptor genes also had P28-specific Otx2 binding sites (Table S4). Notably, the two NMDA receptor subunit genes Grin2a and Grin2b possessed P28-specific and common Otx2 binding sites, respectively, consistent with the activity-dependent switch in predominant subunit composition from NR2B to NR2A during development (Williams et al., 1993; Flint et al., 1997). Thus, the list of genes that were bound by Otx2 specifically in postnatal (P28) cortex would indicate their functions to promote activity-dependent development of PV cell characteristics.

Postnatal Otx2 induces the critical period of ocular dominance plasticity (ODP) in the V1 (Sugiyama et al., 2008). Therefore, we focused on genes implicated in ODP from mouse phenotypes and 32 candidate genes previously discovered by comparing three different data sets (components of ODP related pathways; genes regulated by monocular deprivation or dark-rearing; genes with correlation between ODP and expression levels; Rietman et al., 2012). As shown in Table 3, 28 out of the 54 known and candidate ODP genes inspected had Otx2 binding sites (full list of inspected genes is shown in Table S5). We noticed that some of the ODP genes with functional evidence such as Gad2 (Hensch et al., 1998), Grin2a (Fagiolini et al., 2003), Arc (McCurry et al., 2010), Clock (Kobayashi et al., 2015), and Csgalnact1 (Hou et al., in press) had strong Otx2 peaks at their promoters, suggesting that Otx2 may directly regulate these important critical-period plasticity genes. Of these, Csgalnact1, which encodes a key chondroitin sulfate (CS)-synthesizing enzyme, showed postnatal binding of Otx2 consistent with its role in increasing CSPG around developing PV cells (Hou et al., in press).

In addition, comparison of ODP candidate genes with our ChIP-seq targets showed that 17 out of 32 (53%) of the candidate genes had Otx2 binding sites (Table 3). These genes contain factors implicated in GABA receptors and signaling pathways involving protein kinase A (PKA), extracellular signal regulated kinase (ERK), and calcineurin. In sum, our list of Otx2 target genes may provide additional criteria for the exploration of candidate genes for critical period plasticity.

Next, we aimed to examine Otx2-dependent gene expression using the conditional KO mice in which Otx2 gene is deleted by Cre expression under control of CamKIIα promoter. This conditional deletion of Otx2 in its expressing regions such as retina, thalamus and midbrain causes reduction of Otx2 protein in the V1 (Sugiyama et al., 2008), while cortical Otx2 is also derived from the choroid plexus (Spatazza et al., 2013). We quantified Otx2 immunoreactivity within PV cells of control (Otx2^fl/fl) and Otx2-deficient (Otx2^fl/fl; CaMKIIα-Cre) mice on P28 and found that the fluorescence intensity of Otx2 was significantly reduced across the areas of Otx2-deficient cortex in addition to the V1 (Figure 4). PV protein was also reduced in these areas, suggesting Otx2 function in PV cell maturation. Thus, we used the KO mice (Otx2^fl/fl; CaMKIIα-Cre) to analyze Otx2-dependent transcriptome in cortical interneurons.

Interneurons tagged by Venus fluorescence under control of the VGAT promoter (VGAT-Venus; Wang et al., 2009) were sorted by flow cytometry from control and Otx2-deficient cortices and subjected to RNA-seq (Figure S3). We found that among all of the genes expressed in control interneurons, Otx2-bound genes were highly expressed in comparison to genes without Otx2 binding sites (Figure S4). To identify differentially expressed (DE) genes between the interneurons of control and Otx2-deficient cortices, we used a bioinformatics program (Cuffdiff) and ranked DE genes according to q-value (top 100 shown in Table S6). Interestingly, prominent DE genes were immediate early genes (Fos, Fosb, Jun, Junb, Arc, Npas4), which can be activated by cAMP or calcium ions in response to extracellular stimuli. Some of these activity-dependent genes (Junb, Jund, Arc) had P28-specific Otx2 binding sites, which suggests that Otx2 is involved in regulating these genes in postnatal interneurons. Regardless of Otx2 binding sites, dysregulation was seen for the genes related to amphetamine addiction (Ppp1r1b, Adcy5, Arc, Fosb, Fos, Jun) that are implicated in acceleration of dopamine transmission, and L-serine biosynthesis (Phgdh, Psph, Psat1) that provides the source for D-serine, a specific co-agonist of the NMDA receptor (Figure S5). Interestingly, these pathways are linked to schizophrenia (Bramness et al., 2012; Balu and Coyle, 2015), implying that Otx2 is involved in this disease.

To further examine whether genes in a defined set of particular functional category tend to be upregulated or downregulated, we applied gene set enrichment analysis (GSEA; Subramanian et al., 2005) to the data from all expressed genes. GSEA in biological processes revealed that several gene categories were significantly dysregulated (Tables S7, S8, Figure 5). Many gene categories enriched for Otx2-bound genes, such as “signal transduction,” “transcription,” “RNA processing,” “translation,” and “proteolysis” (Tables S2, S3), were dysregulated in Otx2-deficient interneurons (Tables S7, S8). Importantly, dysregulation of cellular respiration and metabolism indicated that mitochondrial function and homeostasis were impaired in Otx2-deficient interneurons. GSEA also revealed a striking dysregulation of the term “mitochondrion” in cellular components under Otx2-reduced conditions (Figure 5C), again suggesting an Otx2-dependent function in mitochondria. Together, these results suggest that a loss of Otx2 in cortical interneurons broadly affects the transcription of genes for several aspects of cellular functions in coordinating the development and maintenance of cortical interneurons.

Among the DE genes in our RNA-seq analysis, a notable gene whose expression was significantly downregulated in Otx2-deficient interneurons was oxidative resistance-1 (Oxr1; Figure 6A). Oxr1 is reported to play an important protective role against oxidative stress-induced neurodegeneration in vitro and in the cerebellum (Oliver et al., 2011), and in regulating mitochondrial morphology (Wu et al., 2016a). Otx2 bound to the promoter region of Oxr1, indicating direct transcriptional regulation (Figure 6B). Interestingly, the Otx2 ChIP-seq peak at the promoter of full-length Oxr1 was prominent only in P28 cortex, whereas the TSS of a shorter isoform was marked in embryonic subpallium and neural retina (Figure 6B, Figure S6), suggesting that expression of Oxr1 transcription isoforms might be regulated differently in P28 cortex. Both full length (Oxr1-FL) and short isoform peptides (Oxr1-C) contain the conserved carboxyl-terminal domain, which is essential and sufficient for anti-oxidative function, but the presence of the extended amino-terminal region in Oxr1-FL confers additional regulatory properties (Finelli et al., 2016). Although Oxr1 is reportedly expressed in postnatal cortex (Allen Mouse Brain Atlas, http://mouse.brain-map.org), cell type specificity is unknown. We performed in situ hybridization and revealed that Oxr1 transcript was enriched in GABAergic neurons in the visual cortex of P28 mice (Figure 6C), pointing to the importance of this gene in interneuron-specific function. Together, our data suggests that Otx2 may contribute to the protection of PV cells against oxidative stress-induced damage through transcriptional regulation of Oxr1.

All animal experiments were conducted in compliance with a protocol approved by the Institutional Animal Care and Use Committee of Niigata University (Permit Number: #27-SHINDAIKENDAI39-3). Conventionally raised C57BL/6J mice (12:12 h light:dark cycle) were purchased from Japan SLC. The Otx2^fl/fl mice (Tian et al., 2002) were crossed with CaMKIIα-Cre mice (Minichiello et al., 1999) to generate mice with a conditional Otx2 KO in postnatal brain (Sugiyama et al., 2008). To tag interneurons for FACS, Otx2^fl/fl; CaMKIIα-Cre mice were further crossed with VGAT-Venus mice (Wang et al., 2009).

A large part of the cerebral cortex (from Bregma +1 mm to posterior end) was dissected from P28 C57BL/6J mice and flash-frozen in liquid nitrogen. Pooled cortices from 10 males and 10 females (total 20 mice) were cross-linked with 1% formaldehyde in PBS for 15 min at room temperature and ChIP was performed using the ChIP-IT High Sensitivity kit (Active Motif) per the manufacturer's instructions, with some modifications: after homogenization using a Dounce homogenizer, nuclei were washed in LB0 buffer (20 mM Tris-HCl [pH7.5], 10 mM NaCl, 1 mM EDTA, 0.2% NP-40, 1 mM PMSF) and resuspended in MNase buffer (10 mM Tris-HCl, pH7.5, 10 mM NaCl, 2.5 mM MgCl₂, 0.1% NP-40, 1 mM DTT, 1 mM PMSF), provided with cOmplete™ EDTA-free protease inhibitor cocktail (Roche Diagnostics). Sonication was carried out using a Misonix 2000 Sonicator (Misonix) at power 7. After four pulses of 25 s each, 5 mM CaCl₂ and MNase (NEB, 3,000 U in 500 μl reaction mixture) were added and incubated for 10 min at 37°C. The reaction was stopped by adding 20 mM EGTA. The samples were diluted by half with 2 × ChIP dilution buffer (2% Triton X-100, 280 mM NaCl, 80 mM Tris-HCl [pH8.0] in ChIP buffer from the Active Motif kit) and further sonicated for two pulses of 25 s each. Insoluble material was removed by centrifugation at 15,000 rpm for 5 min at 4°C. Fragmented chromatin sample was divided into four aliquots, of which three were immunoprecipitated with three independent anti-Otx2 antibodies: CH (Millipore AB9566, 4 μl polyclonal rabbit serum), R&D (R&D systems AF1979, 4 μg polyclonal goat IgG) and Operon2 (generated in this study, 6 μg polyclonal rabbit IgG raised against full-length Otx2). Normal rabbit IgG (Abcam ab46540) and normal goat IgG (Santa Cruz sc-2028) were used as negative controls. For immunoprecipitation with CH antibody only, which is not affinity-purified, chromatin was pre-cleared by incubating with protein G agarose-bound normal rabbit IgG for 1 h to increase the signal-to-noise ratio. Samples were assessed for enrichment by qPCR with SsoAdvanced Universal SYBR Green SuperMix (BioRad). Primers are listed in Table S9.

DNA from whole cell extracts (input) and ChIP fractions was further sheared to an average size of ~150 bp by ultrasonication (Covaris), end-repaired, ligated to sequencing adapters and amplified using the NEBNext ChIP-seq Library kit (New England Biolabs) according to the manufacturer's instructions. The DNA library was sequenced on the Illumina HiSeq2000 platform to generate single-end 65-bp reads. Sequenced reads of both ChIP fractions and input were aligned to the mouse genome (UCSC mm10) using Bowtie version 1.1.0 (Langmead et al., 2009), allowing two mismatches in the first 28 bases per read (-n2 option). All duplicate reads and those without unique alignment were removed from further analysis. In average, total of 46,645,691 reads (80.1%) were mapped uniquely (Table S10). Peak calling was performed using MACS software (Zhang et al., 2008; Feng et al., 2012) in version 2.1.1, using the nomodel option at a q-value threshold of <0.1. For visualization of peaks on the genomic map, DROMPA2 tool (Nakato et al., 2013) was used. Peaks that overlapped among different samples were calculated using BEDTools (Quinlan and Hall, 2010) and DROMPA2 (Nakato et al., 2013). As for defining the consensus set of Otx2 peaks from ChIP-seq using three different antibodies, each peak detected by MACS was first extended to 1,000 bp and then overlaps were calculated. Overlapped peak regions <500 bp were then extended to 500 bp. As one peak from a dataset can overlap with two peaks from another dataset, total peak number may vary in some cases. For comparison with published Otx2 ChIP-seq studies, fastq files of the following data from the database were used to perform mapping and peak calling in the same way as our data: GSE69724 for embryonic subpallium (Hoch et al., 2015) and GSE54084 for neural retina and RPE (Samuel et al., 2014). Number of reads and mapped ratio of each sample are summarized in Table S10. To compare the peak position to genes and TSS, DROMPA2 (Nakato et al., 2013; Figures 1B, 3B) and Genomic Regions Enrichment of Annotations Tool (GREAT; McLean et al., 2010; Figure 1C) were used, respectively.

Interneurons were isolated using a trehalose-supplemented protocol (Saxena et al., 2012) from Otx2^fl/fl; VGAT-Venus, or Otx2^fl/fl; CamKIIα-cre; VGAT-Venus juvenile mice (P27–30). In brief, digestion and dissociation of the same region of cerebral cortex as described in the ChIP procedure were carried out using the Papain Dissociation System (Worthington Biochemical Corporation) in the presence of 10% (vol/vol) D-trehalose (Wako Pure Chemical Industries). Digested tissue was dissociated with trituration to obtain a single-cell suspension. Hoechst 33342 (5 μg/ml) and propidium iodide (PI; 1 μg/ml) were added to the medium to identify viable cells. Using FACSAriaII (BD Biosciences), the Venus-fluorescent, PI-negative interneurons were sorted directly into Trizol LS reagent (Ambion) and flash-frozen in liquid nitrogen until RNA purification.

RNA isolations were performed using the NucleoSpin RNA kit (Macherey-Nagel) or RNeasy plus universal MiniElute Kit (Qiagen). Eluted RNA was concentrated with NucleoSpin RNA Clean-up XS (Macherey-Nagel). RNA quality was assessed using the Agilent 2100 Bioanalyzer with the Nano LabChip Kit (Agilent Technologies). RNA of interneurons from one or two mice was mixed to obtain ~100 ng of total RNA for construction of one library. The mouse Truseq Stranded Total RNA with riboZERO kit (Illumina) was used to construct four libraries, which were then sequenced on the Illumina HiSeq2000 platform to generate single-end 65-bp reads. The resulting reads were aligned to the mouse genome (UCSC mm10) using TopHat v2.1.0 (Trapnell et al., 2009, 2012; Kim et al., 2013). The number of reads and mapped ratio of each sample, and the sex of the animals used, are summarized in Table S10. To identify differentially expressed genes (DEGs), we used Cuffdiff v.2.2.1 (Trapnell et al., 2009, 2012) to compare control and Otx2 KO samples, each of which had four biological replicates. Expression of each transcript was quantified as fragments per kilobase of transcript per million fragments mapped (FPKM) by Cufflinks v.2.2.1 (Trapnell et al., 2009, 2012). To calculate differences in gene expression among Otx2 bound- and unbound- genes, those with FPKM > 0.3 were considered as expressed and the ANOVA p-value was computed using the statistical toolkit R.

GO and KEGG pathway functional annotations were performed using the NIH web based tool DAVID v.6.8 (Huang et al., 2009a,b). The web tool FunDo (Osborne et al., 2009) was used for annotation to Disease Ontology. GSEA (Subramanian et al., 2005) was used to identify functionally related groups of genes enriched toward the top or bottom of a ranked list. Genes were ranked in order of log₂ (fold change) calculated by Cuffdiff, and genes with a status of “NOTEST” excluded from analysis. Gene names were converted from mouse to human, and the pre-ranked list was submitted to GSEA using the default parameters and the Molecular Signature Database (MSigDB; Subramanian et al., 2005) as the gene sets database.

For immunohistochemistry, deeply anesthetized mice were perfused transcardially and post- fixed with 4% paraformaldehyde (Sugiyama et al., 2008). Floating coronal sections (20 μm) were incubated with the following primary antibodies: rabbit anti-Otx2 (Millipore AB9566), mouse anti-PV (Swant PV235), mouse anti-GAD67 (Millipore MAB5406). Alexa-conjugated anti-mouse and anti-rabbit IgG (Invitrogen) were used for secondary detection.

Quantitative analysis of fluorescence intensity was performed as described previously (Sugiyama et al., 2008). Immunostaining for experimental control and sample sections was performed concurrently with the same solutions, and images were photographed in one sitting with the same gain and exposure time. The fluorescence intensity for each cell was measured with the spot module of NIS-Elements AR Analysis software (Nikon) in the supragranular layers (a 600 × 350 μm area for V1 or A1, a 600 × 400 μm area for M1, a 600 × 500 μm area for S1 or Cg2). Otx2+/PV+ cells were defined by combining threshold (between the intensity values of 306–4,096) and area size (above 90 μm²) to distinguish positive cells from background signal. For quantification, we compared the data between two groups (Mann–Whitney U-test).

In situ hybridization on sections mounted on glass slides was performed according to methods described previously (Sugiyama et al., 2008). In brief, coronal sections (20 μm) were hybridized at 65°C with a Digoxigenin (DIG)-labeled antisense RNA probe for Oxr1, then detected by alkaline phosphatase-conjugated anti-DIG antibody. For immunostaining of Gad67, sections were incubated with mouse anti-Gad67 antibody (Millipore MAB5406), and Alexa-conjugated anti-mouse IgG (Invitrogen) was used as a secondary antibody.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.