Loss of Kdm5c causes spurious transcription and prevents the fine-tuning of activity-regulated enhancers in neurons

Article Loss of Kdm5c Causes Spurious Transcription and Prevents the Fine-Tuning of Activity-Regulated Enhancers in Neurons Graphical Abstract Highlights dDoseand time-dependent control of cognitive development by Kdm5c dKdm5c is required for germline gene silencing during early development dKdm5c functions as a fine-tuner of enhancers in maturing neurons dKdm5c retains a genome surveillance role in adult neurons Authors Marilyn Scandaglia, Jose P. Lopez-Atalaya, Alejandro Medrano-Fernandez, ..., Shigeki Iwase, Yang Shi, Angel Barco Correspondence [email protected] In Brief Scandaglia et al. show that Kdm5c plays critical roles constraining transcription during neuronal differentiation and maturation. Although Kdm5c contribution to neuronal transcription regulation later declines, it retains a genome surveillance role precluding spurious transcription in adult neurons. These functions likely contribute to the pathoetiology of ClaesJensen-type X-linked intellectual disability. Data and Software Availability GSE85874 GSE85873 Scandaglia et al., 2017, Cell Reports 21, 47–59 October 3, 2017 ª2017 The Author(s). http://dx.doi.org/10.1016/j.celrep.2017.09.014 Cell Reports Article Loss of Kdm5c Causes Spurious Transcription and Prevents the Fine-Tuning of Activity-Regulated Enhancers in Neurons Marilyn Scandaglia, 1 Jose P. Lopez-Atalaya, 1 Alejandro Medrano-Fernandez, 1 Maria T. Lopez-Cascales, 1 Beatriz del Blanco, 1 Michal Lipinski, 1 Eva Benito, 1,5 Roman Olivares, 1 Shigeki Iwase, 2 Yang Shi, 3,4 and Angel Barco 1,6, * 1 Instituto de Neurociencias (Universidad Miguel Herna ´ndez-Consejo Superior de Investigaciones Cientı ´ficas), Molecular Neurobiology and Neuropathology Unit, Av. Santiago Ramo ´n y Cajal s/n, Sant Joan d’Alacant, 03550 Alicante, Spain 2 Department of Human Genetics, University of Michigan, 5815 Medical Science II, Ann Arbor, MI 48109, USA 3 Division of Newborn Medicine, Boston Children’s Hospital, Boston, MA 02115, USA 4 Department of Cell Biology, Harvard Medical School, 300 Longwood Avenue, Boston, MA 02115, USA 5 Present address: European Molecular Biology Organization (EMBO), Meyerhofstrasse 1, 69117 Heidelberg, Germany 6 Lead Contact *Correspondence: [email protected] http://dx.doi.org/10.1016/j.celrep.2017.09.014 SUMMARY During development, chromatin-modifying enzymes regulate both the timely establishment of cell-typespecific gene programs and the coordinated repression of alternative cell fates. To dissect the role of one such enzyme, the intellectual-disability-linked lysine demethylase 5C (Kdm5c), in the developing and adult brain, we conducted parallel behavioral, transcriptomic, and epigenomic studies in Kdm5c-null and forebrain-restricted inducible knockout mice. Together, genomic analyses and functional assays demonstrate that Kdm5c plays a critical role as a repressor responsible for the developmental silencing of germline genes during cellular differentiation and in fine-tuning activity-regulated enhancers during neuronal maturation. Although the importance of these functions declines after birth, Kdm5c retains an important genome surveillance role preventing the incorrect activation of non-neuronal and cryptic promoters in adult neurons. INTRODUCTION The development of the nervous system is a highly organized process that requires precise spatial and temporal regulation of gene programs involved in not only the differentiation, maturation, and survival of neurons but also the repression of alternative cell fates and restriction of cell-type-specific gene expression (Lilja et al., 2013). Such dynamic expression patterns are sustained by extensive changes in the epigenome, and consequently, mutations of genes encoding chromatin-modifying enzymes can lead to severe neurodevelopmental disorders (Bjornsson, 2015; Kleefstra et al., 2014). In particular, mutations in the X-linked gene encoding the lysine-specific demethylase 5C (KDM5C, also known as JARID1C or SMCX) can cause Claes-Jensentype X-linked intellectual disability (CJ-XLID) (Claes et al., 2000; Jensen et al., 2005). This rare syndrome, which accounts for 1%–3% of all XLID cases, produces severe intellectual disability and is characterized by autistic behavior, short stature, hyperreflexia, emotional outbursts, spastic paraplegia, and epileptic seizures (Adegbola et al., 2008; Gonc¸ alves et al., 2014). KDM5C mediates the demethylation of triand di-methylated lysines in position four of histone H3 (H3K4me3 and H3K4me2, respectively), functioning in non-neuronal cells as a transcriptional co-repressor of the RE1-silencing transcription factor (REST) complex (Iwase et al., 2007; Tahiliani et al., 2007) and an enhancer modulator (Outchkourov et al., 2013; Shen et al., 2016). Notably, several other intellectual disability disorders (IDDs) also originate from mutations in enzymes that either modify or interact with the different species of K4-methylated histone H3 (Parkel et al., 2013; Vallianatos and Iwase, 2015). Furthermore, changes in H3K4 methylation are correlated with memory acquisition (Gra ¨ff et al., 2011; Gupta et al., 2010; Kerimoglu et al., 2013), but whether this histone posttranslational modification (HPTM) has a functional role in the process is still unknown. Although KDM5C is mainly expressed in brain and skeletal muscle in adult human tissues (Jensen et al., 2005), the mechanisms that link transcriptional deregulation with impaired Kdm5c function in the developing and adult brain remain largely unexplored. A recent and seminal study, upon developing the first animal model of CJ-XLID, demonstrated that the germinal loss of Kdm5c in mice causes dendritic and spine anomalies, alterations of the neural transcriptome, and behavioral deficits that resembled clinical symptoms (Iwase et al., 2016), but did not determine the specific genomic actions of Kdm5c in neurons throughout life. Shortly afterward, Shen et al. (2016) demonstrated that loss of Kdm5c results in the activation of a set of enhancers in human breast cancer cells, but whether enhancer over-activation also occurs in neurons and contributes to intellectual disability (ID) remains unexplored. To address these questions, we conducted parallel behavioral and genomics screens in mice exhibiting either germinal or adult forebrain ablation of Kdm5c. Our comprehensive analyses, together with lossand gain-of-function experiments in neuronal cultures, Cell Reports 21, 47–59, October 3, 2017 ª2017 The Author(s). 47 This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Figure 1. Kdm5c-KOs Recapitulate CJ-XLID Clinical Symptoms and Exhibit Stronger Phenotypes Than ifKOs (A) Schematics of the genetic strategy used to obtain Kdm5c-KOs and ifKOs. (B) Kdm5c transcript level in hippocampus (hp), cortex (ctx), and cerebellum (cb). qRT-PCR assays were conducted using primers specific for the deleted exons (hp: KO t 2 = 5.93, p < 0.0001; ifKO: t 2 = 10.23, p = 0.009; ctx: KO t 2 = 82.27, p = 0.0001; ifKO: t 8 = 2.21, p = 0.06; cb KO: t 2 = 7.43, p = 0.02; and ifKO: t 2 = 1.63, p = 0.2). (C) Average weight at different ages (KOs: F (1,23) = 67.79, p < 0.0001, post hoc 4 months p < 0.05, post hoc 15 months p < 0.0001; ifKOs: F 1,33 = 0.13, p = 0.7). (D) Fear conditioning task in KOs (left: pre/post shock comparison, WT: U = 4.00, p = 0.01; KO: U = 18.50, p = 0.3; context 24 hr: U = 0.00, p = 0.0006; pre-cue 24 hr: U = 8.00, p = 0.02; post-cue: t 12 = 5.93, p < 0.0001) and ifKOs (right: pre-shock, U = 74.00, p = 0.3; post-shock, t 26 = 0.99, p = 0.3; context: t 27 = 0.38, p = 0.7; pre-cue, U = 85.00, p = 0.5; post-cue U = 84.50, p = 0.5). (legend continued on next page) 48 Cell Reports 21, 47–59, October 3, 2017 demonstrate the doseand time-dependent control of cognitive development by Kdm5c and its distinct functions in developing and mature neurons. RESULTS Temporal Dissection of CJ-XLID Endophenotypes in Mice To determine the specific role of Kdm5c in the adult brain and its contribution to cognitive processes, we generated CaMKIIacreERT2::Kdm5c f/f mice (hereafter referred to as ifKO and their control littermates as ifWT) in which gene ablation was spatially restricted to principal neurons of the forebrain and temporarily regulated by tamoxifen administration in the adult stage (Figures 1A and 1B). In order to distinguish between the adult and developmental effects of Kdm5c ablation, we conducted parallel neurological and behavioral assessments of ifKOs and Kdm5cnull mice (KOs). Kdm5c /y males had smaller bodies and brains (Figures 1C, S1A, and S1B) but normal brain histology and anatomy (Figures S1C–S1G), while ifKO males displayed normal weight and brain size (Figures 1C and S1H–S1J). Exhaustive behavioral testing confirmed the cognitive impairments of Kdm5c KOs (Iwase et al., 2016)(Figures 1D and 1E) and revealed traits related to CJ-XLID endophenotypes (Claes et al., 2000; Fujita et al., 2016; Gonc¸ alves et al., 2014) that had not been described before. These included hyperreflexia (Figure 1F), increased impulsivity and emotional responses (Figures 1G–1I), impaired motor coordination (Figure 1J), and epileptic seizure propensity (Figures 1K and S2A–S2L; Table S1). In contrast, ifKOs were indistinguishable from control littermates in our battery of behavioral tasks with the exception of a significant learning delay in the Morris water maze (MWM), both in the hidden platform and reversal phases (Figures 1D–1J, 1L, and S2M–S2O; Table S2). Clinical investigations indicate that females carrying KDM5C mutations suffer from mild cognitive impairments (Rujirabanjerd et al., 2010; Simensen et al., 2012). To model this aspect of KDM5C-related pathology, we evaluated Kdm5c +/ females and found that they were slightly smaller than their control littermates (Figure 2A), showed hindpaw clasping (although to a lesser extent than males; Figures 2B and S2A), exhibited memory deficits in the fear-conditioning task (Figure 2C), and presented a learning delay in the cued phase of the MWM task (Figures 2D and 2E). Other behavioral abnormalities, however, were corrected for by the presence of the wild-type (WT) allele (Tables S1 and S2). We also searched for behavioral impairments in ifKO females but did not observe any significant differences in basal conditions or after stimulating the animals through environmental enrichment (Figures S2P–S2R). Overall, these experiments validate Kdm5c-KO mice as a suitable model to investigate CJ-XLID, including the milder impairments seen in female carriers. Furthermore, the comparison of conventional and conditional KOs indicate that (1) Kdm5c plays a more prominent role during development than in the adult brain and (2) Kdm5c still retains some function in mature principal neurons responsible for the spatial navigation defects observed in ifKO males. We cannot, however, discard that Kdm5c ablation in other cells or brain regions could cause stronger phenotypes than those reported here for ifKOs. Kdm5c Restrains H3K4me3 Content at Specific Promoters and Enhancers in Adult Neurons To identify the molecular causes of these phenotypes, we next conducted parallel genomic screens in KOs and ifKOs. We reasoned that the analysis of KOs would clarify CJ-XLID pathoetiology and the roles of Kdm5c during development, whereas the investigation of ifKOs could provide novel insights into the specific genomic actions of Kdm5c in mature neurons. We focused on epigenetic and transcriptional alterations in the hippocampus, because this brain region is highly relevant for cognitive processes (including those affected in both KOs and ifKOs) and has a high level of Kdm5c expression (Xu et al., 2002). The occupancy profile of Kdm5c in hippocampal chromatin confirmed the loss of Kdm5c binding in KOs (Figure 3A). No global difference was observed in bulk H3K4me3 levels for KOs and ifKOs (Figure S3A). However, chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) revealed a prominent increase (12%) in the number of H3K4me3-enriched regions in KOs (Figures S3B and S3C), as well as local increases in H3K4me3 levels in both strains (Figures 3B and S3D). Differential profiling analysis identified 1,423 and 540 differentially methylated H3K4me3 peaks (DHMPs) in KOs and ifKOs, respectively (Figure 3C; Table S3). These changes largely corresponded to increases in H3K4 methylation, with each strain having a negligible number of regions showing reduced methylation. The overlap between these epigenomic changes in KOs and ifKOs was very high. For instance, 85% of the changes found in ifKOs were also observed in KOs, and approximately one-third of the changes detected in KOs were observed in ifKOs, albeit with smaller magnitude (Figure 3D). In addition to H3K4me3 profiles, we also determined H3K4me1 and H3K27ac enrichment in hippocampal chromatin from WT mice to identify putative enhancers across the genome. H3K4me3 peaks preferentially locate at transcription start sites (E) MWM escape latency graph. Left: KOs showed deficits in the visible platform (F (1,12) = 4.913, p = 0.05), hidden platform (F (1,108) = 31.43, p < 0.0001), and reversal (F (1,66) = 5.60, p = 0.02) tests, as well as in the different probe trials (PT; see Figure S2). Right: ifKOs showed milder impairments in the hidden (F (1,216) = 6.26, p = 0.01) and reversal (F (1,135) = 7.22, p = 0.008) tests. (F) Reaction latency in the hot plate (KOs: t 12 = 3.05, p = 0.01; ifKOs: t 21 =1.26, p = 0.2). (G) Marble burying test (KOs: t 12 = 2.68, p = 0.02; ifKOs: t 21 =0.97, p = 0.3). (H) Novelty suppressed feeding test (KOs: U = 49, p = 0.001; ifKOs: U = 92.00, p = 0.5). (I) Mobility in the tail suspension test (KOs: t 26 = 2.51, p = 0.01; ifKOs: t 27 = 0.33, p = 0.7). (J) Latency to fall in the accelerating rotarod task (KOs: t 12 = 2.66, p = 0.02; ifKOs: t 27 = 1.45, p = 0.2). (K) Kainate-induced seizures measured in the Racine scale (left), and number of injections needed for exclusion (right: t 6 = 6.65, p = 0.0006). (L) Sensitivity to pentylenetetrazol (PTZ)-induced seizures. Data are expressed as means + SEM or means ±SEM. ns, non significant; *p < 0.05; **p < 0.005; ***p < 0.0005 (Mann-Whitney test, Student’s t test, or two-way ANOVA). Cell Reports 21, 47–59, October 3, 2017 49 (TSSs) and putative enhancers (i.e., at intraand intergenic regions that show a concomitant enrichment for H3K27ac and H3K4me1; Figure S3E). However, DHMPs were overrepresented at intraand intergenic regions, particularly at de novo peaks (Figures 3EandS3F), suggesting that enhancers are particularly sensitive to the absence of Kdm5c. The impact of Kdm5c loss on H3K4me3 at enhancers followed an inverted-U function, in which H3K27ac-rich enhancers with a moderate H3K4me3 content were the most affected (Figures 3FandS3G). The same pattern was also observed in ifKOs (Figure S3H). This indicates that regions depleted of H3K4me3 or with very high H3K4me3 content are both resilient to Kdm5c absence. Consistent with this view, the regions showing stronger Kdm5c binding in WT mice exhibited reduced H3K4me1 signal in KOs, while the increase was rather modest in the case of H3K4me3 (Figure S3I). Therefore, Kdm5c seems to be required in regions in which it is necessary to maintain a given tri-methylation to mono-methylation balance. Kdm5c Modulates the Expression of Plasticity-Related Genes and Fine-Tunes Activity-Dependent Enhancers during Neuronal Maturation RNA sequencing (RNA-seq)-based differential expression (DE) screens revealed that transcriptional changes were both larger and more numerous in KOs than in ifKOs (248 and 107differentially expressed genes [DEGs], respectively) (Figures 4A, 4B, and S4A; Table S4). In fact, most of the DEGs in ifKOs were also altered in KOs with larger fold changes (Figure S4B). In both strains, the changes consisted predominantly of upregulations, thereby supporting Kdm5c’s role in transcriptional repression. The correlation between H3K4me3 gain atTSSs and transcript levels was remarkably high for both mutants (Figure 4C), indicating that the increment of H3K4me3 resulting from the loss of Kdm5c is highly predictive of transcriptional upregulation. In fact, in both KOs and ifKOs, 97% of the upregulated genes associated with H3K4me3 peaks showed an increase of H3K4me3 signal. To link genomic and phenotypic alterations in Kdm5c-deficient mice, we first focused on genes exclusively altered in KOs (Figure 4D), because we reasoned that these changes should be accountable for the severe phenotypes observed in those mice. This DEG set included a number of genes linked to neuronal terms according to Gene Ontology (GO) enrichment analysis. These genes showed appreciable hippocampal expression and relatively modest changes in transcription and H3K4 methylation (Figure 4E, light green dots). Downregulations affected genes important for synaptic transmission and neurodevelopment (e.g., Grik3,Nrp2, and Syt2), while upregulated genes were involved in neuronal signal transduction (e.g., Hexim1,Hexim2,Dusp5, and Dusp18). Consistent with our results, transcriptome analyses in the amygdala and frontal cortex of Kdm5c-KO mice had also revealed misexpression of neuronal genes, but whether these changes were directly linked to altered H3K4 methylation was not explored (Iwase et al., 2016). Our parallel epigenome and transcriptome screens enabled the direct comparison of both effects in vivo. The correlation between differential H3K4me3 methylation at TSSs and changes of transcript levels for DE neuronal genes was modest, suggesting that the impact of Kdm5c loss in some of these loci might rely on changes at more distant regulatory sequences. Figure 2. Cognitive Impairments Are Kdm5c Dose Dependent (A) Kdm5c +/ females are smaller than their WT female littermates (F (1,54) = 17.86, p < 0.0001; post hoc 12-month p < 0.0001). (B) Hemizygous females show hindpaw-clasping (U = 31.50, p = 0.001), although it is less severe than that observed in male KOs (see Figures S2A and S2B). (C) Fear conditioning task: Kdm5c +/ females exhibit fear memory deficits (pre-shock, t 27 = 0.59, p = 0.6; post-shock, t 27 = 0.7, p = 0.5; context 24 hr, U = 57.00, p = 0.04; pre-cue 24 hr, t 27 = 2.10, p = 0.05; post-cue 24 hr, t 27 = 2.49, p = 0.02). (D and E) Morris water maze. The escape latency curve (D) revealed mild learning difficulties during the first day of the task (F (1,54) = 8.91, p = 0.004;post hoc day 1 p < 0.01) but no difference in performance during the PTs (E). Data are expressed as means + SEM or means ±SEM. *p < 0.05; **p < 0.005; ***p < 0.0005 (Mann-Whitney test, Student’s t test, and two-way ANOVA). 50 Cell Reports 21, 47–59, October 3, 2017 Given the role of Kdm5c in regulating enhancer function, as well as the potential causal link between KO phenotypes (such as learning impairment and epilepsy propensity) and altered activity-driven transcription, we next investigated the contribution of Kdm5c to gene induction upon exploration of a novel environment (NE). This experience is known to trigger hippocampal transcription of numerous immediate early genes (IEGs) involved in synaptic plasticity and memory consolidation (Figures 5A and S5A; Table S5)(Flavell and Greenberg, 2008). Strikingly, NEinduced changes in KOs represented approximately one-third of those observed in WTs (66 versus 186 genes). This was because a significant proportion of the NE transcriptional program was moderately overexpressed in KOs in naive situation (homecage [HC]) (Figures 5A and 5B), suggesting a role for Kdm5c fine-tuning activity-driven transcription. Independent qRT-PCR for the IEGs Fos and Arc confirmed this observation (Figure 5C). We next investigated the impact of NE on H3K4me3 profiles and its relationship with Kdm5c loss. Our ChIP-seq experiment revealed that NE caused a 5% increase in the number of detected peaks in hippocampal chromatin (Figure 5D), demonstrating that the levels of H3K4me3, like the transcriptome, are susceptible to modulation by this experience. Notably, NE-induced changes in H3K4me3 profiles were partially occluded in KOs (Figures 5D and 5E), which is consistent with Kdm5c playing a role in the regulation of activitydependent enhancers. H3K4me3 levels at proximal enhancers of important IEGs involved in neuroplasticity were significantly elevated in naive Kdm5c-KO mice, even when the TSS was not affected. A clear example of this is the IEG Npas4,which encodes a transcription factor that regulates cognition-related transcription and GABAergic synapse formation (Ramamoorthi et al., 2011). H3K4me3 levels at the enhancer region 10 kb upstream of the Npas4 TSS were higher in KOs than in WTs, and this increase was associated with augmented transcription of the enhancer RNA (eRNA) at the basal state (Figures 5F and 5G). A similar scenario was also observed at enhancers 2 and 5 of Fos (Figures S5BandS5C)andatthe Arc enhancer (Figures S5D and S5E), although for the latter, H3K4me3 levels at the promoter were also higher in KO chromatin. In contrast to KOs, basal expression and induction of IEGs and H3K4me3 profiles at these activity-regulated enhancers were normal in ifKOs (Figures S5F–S5H). This suggests that Kdm5c Figure 3. Lack of Kdm5c Increases H3K4me3 Levels throughout the Neuronal Epigenome (A) Heatmap (right) and line plot (left) showing Kdm5c density at Kdm5c-enriched regions (as detected in hippocampal chromatin of WT mice) in KO and control littermates. Color intensity in the heatmap is proportional to read density. (B) Density plot showing the average H3K4me3 signal (expressed as reads per million [RPM]) of DHMPs for WT, KO, ifWT, and ifKO replicates. Input values are also shown. Shaded areas represent the 95% confidence interval (CI). (C) Cumulative graph showing the number of DHMPs (Benjamini-Hochberg adjusted p < 0.05, LFC > 0.58) and their FC with respect to the relative control (i.e., KO versus WT; ifKO versus ifWT). DHMPs were ordered by Log2FC and plotted as separate entities on the x axis. (D) Venn diagram depicting the overlap between the DHMPs of KO (versus WT) and ifKO (versus ifWT) mice. (E) Distribution of DHMPs in function of gene feature, in Kdm5c-KOs and ifKOs. ‘‘Gene start’’ refers to peaks located within ±1 kb of the first TSS, ‘‘gene body’’ denotes internal peaks (including alternative TSSs), and ‘‘intergenic’’ refers to peaks that are farther than 1 Kb from the TSS of any RefSeq annotated gene. (F) H3K4me3 changes at putative enhancers (intraand intergenic H3K27ac-rich regions not overlapping with a TSS) in KOs and ifKOs. Enhancers were classified according to their H3K4me3 content (see Figures S3G and S3H for additional details). Cell Reports 21, 47–59, October 3, 2017 51 plays a specific role in establishing the ground activity of IEGs during neuronal maturation. To tackle this model, we investigated the fine-tuning of activity-regulated enhancers using primary neuronal cultures (PNCs) and found that Kdm5c ablation in PNCs from Kdm5c f/f embryos using a Cre-recombinase-expressing lentivirus (LV) enhanced basal expression of IEGs such as Fos,Arc, and Npas4 (Figures 5H–5I). Moreover, normal basal IEG expression was observed upon co-infection with a second LV driving the expression of WT human KDM5C (hKDM5C) (Figures 5H–5I). These experiments show that these loci are still susceptible to epigenetic tuning at this stage and demonstrate that Kdm5c is both necessary and sufficient for proper modulation of activity-regulated enhancers during neuronal maturation. Figure 4. Transcriptional Deregulation in Kdm5c-Deficient Mice Correlates with Altered H3K4me3 Levels (A) RNA-seq profile of the Kdm5c gene in the hippocampus of KOs and ifKOs confirms the absence of exons 11 and 12 (red box). Note that only neuronal cells are affected in the hippocampus of ifKOs (70% reduction). (B) Gene expression changes in KOs and ifKOs (Benjamini-Hochberg adjusted p < 0.05, absolute log2 FC > 0.3). Genes were ordered by log2 FC and plotted as separate entities on the x axis. (C) Scatterplots for KOs (top) and ifKOs (bottom) showing the correlation between H3K4me3 levels at peaks located in close proximity to annotated DEGs and the transcript changes. Filled dots represent DHMPs (adjusted p < 0.1), and colors indicate the location of the peak with respect to the gene feature. (D) Heatmap showing the DEGs common to both KOs and ifKOs (common), as well as the DEGs altered exclusively in either strain. The number of genes in each category is indicated. (E) Boxplot showing log2(TPM WT +1), log2 FC of transcript level, and log2 FC of H3K4me3 at TSSs for KO-exclusive DEGs. Light green dots correspond to genes with significant hippocampal expression and related to neuronal function. Pink dots correspond to germline-related genes associated with a Kdm5c peak observed in embryonic stem cells (ESCs), but not in neural precursor cells (NPCs). Purple dots correspond to germline-related genes that are not associated with an ESC-specific Kdm5c peak. Grey dots correspond to DEGs not falling in the previous categories. Note that the genes with the largest transcript changes (germline-related genes) have very low TPM values. Bars represent means ±SD. 52 Cell Reports 21, 47–59, October 3, 2017 Germinal Loss of Kdm5c Precludes the Silencing of Germline Genes The exploration of DEGs exclusive of KOs also revealed a subset comprising genes related to reproduction and gametogenesis, such as D1Pas1,Naa11,Ddx4, and Ccnb1iP1, that had not been reported in previous screens. These genes had very low or undetectable expression in the hippocampus of control mice and presented large, highly correlated changes in transcript and H3K4me3 levels in the hippocampus of Kdm5c-KO mice (Figure 4E, pink and purple dots). In fact, most of the genes Figure 5. Kdm5c Modulates Neuronal-Activity-Dependent Enhancers (A) Heatmap of DEGs upon novelty exposure (NE) in WT mice together with their values for naive (HC) and NE KOs. (B) Scatterplot comparing NE effect with genotype effect in KOs for the set of NE-induced genes. (C) qRT-PCR assays for Arc (F (1,20) condition = 8.73, p = 0.008; F (1,20) genotype = 2.22, p = 0.2; post hoc WT p < 0.05) and Fos (F (1,20) condition = 12.67, p = 0.002; F (1,20) genotype = 0.72, p = 0.4; post hoc WT p < 0.01) using hippocampal RNA. (D) Number of H3K4me3 peaks detected in WT-HC, WT-NE, KO-HC, and KO-NE replicates. (E) Venn diagram depicting the overlap between the DHMPs exclusive of WT-NE (versus WT-HC) and KO-HC (versus WT-HC). (F) Left: genomic profiles for Npas4. Right: bar graphs representing the read density for H3K4me3 at the TSS (F (1,4)condition = 11.56, p = 0.03; F (1,4)genotype = 0.0073, p = 0.9; post hoc ns) and upstream enhancer (F (1,4) condition = 1.31, p = 0.3; F (1,4) genotype = 100.9, p = 0.0006; post hoc HC and NE p < 0.01) of Npas4; and for the transcripts corresponding to Npas4 mRNA (F (1,8) condition = 19.33, p = 0.002; F (1,8) genotype = 0.05, p = 0.8; post hoc WT p < 0.01, post hoc KO p > 0.05) and its putative eRNA (F (1,8) condition = 0.70, p = 0.4; F (1,8) genotype = 20.03, p = 0.002, post hoc ns). (G) qRT-PCR assays of IEG-associated eRNAs. Npas4-enhancer (t 4 = 1.93, p = 0.13); Fos-enhancer 5 (t 4 = 4.74, p = 0.009). (H) Top: scheme of PNC experiment. Bottom: immunostaining of Kdm5c f/f PNC co-infected with LVs expressing Cre and hKDM5C. (I) qRT-PCR assays confirmed the loss of the endogenous Kdm5c, the expression of hKDM5C, and their modulation of IEG expression (Kdm5c: Cre  versus Cre + , t 4 = 4.92, p = 0.008; Cre  versus Cre + +hKdm5c, t 4 = 6.54, p = 0.003; Cre + versus Cre + +hKdm5c, t 4 = 13.74, p = 0.0002; Arc: Cre  versus Cre + ,t 4 = 3.62, p = 0.02; Cre + versus Cre + +hKdm5c, t 4 = 5.43, p = 0.006; Fos: Cre  versus Cre + ,t 4 = 4.28, p = 0.01; Cre + versus Cre + +hKdm5c, t 4 = 4.17, p = 0.01. Npas4: Cre  versus Cre + ,t 4 = 3.58, p = 0.02; Cre + versus Cre + +hKdm5c, t 4 = 3.07, p = 0.04). Data are expressed as means + SEM or means ±SEM. *p < 0.05, **p < 0.005; ***p < 0.0005 (genotype effect); #p < 0.05, ##p < 0.005; ###p < 0.0005 (condition effect) in two-way ANOVA (C and F) and Student’s t test (G and I). Cell Reports 21, 47–59, October 3, 2017 53 presenting the largest differences between KO and WT littermates belonged to this category (Figure 4D). Interestingly, some of these genes (pink dots) showed robust Kdm5c binding at their promoters in the chromatin of embryonic stem cells (ESCs), although this occupancy decreased over time, from neuroprogenitor cells (NPCs) to mature neurons (MNs) (Figures 6A and 6B). Kdm5c’s disappearance coincides with the strong DNA methylation of these loci in WT mature neurons, according to available MeDIP data (Halder et al., 2016)(Figures 6B and S6A, pink track). To confirm this model, we next examined DNA methylation levels in the hippocampi of KOs and control littermates. Reduced CpG methylation at the promoter of the germline genes D1Pas1 and Naa11 (Figure 6C) was concomitant with their enhanced transcription (Figure 6D). Further supporting our hypothesis, PNCs from Kdm5c /y embryos also showed derepression of D1Pas1 and Naa11 (Figures 6E and 6F), whereas PNCs from Kdm5c f/f embryos transduced with Cre recombinase showed normal germline silencing despite a rapid and strong reduction in the levels of Kdm5c (Figure S6B). These results indicate that Kdm5c is necessary for germline gene silencing during early development but becomes dispensable at later developmental stages once a repressed status has been established through other mechanisms such as DNA methylation (Figure S6C). To examine whether the spurious expression of germline genes in neurons can be rescued by restoring Kdm5c activity, PNCs from Kdm5c /y embryos were transduced with a hKDM5C-expressing LV (Figure 6E). Interestingly, hKDM5C Figure 6. Germinal Loss of Kdm5c Prevents the Silencing of Germline Genes (A) Developmental loss of Kdm5c occupancy at the TSS of germline genes (pink dots in Figure 4E). RPM, reads per million. (B) Genomic profiles for representative germline genes. From top to bottom, we show (a) Kdm5c-ChIP-seq profiles in ESC, NPC (Outchkourov et al., 2013), PNC (Iwase et al., 2016), and adult brain (MN, this study); (b) MeDIP profiles in adult brain of WT mice (Halder et al., 2016), and (c) RNA-seq and H3K4me3-ChIP-seq experiments in adult hippocampus of WT and KO mice (this study). (C) DNA methylation at specific CpGs (red boxes in Figure 6B) at the promoter of D1Pas1 and Naa11 in hippocampal chromatin of WT and KO littermates (star symbols label significant DNA methylation reduction in KOs). Testis DNA was used as control for un-methylated DNA. (D) RT-qPCR assays confirm the expression of germline genes in the hippocampi of KOs (D1Pas1:t 10 = 8.53, p < 0.0001; Naa11:t 10 = 3.90, p = 0.003). (E) Scheme of the experiment (top). Functionality of the hKDM5C construct was confirmed by transient transfection and examination of H3K4me3 levels in EYFPexpressing neurons (bottom). (F) qRT-PCR assay confirmed the overexpression of germline genes in PNCs from Kdm5c /y embryos infected or not with a hKDM5C-expressing LV (Kdm5c,WT versus KO: t 8 = 4.64, p = 0.002, KO versus KO+hKDM5C: t 7 = 15.57, p < 0.0001, WT versus KO+hKDM5C: t 9 = 18.54, p < 0.0001; D1Pas1, WT versus KO: t 8 = 3.49, p = 0.008; KO versus KO+hKDM5C: t 8 = 1.27, p = 0.2; Naa11, WT versus KO: t 9 = 5.05, p = 0.0007; KO versus KO+hKDM5C: t 8 = 0.0007, p = 1.0). Data are expressed as mean + SEM or means ±SEM. ns, non-significant; *p < 0.05; **p < 0.005; ***p < 0.0005 (Student’s t test). 54 Cell Reports 21, 47–59, October 3, 2017