The Krüppel-like factor 9 cistrome in mouse hippocampal neurons reveals predominant transcriptional repression via proximal promoter binding

Krüppel-like factor 9 (Klf9) is a zinc finger transcription factor that functions in neural cell differentiation, but little is known about its genomic targets or mechanism of action in neurons. We used the mouse hippocampus-derived neuronal cell line HT22 to identify genes regulated by Klf9, and we validated our findings in mouse hippocampus. We engineered HT22 cells to express a Klf9 transgene under control of the tetracycline repressor, and used RNA sequencing to identify genes modulated by Klf9. We found 217 genes repressed and 21 induced by Klf9. We also engineered HT22 cells to co-express biotin ligase and a Klf9 fusion protein containing an N-terminal biotin ligase recognition peptide. Using chromatin-streptavidin precipitation (ChSP) sequencing we identified 3,514 genomic regions where Klf9 associated. Seventy-five percent of these were within 1 kb of transcription start sites, and Klf9 associated in chromatin with 60% of the repressed genes. We analyzed the promoters of several repressed genes containing Klf9 ChSP peaks using transient transfection reporter assays and found that Klf9 repressed promoter activity, which was abolished after mutation of Sp/Klf-like motifs. Knockdown or knockout of Klf9 in HT22 cells caused dysregulation of Klf9 target genes. Chromatin immunoprecipitation assays showed that Klf9 associated in chromatin from mouse hippocampus with genes identified by ChSP sequencing on HT22 cells, and expression of Klf9 target genes was dysregulated in the hippocampus of neonatal Klf9-null mice. Gene ontology analysis revealed that Klf9 genomic targets include genes involved in cystokeletal remodeling, Wnt signaling and inflammation. We have identified genomic targets of Klf9 in hippocampal neurons and created a foundation for future studies on how it functions in chromatin, and regulates neuronal morphology and survival across the lifespan.


Background
Krüppel-like factors (Klfs) comprise a family of zinc-finger transcription factors (TFs) that function in metabolism, development and oncogenesis [1]. They have a highly conserved DNA binding domain comprised of three zinc fingers that binds GC/GT rich regions in the genome. Members of this family are distinguished by their highly divergent N-terminal domains which recruit different chromatin modifying factors that in part govern whether the Klf functions as a transcriptional activator or repressor [1]. Krüppel-like factor 9 (Klf9; formerly basic transcription element binding protein 1 -BTEB1) was identified in a screen of a rat liver cDNA library for proteins that bind the basic transcription element (BTE), a GC-rich motif in the promoter of the rat Cyp1a1 (Cytochrome P450) gene [2]. The zinc fingers of Klf9 have high sequence identity with those of Specificity protein 1 (Sp1), which binds to similar motifs and typically activates transcription [2]. Transient transfection assays showed that Klf9 repressed transcription from a reporter construct containing the BTE sequence. However, Klf9 activated transcription from a reporter containing six tandem repeats of the BTE, suggesting that its activity may be governed by the number of binding sites at a locus [2]. The N-terminal region of Klf9 contains two separable transactivation domains required for full activation of the six-repeat BTE promoter, and an α-helical motif that interacts with the repressor protein Swi-independent 3a (Sin3a) [3,4].
In mouse central nervous system (CNS) Klf9 expression is low at birth, rises postnatally, and peaks at approximately postnatal day (PND) 30 with highest expression in the hippocampus and cerebellum [5,6]. This postnatal increase in Klf9 expression depends on thyroid hormone (TH), which acts via its nuclear receptors to directly regulate transcription of the Klf9 gene [7][8][9][10]. In immature and developing neurons, Klf9 mediates actions of TH on neurite extension [8,11,12]. However, in mature cortical neurons Klf9 inhibits neurite outgrowth [13], while in the cerebellum it mediates TH-dependent inhibition of axonal regeneration [14]. Thus, in neuronal cells Klf9 first promotes, then maintains the differentiated state. It is also required for survival of adult-born dentate granule neurons and Purkinje neurons of the cerebellum [15,16], and it promotes oligodendrocyte differentiation and the expression of myelinating genes [17]. Consistent with these findings, Klf9-null mice show neurological defects, including deficits in fear conditioning and late-stage neurogenesis [6,16]. In addition to its developmental role, Klf9 is also an immediate-early gene that is upregulated in neurons by many extracellular stimuli, including TH [8], glucocorticoids [18], and electrical activity [16].
Despite evidence for a diversity of developmental and physiological roles for Klf9, very little is known about Klf9 genomic targets in any cell type, and nothing is known about the genes that Klf9 regulates in neurons. In the present study, we identified direct genomic targets of Klf9 in the mouse hippocampus-derived neuronal cell line HT22, which is a model for mature hippocampal neurons. This cell line expresses neuronal markers such as enolase and neurofilament proteins, but does not express the glial marker glial fibrillary acidic protein [19,20]. We engineered several HT22 cell lines to allow for control of Klf9 expression to study Klf9-dependent transcriptional responses by RNA sequencing (RNA-seq); to identify genomic regions where Klf9 associates in chromatin by chromatin-streptavidin precipitation followed by deep sequencing (ChSP-seq); and to investigate the consequences of Klf9 deletion for gene transcription and cell proliferation (using CRISPR/Cas9 genome editing). We validated our findings in HT22 cells by analyzing Klf9 association in chromatin in mouse hippocampus by chromatin immunoprecipitation (ChIP) assay, and mRNA levels for putative Klf9 target genes in wild type and Klf9-null mouse hippocampus.

Results
Validation of engineered HT22 cell lines for the identification of Klf9 genomic targets We stably transfected HT22 cells with pCDNA4:TO-Klf9 and pCDNA6:TR vectors (see Methods), then isolated and screened seven clonal lines for baseline and doxycycline (dox)-inducible Klf9 mRNA. One clonal cell line (2-1) was selected for further analysis. The mean baseline Klf9 mRNA level of this cell line (hereafter referred to as HT22 [TR/TO-Klf9]) was slightly higher than that of the parent line, but this was not statistically significant (Fig. 1a). After treatment with dox for 8 h Klf9 mRNA increased by~10-fold ( Fig. 1a; F (3,8) = 480.974, p < .001; n = 3/treatment; ANOVA), which is within the physiological range seen following hormone treatment in HT22 cells, in neonatal mouse brain following hormone treatment [7], and in mouse brain during development [5].
We were unable to detect the endogenous (native) or recombinant (dox-induced) Klf9 protein by Western blotting on nuclear extracts of HT22 cells (data not shown), or with extracts from different cells/tissues of mouse or Xenopus using different antiserums (J.R. Knoedler, P. Bagamasbad and R.J. Denver, unpublished data). We therefore developed a bioassay that served as a proxy for the level of functional Klf9 protein in the cell. This assay comprised transient transfection of HT22 [TR/TO-Klf9] cells with a luciferase reporter vector containing three tandem repeats of the BTE sequence (pGL4.23-3xBTE), which supports Klf9-dependent transactivation or transrepression depending on the cell type [2,10]; a promoter-less luciferase vector (pGL4.23) served as control. Treatment with dox for 8 h reduced luciferase activity by 20% in cells transfected with pGL4.23-3xBTE, but did not alter luciferase activity in empty pGL4.23 vector-transfected cells ( Fig. 1b; t (5) = 3.752, p < .05; Student's two-sample t-test). To independently confirm that Klf9 represses activity from this promoter we co-transfected the parent HT22 cell line with pGL4.23-3×BTE and the pCS2-Klf9 expression vector. This produced a statistically significant reduction (39%) in luciferase activity compared with cells transfected with empty pCS2 vector (Additional file 1: Figure S1; t (6) = 3.292p < .05; Student's two-sample t-test). Taken together, our results show that Klf9 mRNA can be induced within the physiological range by dox treatment of HT22 [TR/TO-Klf9] cells, and that this leads to the production of functional Klf9 protein.

Identification of Klf9-regulated genes in HT22 [TR/TO-Klf9] cells by RNA sequencing
We conducted RNA sequencing (RNA-seq) on parent HT22 cells and the HT22 [TR/TO-Klf9] cell line treated with vehicle or dox for 8 h (n = 3/treatment for the parent line, 3 for HT22 [TR/TO-Klf9] vehicle treated and 2 for HT22 [TR/TO-Klf9] dox treated; a third replicate had to be discarded due to a technical error). We aligned sequencing reads to the mm8 build of the mouse genome using Bowtie [21], and differences in transcript abundance were quantified using DESeq [21,22] [23]. This allowed for high-affinity purification of Klf9 in chromatin by streptavidin precipitation (ChSP). We used Western blotting to detect the biotinylated fusion protein with streptavidin-HRP in HT22 nuclear extract (Fig. 2a). Previous work showed that Klf9 associates in chromatin with the Klf13 5′ upstream region in NIH 3 T3 cells (M. Nikiforov, unpublished results). We therefore investigated if Klf9 associated in chromatin with this genomic region in HT22 cells as proof-ofprinciple for the ChSP technique. First, we conducted ChIP assay for Klf9 on chromatin from the parent HT22 cell line. This resulted in~5 fold enrichment above background (determined by ChIP with normal goat IgG) at the Klf13 promoter but not at a Klf13 intronic region which lacks Sp/Klf motifs (Additional file 3: Figure S2; intronic region, t (6) = .850, p = .428; promoter region, t (6)  We restricted further analysis to only those peaks called by both programs (total = 3,378), and then applied the program PeakSplitter to identify and subdivide regions with multiple closely spaced peaks [26]. Peaksplitter is only compatible with MACS; we therefore analyzed the MACS dataset with Peaksplitter, then used the program BedTools to extract the overlap of these split peaks with the peaks called by PePr [27]. Examples of how this approach classifies peaks are shown in Additional file 4: Figure S3. This approach gave a final count of 3,514 Klf9 peaks, which ranged from 8 to 2,429 bp in length (average length 881 bp). All peak coordinates, nearest gene and average sequencing read density across the peak (based on build mm10 of the mouse genome) are given in Additional file 5: Table S2. Sequencing and analysis of ChSP DNA from HT22 [BirA] cells showed very few regions (20 by MACS, 52 by PePr) with higher mapped sequencing read density compared with HT22 [BirA/FLBIO-Klf9] cells. This demonstrates that the BirA-FLBIO platform allowed for identification of Klf9-associated genomic regions with very low background.

Klf9 associates in chromatin from mouse hippocampus with genomic regions identified by ChSP-seq in HT22 cells
To determine if Klf9 associates in chromatin in mouse hippocampus at genomic sites identified in HT22 [BirA/ FLBIO-Klf9] cells, we conducted targeted ChIP assays using chromatin isolated from the hippocampal region of the brain of adult wild type mice (five male and five female). We analyzed the same 8 genomic regions described above for HT22 cells and found statistically significant Klf9 ChIP signal at 7 of the 8 regions in both males and females ( Fig. 2e; Slc11a2: t (13) = 2.260, p < .05; Klf16: t (12) = .4.458, p < .001; Klf13: t (13) = 3.752, p < .005; Sin3a: t (12) = 5.659, p < .0005; Nr3c1: t (10) = −2.684, p < .05; Limk1: t (11) = 3.887, p < .005; Klf11: t (16) = 2.246, p < .05; Mapk11: t (16) = −.0265, p = .979; n = 10; Student's two-sample ttest). Since there were no statistically significant differences between the sexes we pooled the data for analysis. The lack of Klf9 ChIP signal at the 5′ upstream region of Mapk11 is consistent with this region having the lowest ChSP signal in HT22 [BirA/FLBIO-Klf9] cells (Fig. 2d). We did not detect Klf9 ChIP signal in chromatin from mouse hippocampus at intronic regions located 10 kb or more downstream from the identified Klf9 peaks in the Klf13, Klf16, Limk1 or Nr3c1 genes (Additional file 7: Figure S5B). Taken together, our findings support that the BirA/FLBIO platform applied to the HT22 cell line is a useful model for identifying Klf9 genomic targets in mouse hippocampus.

Klf9 associates in chromatin primarily with proximal promoter regions
We used the ChIP-enrich web tool [28] to assign Klf9 peaks to genes based on the nearest transcription start site (TSS), and to identify where the peaks were distributed with respect to the TSSs of annotated genes. Based on this analysis, 89.5% of the peaks fell within 10 kb of a TSS, and of these, 86% (or 77.1% of total peaks) were 1 kb or less from a TSS (Fig. 3a). The ChIP-enrich software assigns peaks to genes based on the closest TSS, which results in some genes having multiple peaks associated with them. After accounting for such double-counted targets, this left 2,847 genes with at least one associated Klf9 peak (defined as being closer to that peak than any other annotated gene). The majority of these genes (2,749) had at least one Klf9 peak within 10 kb of their TSS. We next used the HOMER peak annotation program to analyze the distribution of Klf9 peaks relative to the TSS. Of the peaks centered within 1 kb of a TSS, 32.7% were centered upstream, 21.7% were centered within the 5′ untranslated region, and 27.1% were centered either in the first exon or the first intron (Fig. 3b). We then used the Cis-regulatory Element Annotation System (CEAS) to map how sequencing reads were distributed with respect to genomic features [29]. Analysis of the distribution of mapped sequencing reads around TSSs revealed a moderate bias towards regions immediately upstream of the TSSs (Additional file 8: Figure S6). Thus, the larger peaks tend to be centered in 5′ flanking regions rather than in 5′ UTRs, exons or introns.
Of the 217 genes that we found to be repressed by Klf9 by RNA-seq, 130 (60%) have a Klf9 peak within 10 kb of their TSS. By contrast, of the 21 genes found to be induced by Klf9 only 1 (5%) had a peak within 10 kb of its TSS (Fig. 3c). This is consistent with Klf9 acting primarily as a transcriptional repressor. Differentially regulated genes without associated Klf9 peaks may be indirect target genes (regulated by other Klf9-responsive genes). If this is correct, it suggests that Klf9-induced genes are mostly indirect targets, while the majority of repressed genes are directly regulated by Klf9.
An additional 2,322 genes had at least one Klf9 peak located within 10 kb of their TSS, but their mRNA levels analyzed by RNA-seq were not significantly affected by forced Klf9 expression. However, when we looked at the mean mRNA level for all polyadenylated transcripts detected by RNA-seq in the HT22 [TR/TO-Klf9] cell line treated with vehicle or dox, and compared genes that did or did not have Klf9 peaks within 10 kb of their TSS, we found evidence for a general repressive action of Klf9 on transcription of genes possessing Klf9 peaks (Fig. 3d). We calculated the ratio of the mean mRNA levels with dox to that with vehicle for all expressed genes in each of the two cell lines ("expression ratio"). An expression ratio of 1 indicates no change caused by dox treatment, >1 indicates an increased mRNA level, and <1 a decreased mRNA level. In comparing the expression ratio of genes with or without Klf9 peaks, we found no statistically significant difference in the parent cell line, but a statistically significant lower median This supports that Klf9 exerts a general repressive action on transcription of genes with which it associates.

Identification of consensus Sp/Klf motifs at regions of Klf9 ChSP-seq peaks
We used the program HOMER to identify enriched DNA sequence motifs in Klf9 ChSP-seq peaks [30]. The most highly enriched sequence was a Sp/Klf motif (GCCACGCCCMCY) that was present in 75.6% of all peaks; hereafter we refer to this sequence as the 'Klf9 consensus motif ' (Fig. 4a). We identified 18 additional motifs enriched at Klf9 peaks that are partially redundant with the Klf9 consensus motif. The top four most frequently observed Sp/Klf motifs (found in >50% of all peaks) are shown in Fig. 4a, and the remainder are listed in Additional file 9: Table S3. The Klf9 consensus motif, and the 18 additional motifs are hereafter collectively referred to as Sp/Klf motifs. At least one Sp/ Klf motif was present in 98% of all Klf9 peaks, supporting that the presence of Sp/Klf motifs is important for targeting Klf9 in the genome. In addition, HOMER identified 18 significantly enriched motifs in Klf9 peaks other than the Sp/Klf motifs, the most significantly enriched of which (p < 1×10 −20 ) matched previously reported binding sites for early growth response 2 (Egr2), ELK4, ETS transcription factor (Elk4), E2F transcription factor 1 (E2f1), Fos-like 2 (Fosl2), basic helix-loop-helix (bHLH), and regulatory factor X domain-containing 2 (Rfxdc2). A complete list of enriched motifs is given in Additional file 10: Table S4.
The Klf9 consensus motif tended to occur at or near the center of peaks, and the average density of mapped sequencing reads across the peaks closely matched the frequency of motif occurrence (Fig. 4b). The peaks contained between 0 and 19 Sp/Klf motifs (one outlier contained 48), with an average of 4.47/peak. There was a weak but statistically significant correlation between the number of nonredundant Sp/Klf motifs and average peak height ( Fig. 4c; R = .245, p < .0001; Spearman Rank Order correlation).
We also used HOMER to analyze the promoters of the 21 genes that were induced by Klf9. Sequences of 1000 bp in length located upstream of the TSSs of these genes were downloaded from the UCSC genome browser. Scanning these sequences for the presence of Sp/Klf motifs found by de novo analysis of the Klf9 ChSP peaks found that only 4 of 21 (19%) contained one copy of the Klf9 consensus motif, while 17 of 21 (81%) had at least one Sp/Klf motif.

Peak shape clustering reveals three separate categories of Klf9 peaks
We analyzed Klf9 peaks using the program SIC-ChIP [31], which clusters peaks into subcategories based on five shape parameters (peak height, peak width at halfmaximum height, peak area, number of local subpeaks, and shape index M (a measure of the peak's topological complexity normalized to height) [32]. Boxplots of the  The Klf9 consensus motif was highly enriched in peaks from all three categories. In genomic regions defined by Clusters 1 and 3, the motifs were located near the center of the peak, while in Cluster 2 the distribution was spread evenly across the peak (Fig. 5b, upper panel). The density of mapped reads (reflective of peak height) closely matched the distribution of Klf9 consensus motifs in all three clusters (Fig. 5b, lower panel). Sequencing read density (peak height) therefore correlates with the presence of Klf9 consensus motifs, even in wide peaks with multiple local maxima such as those seen in cluster 2. Clusters 2 and 3 had a larger number of Sp/Klf motifs per peak than Cluster 1 (Fig. 5c, upper panel; H (2) = 528.802, p < .001; Kruskal-Wallis nonparametric ANOVA). There was no significant difference between Clusters 2 and 3 in average number of Sp/Klf motifs, but peaks from Cluster 3 contained more copies of the Klf9 consensus motif (Fig. 5c, lower panel; H (2) = 245.055, p < .001; Kruskal-Wallis non-parametric ANOVA).
The proportion of peaks associated with Klf9-repressed genes differed among clusters, with Cluster 3 having the highest percentage, and Cluster 1 the lowest (Fig. 5d). This supports that large peaks with larger numbers of Sp/Klf and Klf9 consensus motifs are more likely to be associated with genes that are repressed by Klf9. In further support of this observation, the distribution of peaks from each cluster associated with repressed genes was nonrandom. Peaks from Cluster 1 accounted for 60.1% of all peaks, and 35% of repressed genes had at least one peak from Cluster 1 associated. In contrast, while peaks from Clusters 2 and 3 accounted for 23.1 and 15.6% of all peaks, respectively, 20.3 and 23% of all repressed genes had a peak from Clusters 2 or 3 associated (Additional file 12: Figure S8). Peaks from Clusters 2 and 3 are therefore more likely to be associated with repression by Klf9 than would be expected by chance (Chi-Square = 28.704 with 2°o f freedom, p < .001).
We also calculated the expression ratio of genes with peaks from each cluster in vehicle vs. dox-treated HT22 [TR/TO-Klf9] cells. Genes with peaks from Cluster 2 show a lower expression ratio (indicating greater repression) than genes with peaks from Cluster 1, and genes with peaks from Cluster 3 have a lower expression ratio than from Cluster 2, supporting that Klf9 exerts a stronger repressive effect on transcription from peaks with either a greater number of consensus motifs or a higher ChSP signal ( Fig. 5e; H Figure S3 for the ChSP peaks determined by PePr and the relative locations of the cloned DNA fragments, and Additional file 13: Table S5 for the DNA sequences and locations of Sp/Klf motifs within them). Treatment with dox for 24 h reduced luciferase activity from pGL4.23-Klf13 by 33%, pGL4.23-Klf16 by 45%, and pGL4.23-Limk1 by 19.5%; luciferase activity from the pGL4.23-Mapk11 vector was unaffected by dox treatment (Fig. 6; Klf13: t (10) = 8.512, p < 10 −5 ; Klf16: t (10) = 4.430, p < .005; Limk1: t (10) = 3.275, p < .01; Mapk11: t (9) = 1.509, p = .166; Klf13 mutant: t (10) = −1.331, p = .213; n = 6/ treatment; Student's two-sample t-test). To determine if Sp/Klf motifs are required for repression by Klf9 we focused on the Klf13 5′ flanking region, which contains 6 Sp/Klf motifs (Additional file 13: Table  S5). We used site-directed mutagenesis to convert these motifs to a series of 7 thymidines; the complete sequence of the Klf13 DNA fragment and the location of the mutated nucleotides are given in Additional file 13: Table S5. Mutation of two of the 6 Sp/Klf motifs (sites 4 and 5), either individually or in combination, did not affect repression by Klf9 (data not shown). However, mutation of all six sites abolished Klf9-dependent transcriptional repression (Fig. 6).
Forced expression of Klf9 promotes recruitment of Sin3a to some genomic regions with Klf9 peaks Our RNA-seq experiment showed that Klf9 acts predominantly as a transcriptional repressor in HT22 cells. The N-terminus of Klf9 has a motif for interaction with the scaffolding repressor protein Sin3a [33] which recruits histone deacetylases to generate a compact chromatin structure and transcriptional repression. We therefore investigated whether Sin3a was recruited to Klf9 peaks by conducting ChIP assay for Sin3a on chromatin isolated from HT22 [TR/TO-Klf9] cells treated with vehicle or dox for 12 h. Treatment with dox increased the mean Klf9 ChIP signal at eight genomic regions corresponding to Klf9 ChSP peaks; this increase was statistically significant for six of the eight peaks (Additional file 14: Figure S9A). There were statistically significant increases in Sin3a ChIP signal following dox treatment at the Klf16, Sin3a, Nr3c1 and Limk1 genes, but it was unchanged at Slc11a2, Klf13 and Klf11, and was reduced at Mapk11 (Additional file 14: Figure S9B

Depletion of Klf9 leads to dysregulation of Klf9 target genes
To determine if loss of Klf9 alters the expression of Klf9 target genes we used CRISPR/Cas9 genome editing to generate Klf9 knockdown (CRISPR line 1) and knockout (CRISPR line 2) HT22 cell lines. A description of the mutations introduced into these cell lines is given in Additional file 15: Table S6. The CRISPR line 1 exhibited significantly higher mRNA levels compared to wild type for all Klf9-repressed genes analyzed except Mapk11 (Fig. 7a; Slc11a2: F (2,12) = 9.091, p < .005; Klf13: F The CRISPR line 2 also had higher mean mRNA levels for all genes analyzed (again, except for Mapk11) and this was statistically significant for Klf13, Limk1 and Klf11.
We also analyzed mRNA levels for a subset of Klf9 target genes identified in HT22 cells in the hippocampus of Klf9-null mice. Klf9 mRNA in the mouse CNS is low at birth, then rises during the first 4 weeks of life, paralleling the postnatal rise in plasma TH concentration [34]. The mRNAs for Klf13, Limk1, Apc2 and Nlgn2 were dysregulated at PND7 in Klf9-null mice, although the direction of change differed among the genes ( Fig. 7b; Klf13: t (5) = −3.785, p < .01; Limk1, t (5) = −2.763, p < .05; Mapk11: t (4) = 2.777, p = .05; n = 4 animals/genotype; Student's two-sample t-test). Mapk11 mRNA was not different between wild type and Klf9-null mice. These expression differences disappeared by PND14 and remained unchanged at PND 60 (data not shown).

Gene ontology analysis supports roles for Klf9 in neuronal morphology and function
We conducted gene ontology (GO) and pathway analysis using GeneCoDis on Klf9-repressed genes ( Table 2) [35][36][37]. Four GO: PANTHER pathways were enriched among Klf9-repressed genes: "Cytoskeletal regulation by Rho GTPase", "Inflammation mediated by chemokine and cytokine signaling pathway", "Wnt signaling pathway", and "B Cell Activation" (Table 2). We excluded the Klf9induced genes from the pathway analysis because of the small number discovered and because we were able to validate only 3 of 6 tested.
We then conducted pathway analysis on the set of all genes with Klf9 ChSP peaks. The most enriched PANTHER pathways among this set of genes were "Apoptosis", "Integrin signaling pathway", "Inflammation mediated by chemokine and cytokine signaling pathway" and "Cytoskeletal regulation by Rho-GTPase" ( Table 2 and Additional file 16: Table S7). The overlap with the categories enriched among Klf9-repressed genes supports an important role for Klf9 in these pathways. We also conducted separate analyses on the genes with peaks of different shape clusters associated (see above). The top PANTHER pathways enriched in genes associated with Cluster 1 peaks were "Apoptosis", "Wnt signaling pathway" and "Egf receptor signaling pathway". In contrast, the top pathways enriched in the set of genes associated with Cluster 2 peaks were "Cytoskeletal regulation by Rho GTPase" and "Metabotropic glutamate receptor group II pathway", while Cluster 3 peaks were associated with genes in the pathways "PDGF signaling", "Fas signaling pathway" and "Cytoskeletal regulation by Rho GTPase" (Additional file 17: Table S8).

Depletion of Klf9 shortens cell cycle in HT22 cells and de-represses genes involved in cell proliferation
Previous work showed that Klf9 can reduce proliferation of different cell types [38][39][40]. One of the top GO: PAN-THER pathways enriched among Klf9-repressed genes was 'Wnt signaling'. The Wnt pathway has been shown to increase cell proliferation in diverse tissue types, including in neurons [41,42]. We therefore looked at whether the cell cycle was altered in Klf9-deficient HT22 cells using flow cytometry. To facilitate our ability to observe differences in the cell cycle we cultured cells in reduced serum (2% vs. 10%; see Methods). We found that both CRISPR cell lines had a significantly higher proportion of cells in M phase (and a lower proportion in G1/G0) compared with the parent HT22 cell line (

Discussion
Here we report the first genome-wide analysis of Klf9 genomic targets in a mammalian neuronal cell type. Klf9 has been implicated in several aspects of neural development and regeneration, and is regulated by hormones [5,7,9,10,18], NFκB [7], oxidative stress [43], and electrical activity [16]. We therefore set out to identify its genomic targets in neurons to understand how it mediates transcriptional responses to these stimuli, and regulates neuronal differentiation, survival and plasticity. We identified Klf9-regulated genes and genomic regions where Klf9 associates in chromatin using the mouse hippocampus-derived neuronal cell line HT22. We show that Klf9 functions primarily as a transcriptional repressor, that it associates in chromatin predominantly near TSSs, and that it regulates transcription of genes involved in cytoskeletal remodeling, Wnt signaling and inflammation, among other pathways. Furthermore, by analyzing a subset of the genes and genomic regions identified in HT22 cells, we verified that these genes are bona fide Klf9 targets in mouse hippocampus in vivo, supporting that the HT22 cell line represents a useful model for mature hippocampal neurons, at least for the purpose of identifying candidate Klf9 target genes. Taken together, our findings represent an important advance in understanding the diverse developmental and physiological roles that Klf9 has in the mammalian central nervous system.

Klf9 acts as a transcriptional repressor through association with proximal promoter regions
Findings from our RNA-seq experiment conducted in HT22 cells support that Klf9 acts predominately as a transcriptional repressor, with 10 times more genes repressed than induced. This is consistent with a recent study in glioblastoma cells that showed that Klf9 functions primarily as a transcriptional repressor [40]. Of the 10 Klf9-repressed genes that we attempted to validate by RTqPCR, we were able to confirm 9. However, of the 6 Klf9-induced genes tested by RTqPCR, we could validate only 3 (see Fig. 1e). Induction of these genes by Klf9 was transient, with all 3 genes returning to baseline by 24 h following dox treatment. By contrast, mRNA levels of 7 of the 9 validated repressed genes remained low at 24 h. Using ChSP-seq we found that Klf9 associates predominantly with genomic regions within 1 kb of TSSs. Genes found to be repressed by Klf9 by RNA-seq were much more likely than induced genes to have Klf9 associated in chromatin (i.e., 60% of repressed genes had Klf9 associated, but only 5% of induced genes had Klf9 associated with their genomic regions). These findings support that Klf9 represses gene transcription by associating in chromatin at or near proximal promoters. In support of this, the presence of Klf9 at proximal promoter regions was associated with weak repression when averaged across all genes with peaks associated (see Fig. 3d). Because only one of the genes induced by Klf9 had a peak associated and the induced genes showed only transient induction, or failed to validate by RTqPCR, we conclude that they are either false positives or indirect targets of Klf9. Importantly, we found that of 10 peaks investigated, 9 had Klf9 associated with them in chromatin isolated from adult mouse hippocampus. This supports that HT22 cells can be used as a model to identify potential Klf9 genomic targets in adult mouse hippocampus, and to investigate Klf9 function in chromatin in mature neurons.

Klf9 association in chromatin is strongly correlated with the presence of Sp/Klf motifs
The most enriched motif in Klf9 peaks was an Sp/Klf motif (GCCACGCCCMCY) that was present in 75% of all peaks; furthermore, 98% of all peaks had at least one Sp/Klf motif that was similar to the consensus motif. The consensus motif tended to occur near the center of peaks and its position correlated with the density of mapped reads, supporting that Klf9 binds directly to these DNA sequences.
Early studies on Klf9 function showed that it could act as a transcriptional activator or repressor depending on the number of Sp/Klf motifs present in the regulatory element [2,3]. In CV-1 cells, Klf9 activated transcription from promoters containing multiple Sp/Klf motifs, but repressed transcription from promoters containing a single Sp/Klf motif. Most Klf9 peaks in HT22 cells (88%) had more than one Sp/Klf motif, but Klf9 acted almost exclusively as a repressor, suggesting that its regulatory activity is not necessarily related to the number of Sp/Klf motifs present. Other studies have shown that Klf9 can activate or repress the same promoter in the same cell type depending on the developmental stage [44]. The chromatin environment and the complement of interacting proteins may be more important than the promoter sequence for determining whether Klf9 acts as an activator or repressor.
An additional Sp/Klf-like motif different from the Klf9 consensus motif was present in 14.84% of peaks (Additional file 10: Table S4). This motif most closely matches the binding site for Sp1, which typically acts as a transcriptional activator [45]. Thus, the relative concentrations of Klf9, which likely acts as a repressor, and Sp1 in the cell may determine the transcriptional activity from these loci [46]. However, it is also possible that these are binding sites for other Sp/Klf proteins (including Klf9), which may act as repressors or activators of transcription.
Seventeen other enriched motifs similar to known TF response elements were identified by HOMER in Klf9 peaks (Additional file 10: Table S4). Several of these motifs are response elements for TFs important for neuronal structure and function. For example, two of the motifs in this list match binding sites for members of the immediate early gene families Egr and Fos (present in 28.55% and 21.21% of peaks respectively), which are involved in neural activity-dependent transcription [47]. Klf9 is also induced by neural activity [16], and thus may modulate the activity of these TFs, or vice versa, when recruited to common genomic regions..
Another enriched motif (present in 24.31% of peaks) was the E-box. The E-box is required for transcriptional activation by the core circadian transcription factors CLOCK and Bmal1, which form a regulatory feedback loop with Per and Cry proteins to orchestrate circadian rhythms in most tissues [48][49][50][51]. Several Klf genes have been found to exhibit circadian oscillations in their expression, including Klf9 [52]. Klf9 mediates circadian variation in cell proliferation in keratinocytes [38], but to our knowledge its role in circadian physiology in other tissues has not been investigated. One of the GO: PANTHER pathways enriched among genes with Klf9 peaks is "Circadian Clock System" (Additional file 16: Table S7). This suggests that Klf9 may co-regulate CLOCK/Bmal1 target genes, and thus modulate the core loop of the cellular circadian clock. While other Klfs have been shown to function as clock-output genes to orchestrate circadian physiology [53,54], no Klfs have been shown to directly participate in the CLOCK/Per regulatory loop.

Clustering analysis provides a basis for differentiating peaks associated with repressed genes
Different types of chromatin-associated proteins have different ChIP-seq profiles (e.g. broad regions for histones, discrete regions for TFs) and different genomic regions associated with the same factor may have different peak shapes [31]. We found that Klf9 peaks could be divided into three groups. Large, sharp peaks (Cluster 3) or broad, complex peaks (Cluster 2) were more likely to be associated with genes repressed by Klf9 than were smaller, less complex peaks (Cluster 1). In addition, the genes associated with Cluster 3 peaks were on average more strongly repressed after dox treatment on HT22 [TR/TO-Klf9] cells (see Fig. 5e).
The different peak characteristics could reflect at least three basic mechanisms: different numbers of Klf9 molecules recruited, perhaps related to the number of Sp/Klf motifs; differences in the chromatin environment (i.e., nucleosome density and histone modifications) that modulates accessibility of the locus; or local differences in the presence of protein interacting partners that may stabilize Klf9 association at the region. Peaks from Clusters 2 and 3 had more Sp/Klf motifs than peaks from Cluster 1 (Fig. 5c), which is consistent with recruitment of more Klf9 molecules as measured by height and area under the curve (Additional file 7: Figure S5C and D). Cluster 3 peaks had the same average number of Sp/Klf motifs as Cluster 2 peaks, but they had more copies of the Klf9 consensus motif (Fig. 5c); the high, sharp signal at these loci could reflect a higher affinity for Klf9 binding to the Klf9 consensus motif leading to more Klf9 association. Alternatively, or in addition, the chromatin environment at these regions may be more accessible, and/or there is greater recruitment of Klf9 interacting proteins that leads to increased crosslinking efficiency. Whether the peak characteristics result in differences in the kinetics and/or magnitude of transcriptional regulation of the associated genes requires investigation.

Repression by Klf9 requires intact Sp/Klf motifs
To determine if the genomic regions where Klf9 associates are capable of supporting transcriptional repression, we isolated and subcloned several DNA fragments corresponding to peaks identified by ChSP-seq that were located in the 5' upstream regions of Klf9-responsive genes. Forced expression of Klf9 repressed transcriptional activity supported by the Klf13, Limk1 and Klf16 upstream regions, but not by Mapk11. For the Klf13 promoter we showed that Sp/Klf motifs are required for Klf9-dependent transrepression. Although the Mapk11 region has five predicted Sp/Klf-like motifs, it lacks any copies of the Klf9 consensus motif. By contrast, the Klf13, Limk1 and Klf16 upstream regions that we investigated have 3, 4 and 6 Klf9 consensus motifs, respectively. Based on these findings we hypothesize that the Klf9 consensus motif is essential for mediating Klf9dependent repression. The presence of multiple Sp/Klf motifs at most Klf9 peaks implies that such redundancy may be a feature of many, if not most, Klf9 genomic targets. These sites could also potentially act as binding sites for other Sp or Klf-like proteins, which suggests the potential for combinatorial regulation by a network of Klfs [55].

Cellular pathways targeted by Klf9 in HT22 cells
To evaluate cellular pathways regulated by Klf9 in HT22 cells we analyzed GO: PANTHER enriched pathways among Klf9-repressed genes identified by RNAseq, and also enriched pathways among genes with Klf9 peaks identified by ChSP-seq. The most enriched GO: PANTHER pathway among Klf9-repressed genes was "Cytoskeletal Regulation by Rho-GTPase". This pathway was number 4 on the list of enriched pathways among Klf9 bound genes (see Table 2). Among the Klf9-regulated genes in this category are Limk1, which has been shown to promote neurite extension through inhibition of actin depolymerization [56], and Mapk7 (ERK5), which positively regulates neurite extension and arborization in hippocampal neurons [57]. Several genes involved in actin polymerization, such as Limk1, were direct genomic targets of Klf9. The repression of positive regulators of actin polymerization such as Limk1 is consistent with published findings showing that Klf9 represses dendritic and axonal remodeling and outgrowth in mature neurons [13,14]. However, this does not explain how Klf9 promotes neurite elaboration and outgrowth in immature neurons [8,12]. It may regulate a different set of genes, perhaps repressing repressors of growth, or it may upregulate rather than repress the same loci identified in this study at earlier developmental stages [44].
The second most enriched pathway for Klf9-repressed genes was "Inflammation mediated by chemokine and cytokine signaling pathway", and this pathway was number 2 on the list of Klf9 bound genes. Several other Klfs have been implicated in immune and inflammatory responses [58]. For example, Klf4 promotes differentiation of monocytes and directly upregulates transcription of the pro-inflammatory cytokine interleukin-6 [59][60][61][62]. It also cooperates with the glucocorticoid receptor to induce expression of anti-inflammatory genes in keratinocytes [63]. Klf13 regulates expression of IL-4 in CD4 (+) T cells [64]. Klf10 has been found to regulate TGF-beta signaling in CD8 (+) T lymphocytes through modulating expression of TGF-beta receptor type II expression [65]. In contrast, Klf2 generally has anti-inflammatory actions [66]; for example, it inhibits inflammatory activation of monocytes through inhibition of the transcriptional activity of NF-κB [67] and blocks CD4 (+) T follicular helper cell differentiation [68]. However, Klf2 has also been shown to be required for activation of vascular endothelial cells in response to proinflammatory factors [69]. Klf6 has been shown to function as an accessory transcription factor for NF-kappa B [70]. To our knowledge, there have been no reports linking Klf9 to inflammatory signaling. However, Klf9 is induced by glucocorticoids in mouse macrophages [71] and in neurons [7,11,18] via conserved glucocorticoid response elements [7,18]. The repressive activity of Klf9 may serve as a counterbalance to pro-inflammatory Klfs such as Klf4, allowing for modulation of the immune response. It is also noteworthy that the Klf9 Synergy Module, which is an ultraconserved enhancer element located 4-6 kb upstream of the TSSs of tetrapod Klf9 genes, contains a NFkB site to which P50 associates, and LPS can modulate Klf9 expression in HT22 cells [7].
The top enriched cellular pathway for genes with Klf9 ChSP peaks was apoptotic signaling. Interestingly, these genes are most strongly associated with peaks from Cluster 1; whereas, peaks from Clusters 2 and 3 are enriched in cytoskeletal and growth factor-related pathways, respectively (see Additional file 17: Table S8). Genes with peaks from Clusters 2 and 3 were more likely to be repressed by 8 h of dox treatment than genes from Cluster 1. If the kinetics and/or magnitude of transcriptional regulation varies among genes from the different peak clusters, with genes from Cluster 1 exhibiting slower kinetics than genes from Clusters 2 and 3, then one can propose the following model: Rapid induction of Klf9 (e.g., in response to an acute stressor) suppresses morphological remodeling and cell proliferation by direct repression of a set of 'first tier' target genes (from Clusters 2 and 3). On the other hand, prolonged induction of Klf9 in response to chronic stress may promote repression of anti-apoptotic genes and induction of pro-apoptotic genes (Cluster 1 peaks or 'second tier' targets), leading to cell death. This model might be supported by findings of a role for Klf9 in promoting cell death in response to chemotherapeutics and oxidative stress [43,72].

Depletion of Klf9 in HT22 cells leads to de-repression of Klf9 target genes
Knock-down or knock-out of Klf9 in HT22 cells using CRISPR/Cas9 genome editing caused dysregulation of several Klf9 target genes. The genes Klf13, Klf16 and Limk1 were repressed by forced Klf9 expression, but were increased in Klf9 deficient HT22 cells, supporting a direct role for Klf9 in regulating transcription of these genes. The genes Slc11a2 and Klf11 were not repressed by forced Klf9 expression, but their mRNAs were increased following Klf9 depletion, and both genes had Klf9 peaks. Mapk11 did not show any change in expression in CRISPR knockout cells despite its weak repression by forced Klf9 expression and the presence of a Klf9 peak at its proximal promoter region. The Klf9 peak at Mapk11 was at the lower limit of detection by ChSP-seq (in fact, it was detected by the MACS algorithm, but not by PePr). 'Weak' peaks such as this may represent genes that are regulated under conditions of high Klf9 expression, but are not dependent on Klf9 for controlling baseline expression.
Despite the changes seen in HT22 cells, few differences in target gene mRNA level were seen in Klf9 KO mice compared to age-matched wild-type animals, and what differences did exist tended to disappear after the first postnatal week (not shown). This suggests that there are compensatory mechanisms for regulating baseline mRNA levels in the absence of Klf9. The Klf family members 3, 5, 7, 10, 12, 13, 14 and 16 are all expressed in the adult mouse hippocampus, based on in situ hybridization data provided in the Allen Brain Atlas [73]. Some or all of these may contribute to compensatory regulation in the absence of Klf9. Functional compensation between Klfs has been demonstrated in embryonic stem cells, in which Klfs 2, 4 and 5 form a partially redundant network that maintains pluripotency. Knockout of all three is required to induce differentiation into fibroblasts [55]. Klf13, 14 and 16 are the Klf family members most similar to Klf9 and are therefore the most likely candidates for compensating for loss of Klf9 in the hippocampus [1]. In the mouse uterus Klf9 and Klf13 have partially overlapping functions such that Klf9 can partially compensate for the absence of Klf13 [74]. We found that Klf9 associates with and represses transcription from the promoters of Klf13 and Klf16 in HT22 cells and associates with these genes in the mouse hippocampus in vivo. To our knowledge this is the first evidence of repression of direct cross-regulation between Klf9 and Klf13 and supports that these Klf proteins may regulate each other's expression as well as activating or repressing shared targets. Investigation of the Klf13 and Klf16 cistromes in HT22 cells and the mouse hippocampus will be important to shed further light on the extent of cooperation or antagonism among Klf9, Klf13 and Klf16 in neurons.

A role for Klf9 in modulation of the cell cycle in HT22 cells
Previous studies showed that Klf9 can suppress cell proliferation and/or promote apotosis [40,72]. Consistent with these findings, we found that knockdown or knockout of Klf9 shortened the cell cycle in HT22 cells and increased expression of genes involved in cell proliferation. The Klf9-repressed genes Bcl-6 and Itpr3 are part of the Wnt signaling pathway. Bcl-6 was shown to suppress apoptosis through induction of cyclin d1, and is sufficient to immortalize mouse embryonic fibroblasts [75]. In a human breast cancer cell line (MCF-7) knockdown of Itpr3 caused cell cycle arrest, supporting that it is important for promoting tumor proliferation [76]. Furthermore, Itpr3 can confer resistance to apoptosis [77]. The increase in expression of these two genes, and possibly other genes involved in cell proliferation and survival, may underlie the increase in cell proliferation that we observed after depletion of Klf9 in HT22 cells. This would be consistent with reports that Klf9 can act as a tumor suppressor and promote apoptosis in response to chemotherapeutics [39,40,72].

Conclusions
We provide the first unbiased analysis of Klf9 genomic targets in mouse hippocampal neurons. We also provide strong evidence for a cross-regulatory network of Klfs that may be important for regulating the transcriptome. Our findings establish a foundation for understanding the molecular basis for Klf9's effects on neuronal morphology and cell proliferation. Finally, we demonstrate the utility of HT22 cells for identifying genomic targets of TFs in the mouse hippocampus.

Plasmids
We obtained the HT22 cell line from Dr. David Schubert, the Salk Institute, La Jolla, CA. To engineer HT22 cells lines for controlled Klf9 expression we used the pRSV-BTEB plasmid (gift of Dr. Fuji-Kuriyama) as template to PCR-amplify the full-length Klf9 cDNA. We then directionally cloned the cDNA into the pCDNA4:TO (Invitrogen) and pEF1α-FLBIO (gift of Jianlong Wang; Kim et al., 2009) expression vectors at the XhoI/XbaI (pCDNA4:TO) or BamhI/XbaI (pEF1α-FLBIO) sites. To construct a vector to express the biotin ligase BirA we used the pEF1α-BirA plasmid (gift of Jianlong Wang) (Kim et al., 2009) as template to PCR-amplify a DNA fragment containing the EF1α promoter, the BirA coding sequence and the polyadenylation sequence. We subcloned this DNA fragment into the pSV40 zeocin plasmid (gift of Dr. Michael Uhler) at the NotI and NheI sites. Note that HT22 cells are resistant to G418 and the pEF1α-BirA plasmid has a neomycin cassette, so we needed to move the BirA sequence to a plasmid containing a different antibiotic resistance gene (zeocin). To express Klf9 in transient transfection assays we used pCS2-Klf9 [78].

Generation and characterization of stable HT22 cell lines with doxycycline (dox)-inducible Klf9
We cultured HT22 cells (gift of Dr. David Schubert) in high-glucose DMEM (Invitrogen) supplemented with 10% fetal bovine serum (Hyclone Laboratories, Inc., Logan, UT), penicillin G (100 U/ml) and streptomycin sulfate (100 μg/ml) under a humidified atmosphere of 5% CO 2 at 37°C. This cell line was originally derived from mouse hippocampus using the Simian Virus 40 T antigen [19]. To generate stably transfected HT22 [TR/TO-Klf9] cell lines we seeded 5 × 10 6 cells in 100 mm plates and transfected them with 5 μg each of pCDNA4:TO-Klf9 and pCDNA6:TR (Invitrogen) using Fugene6 (Invitrogen). Twenty four hr after transfection we changed to selective medium containing 100 μg/ml zeocin plus 5 μg/ml blasticidin (Research Products International, Mt. Prospect, IL). Louis, MO) and passaging with 0.25% trypsin (Gibco; Thermo-Fisher). We then seeded trypsinized cells in 6well plates and expanded the clonal lines. We tested seven clonal HT22 [TR/TO-Klf9] lines for dox-inducible Klf9 mRNA by culturing them in 6-well plates and treating with vehicle (0.1% DMSO) or 1 μg/ml dox (Sigma; all dox treatments were done at this concentration) for 8 h, after which cells were harvested for RNA isolation and analysis for Klf9 mRNA levels by RTqPCR (described below). To test if the stable cell lines expressed functional Klf9 we seeded cells at 5 × 10 4 per well in 24 well plates and co-transfected them with pGL4.23-3xBTE (200 μg) plus pRenilla (10 μg) (Promega Corporation, Madison, WI). We treated transfected cells with vehicle or dox for 8 h and harvested for dual luciferase assay (Promega). As a control, we co-transfected parent HT22 cells with pGL4.23-3xBTE plus a Klf9 expression vector (pCS2-Klf9) to independently investigate the action of Klf9 on this reporter.
For gene expression analysis we plated HT22 cells in either 6-or 12-well plates at densities of either 2.5 × 10 5 or 1.25 × 10 5 , respectively, and began treatment with dox 24 h after plating. We then harvested cells and extracted RNA at different times after dox treatment for analysis by RTqPCR.
For luciferase assays, we cultured cells in 24-well plates at a density of 5 × 10 4 cells/well. Twenty-four hr after plating we transfected with 200 ng/well of the luciferase vector and 10 ng/well pRenilla plasmid to monitor transfection efficiency, and 24 h later we treated cells with vehicle or dox for 8 h (pGL4.23-3xBTE vector) or 24 h (all other luciferase vectors). We harvested cells for luciferase assay using the Dual Luciferase Reporter Assay System (Promega) according to the manufacturer's instructions. Firefly luciferase activity was quantified using a luminometer (Femtometer FB 12; Zylux Corp) and normalized to Renilla luciferase activity. All transfection reporter assays were repeated at least two times with 5-6 replicates/ treatment.

RNA extraction, reverse transcription and quantitative PCR
We extracted total RNA from HT22 cells or mouse brain (hippocampal region; see [18] for method) using the TRIzol reagent (Invitrogen) following the manufacturer's instructions. We treated total RNA with DNase 1 (20U; Roche, Basel, Switzerland) to remove contaminating genomic DNA and conducted reverse transcription with 1 μg RNA using the High Capacity Reverse Transcription kit with ribonuclease inhibitor (Applied Biosystems, Life Technologies Corp, Foster City, CA). For RTqPCR we used Taqman assays for Gapdh and Klf9 [7] and SYBR green assays for all other genes. All oligonucleotide primer sequences are given in Additional file 18: Table S9. We conducted RTqPCR using an ABI 7500 fast real-time PCR machine with Absolute qPCR low ROX mix (for Taqman assays) or Absolute qPCR SYBR low ROX mix (ABgene, Epsom, UK). We designed SYBR green assays using Integrated DNA Technology's RealTime qPCR Assay tool; where possible we designed assays to span exon-exon boundaries. We used a relative quantitation method using serial dilutions of a cDNA pool to generate standard curves. We normalized all genes to the reference gene Gapdh whose mRNA was unaffected by treatments (data not shown.)

Generation of stable HT22 cell lines that express biotinylated Klf9
We produced stably transfected HT22 cell lines that express biotinylated Klf9 by culturing cells in 100 mm plates and transfecting them with 5 μg each of pEF1α-BirA and pEF1α-FLBIO-Klf9 (HT22 [BirA/FLBIO-Klf9]). Twenty four hr after transfection we treated cells with 100 ug/ml zeocin plus 1 μg/ml puromycin. To make stably transfected HT22 cell lines that express only BirA (HT22 [BirA]) we transfected cells with 5 μg pEF1α-BirA. Twenty four hr after transfection we treated cells with 100 ug/ml zeocin. Following antibiotic selection for 5 days we isolated individual cells by trypsinization using cloning cylinders and subcultured them as described above.
To determine if the stably transfected cell lines expressed the FLBIO-Klf9 fusion protein and/or BirA-V5 we analyzed nuclear extracts by Western blotting. For nuclear extraction we cultured HT22 [BirA] or HT22 [BirA/ FLBIO-Klf9] cells in 100-mm plates until they reached 95% confluency, collected cells with a cell scraper into hypotonic buffer (10 mM HEPES pH 7.9, 10 mM KCl, 1 mM DTT), allowed them to swell for 30 min, then lysed them with a motorized homogenizer and added 0.1 volume of sucrose restore buffer (50 mM HEPES pH7.9, 10 mM KCl, 1 mM DTT, 6.75% sucrose). We isolated nuclei by centrifugation at 4000 × g for 15 min, removed the supernatant (the cytoplasmic fraction), resuspended nuclei in nuclear extraction buffer (50 mM HEPES pH 7.9, .5 M KCl, 1 mM DTT) and incubated on ice for 40 min with vortexing every 10 min, then centrifuged for 1 h at 100,000 × g. We quantified the protein concentration of the extracts using the Pierce assay (Thermo Scientific). We then fractionated 40 μg of nuclear protein per lane of a 10% SDS-PAGE gel, transferred proteins to nitrocellulose membrane and blocked the membrane with SuperBlock (Thermo; following the manufacturer's instructions; for BirA-V5), or with PBS containing 5% bovine serum albumin, 10% normal goat serum, 0.5% Triton-X (for FLBIO-Klf9). To detect BirA-V5 we incubated the membrane with V5 antibody (Millipore; 1:5000 dilution) overnight before washing and incubation with HRP-conjugated goat anti-rabbit secondary antibody for 1 h (Jackson ImmunoResearch Laboratories, Inc.); diluted 1:30,000). To detect FLBIO-Klf9 we incubated the membrane with Streptavidin-HRP (Thermo; diluted 1:1000) for 1 h. Immune or streptavidin-HRP complexes were revealed by chemiluminescence detection using Pierce ECL Substrate (ThermoFisher Scientific).

RNA Sequencing
For RNA-seq analysis we treated the HT22 parent line or the HT22 [TR/TO-Klf9] line 2-1 with vehicle or dox for 8 h. We extracted total RNA using TRIzol reagent, then purified it using the QIAgen RNEasy kit. Library construction and next generation sequencing was done at the University of Michigan DNA Sequencing Core on twelve RNA samples (1 μg/sample) representing the four treatments (n = 3/treatment). The twelve samples were analyzed in two lanes using an Illumina 2000 HI-seq machine, which generated between 23,555,652 and 50,630,253 million reads per sample. The sequencing reads were de-multiplexed by the University of Michigan DNA Sequencing Core, evaluated and filtered using FastQC software (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and aligned to the mouse genome (build mm8) using the program Bowtie [21]. We quantified mRNA levels using DESeq [22]. To calculate gene expression ratios, counts from all genes that were detected by RNA-seq in each cell line were average by treatment, and average counts from dox-treated cells were divided by average counts from vehicle-treated cells. Genes where no reads were mapped in one or both treatments were excluded from the analysis.

Chromatin extraction and precipitation
Chromatin extraction and chromatin immunoprecipitation (ChIP) were done as described previously [5,7,18]. We grew HT22 cells in 100 mm plates, and for some experiments (e.g., preparation of samples for ChSP-seq libraries) we pooled cells from two plates to increase the amount of chromatin recovered. After washing with Dulbecco's Phosphate Buffered Saline (DPBS), we treated cells with 1% formaldehyde for 10 min followed by treatment for 10 min with the bifunctional crosslinking agent dithiobis (succinimidyl propionate) (DSP; 200 nM; Thermo Scientific) before extracting chromatin. We sonicated the chromatin using an M220 Focused-Ultrasonicator (Covaris, Woburn, MA) for 20 min using a 2% duty factor, and checked that the DNA had been sheared to 500-600 bp by electrophoresis on a 1% Tris acetate EDTA agarose gel. We then flash-froze the chromatin with liquid nitrogen and stored it at − 80°C until analysis. For each ChIP reaction we used 5 μg purified IgG of goat anti-mBTEB-C17, goat anti-mSin3a-K20, or normal goat IgG (Santa Cruz Biosciences, Santa Cruz, CA).
We conducted ChSP following the method of Ramadoss and colleagues [79]. We first washed MyOne T1 streptavidin-conjugated Dynabeads (Invitrogen; 50 ul/ reaction) three times with phosphate buffered saline (PBS; pH 7.4), then mixed the beads with 50 μg of chromatin (5 μg of chromatin was reserved for input) and the total volume was brought to 1 ml with ChIP dilution buffer (.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-HCl pH 8.1, 150 mM NaCl). The chromatin and beads were rocked overnight at 4°C. The beads were then collected by magnet and the supernatant removed, then washed three times with 1 ml of 0.5× RIPA buffer (5 mM Tris-Cl pH 8.0, 0.5 mM EDTA, 0.25 mM EGTA, 70 mM NaCl, 0.5% Triton X-100, 0.05% sodium deoxycholate, 0.05% SDS). After the final wash, we resuspended the beads in decrosslinking buffer (50 mM Tris-Cl pH 8.0, 1 mM EDTA, 100 mM NaCl, 0.5% SDS) and removed crosslinks by incubation at 65°C overnight (after this step the input samples were processed simultaneously with the ChSP samples). The DNA was then extracted with phenol:choloroform:isoamyl alcohol (Invitrogen) and precipitated with 0.3 M sodium acetate and 100% ethanol. We added 20 μg (1 μl) molecular biology-grade glycogen (Roche) before precipitation to visualize DNA pellets. After precipitation we resuspended the DNA in 25 μl nuclease-free water and analyzed by qPCR for the Klf13 promoter and intronic regions. Samples were quantified as a percentage of corresponding input sample.

Chromatin streptavidin precipitation sequencing
For ChSP sequencing (ChSP-seq) we prepared DNA precipitated from 150 μg of chromatin per sample as described above, and submitted ten samples to the University of Michigan DNA Sequencing Core for next generation sequencing: one input DNA sample (pooled from four input samples) from each of the two cell lines HT22 [BirA] and HT22 [BirA/FLBIO-Klf9], and four ChSP DNA samples from each cell line. We ran 5 samples per lane which generated between 18,818,940 and 27,816,127 million reads per sample. We filtered the sequenced reads using FastQC and aligned them to the mouse mm8 genome using Bowtie [21]. We then used the NCBI Genome Remapping Service to transfer peaks to the mm10 build. We identified peaks using MACS [25] and PePr software [24] and assigned peaks to genes using ChIP-enrich [28]. Additional analysis of the locations of peaks and mapping of regions of highest peak density was conducted using Cis-regulatory Element Annotation System software [29]. We clustered peaks by shape using the program SIC-ChIP [31]. Motif enrichment was analyzed using HOMER [30], and gene ontology analysis was done using GeneCoDis [35][36][37].
Generation of HT22 Klf9-knockdown and knockout cells using CRISPR/Cas9 genome editing We constructed a guide RNA (gRNA)/Cas9 expression plasmid (OriGene) containing the gRNA sequence 5′ ggggcgctccggaagccgag 3′ (this gRNA targets a sequence in the 5′ region of the Klf9 gene). We transfected the parent HT22 cell line with this vector, which expresses enhanced green fluorescent protein (EGFP), and conducted fluorescence-assisted-cell sorting (FACS) at the University of Michigan Flow Cytometry core to isolate EGFP positive cells. We isolated clonal lines of the FACS-sorted cells as described above, extracted genomic DNA and screened for mutations by direct DNA sequencing of PCRamplified DNA corresponding to Klf9 exon 1 subcloned into the pGEM T-easy vector (Promega). We selected two Klf9-null lines (defined by the absence of a wild-type Klf9 allele and the presence of mutations predicted to create nonsense or prematurely truncated proteins), cultured them simultaneously with the parent HT22 line in 6-well tissue culture plates, harvested cells, extracted RNA and analyzed gene expression by RTqPCR.

Analysis of cell cycle by flow cytometry
Approximately 150,000 cells from the HT22 parent, CRISPR Line 1 and CRISPR Line 2 cell lines were plated in 10 cm plates in DMEM supplemented with 10% FBS. Sixteen hr after plating we changed the medium to DMEM containing 2% serum. After a further 72 h we trypsinized cells and pelleted them by centrifugation for 5 min at 200 × g. The cells were then washed by resuspending the pellet in DPBS containing 1% BSA and pelleting by centrifugation at 1000 rpm for 5 min. Cells were then fixed for 15 min at room temperature by resuspending the pellet in 100 ul of 4% paraformaldehyde in DPBS. The fixed cells were then centrifuged at 1000 rpm for 5 min, then washed in DPBS containing 1% BSA and centrifuged again. Cells were permeabilized by resuspension in 100 μL of 1× Saponin buffer (Click-iT EdU Flow Cytometry Assay Kit, Invitrogen) containing 1% BSA. Cells were stained in darkness for 30 min at room temperature using FxCycle Violet (Thermo Fisher) diluted 1:1000 in DPBS, and analyzed using an Attune Cytometer (Applied Biosystems) at 405 nm wavelength. The relative proportion of cells in G1/G0 and M phases was determined using ModFit LT (Verity Software House, Topsham, ME).

Animals
We purchased C57/BL6 mice from Jackson Laboratories (Sacramento, CA) and maintained them on a 12 L:12D photoperiod with food and water provided ad libitum. Animals were killed by rapid decapitation and a section of the brain that included the hippocampus was dissected and flash-frozen for subsequent chromatin or RNA extraction (see above). Mice null for Klf9 were bred from animals provided by Dr. Yoshiaki Fuji-Kuriyama and Dr. Frank Simmen [6]. All procedures involving animals were approved by the Institutional Animal Care and Use Committee of the University of Michigan.

Statistical analysis
Derived (normalized) values from RTqPCR, ChIP, ChSP and dual luciferase assays were log 10 transformed before analysis by one-way ANOVA followed by the Holm-Sidak