RED HOT Contributors

 

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

0
CREDIT: This post was originally published on this site

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

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

].

Fig. 1

Identification of Klf9-regulated genes in HT22 cells by RNA-sequencing. a Treatment of HT22 [TR/TO-Klf9] cells with doxycycline (dox; 1 μg/ml) for 8 h increased Klf9 mRNA ~10 fold compared to vehicle treated cells, but had no effect in parent HT22 cells. The baseline Klf9 mRNA level did not differ between parent and [TR/TO-Klf9] line. Means with the same letter are not significantly different (p < .05; ANOVA followed by Tukey’s post-hoc test). b Dox-induced expression of Klf9 reduces luciferase activity from the pGL4.23-3xBTE plasmid, but not from pGL4.23-empty. The asterisk indicates a statistically significant difference by Student’s two-sample t-test (p < .05). c Validation by RTqPCR of four genes found to be repressed by Klf9 by RNA-seq. Treatment with dox for 8 h reduced mRNA levels for the four genes in HT22 [TR/TO-Klf9] but not in parent HT22 cells. The asterisks indicate statistically significant differences from parent cells treated with vehicle or dox, and from HT22 [TR/TO-Klf9] cells treated with vehicle (p < .05; ANOVA followed by Tukey’s post-hoc test). d Time-course showing induction of Klf9 mRNA following treatment of HT22 [TR/TO-Klf9] cells with dox. e Validation of repression (9 genes) or induction (3 genes) of Klf9 target genes after treatment of HT22 [TR/TO-Klf9] cells with dox for different times. The asterisks indicate statistically significant differences from the zero time point (p < .05; ANOVA followed by Tukey’s post-hoc test)

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

]. The parent HT22 cell line treated with dox showed no gene expression differences compared to parent cells treated with vehicle (false discovery rate [FDR]-adjusted

p

p

2

fold change). The top 10 most strongly induced and repressed genes are listed in Table 

1

, and a list of all differentially regulated genes is given in Additional file

2

: Table S1. We validated repression by Klf9 of 4 genes by RTqPCR on RNA isolated from parent and HT22 [TO/TR-Klf9] cells treated with vehicle or dox for 8 h (Fig. 

1c

;

Klf13

:

F(3,19)

 = 7.708,

p

Limk1

:

F(3,17)

 = 6.417,

p

Mapk11

: F

(3,19)

 = 22.107,

p

Pou6f1

:

F(3,19)

 = 4.286,

p

n

 = 4–6/treatment; ANOVA).

Table 1

The top ten most up- or down-regulated genes by eight hr of forced Klf9 expression in HT22 [TR/TO-Klf9] cells (FDR-adjusted p value cutoff of .005)

Klf13

Krüppel-like factor 13

−1.21

Rptoros

Regulatory associated protein of MTOR, complex 1, opposite strand

−1.12

Gpr161

G Protein-coupled receptor 161

−1.11

Apc2

Adenomatosis polyposis coli 2

−1.04

Zfp704

Zinc finger protein 704

−1.02

Arhgap39

Rho GTPase Activating Protein 39

−0.96

Armc7

Armadillo Repeat Containing 7

−0.95

Mex3a

Mex-3 RNA Binding Family Member A

−0.92

Wdfy2

WD Repeat And FYVE Domain Containing 2

−0.89

Nlgn2

Neuroligin2

−0.88

Frk

Fyn-Related Src Family Tyrosine Kinase

0.43

1810043G02Rik

RIKEN cDNA 1810043G02

0.46

4930430 F08Rik

RIKEN cDNA 4930430 F08

0.47

2610301B20Rik

RIKEN cDNA 2610301B20

0.48

Tctn2

Tectonic Family Member 2

0.49

Arxes2

adipocyte-related X-chromosome expressed sequence 2

0.51

Mitd1

Microtubule Interacting and Transport Domain Containing 1

0.51

Rilpl2

Rab Interacting Lysosomal Protein-Like 2

0.51

Zfp930

Zinc-finger protein 930

0.52

Cdkn3

Cyclin-Dependent Kinase Inhibitor 3

0.56

To further validate our RNA-seq data set, and to investigate the kinetics of Klf9-dependent gene repression, we conducted RTqPCR on HT22 [TO/TR-Klf9] cells treated with dox for different times. We observed a statistically significant increase in Klf9 mRNA by 2 h, which peaked at 4 h and remained elevated through 24 h of dox treatment (Fig. 1d; F (5,16) = 14.97, p < .001; n = 4/time point; ANOVA). We then conducted RTqPCR on 10 Klf9-repressed and 6 Klf9-induced genes with log2 fold changes ranging from − 0.44 to − 1.12 for repressed, or 0.38 to 0.56 for induced genes. We validated 9 repressed genes (Fig. 1e; Klf13: F (5,16) = 6.642, p < .001; Limk1: F (5,15) = 4.048, p < .05; Mapk11: F (5,16) = 4.787, p < .01; Pou6f1: F (5,15) = 4.527, p < .05; Klf16: F (5,14) = 6.914, p < 0.05; Apc2: F (5,16) = 5.548, p < .005; Nlgn2: F (5,16) = 3.185, p < .05; Smurf1: F (5,16) = 4.689, p < .01; Nyap1: F (5,16) = 3.474, p < .05; n = 4/time point; ANOVA). Messenger RNA for Hhipl was unaffected (data not shown). However, of the 6 Klf9-induced genes tested, we could validate only 3 (Frk, Tctn2 and Cdkn3; Mitd1, Rilpl2 and Ostm1 mRNAs were unaffected; data not shown). The mRNAs of the 3 induced genes that we validated were increased following 8–12 h of dox treatment, but returned to baseline by 24 h (Fig. 1e; Frk: F (5,12 = 4.721, p < .05; Tctn2: F (5,16) = 3.055, p < .05; Cdkn3: F (5,14) = 3.736, p < .05; n = 4/time point; ANOVA).

Identification of sites across the HT22 genome where Klf9 associates in chromatin

We engineered HT22 cells to express the

E. coli

biotin ligase BirA (HT22 [BirA]) or BirA plus a Klf9 fusion protein with an N-terminal FLAG tag and biotin ligase recognition peptide (HT22 [BirA/FLBIO-Klf9]) [

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-of-principle 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)

 = 3.607,

p

n

 = 4/treatment; Student’s two-sample

t

-test). Next, we conducted ChSP on chromatin from HT22 [BirA] and HT22 [BirA/FLBIO-Klf9] cells. This resulted in ~25 fold enrichment above background (ChSP on chromatin from HT22 [BirA]) at the

Klf13

promoter but not at the same

Klf13

intronic region (Fig. 

2b

; t

(6)

 = −8.315,

p

n

 = 4/cell line; Student’s two-sample

t

-test). These findings support that both native Klf9 and the FLBIO-Klf9 fusion protein associate in chromatin at the

Klf13

locus, and that the ChSP technique results in a greater signal/noise ratio than ChIP assay.

Fig. 2

Identification of genome-wide association of Klf9 in HT22 cell chromatin using chromatin-streptavidin precipitation sequencing. a HT22 [BirA/FLBIO-Klf9] cells express biotinylated Klf9. Whole cell extracts from the HT22 parent cell line (lane 1), HT22 [BirA] cells (lane 2) or HT22 [BirA/FLBIO-Klf9] (lane 3) were fractionated on 10% SDS-PAGE and analyzed by Western blotting using streptavidin-HRP. b Chromatin-streptavidin precipitation gives ~25-fold enrichment at the Klf13 promoter in HT22 [BirA/FLBIO-Klf9] cells compared with HT22 [BirA] cells. Precipitated DNA was analyzed by qPCR at the Klf13 promoter and intron (negative control region). The asterisk indicates a statistically significant difference by Student’s two-sample t-test (p < .0005). c Genome Browser (University of California, Santa Cruz) views showing the location of Klf9 peaks at eight Klf9-repressed genes. Top track = reads from cells expressing BirA alone; bottom track = cells expressing BirA + FLBIO-Klf9. The 5′ flanking region of each locus is shown; bars below the peaks represent exons, lines represent introns. d Validation of ChSP-seq peaks (shown in C) by targeted ChSP with quantitative PCR. Chromatin isolated from HT22 [BirA] and HT22 [BirA/FLBIO-Klf9] cells is compared. The asterisks indicate statistically significant differences analyzed by Student’s two-sample t-test (p < .005). e Klf9 associates with the same genomic loci in mouse hippocampus in vivo as in HT22 cells. Targeted chromatin immunoprecipitation (ChIP) assays for Klf9 were conducted on chromatin isolated from adult mouse hippocampus. The asterisks indicate statistically significant differences from the normal goat serum (NGS) IgG control analyzed by Student’s two-sample t-test (p < .05)

We conducted ChSP sequencing (ChSP-seq) on chromatin isolated from HT22 [BirA] and HT22 [BirA/FLBIO-Klf9] cells to identify sites of Klf9 association in chromatin. We used two independent peak calling programs (MACS and PePr) to identify genomic regions with higher densities of mapped reads [24, 25]. MACS analysis identified 8,841 peaks, while PePr analysis identified 3,382 peaks present in HT22 [BirA/FLBIO-Klf9] but not in HT22 [BirA] cells. All except four of the peaks called by PePr were also called by MACS; the peaks called by both programs were more enriched (larger difference in the number of mapped reads in HT22 [BirA/FLBIO-Klf9] compared to HT22 [BirA] cells) than those called only by MACS (not shown). 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.

Validation of Klf9 peaks identified by ChSP in HT22 cells

To validate the ChSP-seq dataset we analyzed 8 Klf9 peaks using targeted ChSP- and ChIP-qPCR assays. The aligned sequencing read densities from the 8 genomic regions are shown in Fig. 2c, with peaks arranged from largest (220 overlapping reads at maximum height for Slc11a2; upper left panel) to smallest (20 overlapping reads for Mapk11; lower right panel). The small peak at the Mapk11 5′ upstream region was detected by MACS but not by PePr (see also Additional file 4: Figure S3); we analyzed this region to investigate the lower limit of detection of the ChSP-seq data set. All genomic regions tested showed significantly higher signal with ChSP DNA from HT22 [BirA/FLBIO-Klf9] cells compared with HT22 [BirA] cells (Fig. 2d; Slc11a2: t (6) = −10.981, p < .001; Klf16: t (6) = −14.417, p < .001; Klf13: t (6) = −10.981, p < .001; Sin3a: t (6) = −10.135, p < .001; Nr3c1: t (6) = −9.402, p < .001; Limk1: t (6) = −8.561, p < .001; Klf11: t (6) = −8.653, p < .001; Mapk11: t (6) = −5.448, p < .005; n = 4/cell line; Student’s two-sample t-test). We also conducted ChIP assay for Klf9 on chromatin extracted from the parent HT22 cell line, which showed statistically significant Klf9 ChIP signal (compared to IgG from normal goat serum) at four genes tested (Additional file 6: Figure S4; Klf16: t (4) = 6.285, p < .005; Limk1: t (4) = 4.093, p < .05; Nr3c1: t (4) = 3.778, p < .05; Sin3a: t (5) = 3.159, p < .05; n = 4/treatment; Student’s two-sample t-test). The Klf9 ChSP signal was at the background level at intronic regions located 10 kb or more downstream from the identified Klf9 peaks at the Klf16, Limk1 and Nr3c1 genes (Additional file 7: Figure S5A).

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 t-test). 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.

Fig. 3

Klf9 associates near transcription sites and is more likely to be associated with repressed than induced genes. a Distribution of Klf9 peaks with respect to transcription start sites (TSS). b Distribution of Klf9 peaks with respect to gene features. c Overlap of Klf9-regulated genes with genes containing Klf9 peaks within 10 kb of their TSS. d The median expression ratio of all genes expressed in HT22 cells, with or without associated Klf9 peaks, in parent HT22 and HT22 [TR/TO-Klf9] cells treated with vehicle or doxycycline (1 μg/ml) for 8 h. In the parent cell line there was no expression difference after dox treatment whether or not the genes have Klf9 peaks associated (within 10 kb of their TSS; p = .057; Mann–Whitney U-test). In the HT22 [TR/TO-Klf9] cell line treatment with dox caused a statistically significant decrease in median expression ratio, but only for genes with associated Klf9 peaks (p < .001; Mann–Whitney U-test)

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 expression ratio for genes containing Klf9 peaks in the HT22 [TR/TO-Klf9] cell line (Fig. 3d; parent line: p = .057; HT22 [TR/TO-Klf9]: p < .001; Mann–Whitney U-test). 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

−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.

Fig. 4

Sp/Klf motifs are enriched at Klf9 peaks, they correlate with peak features, and they are required for transcriptional repression by Klf9. a Position weight matrices showing the Klf9 consensus motif (left) and the four most commonly occurring Sp/Klf motifs that were partially redundant with the consensus motif (right). b Histogram showing the probability of the Klf9 consensus motif occurrence (solid) and average enrichment (number of mapped reads; dashed) across all Klf9 peaks. c Number of non-redundant Sp/Klf motifs correlates with the mean tag density across a peak (R = .245, p = .0000002)

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 non-redundant 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 half-maximum 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 distribution of each parameter in each cluster, and a scatterplot of how shape parameters correlate with each other and group peaks by cluster are shown in Additional file

11

: Figure S7. By these criteria we divided the peaks into three categories, examples of which are shown in Fig. 

5a

. Peaks in Cluster 1 (2115 peaks) are of low average height and low complexity; peaks in Cluster 2 (812 peaks) are of low height but greater complexity (as measured by a higher M index and larger number of local subpeaks); peaks in Cluster 3 (549 peaks) are large and of low complexity.

Fig. 5

Klf9 peaks group into shape clusters that show different likelihood of association with repressed genes, and degree of transcriptional repression. a Klf9 peaks in HT22 cells can be grouped into three clusters based on shape and height characteristics. Example peaks from each cluster are shown. The graph height (200 pixels on UCSC genome browser) and viewing window size (2 kb) are held constant. b Distribution of the Klf9 consensus motif (top) and mapped sequencing read density (bottom) within peaks from each shape cluster. Sequencing read density closely tracks the probability of the occurrence of consensus motifs in all clusters. c Average number of Sp/Klf motifs and number of Klf9 consensus motifs present in peaks belonging to each cluster. Boxplots indicate 75% and 25% quantiles; dots indicate 5% and 95% quantiles. Means with the same letter are not significantly different (p < .05; Kruskal-Wallis non-parametric ANOVA). Top: peaks of clusters 2 and 3 contain more Sp/Klf-like motifs than those of cluster 1; bottom: peaks of cluster 3 have more copies of the Klf9 consensus motif than peaks of clusters 1 and 2. d Percentage of peaks of each cluster associated with repressed genes. e Median ratio of mRNA levels in HT22 [TR/TO-Klf9] cells treated +/− dox of genes associated with peaks from each cluster. Means with the same letter are not significantly different (p < .001; Kruskal-Wallis non-parametric ANOVA)

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 non-parametric 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° of 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 (2) = 45.181, p < .001; Kruskal-Wallis non-parametric ANOVA).

Genomic regions where Klf9 associates support transcriptional repression by Klf9, and this requires intact Sp/Klf motifs

To determine if Klf9 can repress transcription of genes with which it associates in chromatin we transfected HT22 [TR/TO-Klf9] cells with pGL4.23 reporter constructs containing DNA fragments corresponding to genomic regions with Klf9 ChSP peaks:

Klf13

(439 bp; 6 Sp/Klf motifs, of which 3 were Klf9 consensus),

Klf16

(2192 bp; 17 Sp/Klf motifs, of which 6 were Klf9 consensus),

Limk1

(802 bp; 11 Sp/Klf motifs, of which 4 were Klf9 consensus) and

Mapk11

(886 bp; 6 Sp/Klf motifs, of which zero were Klf9 consensus) (see Additional file

4

: 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

−5

;

Klf16

: t

(10)

 = 4.430,

p

Limk1

: t

(10)

 = 3.275,

p

Mapk11

: t

(9)

 = 1.509,

p

 = .166;

Klf13 mutant

: t

(10)

 = −1.331,

p

 = .213;

n

 = 6/treatment; Student’s two-sample

t

-test).

Fig. 6

Klf9 represses transcription from synthetic promoters of Klf9 target genes, and Sp/Klf-like motifs are required for transcriptional repression by Klf9. We transfected HT22 [TR/TO-Klf9] cells with reporter constructs containing cloned DNA fragments corresponding to genomic regions with Klf9 peaks associated with the indicated genes (See Additional file 4: Figure S3 for genomic ranges). Forced expression of Klf9 repressed transcriptional activity from the Klf13, Klf16 and Limk1, but not from the Mapk11 promoter. Mutation of six Sp/Klf-like motifs in the Klf13 promoter abrogated Klf9-dependent repression. The asterisks indicate a statistically significant difference from control by Student’s two-sample t-test (p < .01)

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; Slc11a2: t (7) = −4.628, p < .005; Klf16: t (8) = −4.628, p < .005; Klf13: t (10) = −2.829 p < .05; Sin3a: t (9) = −2.138, p = .06; Nr3c1: t (10) = −2.684, p < .05; Limk1: t (8) = −2.334, p < .05; Klf11: t (7) = −2.371, p < .05; Mapk11: t (6) = −2.299, p = .06.; n = 6/treatment; Student’s two-sample t-test).

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

Klf13

:

F(2,12)

 = 14.267,

p

Klf16

:

F(2,13)

 = 14.135,

p

Limk1

:

F(2,13)

 = 5.875,

p

Klf11

:

F(2,11)

 = 9.1,

p

Mapk11

:

F(2,13)

 = .922,

p

 = .422;

n

 = 6/cell line; ANOVA). 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

.

Fig. 7

Klf9 target genes are dysregulated in Klf9 HT22 depleted cells and in Klf9 knockout mice, and depletion of Klf9 accelerates the cell cycle in HT22 cells. a Klf9 target genes are dysregulated in HT22 cells with Klf9 depleted by CRISPR/Cas9 genome editing. Asterisks indicate a statistically significant difference from the parent HT22 cell line (p < 0.05; ANOVA followed by Holm-Sidak post-hoc test). b Klf9 target genes are dysregulated in the hippocampus of postnatal day 7 Klf9-null mice. Asterisks indicate a statistically significant difference from wild-type mice by Student’s two-sample t-test (p < .05). c Cells with Klf9 depleted by CRISPR/Cas9 genome editing showed a higher percentage of cells in M phase (gray bars, lowercase letters) and a lower percentage in G1/G0 phase (black bars, uppercase letters). Means with the same letter are not significantly different (p < .05; ANOVA followed by Holm-Sidak post-hoc test). d The mRNA levels for two Klf9 target genes involved with cell cycle control are increased in Klf9 mutant HT22 cells. Means with the same letter are not significantly different (p < .05; ANOVA followed by Holm-Sidak post-hoc test)

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

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 Klf9-induced genes from the pathway analysis because of the small number discovered and because we were able to validate only 3 of 6 tested.

Table 2

Genes repressed by forced Klf9 expression in HT22 [TR/TO-Klf9] cells and genes with Klf9 peaks associated in HT22 [BirA/FLBIO-Klf9] cells were subjected to pathway analysis using GeneCoDis

All enriched pathways among genes repressed by Klf9

 Panther:P00016

Cytoskeletal regulation by Rho GTPase

3

0.034087

 Panther:P00031

Inflammation mediated by chemokine and cytokine signaling pathway

5

0.041865

 Panther:P00057

Wnt signaling pathway

5

0.041865

 Panther:P00010

B cell activation

3

0.046343

The top ten most enriched pathways among genes with Klf9 peaks associateda

 Panther:P00006

Apoptosis signaling pathway

31

8.67E-11

 Panther:P00034

Integrin signalling pathway

36

2.79E-10

 Panther:P00031

Inflammation mediated by chemokine and cytokine signaling pathway

46

3.16E-10

 Panther:P00016

Cytoskeletal regulation by Rho GTPase

23

1.28E-09

 Panther:P00005

Angiogenesis

33

1.51E-09

 Panther:P00057

Wnt signaling pathway

42

1.75E-08

 Panther:P00056

VEGF signaling pathway

19

2.64E-08

 Panther:P00047

PDGF signaling pathway

28

2.82E-08

 Panther:P04393

Ras Pathway

20

7.85E-08

 Panther:P00021

FGF signaling pathway

24

4.71E-07

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 [3840]. One of the top GO: PANTHER 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 (Fig. 7c; G1/G0 (lowercase): F (2,8) = 56.27, p < .001; M (uppercase): F (2,8) = 408.754, p < .001; n = 4/cell line; ANOVA). The mRNA levels for two confirmed Klf9 targets that are classified as Wnt-pathway related and are implicated in promoting mitosis, B cell CLL/Lymphoma 6 (Bcl6) and inositol triphosphate receptor 3 (Itpr3), were significantly elevated in both CRISPR lines (Fig. 7d; Bcl6: F (2,9) = 19.797, p = 0.001; Iprt3: F (2,9) = 13.346, p = 0.002; n = 6/cell line; ANOVA).

About

We Support OUR Contributors

Get Our Newsletter

 Receive podcast updates
Exclusive insights
Patient Engagement Tips from industry experts
We hate SPAM as much as you do and promise to keep your email address safe.
  • Subscribe to the Podcast