Direct targets of Klf5 transcription factor contribute to the maintenance of mouse embryonic stem cell undifferentiated state

Identification of Klf5 direct targets

To identify genes directly regulated by Klf5, we first analyzed the transcriptome changes upon Klf5 KD in ESCs by microarray analysis. To minimize indirect effects due to phenotypic changes induced by Klf5 KD, we performed a time course analysis both by qPCR and by Western blot analysis, which showed that Klf5 expression is significantly decreased already 12 hr after siRNA transfection (Figure 1a). On the contrary, the expression of stemness genes such as Oct3/4 and Nanog, previously demonstrated to be modified by Klf5 KD [16], is still unaffected at 12 hr (Figure 1b). Therefore, it is possible to capture early effects on transcriptome following Klf5 KD at 12 hr after transfection.

Statistically significant probes (FDR < 0.1) were classified as upregulated and downregulated following Klf5 KD by using as cut-off > 1.25- and < 0.75-fold changes, respectively. On the basis of these constraints, we identified 239 upregulated and 74 downregulated genes (Additional file 1).

These predicted genes encode a set of proteins involved mainly in development and differentiation (Additional file 2), in agreement with the Klf5 role in ESCs and embryonic development [16, 17]. Interestingly, 52 genes controlled by Klf5 encode regulators of transcription, indicating that Klf5 regulates many transcription factors, which in turn regulate their downstream targets. Moreover, among the genes regulated by Klf5, there are factors that have already demonstrated a role in ESC pluripotency, such as Tcl1 [8, 22], BMP4 [23], Nr0b1 [24] and CD9 [25]. We also compared our results with available data [15, 17, 26, 27], and we found that 86 of 96 transcripts showed the same behavior upon Klf5 KD and upon triple KD of Klf2, Klf4 and Klf5 [15] (Additional file 3). Moreover, we found that 75% of the transcripts showed the same trend in our analysis and in Klf4 KD cells [26] (Additional file 3). Instead, no significant correlation was found when our data were compared with the data of LIF deprivation or expression of STAT3 dominant-negative form [27] (Additional file 3).

It is reasonable to assume that Klf5-regulated genes, identified by gene expression profiling, include both direct and indirect transcriptional targets. To distinguish between direct target genes and those indirectly regulated through alterations in transcriptional networks governed by Klf5, we employed genome-wide chromatin immunoprecipitation (ChIP). We introduced a FLAG-tagged Klf5 in ESCs and generated two independent pools of stable clones. The levels of Klf5 were assessed by Western blot analysis with anti-Klf5 and anti-FLAG antibody to detect both endogenous and exogenous Klf5 (Additional file 4). Chromatin from these cells was immunoprecipitated with anti-FLAG antibody, and purified DNA fragments were analyzed by direct high-throughput sequencing. We identified 6480 putative binding sites, and among these 5820 had an FDR lower than 5% (Additional file 5). ChIP-qPCR validations were carried out on 15 Klf5-binding sites with different numbers of ChIP tag counts within the defined overlap region (Additional file 4). Twelve of 15 target loci subjected to validation showed a significant enrichment over three independent control regions (Additional file 4).

Next, we analyzed Klf5-ChIP target sequences with the motif-finding CisFinder software to search for over-represented motifs. The three best results were represented by CG-rich elements (Additional file 6), in agreement with previous reports [28]. Moreover, all three motifs contained the CTGC sequence, suggesting a putative binding site for Klf5.

To gain insights into the primary targets of Klf5, we analyzed the expression pattern of the 53 validated genes (Figures 1c and 1d), during neural differentiation of ESCs. Klf5, such as the other two ESC-specific Klfs, is highly expressed in undifferentiated ESCs, and its expression dramatically decreases when differentiation occurs (Additional file 2). We reasoned that genes activated by Klf5 are expected to show the same expression profile of Klf5, i.e., expressed in undifferentiated ESCs and downregulated when differentiation occurs. On the contrary, the genes that are negatively controlled by Klf5 are expected to show an opposite expression profile, i.e., not expressed in undifferentiated ESCs and upregulated when differentiation occurs. As expected, almost all the genes activated by Klf5 are expressed in ESCs and repressed during differentiation (Figure 1). On the other hand, almost all the genes that are inactive or repressed in ESCs, but are expressed during differentiation, appear to be negatively regulated by Klf5 (Figure 1d).

Suppression of Klf5 target genes impairs ESC undifferentiated state

The first point we asked is that of the relevance of Klf5 direct targets in the maintenance of ESC phenotype. Among the putative Klf5 direct targets, we decided to start by analyzing 23 genes that were selected on the basis of validation by qPCR (Additional file 7) and their expression profile during ESC differentiation (Figure 1). To this aim, we selected pools of stably transfected ESC clones, where the expression of these 23 genes was knocked down by specific shRNAs. A reduction of at least 20% in transcription levels was observed after selection (Additional file 8). The stably transfected cells were plated at clonal density and further grown for 7 days. The resulting colonies were stained for alkaline phosphatase (AP) activity (Figure 2a) and the percentage of undifferentiated and differentiated colonies was calculated. KD of Klf5 (positive control) resulted in at least 40% decrease of AP-positive colonies compared to control shRNA transfected cells (Figure 2b). Knockdown of 8 of 23 target genes analyzed resulted in a significant decrease of AP-positive colonies (Figure 2b). To confirm that these decreases in the number of AP-positive colonies were due to an impairment of ESC undifferentiated state, we measured the level of stemness markers in cells KD for these eight genes soon after selection. Expression of both Oct3/4 and Nanog was significantly reduced compared to control after knockdown of all these genes (Figure 2c). The observed decrease in Oct3/4 and Nanog expression corresponded to a significant increase of early differentiation markers (Additional file 9). These results were confirmed by further independent shRNAs (Additional file 8). Taken together, these results suggest that Klf5 controls the expression of at least 8 of 23 genes required to maintain the undifferentiated state of ESCs.

Ectopic expression of Serpine1 impairs ESC undifferentiated state

As shown above, we found many genes whose expression in ESCs seems to be negatively regulated by Klf5. Two of them, Serpine1 and Runx1, were selected for functional studies. Their Klf5-dependent regulation is confirmed by the observation that they are expressed at very low levels in undifferentiated ESCs. Furthermore, these genes that seem to be negatively regulated by Klf5 in undifferentiated ESCs showed a strong induction upon ESC differentiation when Klf5 disappeared (Figure 1d) [16]. Thus, we investigated their effect on stemness by forced expression in undifferentiated ESCs. We transfected ESCs with expression vectors bearing FLAG-tagged Serpine1 or Runx1 (Figure 3a); these cells, grown at clonal density for 7 days, were stained to assay AP activity. Alkaline phosphatase staining showed that forced expression of FLAG-Serpine1 is able to induce a significant decrease in the number of undifferentiated colonies, whereas FLAG-Runx1-transfected cells did not show a significant decrease in AP expression (Figure 3b). To further study the effect of forced expression of Serpine1 and Runx1, we measured the expression level of Oct3/4 and Nanog. As shown in Figure 3c, the expression of Oct3/4 was impaired upon Serpine1 constitutive expression in ESCs, and also Nanog showed a moderate but significant decrease, whereas no significant changes were detectable in FLAG-Runx1-expressing cells. This decrease of Oct3/4 and Nanog levels corresponded to a significant increase of early differentiation markers (Figure 3d) even in the presence of LIF, indicating that Serpine1 alone is able to impair the undifferentiated state by inducing an uncontrolled differentiation.

ESC-specific targets of Klf5

Next, we asked the question of the specificity of Klf5. Previous works demonstrated that Klf5 is highly expressed in skin [19] and that alterations in Klf5 expression level may affect the epidermis-differentiated phenotype [29], suggesting that Klf5 could regulate the regenerative potential of stem cells in the epidermis. We analyzed expression changes of the 53 qPCR-validated Klf5 targets upon Klf5 KD in primary keratinocytes (Additional file 10). In fact, we found that the transcripts of six genes expressed in ESCs were not detectable in primary keratinocytes, both in Klf5 KD and control cells, whereas there were 21 of 47 genes similarly regulated by Klf5 in primary keratinocytes and ESCs. On the other hand, 14 of 47 genes showed opposite changes in ESCs versus keratinocytes, while 12 of 47 genes were regulated by Klf5 only in ESCs (Figure 4). Although the extent of Klf5 suppression is different in ESCs versus keratinocytes, the comparison of expression profiles of these two cells indicated a cell type-specific gene regulation by Klf5.

A further aspect of the specificity of the Klf5-based regulation concerns the possibility that Klf2, Klf4 and Klf5 have redundant functions by binding and thus regulating common targets in ESCs [15]. To address this point, we investigated gene expression changes of the 53 qPCR-validated Klf5 targets, following KD of each single Klf. We transfected Klf2 or Klf4 siRNA in ESCs, and 12 hr after transfection we found that the level of Klf2 and Klf4 were significantly reduced, although with different extent of KD (Additional file 10). In these conditions, we found that only a low percentage (<10%) of examined genes showed the same trend upon KD of all these Klfs (Figure 5). The same low percentage was observed by comparing Klf5- versus Klf4-regulated genes. Instead, we found that the genes that showed the same behavior upon Klf2 and Klf5 KD represent about 40%. Finally, we observed that about 45% of the genes show a Klf5-specific gene regulation, different from that dependent on Klf2 or Klf4 KD (Figure 5), indicating that some redundancy could exist in the genes controlled by Klf5 and Klf2, rather than Klf4.

Plasmid construction

Serpine1 and Runx1 cDNA were derived from pSport vector (NIH Mammalian Gene Collection, Open Biosystems, National Institutes of Health, Bethesda, MD, USA) by PCR [16] with the following oligonucleotides:

HindIII-Serpine1: 5'-GATGACAAGCTTCAGATGTCTTCAGCCCTTGCTTGCCTCATCC-3'

NotI-Serpine1: 5'-GGCGATGAGCGGCCGCTCAAGGCTCCATCACTTGGCCCATGAAGAGG-3'

HindIII-Runx1: 5'-GATGACAAGCTTGCTTCAGACAGCATTTTTGAGTCATTTCCTTCATATCC-3'

NotI-Runx1: 5'-GGCGATGAGCGGCCGCTCAGTAGGGCCGCCACACGGCCTCCTCC-3'

Next, these cDNAs were cloned downstream FLAG-tag in the p-CBA-FLAG vector by using HindIII and NotI restriction sites.

Cell culture, transfection, differentiation and alkaline phosphatase staining

E14Tg2a (BayGenomics, San Francisco, CA, USA) mouse ESCs were maintained on feeder-free, gelatin-coated plates in the following medium: GMEM (Glasgow Minimum Essential Medium) (Sigma, St. Louis, MO, USA) supplemented with 2 mM glutamine, 100 U/ml penicillin/streptomycin, 1 mM sodium pyruvate, 1× nonessential amino acids (all from Invitrogen, Carlsbad, CA, USA), 0.1 mM β-mercaptoethanol (Sigma), 10% fetal bovine serum (HyClone Laboratories, Logan, UT, USA), and 10³ U/ml leukemia inhibitory factor (LIF; Millipore, Billerica, MA, USA). Neural differentiation was induced as previously described [16]. Briefly, undifferentiated ESCs were trypsinized into a single-cell suspension and plated at low density (1-5 × 10³ cells/cm²) on gelatin-coated dishes in the following medium: knockout Dulbecco's minimal essential medium supplemented with 10% knockout serum replacement (both from Invitrogen), 0.1 mM β-mercaptoethanol (Sigma), 2 mM glutamine (Invitrogen), 100 U/mL penicillin/streptomycin (Invitrogen). Medium was changed on alternate days.

Primary mouse keratinocytes were isolated from 2-day-old Swiss ICR(CD-1) mice (Harlan Laboratories, Correzzana, Italy) and cultured as previously described [38]. Transfections were performed 5 days after plating.

Transfection of expression plasmids and shRNA plasmids (Open Biosystems; see Additional file 11) both in ESCs and in primary keratinocytes were performed using Lipofectamine 2000 (Invitrogen) following the manufacturer's instructions. To generate the stable cell lines, E14Tg2a cells were transfected with shRNA plasmids and recombinant clones were selected with Puromycin (Sigma).

For alkaline phosphatase staining, ESCs were cultured at clonal density (30 cells/cm²). The cells were fixed in 10% cold neutral formalin buffer (10% formalin, 110 mM Na₂HPO₄, 30 mM NaH₂PO₄ H₂O) for 15 min and then rinsed in distilled water for 15 min. The staining was obtained by incubation for 45 min at room temperature with the following staining solution: 0.1 M Tris·HCl, 0.01% naphthol AS MX-PO₄ (Sigma), 0.4% N, N-dimethylformamide (Sigma), 0.06% red violet LB salt (Sigma).

RNA isolation and qPCR

RNA from ESCs and primary keratinocytes was isolated by using the Tri reagent (Sigma) and then reverse transcribed using MuMLV-RT (New England Biolabs, Ipswich, MA, USA) according to the manufacturer's instructions. Real-time RT-PCR was performed using PRISM 7900HT Sequence Detection System (Applied Biosystems, Foster City, CA USA) and Power SYBR Green PCR Master Mix (Applied Biosystems). Gene-specific primers used for amplification are listed in Additional file 12.

Microarrays and bioinformatics analysis

Total RNA from ESCs was isolated with Tri reagent (Sigma) and further purified with the Qiagen column kit (Qiagen, Milan, Italy). Then samples from three independent experiments were sent to Coriell Genotyping and Microarray Center (Coriell Institute for Medical Research, Camden, NJ, USA), where, after proper sample processing, cRNA were hybridized with the Affymetrix Mouse Genome 430 2.0 Array (Affymetrix, Santa Clara, CA, USA).

For bioinformatics analysis, raw probe intensities for each of the hybridized microarrays were normalized to gene expression levels using the dChip algorithm [39]. To identify genes significantly responding in the experiment, we computed the P values and false discovery rate (FDR). A total of 313 probes, corresponding to 313 different transcripts, have been identified (FDR < 0.1 corresponding to P < 0.005) that responded significantly.

Preparation of cell lysates and Western blot analysis

ESCs and primary keratinocytes were lysed in a buffer containing 1 mM EDTA, 50 mM Tris·HCl, pH 7.5, 70 mM NaCl, 1% Triton protease inhibitor cocktail (Sigma) and analyzed by Western blot analysis. The following primary antibodies were used: rabbit anti-Klf5 (Santa Cruz Biotechnology, Santa Cruz, CA, USA), rabbit anti-Nanog (Calbiochem, San Diego, CA, USA), mouse anti-Oct3/4 (Santa Cruz Biotechnology), mouse anti-GAPDH (Santa Cruz Biotechnology) and mouse anti-FLAG (Sigma). Antibody-protein complexes were detected by HRP-conjugated antibodies and ECL (both from Amersham Pharmacia, Milan, Italy).

Chromatin immunoprecipitation

For ChIP-seq analysis, ESCs stably transfected with FLAG-Klf5 were cross-linked with 1% formaldehyde for 10 min at room temperature, and formaldehyde was then inactivated by the addition of 125 mM glycine. Then the chromatin was sonicated to an average DNA fragment length of 200 to 500 bp. Soluble chromatin extracts were immunoprecipitated using the mouse monoclonal anti-FLAG (Sigma) or mouse IgG (Santa Cruz Biotechnology) as control. Then samples from two independent experiments were sent to the DNA sequencing service of EMBL (Heidelberg, Germany) and subjected to high-throughput sequencing with Illumina Genome Analyzer platform (Illumina, San Diego, CA, USA).

For ChIP-qPCR, samples were prepared as described above. Supernatant obtained without antibody was used as an input control. qPCR analyses were performed using the ABI PRISM 7900HT sequence detection system and SYBR Green PCR Master Mix (Applied Biosystems). Primers used for ChIP-qPCR are listed in Additional file 13.

The amount of precipitated DNA was calculated relative to the total input chromatin and expressed as the fold enrichment relative to total input according to the following formula [40]: fold enrichment = 2{Delta}Ct × 10, where {Delta}Ct = Ct(input) - Ct(immunoprecipitation), where Ct refers to cycle threshold.

ChIP-seq bioinformatics analysis

More than 13 million sequences were produced and aligned to the mouse genome (version m37) masked for DNA repeat by using Bowtie tool version 0.9.9.2 (Center for Bioinformatics and Computational Biology, Institute for Advanced Computer Studies, University of Maryland, College Park, MD, USA) [41]. About 50% of sequences were univocally aligned, and the resulting coordinates were fed to MACS software version 1.3.5(Department of Biostatistics and Computational Biology, Dana-Farber Cancer Institute and Harvard School of Public Health, Boston, MA USA) [42] to detect genomic regions enriched for multiple overlapping DNA fragments (peaks) that we considered as putative binding sites. False discovery rate (FDR) was estimated by MACS, by comparing the peaks from anti-FLAG samples with those from control (anti-IgG ChIP) at the same P value cutoff. Motif analysis was performed by using CisFinder tool (Developmental Genomics and Aging Section, Laboratory of Genetics, National Institute on Aging, NIH, Baltimore, MD, USA) [43] on 200-bp sequences centered at the expected binding site indicated by peak summit calculated using the MACS tool. Flanking sequences 1000 bp away from the peak summit have been used as control sequences.