Kruppel-like Factor-9 (KLF9) Inhibits Glioblastoma Stemness through Global Transcription Repression and Integrin α6 Inhibition*

Background: Cancer cell stemness determines tumor propagation and malignancy by poorly defined mechanisms. Results: The KLF9 transcription factor regulates cell fate and oncogenic pathways by repressing gene transcription in glioblastoma stem cells. Conclusion: KLF9 inhibits glioblastoma cell stemness and tumor growth by directly repressing genes, including ITGA6. Significance: KLF9 and its transcriptional network are potential targets for regulating cancer stem cells and glial malignancies. It is increasingly important to understand the molecular basis for the plasticity of neoplastic cells and their capacity to transition between differentiated and stemlike phenotypes. Kruppel-like factor-9 (KLF9), a member of the large KLF transcription factor family, has emerged as a regulator of oncogenesis, cell differentiation, and neural development; however, the molecular basis for the diverse contextual functions of KLF9 remains unclear. This study focused on the functions of KLF9 in human glioblastoma stemlike cells. We established for the first time a genome-wide map of KLF9-regulated targets in human glioblastoma stemlike cells and show that KLF9 functions as a transcriptional repressor and thereby regulates multiple signaling pathways involved in oncogenesis and stem cell regulation. A detailed analysis of one such pathway, integrin signaling, showed that the capacity of KLF9 to inhibit glioblastoma cell stemness and tumorigenicity requires ITGA6 repression. These findings enhance our understanding of the transcriptional networks underlying cancer cell stemness and differentiation and identify KLF9-regulated molecular targets applicable to cancer therapeutics.

Mounting evidence indicates that neoplastic stem-like cells, often referred to as cancer stem cells (CSCs), 3 play an important role in several cancers, including glioblastoma (GBM), an aggressive malignancy in the adult central nervous system (1). CSCs are a minority population of slow cycling cells in tumors. They display various stemlike properties (hereby referred to as "stemness"), including long term self-renewal and the capacity to generate phenotypically diverse hierarchical neoplastic progeny and stromal cells (2)(3)(4). A critically important characteristic of CSCs is their ability to efficiently propagate tumor xenografts that recapitulate essential phenotypes of the original tumor, such as tumor cell heterogeneity, invasiveness, and vascularity (2). Many CSCs, including glioblastoma stem cells (GSCs), are particularly resistant to chemotherapy and radiation (5)(6)(7). Thus, neoplastic cells displaying stemlike phenotypes are currently believed to contribute disproportionately to tumor growth patterns and recurrence after therapy (8).
Because CSCs and normal stem cells share phenotypic properties, it is not unexpected that they also share signaling pathways (e.g. Notch, BMI1, and Hedgehog) that maintain their stemness and regulate their tumor propagating capacity (9 -12). The self-renewal of non-neoplastic pluripotent stem cells is regulated by a pluripotency-driving transcription factor network, involving Oct4, Sox2, and Nanog (13,14), that also drives the stemness and tumorigenicity of CSCs (15)(16)(17). Growing evidence is revealing that cancer cells are highly plastic with the capacity to dynamically and bidirectionally transition between more differentiated and more stemlike phenotypes in response to contextual autocrine/paracrine signals (18,19). A more complete understanding of the gene regulatory networks that control neoplastic cell stemness is vital for understanding mechanisms behind the generation and maintenance of CSCs toward the goal of developing strategies for targeting CSC pools therapeutically.
One transcription factor family of particular interest in stem cell regulation is the family of Kruppel-like transcription factors (KLFs). KLFs consist of 17 zinc finger domain transcription factors that bind to GC-or GT-rich DNA regions to regulate transcription and thereby influence multiple biological events (20,21). KLFs are also involved in regulating stem cell selfrenewal and differentiation. KLF4 is one of four Yamanaka transcription factors that, when ectopically expressed, reprogram somatic cells to pluripotent stem cells (22,23). Certain KLFs function as tumor suppressors and/or oncogenes depending on the cellular context. For example, the localization of KLF4 to the nucleus in early stage ductal carcinoma is associated with a more aggressive phenotype (24). KLF6 functions as a tumor suppressor that is mutated and/or suppressed in several cancers, including breast cancer, prostate cancer, and GBM (24,25).
KLF9 (also known as basic transcription element-binding protein), a relatively unexplored KLF family member, is expressed in various tissues, most abundantly in the brain, kidney, lung, and testis (26), and regulates a variety of cellular functions, including stem cell self-renewal and differentiation. For example, KLF9 has been shown to regulate differentiation of intestinal cells and oligodendrocytes and is essential for late phase neuronal maturation in the developing dentate gyrus (27,28). KLF9 has been linked to cancers as it is down-regulated in colorectal cancer and endometrial carcinoma (29,30). KLF9 was also found to be induced during the differentiation of GSCs, and enforced KLF9 expression inhibits GBM cell stemness in part by suppressing Notch1 expression and its downstream signaling (31). The genome-wide gene targets of KLF9 have not been identified, and the molecular mechanism underlying the functional contributions of KLF9 to the regulation of cancer cell stemness and oncogenesis remains elusive.
Here, we used RNA-sequencing (RNA-Seq) and chromatin immunoprecipitation followed by deep sequencing (ChIP-Seq) to investigate genome-wide KLF9 binding sites and KLF9-regulated genes in human GBM-derived neurospheres enriched for GSCs. This first comprehensive genome-wide analysis of KLF9 targets revealed a significant role for KLF9 in regulating several pathways involved in oncogenesis and stem cell regulation.

EXPERIMENTAL PROCEDURES
Reagents-All reagents were purchased from Sigma-Aldrich. Stock of all-trans-retinoic acid (RA) was prepared in DMSO and diluted to 1 M in cell culture medium as a working concentration. Doxycycline (Dox) was diluted to a concentration of 0.5 g/ml in cell culture medium as a working concentration. In all experiments, the final DMSO concentration was Ͻ0.1%, and DMSO had no demonstrable effect on neurosphere cultures. Laminin was diluted to 10 g/ml as a working concentration.
Cell Culture-The human GBM neurosphere lines GBM1a (0913) and GBM1b (0627), originally derived and characterized by Vescovi and co-workers (32) and used extensively by us (17,31,33), were maintained in serum-free medium containing EGF and FGF. Cells were differentiated in EGF/FGF-free medium with 1 M RA or 1% fetal bovine serum (32,34,35). Primary GBM neurospheres 551 (JHH551) and 612 (JHH612) were derived from clinical GBM specimens obtained at The Johns Hopkins Hospital following a published method (32). All human materials were obtained and used in compliance with The Johns Hopkins Institutional Review Board.
Lentiviral Vectors and Cell Infection-N-terminal 3xFLAGtagged KLF9 was constructed by high fidelity PCR and cloned into pTRIPZ and pLEX vectors (Thermo Scientific) using AgeI and MluI. ITGA6 was cloned into pLEX vector with AgeI and MluI. Lentiviral packaging followed a second generation lentivirus packaging protocol using psPAX2 and pMD2.G vectors (Addgene). Cells were infected with lentivirus at a multiplicity of infection of 5 along with Transdux (System Biosciences) and selected with puromycin (1 g/ml) for stable cell lines.
Cell Adhesion and Migration Assay-GBM neurosphere lines were infected with lentivirus. 24 h after infection, cells were treated ϮDox for 96 h to induce KLF9 expression. Primary GBM neurospheres were infected with KLF9, ITGA6 lentivirus, or both. Cells were dissociated and plated on laminin-coated wells for 2-6 h. Adherent cells were stained with crystal violet, dissolved with 2% SDS, and quantified spectrophotometrically at 550 nm using a SpectraMAX 340pc (Molecular Devices) plate reader. Results show relative adhesion measured after subtracting the background absorbance from all values.
Cell migration assays were performed using laminin-coated Transwell chambers. GBM neurosphere lines were infected by ITGA6 lentivirus for 24 h, and cells were treated ϮDox for 48 h to induce KLF9. Primary GBM neurospheres were directly infected with KLF9 lentivirus. Upper chamber medium consisted of neurosphere culture medium without EGF/FGF, and lower chamber medium consisted of DMEM with 10% FBS. After 24 h, cells that had migrated through the filter were fixed and stained with Hoechst 33342 (Invitrogen). Migration was quantified by counting cells from eight random fields.
Immunofluorescence-Neurosphere cells were collected by cytospin onto glass slides and fixed with 4% paraformaldehyde. The cells were immunostained with anti-FLAG (Sigma-Aldrich) and anti-integrin ␣6 (Cell Signaling Technology) antibodies and Hoechst 33342 nucleic acid stain (Invitrogen). Secondary antibodies were conjugated with Alexa Fluor 488 or cyanine Cy3. Immunofluorescent images were captured and analyzed using AxioVision software (Zeiss).
Tumor Xenografts-All animal protocols were approved by The Johns Hopkins School of Medicine Animal Care and Use Committee. For intracranial xenografts, SCID mice received 5,000 viable cells in 2 l of DMEM by stereotactic injection to the right caudate/putamen. Cell viability prior to implantation was determined by trypan blue dye exclusion. Mice were perfused with 4% paraformaldehyde, and the brains were sectioned for histological analysis as described previously (31). Tumor size was quantified on H&E-stained coronal sections using parameters. For each gene, the number of reads aligned to its exons were counted and summarized into gene level counts by the R Bioconductor package GenomicFeatures (37) based on the UCSC refFlat table for hg18. Normalization between samples was carried out by R package edgeR (38,39), which controls sequencing depth and RNA composition effects.
Reproducibility Plots-Reads per kilobase per million mapped reads (RPKM) for each differentially expressed gene was calculated by the R Bioconductor package GenomicRanges (37). The heat map was generated according to the count table with scaling across the samples for each gene. The log 2 -fold change, log 2 ((IP RMKM ϩ 1)/(Control RPKM ϩ 1)), for each gene in each cell line was calculated. The log 2 -fold change for KLF9-GBM1b is plotted against that of KLF9-GBM1a.
ChIP and Sequencing-Cells were subjected to ChIP using the MAGnify ChIP system (Invitrogen) following the manufacturer's protocol. KLF9-bound DNA was immunoprecipitated using mouse anti-FLAG M2 antibody (Sigma-Aldrich) and Dynabeads magnetic beads (Invitrogen). Mouse IgG served as the control. ChIP-enriched DNA was used for PCR and deep sequencing. ChIP-enriched DNA or input DNA (10 ng each) was subjected to library preparation using a ChIP-Seq DNA Sample Prep kit (Illumina) following the manufacturer's protocol. Sequencing was performed using an Illumina HiSeq 2500 platform.
Peak Calling-The 50-bp-long raw KLF9 ChIP-Seq reads were aligned to the reference human genome build hg18 using Bowtie (40) allowing at most two mismatches in the first 28-bp "seed" bases. KLF9 binding sites were called using CisGenome (41) with default settings by comparing the two IP samples against the two control samples.
Reproducibility Plots-The number of reads aligned to the peak regions by each of the four ChIP samples was counted by the R Bioconductor package GenomicRanges (37) and then normalized for the library size for each sample. For each peak region, its normalized counts in a given sample were further subtracted by its mean normalized counts across samples and then divided by its standard deviation, which gave the scaled binding intensity.
De Novo Motif Discovery-The 150-bp-long sequences centered at the peak summits for the top ranked 500 peaks were extracted and fed as the input for the de novo motif discovery algorithm of CisGenome. Ten motifs of varying length with a mean motif length of 12 were searched simultaneously. To obtain the truly KLF9 enriched motifs, the occurrence rate of a motif in the 150-bp-long sequences centered at the peak summits for all peaks was compared with its occurrence rate in control genomic regions. The control regions were randomly chosen to match the GC content and distributional properties of ChIP-Seq peak regions (42).
KLF9 Target Gene Detection-The differential gene expression detection was carried out by the R Bioconductor package edgeR (38,43,44) with tagwise dispersion at a false discovery rate (FDR) of 5%. A matched study design was used because the four samples came from two cell lines. The list of differentially expressed transcripts was further filtered if the absolute values of log 2 -fold changes for differentially expressed transcripts comparing the case versus control exceeded 0.8. KLF9 target genes were defined as differentially expressed transcripts with Ն1 KLF9 binding peaks in the Ϫ20 to ϩ10 kb window surrounding the transcription start sites (TSSs).
Pathway Analyses-Canonical pathway analysis was performed using Ingenuity Pathway Analysis (Ingenuity). The significance of association between KLF9 targets and a canonical pathway was measured using the ratio between the number of KLF9 targets in the pathway and the total number of molecules in the pathway database. Fisher's exact test was performed to determine the association between KLF9 targets and canonical pathways.
Bioinformatics and Statistical Analyses-Statistical differences for other experiments were evaluated by Student's t test, Wilcoxon paired t test, or analysis of variance followed by a Tukey multiple comparison tests as appropriate. All data are represented as mean values ϮS.E. All results reported are representative of at least three independent replications.

KLF9 Expression and GBM Cell
Stemness-We previously found that induction of endogenous KLF9 expression is required for the responses of GBM neurospheres to forced differentiation stimuli and that enforced KLF9 expression inhibits GBM cell stemness (i.e. self-renewal, multipotency, and stem cell marker expression) (31). To identify genome-wide KLF9 targets linked to stemness regulation in GBM, we first established appropriate KLF9 induction conditions in these GSC models. Two human GBM neurosphere lines (designated as KLF9-GBM1a and KLF9-GBM1b) were engineered to express a Dox-inducible KLF9 transgene with a 3xFLAG tag (Fig. 1A), and Dox treatment (0.5 g/ml) for 48 h induced KLF9 mRNA expression ϳ28-fold (Fig. 1B, left). This KLF9 induction was comparable with the magnitude of endogenous KLF9 induction (10 -12-fold) in response to two inducers of GSC differentiation (RA and serum) following withdrawal of growth factors (Fig. 1B, right). Ectopic KLF9 protein levels plateaued by 2 days of Dox treatment and quickly diminished after Dox withdrawal (Fig. 1C). Next, we examined the effects of ectopic KLF9 induction on the expression of molecular markers and drivers of GBM cell stemness. The expression of various marker and drivers of GBM cell stemness, including BMI1, Nestin, Olig2, and Sox2, decreased in response to KLF9 induction consistent with our previously reported findings (31) (Fig. 1D). These GSC models with biologically relevant KLF9 induction that inhibits GBM cell stemness are suitable for identifying genome-wide KLF9 targets in human cancer models.
KLF9 Gene Binding and Gene Expression Signatures-A genome-wide analysis of KLF9 targets was performed by combining gene expression profiling using RNA-Seq and mapping of transcription factor binding sites using ChIP-Seq in GBM neurospheres with ectopic KLF9 induction (the strategy is outlined in Fig. 2A). KLF9-GBM1a and KLF9-GBM1b cells were treated ϮDox for 48 h after which RNA was isolated and subjected to cDNA library preparation and deep sequencing. Over 16 million cDNA reads were generated for each of four samples representing two biological replicates (KLF9-GBM1a and KLF9-GBM1b) with or without KLF9 induction. More than 82% of the reads aligned to the human genome. Consistency NOVEMBER 21, 2014 • VOLUME 289 • NUMBER 47 JOURNAL OF BIOLOGICAL CHEMISTRY 32745 between two biological replicates was confirmed by heat map clustering (Fig. 2B) and a correlation assay ( Fig. 2C; correlation coefficient ϭ 0.63). Of all the 24,524 RefSeq transcripts (18,275 genes), 1,550 transcripts (1,161 genes; 6.4%) were up-regulated (FDR Յ 5%, log 2 (-fold change)) Ն 0.8; supplemental Table S1) and 2,843 transcripts (2,092 genes; 11.4%) were down-regulated (FDR Յ 5%, log 2 (-fold change)) Յ Ϫ0.8; supplemental Table  S2) in response to KLF9 induction (Fig. 2D).

KLF9 Represses Cancer Stem Cell Transcription
Genome-wide KLF9 binding sites were identified using ChIP-Seq in GBM neurospheres. Cellular chromatin bound by FLAG-tagged KLF9 was specifically precipitated using anti-FLAG M2 antibody and Dynabeads (Fig. 2E), and the specific ChIP-enriched DNAs were further processed to produce libraries for ChIP-Seq. Over 150 million reads were generated from ChIP-enriched DNA libraries and input controls derived from two biological replicates. More than 71% of the reads aligned to the human genome. Using CisGenome (41), 31,261 KLF9 binding peaks were called at an FDR of 1% (supplemental Table S3). The ChIP intensities within KLF9 binding peaks were consistent for both biological replicates as shown by heat map clustering (Fig. 2F) and a correlation assay ( Fig. 2G; correlation coefficient ϭ 0.875). The majority of KLF9 binding sites localized to the region Ϫ20 to ϩ10 kb around TSS (58.9%), the region Ϫ2 to ϩ1 kb around TSS (37.9%), intergenic regions (54.0%), and introns (35.9%). Binding sites localized less frequently to 5Ј-untranslated regions (7.3%), exons (10.9%), and 3Ј-untranslated regions (1.2%) (Fig. 2H). In comparison, ran-domly selected genomic regions differed substantially from the KLF9 binding pattern (Fig. 2H). Furthermore, KLF9 binding sites, but not randomly selected genomic regions, were found to be highly enriched around TSSs (Fig. 2I). The top 500 KLF9 ChIP-Seq peaks as ranked by their FDR were used to perform de novo motif discovery analysis (41). The consensus sequence 5Ј-(G/A)(G/T)GGG(C/T)G(G/T)GGCN-3Ј was identified as the most enriched KLF9 binding motif (Fig. 2J). This motif closely resembles the motifs of two Sp1/KLF family members, KLF4 (5Ј-NGGG(T/C)G(G/T)GG-3Ј) and Sp1 (5Ј-GGGGGNG-GGG-3Ј) (Fig. 2K) identified using the transcription factor binding site profile database JASPAR (45).
Analysis of KLF9-regulated Gene Targets-KLF9 ChIP-Seq and RNA-Seq data sets were combined to establish gene targets directly regulated by KLF9 (i.e. genes differentially expressed in response to KLF9 induction and having one or more KLF9 binding peaks within Ϫ20 to ϩ10 kb of their TSSs). The expression changes of KLF9-bound and KLF9-unbound genes are summarized in Fig. 3A. Among the 2,092 genes down-regulated by KLF9, 1,849 (88.4%) fulfilled these criteria (supplemental Table S4; designated as KLF9 down-regulated targets). This number was significantly higher than that calculated by random gene sampling (the number of KLF9-bound genes found among 2,092genes randomly selected 1,000 times from the human genome; Fig. 3B, top panel; p value ϭ 0). In contrast, among the 1,161 up-regulated genes, only 726 (62.5%) were directly bound by KLF9, a number indistinguishable from that  A, strategy for identifying KLF9 gene targets in two GBM neurosphere lines with KLF9 induction after 48-h Dox treatment. KLF9 gene targets represent genes that were bound by KLF9 and showed differential expression in response to KLF9 induction. B, heat map of expression pattern for differentially expressed genes from RNA-Seq data. Gene expression was calculated by RPKM. C, differentially expressed genes for KLF9-GBM1b cells were plotted against those for KLF9-GBM1a cells to show the reproducibility of RNA-Seq (Spearman's correlation coefficient (R) ϭ 0.63, p Ͻ 0.001). D, volcano plot of all genes analyzed for differential expression (red dots, genes with FDR Յ0.05 and absolute log 2 (-fold change) Ͼ0.8). E, KLF9-GBM1a cells were treated with Dox for 48 h. Anti-FLAG antibodies but not nonimmune IgG specifically precipitated 3xFLAG-tagged KLF9 from fragmented DNA-protein complexes. Anti-FLAG M2 antibody bound to Dynabeads (the FLAG A combination) showed higher affinity than anti-FLAG M2 antibody directly cross-linked to magnetic beads (the FLAG B combination). F, heat map of ChIP intensities for identified peaks in KLF9 ChIP-Seq data. The two control samples from both cell lines are clustered together, and the two IP samples are clustered together by hierarchical clustering. G, ChIP-Seq reproducibility as determined by scatter plot comparing peak intensities for KLF9-GBM1b and KLF9-GBM1a cells (Spearman's correlation coefficient (R) ϭ 0.875, p Ͻ 0.001). H, genomic distribution of KLF9 binding peaks and randomly selected genomic regions. I, distribution around TSSs of KLF9 binding peaks and randomly selected genomic regions. J and K, KLF9 binding motif identified using the top ranked 500 KLF9 binding peaks (J). KLF4 and Sp1 motifs are the most similar to our discovered KLF9 motif (K). Enrichment score (E) was calculated by dividing its occurrence frequency in ChIP-Seq peaks by its frequency in matched control sequences randomly sampled from the genome. CDS, coding DNA sequence. NOVEMBER 21, 2014 • VOLUME 289 • NUMBER 47 calculated by random gene sampling (the number of KLF9bound genes found among 1,161 genes randomly selected 1,000 times from the human genome; Fig. 3B, bottom panel). These results indicate that the predominant role of KLF9 is to serve as a transcriptional repressor. Therefore, we focused the functional analyses described below exclusively on the 1,849 KLF9 down-regulated genes.

KLF9 Represses Cancer Stem Cell Transcription
Gene function annotation and Ingenuity pathway enrichment analysis (46,47) were performed using KLF9 down-regulated targets. The top pathways enriched with KLF9 down-regulated targets included those active in cancer regulation (e.g. molecular mechanisms of cancer, CXCR4 signaling, integrin signaling, and Notch signaling), stem cell pluripotency signaling (mouse embryonic stem cell pluripotency), and signaling in neuron development (axonal guidance signaling, cAMP-response element-binding protein signaling, and semaphorin signaling in neurons) ( Fig. 3C and supplemental Table S5).
A subset of the KLF9 down-regulated targets selected from pathways highly ranked by Ingenuity Pathway Analysis (e.g. molecular mechanisms of cancer, CXCR4 signaling, integrin signaling, and Notch signaling) was further validated by qPCR in KLF9-GBM1b neurospheres and low passage primary GBM neurospheres (GBM 551) (Fig. 4A). The validated genes include Notch pathway members NOTCH1, PSEN2, and NUMBL; CXCR4 (encoding a chemokine receptor that contributes to cytoskeletal organization and mediates metastasis in various cancers); and CAMK2G (encoding a serine/threonine protein kinase associated with the proliferation, resistance, and survival of cancer cells). Integrin signaling, which modulates multiple cellular processes, including cell adhesion and migration, tumor cell invasion, and cell stemness (48,49), was one of the top ranked and validated pathways enriched in KLF9 downregulated targets. ITGA6, encoding integrin ␣6 receptor subunit, was down-regulated up to 89% following KLF9 induction (Fig. 4A). Other downstream components of the integrin signaling pathway, such as ARPC1B, CAPN5, GIT1, and MYLK, were also validated to be down-regulated following KLF9 induction up to 88, 74, 72, and 71%, respectively (Fig. 4A). We further tested the expression of KLF9 target genes in response to different levels of KLF9 induction as controlled by Dox concentration (Fig. 4B). Dox treatment at 0.5 g/ml induced KLF9 11-fold, which is comparable with the magnitude of endogenous KLF9 induction in response to two inducers of GSC differentiation (RA and serum) (Fig. 1B, right). This treatment inhibited the expression of 11 KLF9 target genes but not three control genes that were not identified as KLF9 targets.
The KLF9 binding peaks near the TSS of 12 KLF9 downregulated targets as marked in Fig. 4B were further validated by quantitative ChIP-PCR (Fig. 5A). We also compared the -fold enrichment of KLF9 binding peaks in seven targets with control genomic regions located either 10 or 20 kb upstream from individual KLF9 binding peaks (Fig. 5B). Peaks identified by ChIP-Seq showed significantly higher enrichment than control regions. These validation results render credibility to the predictive value of our genome-wide data sets.
Regulation of Integrin ␣6 Expression and Function by KLF9 -Integrin ␣6 is the most upstream component of the integrin signaling pathway that was identified to be highly regulated by KLF9. Also, among the integrin pathway genes, ITGA6 was found to be one of the most down-regulated by KLF9 (Fig. 4A), and the ITGA6 promoter was validated by ChIP-PCR to have one of the most highly enriched (ϳ75-fold) KLF9-bound chromatin peaks (Fig. 5B). One KLF9 binding motif (referred to as basic transcription element (BTE) site) was identified in the human ITGA6 promoter at Ϫ396 bp relative to the TSS (Fig.  6A). This site was found to be conserved in mouse ITGA6 promoter (data not shown). KLF9 binding to this ITGA6 promoter region was confirmed in KLF9-GBM1b neurospheres by ChIP-PCR. FLAG-KLF9 co-precipitated with ITGA6 promoter regions containing the BTE site (segments A and B) but not with control regions lacking a BTE site (segments C and D) (Fig.  6, A and B). BTE-containing segments A and B were enriched compared with control segments C and D (Fig. 6B; 16.4-and 67.8-fold enrichment for segments A and B versus 4.3-and 0.9fold enrichment for segments C and D, respectively).
The effects of KLF9 induction on the level of integrin ␣6 mRNA and protein were quantified in two GBM neurosphere lines and two low passage primary GBM neurosphere isolates. KLF9 inhibited integrin ␣6 mRNA levels in all cell cultures (Fig. 6C). KLF9 induction also inhibited the expression of integrin ␣6 isoforms A and B (50) (Fig. 6D). We further examined the dynamics of ITGA6 expression in response to ectopic KLF9 expression controlled by Dox. ITGA6 expression levels changed inversely with KLF9 levels; ITGA6 expression decreased along with KLF9 induction and rapidly returned to baseline levels after Dox withdrawal (Fig.  6E). The effect of KLF9 expression on the number of integrin ␣6-positive cells within neurospheres mirrored its effects on bulk culture integrin ␣6 levels. KLF9 induction reduced integrin ␣6-positive neurosphere cells within neurospheres from 14.4 to 7.0% (Fig. 6F).
Integrin ␣6 Repression by KLF9 Inhibits GBM Cell Stemness and Tumorigenicity-We asked whether KLF9 modulates GBM cell behavior and stemness by regulating integrin ␣6 expression. To this end, we examined how KLF9 induction modulates GSC adhesion and migration on laminin, the integrin ␣6 ligand and an essential component of the perivascular niche that supports normal and neoplastic neural stem cells (51,52). KLF9 was induced in GBM neurospheres for 4 days, and cells were then transferred to laminin-coated tissue culture FIGURE 6. KLF9 down-regulates integrin ␣6 by promoter binding. A, schematic of the human ITGA6 promoter with a KLF9 binding peak identified by ChIP-Seq (Ϫ25,000 to ϩ3,000 bp relative to TSS). Primers were designed for segments A-D. B, Dox-treated KLF9-GBM1b cells were subjected to ChIP using FLAG antibody and mouse IgG. Selective enrichment in qPCR (left panel) and conventional PCR (right panel) was detected for BTE-containing segments A and B but not for control segments C and D. C, ITGA6 expression was measured by qPCR in GBM neurosphere lines and primary GBM neurospheres after KLF9 induction for 48 h. D, KLF9 was induced by Dox for 96 h in GBM1a and GBM1b cells. Membrane protein extraction was subjected to immunoblotting against integrin ␣6, which detected isoforms A and B. -Fold expression normalized to actin is shown below each lane. E, KLF9-GBM1b cells were treated with Dox for 6 days, passaged to Dox-free medium for 6 days, and then passaged again to Dox-containing medium for 6 days. Cells were collected on the days indicated and subjected to qPCR for ITGA6. ITGA6 expression levels changed inversely with Dox-induced KLF9 expression. F, KLF9-GBM1b cells were treated ϮDox for 96 h and subjected to flow cytometry using anti-integrin ␣6-FITC antibody or isotype IgG control. Representative dot plots and the percentages of integrin ␣6-positive cells are shown. Data represent mean Ϯ S.E. (error bars). *, p Ͻ 0.01, t test.
substrata. Control cells rapidly attached and spread within 2-6 h. KLF9 induction inhibited cell spreading (Fig. 7A, left panel) and decreased cell adhesion by 25-71% (Fig. 7A, right panel). We asked whether enforced integrin ␣6 expression could rescue the cell adhesion defects induced by KLF9. Lentivirus vectors were used to express KLF9, integrin ␣6, or both in GBM neurospheres for 48 -96 h prior to assessing cell adhesion and spreading on laminin. Enforced integrin ␣6 expression partially rescued the cell spreading defect (Fig. 7A, left panel) and abro-gated the ability of KLF9 to inhibit cell adhesion by ϳ71, 48, and 64%, respectively, in the three GBM neurosphere cultures (Fig.  7A, right panel). Migration through laminin-coated Transwell membranes was reduced in response to KLF9 induction by 62-86% (Fig. 7B). Enforced expression of integrin ␣6 also rescued this KLF9-induced cell migration defect (Fig. 7B).
Integrin ␣6 is highly expressed in various stem cells (48,53), and cell surface expression was found to be enriched in GSCs and to maintain GSC self-renewal and tumorigenicity (48). We KLF9 expression inhibited cell adhesion. Co-expressing integrin ␣6 partially rescued cell adhesion inhibition by KLF9. B, cells were passaged to laminin-coated Transwell membranes. Cell migration was evaluated 24 h later by analyzing DAPI-stained cells. KLF9 expression inhibited transmembrane migration, which was rescued by integrin ␣6. C, transfected KLF9-GBM cells were maintained in neurosphere growth medium for 96 h. Whole cell lysates were subjected to immunoblotting to assess the expression of FLAG-KLF9, integrin ␣6, and the stem cell markers CD133, Nestin, Sox2, BMI1, and Olig2. Relative expression of each protein (normalized to actin) is shown below each band. KLF9 inhibited the expression of stem cell markers, and this response was partially rescued by enforced integrin ␣6 expression. Scale bars, 50 m. Data represent mean Ϯ S.E. (error bars). *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001, one-way analysis of variance followed by Tukey test.
hypothesized that KLF9 inhibits glioma cell stemness in part by repressing integrin ␣6 expression. KLF9 down-regulated the expression of stemness regulators and stem cell markers in GBM neurospheres (Fig. 7C). Enforced integrin ␣6 expression reversed KLF9-induced suppression of CD133, Nestin, Sox2, BMI1, and Olig2 expression (Fig. 7C). These results support the conclusion that KLF9 regulates molecular markers and drivers of GSC stemness in part by repressing ITGA6.
We found previously that KLF9 induction reduces the growth of intracranial tumor xenografts established from GBM neurospheres and extends the survival of mice bearing xenograft tumors (31). We hypothesized that KLF9 reduces xenograft tumor growth in part by repressing integrin ␣6 expression. We examined the effects of expressing KLF9, integrin ␣6, or both on the growth of intracranial xenografts established from GBM-derived neurospheres. KLF9, integrin ␣6, or both were expressed in GBM neurosphere cells by lentivirus infection (Fig. 8A), and cells were implanted to the brains of immunocompromised mice 48 h after virus infection. Mice (n ϭ 4 for each group) were sacrificed 60 days post-transplantation, and coronal histological brain sections were examined for tumor size. Enforced KLF9 expression significantly inhibited tumor xenograft growth compared with control xenografts (tumor volume Ϯ S.E., 9.0 Ϯ 2.9 versus 34.4 Ϯ 6.1 mm 3 ; p Ͻ 0.01). Expressing integrin ␣6 alone as an additional control had no effect on xenograft growth. However, integrin ␣6 expression rescued tumor growth inhibition induced by KLF9 (Fig. 8, B and C).

DISCUSSION
Cancer cells are highly plastic in their ability to shift between more differentiated and more stemlike states. This phenotypic plasticity is dynamically controlled by transcription factor networks that regulate cell multipotency, differentiation, and tumor propagating capacity through coding and noncoding RNAs (18). Identification of these regulatory networks is paramount to understanding cancer cell hierarchy, malignant progression, and mechanisms of therapeutic resistance.
We reported previously that the KLF9 transcription factor inhibits glioma cell stemness and that KLF9 induction was found to inhibit GSCs and the growth of GSC-derived tumor xenografts (31). The inhibitory effect of KLF9 on stemness diverges from the stemness-supporting effects of other KLFs, such as KLF2, -4, and -5 (13,14,54). The molecular mechanism underlying the function of KLF9 in human cancer models had not been extensively examined. This current study has established for the first time a genome-wide map of KLF9-regulated targets in a human GSC model. 31,261 genome-wide KLF9 binding peaks were identified in GBM neurosphere cells under biological conditions of KLF9-induced stemness inhibition. Among genes around these peaks, 1,849 gene targets were found to be directly down-regulated by KLF9. The predictive value of these data sets is supported by our validation of KLF9regulated gene expression and KLF9 binding sites in the KLF9 targets we identified. The majority of KLF9 binding sites were found to be located in intergenic regions and introns, and KLF9 binding peaks were found to be most enriched proximal to the TSS, patterns that support the transcriptional regulatory function for KLF9. Statistical analyses of the differentially expressed genes bound by KLF9 proximal to TSS showed that KLF9 functions primarily as a transcriptional repressor in the context of glioma neurospheres. This does not absolutely rule out the possibility that KLF9 has transactivating activity under defined biological contexts, particularly given the strong influence of biological context and genetic background on the gene regulatory activities, including those of KLF family members (24,55). A particularly relevant example is from Mitchell and DiMario (56), who found that KLF9 transactivated an FGFR1 promoterreporter construct in human myoblasts but repressed the same reporter in differentiated myotubes. However, the capacity for these reporters to accurately mirror the regulation of endogenous FGFR1 expression in the same models was not addressed. KLF9 contains domains known to mediate complex formation with Sin3 histone deacetylase and with histone acetylase, highlighting its potential for gene repression and activation, respectively (57,58).
Ours is the first application of genome-wide de novo motif discovery to define the KLF9 binding DNA consensus sequence, 5Ј-(G/A)(G/T)GGG(C/T)G(G/T)GGCN-3Ј. This result is consistent with a previous report showing that KLF9 (also known as basic transcription element-binding protein) binds to a GC-rich DNA sequence (26). We also confirmed the similarity of this KLF9 binding motif to the motifs of Sp/KLF family members, such as Sp1 and KLF4. KLF2, -4, and -5 have been shown to extensively share binding sites  , 1 mm). C, quantification of tumor xenograft volumes shows that KLF9 induction inhibited xenograft growth and that integrin ␣6 expression rescued KLF9-induced antitumor response. Data represent mean Ϯ S.E. *,p Ͻ 0.01; **, p Ͻ 0.001, one-way analysis of variance followed by Tukey test.
across the genome and collaboratively support the self-renewal of pluripotent stem cells (54). These and other Sp/KLF family members share DNA binding motifs by virtue of highly conserved zinc finger DNA-binding domains located in the C terminus (21). The activating or repressing transcriptional effects of KLF family members can be attributed at least in part to their divergent N-terminal domains that are believed to mediate complex formation with other transcription co-regulators (21). Considerably more information is needed to fully understand mechanisms behind the collaborative or competitive interplay among Sp/KLF family members and the context-dependent effects of this transcriptional network on cell stemness and malignancy.
Gene function annotation and pathway analyses revealed that signaling pathways relevant to cancer signaling, stem cell regulation, and neural cell function are enriched in KLF9 downregulated gene targets. Integrin pathway members, including ITGA6, ITGA9, ARPC1B, CAPN5, GIT1, and MYLK, were found to be particularly enriched for KLF9 down-regulated targets. Integrin signaling modulates various cellular processes, such as proliferation, migration, and differentiation, through cell-cell and cell-matrix interactions (59). Integrins, a family of cell adhesion receptors, are up-regulated in stem cells and are believed to support cell stemness by mediating adhesive interactions between stem cells and their niches (60). Integrin ␣6, a subunit for two laminin receptors, is also highly expressed in several types of stem cells, including subventricular neural stem cells (61,62), and facilitates neural stem cell adherence to endothelial cells in the perivascular niche (62). Integrin ␣6 has also been identified as a marker for cancer stem cells, including GSCs (48,53). It has been proposed that integrin ␣6 is essential for the maintenance of GSC self-renewal and tumor-initiating cell phenotype, providing a potential target for anti-GBM therapies (48). Despite the involvement of integrin ␣6 in stem cell maintenance, little was known about the mechanisms that regulate its expression in cancer cells or how it regulates stem cell phenotypes. Within this context, it is interesting that our genome-wide analysis identified KLF9-regulated targets involved with cell adhesion and migration mechanisms but not previously linked to cell stemness regulation. Examples include RAP2A, which codes for a Ras-related protein that regulates cytoskeletal arrangement, adhesion, and migration (63,64); ARPC1B, which codes for subunits of the Arp2/3 complex implicated in actin polymerization (65); and RhoV, which regulates expression of proinvasion transcription factors (66). This suggests that KLF9 may be particularly important in regulating a transcriptional network directed at cytoskeletal, adhesive, and migratory function.
We present novel findings that specifically link the repression of integrin ␣6 expression to the mechanism by which KLF9 inhibits GBM cell stemness. Multiple complementary criteria were used to show that KLF9 induction inhibits integrin ␣6 expression by directly repressing ITGA6 promoter activity. These include KLF9-induced reductions in integrin ␣6 mRNA and protein and the interaction between KLF9 and the ITGA6 promoter. The mechanistic relevance of integrin ␣6 repression by KLF9 is supported by the concurrent inhibition of laminindependent GBM neurosphere cell adhesion, cell migration, and cell stemness in response to KLF9 induction and their normalization by enforced integrin ␣6 expression. The capacity for enforced integrin ␣6 expression to reverse the KLF9-induced inhibition of stem cell markers supports that KLF9 regulates stemness in part by down-regulating integrin ␣6 expression. GBM is a very hypervascular neoplasm, and integrin ␣6-expressing GSCs preferentially localize to the perivascular niche that provides tropic signals and adhesive matrix proteins (e.g. laminin) to promote tumor cell stemness (67). Consistent with this trophic interaction, Lathia et al. (48) found that silencing integrin ␣6 reduces the self-renewal and tumor propagating capacity of GSCs. Further investigation into the effects of KLF9 on the interaction between GSCs and the perivascular niche could provide further insight into the mechanisms behind the inhibitory role of KLF9 in GSC stemness and tumor suppression.
Two other signaling pathways, CXCR4 and Notch, which regulate malignancy and stem cell biology, were also among those significantly down-regulated by KLF9. CXCR4 is a chemokine receptor with multiple functions, such as regulating the migration of stem cells and cancer cells (68,69). CXCR4 signaling is active in over 20 human tumors, including gliomas, and contributes to tumor promotion in multiple ways (69,70). CXCR4 is highly expressed in GSCs (71) and was recently reported to facilitate the recruitment of GSC-derived pericytes to CXCL12-positive endothelial cells to support the tumor vasculature and tumor growth (3). In addition to our novel findings with KLF9, several other KLFs have been reported to regulate CXCR4 expression in cancer cells. KLF2 down-regulates CXCR4 in oral cancer cells (72), whereas KLF5 has been shown to up-regulate CXCR4 in prostate cancer (73). Notch signaling has a prominent role in maintaining stem cell phenotypes within a variety of developmental and neoplastic contexts (e.g. neural stem cells, mammary stem cells, leukemia, GBM, and mammary carcinoma) by promoting self-renewal and inhibiting differentiating programs (74 -76). Our current finding that KLF9 represses multiple Notch pathway components further supports our earlier discovery that KLF9 represses Notch1 expression and its downstream signaling (31). We now broaden these earlier findings by showing that KLF9 also down-regulates NUMBL, a component of the Notch signaling pathway (77). Evidence supports a regulatory role for NUMBL in cell stemness. For example, NUMBL increases pluripotency marker expression in embryonic stem cells by synergizing with Hedgehog signaling (77).
In conclusion, we have identified gene targets directly regulated by KLF9 on a genome-wide scale in a GBM-derived neurosphere model. We have shown that KLF9 inhibits GBM cell stemness and functions predominantly as a transcriptional repressor within this cellular context. KLF9 was shown to regulate multiple signaling pathways, including those involved in oncogenesis, stem cell regulation, neuronal cell signaling, and integrin signaling. KLF9 was shown to inhibit GBM cell stemness and tumorigenicity, and these antitumor effects result from integrin ␣6 repression. These findings enhance our understanding of the transcriptional networks underlying cancer cell stemness and differentia-tion and identify KLF9-regulated molecular targets applicable to cancer therapeutics.