Defining a role for Sonic hedgehog pathway activation in desmoplastic medulloblastoma by identifying GLI1 target genes

Preparation of cell lines

Rat kidney epithelial cells stably transfected with E1A (RK3E cells, American Type Culture Collection, CRL 1895) were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM, GIBCO-BRL, Grand Island, NY) supplemented with 10% fetal bovine serum, penicillin (100 U/ml) and streptomycin (100 µ/ml). Human GLI1 in a retroviral vector (pLTR-GLI1) was introduced into RK3E cells by liposome-mediated transfection (RK3E + GLI1). Transformed foci formed after 2 weeks in culture. Two cell lines were established by trypsinization of independent foci and called RK3E + GLI1-I and RK3E + GLI1-II.

Cell counting

RK3E and RK3E + GLI1 cells were counted using a hemocy-tometer following culture for up to 5 days. Triplicate counts were obtained and differences between cell types assessed using the Student’s t-test.

Western blot analysis

Cell lysates were prepared using 1× lysis buffer (Promega, Madison, WI). Fifty to one hundred micrograms of total protein was separated using 7.5–12% SDS-PAGE and transferred to nitrocellulose membranes. GLI1 protein was detected using polyclonal anti-GLI1 antibody (Rockland, Gilbertsville, PA); p53 using a pol-yclonal anti-p53 antibody (Santa Cruz Biotech, Santa Cruz, CA); and β-actin using a monoclonal antibody (Sigma, St. Louis, MO). Antibodies were diluted 2,000- to 4,000-fold in PBS with 0.3% Tween-20. Protein was visualized using a chemiluminescence kit (Pierce, Rockford, IL).

Microarray assay

Total RNA was isolated from RK3E and RK3E + GLI1 cells using the Qiagen RNeasy RNA isolation kit (Qiagen, Valencia, CA). cDNA was synthesized using the SuperScript Choice system (Invitrogen, Carlsbad, CA). The manufacturer’s protocol was modified slightly to include a high-performance liquid chromatog-raphy-purified T7-(dT) 24mer primer (Genset Corp., LaJolla, CA). Double-stranded cDNA was purified by phenol/chloroform extraction using phase lock gels (Eppendorf-5 Prime, Gaithersburg, MD) followed by ethanol precipitation. Biotin-labeled cRNA was prepared using the Enzo BioArray high yield RNA transcript labeling kit (Affymetrix, Santa Clara, CA). The in vitro transcription product was purified using RNeasy spin columns (Qiagen, Valencia, CA). We used oligonucleotide arrays (Affymetrix Rat 230 2.0) containing over 31,000 probe sets (Affymetrix, Santa Clara, CA). Hybridization, washing, staining and scanning were carried out using preprogrammed Affymetrix protocols. Data files (RK3E, RK3E + GLI1-I and RK3E + GLI1-II) were uploaded into Affymetrix Microarray Suite 5.0 (MAS5) software. Genes showing more than ±2-fold changes compared with RK3E cells were organized into a list. Genes that were upregulated in one cell line but downregulated in the other were excluded. The resulting list of genes was studied, and further information on genes of interest was compiled as tables. MIAME (minimum information about a microarray experiment) compliant data from our experiments has been deposited with GEO (Gene Expression Omnibus) at the National Center for Biotechnology Information (NCBI), GEO accession # GSE11987.

RT-PCR analysis

RT-PCR was performed using the Qiagen RT-PCR kit (Qiagen, Valencia, CA). cDNA was synthesized using random hexamers with 0.1 to 0.5 µg of total RNA. Identical reactions without RNA or without reverse transcriptase served as negative controls. cDNA was amplified using 32–35 PCR cycles. Amplified DNA was analyzed by agarose gel electrophoresis. Primers were designed using MacVector software (MacVector, Cary, NC). The primer sequences are shown, listing the forward primer (5′-3′) followed by the reverse primer (5′-3′ ): Activated leukocyte cell adhesion molecule (Alcam) TTTTACTTACCAGGGCAGGCTG and CCTTCATCCACACCACAGTTG, chemokine (C-X-C motif) receptor 4 (Cxcr4) CAGAAGAAACCACACAGCACAACC and AAGGAACTGAACGCTCCAGAATG, desmuslin (Dmn) GGAA GGCTAAGAAACATTCAGTATC and TAGGTCACTAAGGGC AGAGC, glycine amidinotransferase (Gatm) CCTTCAACATC ATTGGACCTGG and TCATTGGCGTCTACCATCACG, Mgmt TCTCCATCACCCTGTGTTCCAG and ATTGCTCCTCATCGC TCCTCCTAC, myosin IE (Myo1E) GCTGACTGATGGT CTTGCTTTCC and ATACTACCTGCGGCATTCATACG, Ntrk2 GTCAAGTTTACCTAAGGGCACCTG and ATGGTTATCCGC ACTGGGAAG, p53 TCTGTCATCCTTCCGTCCCTTCTC and GGCACAAACACGAACCTCAAAG, phosphoribosyl pyrophosphate synthetase-associated protein 1 (Prpsap1) AGACCAATGG AGAAACCAGAGTTG and GGGGAAGCAGAAAAAGCCTTG, phospholipid transfer protein (Pltp) CGGTTCTTGTCAATCA CTTCTCG and AATGAGCGTATCTGGCGTGG, protein kinase C, eta (Prkch) GCAGGATGACGATGTGGAATG and GTCT CTGTAGATGATGCCTTTCTCG, solute carrier family 19, member 1 (Slc19a1) TGACCTTCGTGCTTTTCCGTG and GAAC TGTTGATGGACTTGGAGGC, slit homolog 2 (Slit2) GCGGAT TATCTCCACACCAACC and TTTCAGGACAAGCCAAGT CTGC and GAPDH ATCACCATCTTCCAGGAGCG and CTAAGCAGTTGGTGGTGC. We performed multiple RT-PCRs using the same batch of RNA and always included a control GAPDH RT-PCR. The levels of GAPDH in RK3E and RK3E+ GLI1 RNAs were comparable. Representative GAPDH RT-PCR bands for RK3E and RK3E+GLI1 RNA are shown.

Identifying GLI1 target genes in medulloblastomas

We compared the genes whose expression was altered more than ±2-fold in RK3E + GLI1 cells to those that define each of the previously described 5 subgroups of medulloblastomas based on analysis using Affymetrix U133 av2 chips.³ Human orthologs of rat genes were obtained using NetAffx analysis (Affymetrix, Santa Clara, CA). Genes present in both rat and human arrays were selected using Excel software if their regulation was altered in the same direction (Microsoft corp., Redmond, WA). Genes whose regulation was in the opposite direction were excluded from this analysis.

Identification of potential direct targets of GLI1

Of the 25 genes whose expression was altered similarly in RK3E + GLI1 cells and medulloblastomas with SHH pathway activation, 11 contained regulatory regions that had been previously characterized. We searched these regulatory regions for the GLI consensus binding motif, GACCACCCA, and for sequences with a mismatch in 1 or 2 loci using MacVector software (MacVector, Cary, NC). The regulatory regions of the remaining 14 genes had not yet been clearly defined and therefore were not analyzed for the GLI-specific DNA binding motif.

Electrophoretic mobility shift assay

Electrophoretic mobility shift assays were performed as described previously.²⁸ Probes were designed using MacVector software (MacVector, Cary, NC). Probe sequences are shown, listing the sense sequence (5′-3′) followed by the anti-sense sequence (5′-3′) used to produce the double-stranded probe: NK6 transcription factor related, locus 1 (hNKX6.1-1) TGTAGTACTT GAC-CACCCACACCTCAATC and GATTGAGGTGTGGGTGGT CAAGTACTACA, hNKX6.1-4 CCAGCTGCTGTGGGTGGAC TTCCAGAGGA and TCCTCTGGAAGTCCACCCACAGCAG CTGG, bone morphogenic protein-7 (hBMP-7–1) CAAACCA AAGGACCACCAAGGAGGGTGCA and TGCACCCTCCTTGG TGGTCCTTTGGTTTG, hBMP-7–3 GGACTCCCCAGACCACT CAGAACCCACCG and CGGTGGGTTCTGAGTGGTCTGGGG AGTCC, hPRPSAP1-1 GGCCTCTGGAGCCCACACACTGCC CTATA and TATAGGGCAGTGTGTGGGCTCCAGAGGCC, hPRPSAP1-2 GGGTAACACGGAGCACTCAGGAATACACG and CGTGTATTCCTGAGTGCTCCGTGTTACCC, hPRPSAP1- 3 TCAGCGCCTTCCCCACCCAAACCGGGGTC and GACCC CGGTTTGGGTGGGGAAGGCGCTGA, hPRPSAP1-4 AAGG GGGCGGTGGGAGGGCCGGGCCGGGG and CCCCGGCCC GGCCCTCCCACCGCCCCCTT and mKrox-20 CCACGCAGCG GACCGCCCAGACACGCCGG and CCGGCGTGTCTGGGCG GTCCGCTGCGTGG. The nonspecific probe sequences were GACTCTCCCGAAGAAGTGGGACAATGATGGTTC and GAC TGAACCATCATTGTCCCACTTCTTCGGGAG. The purified GLI1 protein used for electrophoretic mobility shift assays included aa 211–1106. The control protein was either PinPoint protein (Promega, Madison, WI) or PinPoint protein fused to GLI1 aa 879–1106, lacking the zinc finger DNA binding domain.

Preparation of reporter constructs that contain regulatory regions of GLI1 target genes

The human Cyclin D2-luciferase reporter was obtained from Milner (CV Therapeutics, Palo Alto, CA). The Cyclin D2-luciferase reporter was modified by inserting 3 tandem copies of the region of the Cyclin D2 promoter that contains the GLI binding site (5′ -GACTTCTGCTCGCCCACCACCCAATCCTCGCCT-3′) into the MluI site to make the pGL3-hCyclin D2 reporter. Nine hundred and twenty-two base pairs of the mouse Krox-20 promoter, including the region containing a GLI binding site (GACCGCCCA), were PCR amplified from genomic DNA (sense 5′-AAGGTGTGTTGAGTGGGTTTGTAAG-3′ and anti-sense 5′-GCTCGCCAGGAACGATTTG-3′) and inserted into the SalI/ XhoI site of the pGL3 basic vector to make an mKrox-20-luciferase reporter (pGL3-mKrox-20). A 2,041 base pair fragment of the human PRPSAP1 gene, containing 4 potential GLI binding sites (GCCCACACA, GAGCACTCA, CCCCACCCA, TGGGAGG GC) was PCR amplified (sense 5′-CTTCAGACTTGTGGTCCTT CAC-3′ and anti-sense 5′-TTGAGAGTGGCGGCGGTCAGA TTC-3′) and inserted into the TA cloning vector. hPRPSAP1 DNA was retrieved by KpnI/EcoRV digestion and then ligated into the KpnI/SmaI site of the pGL3 basic vector to make an hPRPSAP1 reporter (pGL3-hPRPSAP1). One thousand five hundred and twelve base pairs of the regulatory sequence of the human NKX6.1 gene, containing a GLI binding site (GACCA CCCA), were PCR amplified (sense 5′-TCAATGCCTAACATCT ACCAGTCG-3′ and anti-sense 5′-GCCTGAATGAGCGGTGAT TACAC-3′) and ligated into the SalI site of the pGL2 promoter vector to produce the hNKX6.1 reporter (pGL2-hNKX6.1).

Cotransfection assays

4 × 10⁵ HeLa cells were grown in minimal essential medium supplemented with 10% fetal bovine serum, 1% L-glutamine and 1% penicillin/streptomycin. Cells were cotransfected using 6 µl of Lipofectamine 2000 reagent (Invitrogen, Carlsbad, CA) with 0 to 1,000 ng of either the pM-GLI1 or pCMV-GLI1 effector plasmid, 600 ng of the reporter construct and 10 ng of Renilla control reporter DNA (Promega, Madison, WI). pM-GLI1 and pCMV-GLI1 were described previously.³⁰ pM (Clontech, Mountain View, CA) or pcDNA3 (Stratagene, LaJolla, CA) vectors served as control effectors. A total of 3 µg of DNA was transfected in each experiment and the difference was made up with pBluescript carrier DNA (Stratagene, LaJolla, CA). Cell lysates were prepared 48 hr after transfection. One hundred microliters of substrate solution (Promega, Madison, WI) was added to 20 µl of cell lysate. The experiments were performed at least in triplicate and results expressed as an average with standard deviation. Statistical significance was assessed using the Student’s t-test.

Establishing interactions among GLI1-responsive genes

To identify functional interactions among GLI1 responsive genes, we used Pathway Architect Software (Stratagene, LaJolla, CA) and NCBI database searches. In particular, we used search modes including gene regulation, protein binding and protein-promoter binding. We expanded the analysis to identify genes important for relating the GLI1 responsive genes using network expansion. Genes originally identified by Pathway Architect were crosschecked with literature searches and unrelated genes were eliminated from the network. A diagram was prepared using Graph Visualization software.

Transient transfection of RK3E cells with pCMV-GLI1

Ten micrograms of pCMV-GLI1 or pcDNA3 were transfected into RK3E cells using Lipofectamine 2000 (Invitrogen, Carlsbad, CA) and OptiMEM (Gibco-BRL, Grand Island, NY). Cell lysates were prepared after 48 hr.

DNA sequence analysis

DNA sequencing was performed using the 3730xl DNA analyzer (Applied Biosystems, Foster City, CA).

The gene expression profile that defines transformed RK3E + GLI1 cells

We established 2 independent cell lines from RK3E cells stably transfected with human GLI1 (RK3E + GLI1-I and RK3E + GLI1-II) that had lost contact inhibition and formed foci in culture (Fig. 1a). The transformed cell lines showed strong GLI1 protein expression and accelerated cell proliferation compared with the parental cells (Figs. 1b and 1c). We used oligonucleotide arrays (Affymetrix Rat 230 2.0), containing over 31,000 probe sets to identify GLI1 targets in the RK3E + GLI1 cell lines, characterized by this transformed phenotype; the phenotype we believed to be most comparable to an established human tumor.

We identified 1,823 genes whose levels of expression were altered ±2-fold in RK3E + GLI1-I and RK3E + GLI1-II cell lines compared with RK3E cells; 572 were upregulated and 1,251 downregulated (Supp. Info. data). We conducted RT-PCR to confirm changes in expression for select genes (Fig. 2). Of interest, the RK3E + GLI1 gene expression profile includes coordinated regulation of several genes that are known to function in the Wnt signaling pathway, including downregulation of pathway activators (Wnt2b, Wnt5b and Frizzled homolog 1 [Fzd1]) and upregulation of negative regulators of the pathway (glypican3 and Wnt inhibitory factor 1 [Wif1]); the Bmp signaling pathway, including downregulation of Bmp-1, 2, 4, 6, and bone morphogenic protein receptor, type II (Bmpr2) and upregulation of Bmp-7; the Notch signaling pathway, including upregulation of Notch1, Notchless (Nle1), and downregulation of Jag1 and Adam10; the chemokine pathway, including upregulation of Cxcr4, chemokine (C-X-C motif) ligand (Cxcl) 4, 9 and 10, and downregulation of Cxcl12; and the receptor tyrosine kinase (RTK) signaling pathway, including upregulation of Stat2 and downregulation of Ntrk2.^31–39

Potential GLI1 targets in human medulloblastomas

To begin to define GLI1 targets in medulloblastomas and therefore the role of GLI1 in medulloblastoma biology, we then compared expression of the GLI1 target genes identified in the RK3E cell transformation model with previously described gene expression profiles that define 5 subgroups of medulloblastomas.³ We selected genes whose expression patterns were identical in RK3E + GLI1 cells and each of the medulloblastoma subsets. Seventy-one genes from those that define the subgroup of medulloblastomas with SHH pathway activation and predominantly of desmoplastic histology (called Subgroup D) were initially identified in this analysis.³ However, only 25 showed the same expression pattern compared to those identified in RK3E + GLI1 cells (Table I). The other previously identified medulloblastoma subgroups showed altered expression of very few of these 25 genes. Subgroups A and C did not show altered expression of any of the genes, while 4 in subgroup B (MGMT, SGK1, PLTP and TST) and 2 in subgroup E (DNAJB9 and KIDINS220) showed comparable altered expression, confirming a close association of the list of 25 genes with the activation of the SHH signaling pathway and GLI1 expression.³

Identification of direct targets of GLI1 in RK3E + GLI1 cells and medulloblastomas

To identify direct targets of GLI1, we searched the promoter and/or enhancer regions of select genes identified in the RK3E + GLI1 cells for the GLI-specific DNA binding motif, GAC-CACCCA, and for sequences with a mismatch in 1 or 2 loci. We found potential consensus binding elements in the regulatory regions of 10 of the 25 genes (Table I). We analyzed the regulatory regions of 3 additional genes (hBMP-7, hNKX6.1 and mKrox-20) whose expression was altered in RK3E + GLI1 cells, but not medulloblastomas, based on their known roles in early development and the fact that their regulatory regions had been previously defined. We found potential consensus binding elements in each of them (Table II). We performed electrophoretic mobility shift assays to determine whether GLI1 physically binds select regulatory regions. The probes with putative GLI binding elements from hBMP-7, hNKX6.1, mKrox-20, and 2 sites in the 5′ region of hPRPSAP1 shifted in the presence of GLI1 protein, suggesting that these genes could represent direct targets (Fig. 3).

To determine whether GLI1 regulates transcription through these regions, we cotransfected HeLa cells with either the pM-GLI1 or pCMV-GLI1 expression vectors and luciferase reporters, containing the regulatory regions of Cyclin D2 (pGL3-hCyclin D2), mKrox-20 (pGL3-mKrox-20), hPRPSAP1 (pGL3-hPRPSAP1) or hNKX6.1 (pGL2-hNKX6.1). We had previously identified a GLI consensus binding element in the regulatory region of human Cyclin D2.²⁸ The 150-kD GLI1 protein was not detected endoge-nously in HeLa cells (data not shown). As increasing amounts of the GLI1 expression constructs were cotransfected with a constant amount of either the pGL3-hCyclin D2 or the pGL3-hPRPSAP1 reporter construct, luciferase activity increased in a concentration dependent manner, suggesting that GLI1 directly activates transcription through these regulatory regions (Figs. 4a and 4b). Surprisingly, as an increasing amount of the GLI1 expression construct was cotransfected with a constant amount of the pGL3-mKrox-20 reporter construct, luciferase activity decreased to a maximum of 2-fold in a concentration dependent manner (Fig. 4c), suggesting that GLI1 directly represses transcription of mKrox-20 through this regulatory region. Since it has generally been accepted that GLI1 functions only as a transcription activator and does not undergo cleavage, we looked for full length GLI1 in HeLa cells and RK3E cells using an anti-GLI1 polyclonal antibody prepared from a C-terminal GLI1 fragment. We identified full-length GLI1 protein in both cell lines (data not shown). This suggests that full-length GLI1 may indeed function as a transcription repressor at least under some circumstances for some promoters. By this method, GLI1 did not change reporter gene transcription through the hNKX6.1 promoter region (Fig. 4d). Therefore, we were frequently able to provide evidence that GLI1 directly regulates transcription through regulatory regions of genes that contain GLI consensus elements, supporting the idea that the 10 genes that contain GLI consensus elements potentially represent direct targets.

Functional relationships among the GLI1-responsive genes in medulloblastoma

In an effort to identify functional relationships among the GLI1-responsive genes in medulloblastomas with SHH pathway activation, we created a regulatory network of the genes using the Direct Interaction Network algorithm of Pathway Architect software (Stratagene, LaJolla, CA) and verified the associations with the primary literature or NCBI databases (Fig. 5). The network appears to converge on genes that regulate cell proliferation, survival and genomic stability, including p53, SGK1 and MGMT. In addition, c-MYC and GLI1 activate several of the same targets, including p53, PRPSAP1, RGS16 and NPM1; MAX and GLI1 regulate PRPSAP1, RGS16 and DNAJB9; and TAF1 interacts with multiple GLI1 targets with diverse functions, including PRPSAP1, NMP1, RGS16, DNAJB9 and PDE4B, suggesting that c-MYC, MAX and TAF1 function cooperatively with GLI1 during tumorigenesis.

Mutant p53 in RK3E + GLI1 cells

It is intriguing that microarray analyses demonstrate upregulation of p53 in RK3E + GLI1 cells and in medulloblastomas with SHH pathway activation. We confirmed this observation at the RNA level (Fig. 2). We also demonstrated upregulation of p53 protein following transfection of GLI1 into RK3E cells (Fig. 6a). Since wild-type p53 inhibits cell survival and the transformed phenotype, yet GLI1 functions as an oncogene, we looked for p53 mutations in RK3E + GLI1 cells, as a possible explanation for GLI1-induced increases in p53 in transformed cells. We identified a p53 point mutation (R246H) in the RK3E + GLI1 cells (Fig. 6b). These data imply that p53 and GLI1 may work cooperatively to ensure normal cellular homeostasis, however, in the setting of p53 mutations the balance may shift toward pathologic cell survival.