Microarray analysis detects novel Pax3 downstream target genes.

Pax3 is a transcription factor that is required for the development of embryonic neural tube, neural crest, and somatic derivatives. Our previous study (Mayanil, C. S. K., George, D., Mania-Farnell, B., Bremer, C. L., McLone, D. G., and Bremer, E. G. (2000) J. Biol. Chem. 275, 23259-23266) reveals that overexpression of Pax3 in a human medulloblastoma cell line, DAOY, resulted in an up-regulation in alpha-2,8-polysialyltransferase (STX) gene expression and an increase in polysialic acid on neural cell adhesion molecule. This finding suggests that STX might be a previously undescribed downstream target of Pax3. Because Pax3 is important in diverse cellular functions during development, we are interested in the identification of additional downstream targets of Pax3. We utilized oligonucleotide arrays and RNA isolated from stable Pax3 transfectants to identify potential target genes. A total of 270 genes were altered in the Pax3 transfectants as compared with the vector control and parental cell line. An independent analysis by cDNA expression array and real-time quantitative polymerase chain reaction of several genes confirmed the changes observed by the oligonucleotide microarray data. Of the genes that displayed significant changes in expression, several contain paired and homeodomain binding motifs of Pax3 in their promoter regions. Using promoter-luciferase reporter transfection assays and electromobility shift assays, we showed at least one previously undescribed downstream target, STX, to be a biological downstream target of Pax3. Thus we report several previously undescribed candidate genes to be potential downstream targets of Pax3.

Regulation of genes during development is tightly modulated by specific transcription factor families, which include the homeobox, T-box, Hox, and Pax families. Pax3 is one such member of a paired homeobox family of an evolutionary conserved transcription factor (1) that plays a pivotal role during neural tube closure. Pax3 is expressed in a spatio-temporally restricted manner during embryogenesis (2), specifically at embryonic day 8.5 in the developing murine spinal cord and brain. After neural tube closure, Pax3 expression is maintained in the dorsal half of the neural tube (2). Deficient Pax3 expression, such as in Splotch mice, results in neural tube defects and an array of neural crest cell related abnormalities (3). Cell adhesion and migration are fundamental processes that contribute to many aspects of neural tube and neural crest dorsoventral patterning and morphogenesis (4 -11). Because Pax3 is implicated in the invasive and metastatic potential of pediatric tumors, it is probable that it does so by modulating cell adhesion and migration (1,(12)(13)(14).
Up-regulation of specific genes and delayed expression of others is expected of a cell that is committed to adhesion or migration (15). Our previous study (16) showed that overexpression of Pax3 in a human medulloblastoma cell line, DAOY, had a direct impact on the cell adhesion properties. Pax3 transfectants were found to have up-regulated ␣-2,8polysialyltransferase (STX) 1 gene expression, which resulted in increased polysialylation of neural cell adhesion molecule (NCAM). Pax3 transfectants expressing high polysialic acid (PSA) on NCAM showed much less NCAM-dependent aggregation than those with less PSA. In addition, these Pax3 transfectants with high PSA-NCAM showed heterophilic adhesion to heparan sulfate proteoglycan (HSPG) and agrin. These experiments suggested that Pax3 might regulate the expression of additional genes involved in cell adhesion and migration. To search for additional downstream targets of Pax3, we employed a microarray approach. Several Pax3 downstream targets such as NCAM (17), MyoD (18), c-MET (19,20), Dep-1 (21), myelin basic protein (MBP) (22), MITF (23,24), and Tyrp-1 (24) have been reported. Even though these Pax3 downstream targets have been described, an unambiguous Pax3 consensus binding motif is lacking (25). As a result, the progress of identifying specific downstream targets of Pax3 has been considerably slowed down. Therefore, a more complete picture of the network of genes altered by Pax3 is required to fully understand its physiological significance. To identify Pax3 downstream targets in this study, we compared the gene expression of three stable Pax3-transfected clones (16) to vector alonetransfected and parental cell lines using oligonucleotide microarrays.

Transfection of Pax3 cDNA into DAOY Cells and Selection of Transfectants
Stable transfection of mouse Pax3 in the human medulloblastomaderived cell line, DAOY, was done as described earlier (16). Briefly, a 2.3-kilobase murine Pax3 cDNA was inserted into the pcDNA3 expression vector (Invitrogen) at the EcoRI sites. The 2.3-kilobase Pax3 cDNA was prepared from pBH3.2, which was kindly provided by Dr. Peter Gruss, Max Planck Institute, Göttingen, Federal Republic of Germany (19). pBH3.2 was digested with EcoRI (Amersham Pharmacia Biotech) to remove the 2.3-kilobase Pax3 cDNA, and this was then ligated into pcDNA3 at the EcoRI site. Restriction digests confirmed orientation of the cDNA insert. The pcDNA3/Pax3 construct or pcDNA3 alone as the vector control was then transfected into the cells using a cationic liposome system, DOTAP (N-[1-(2,3-dioleoyloxy)propyl]-N,N,N-trimethylammonium salts; Roche Molecular Biochemicals). Transfectants were selected by antibiotic resistance in cell medium containing 900 g/ml Geneticin G418 (Life Technologies Inc.). Colonies were cloned by limiting dilution. After 4 weeks in culture in the presence of G418, surviving colonies were tested for the presence of Pax3 mRNA. Cells were seeded in 96-well plates with a cell density of 1 cell/well. The cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Life Technologies Inc.), penicillin (50 units/ml), streptomycin (50 g/ml), and L-glutamine (2 mM) supplemented with Geneticin G418 (900 mg/ml) then incubated at 37°C in a humidified 95% air, 5% CO 2 incubator. Replicate plates of 96-well plates were made from the colonies formed from the initial cell in each well. When cells in the replicate plates were near confluence, mRNA was isolated using the Poly(A)Tract series 9600 system (Promega) following the instructions in the Promega manual.
To synthesize cDNA, 5 l of mRNA from each well was added to each well in a 96-well PCR reaction plate. The RNA was treated with DNase for 20 min at 25°C, and then the DNase was inactivated at 75°C for 10 min. The following were then also added to the wells in a total volume of 50 l for each RT-PCR reaction (PerkinElmer Life Sciences): 5 l of 10ϫ TaqMan buffer A (composed of 500 mM KCl, 100 mM Tris-HCl, 0.1 M EDTA, 600 nM passive reference dye, pH 8.3, at room temperature), 10 l of 25 mM MgCl 2 , 1.5 l of each dNTP (10 mM), 0.5 l of forward and reverse primers (10 M), 1 l of appropriate TaqMan probe (5000 nM), 0.25 l of AmpliTaq Gold supplied at 5 units/l, and 0.25 l of Moloney murine leukemia virus. RT-PCR cycle parameters were 48°C for 30 min and 95°C for 15 min followed by 40 cycles at 95°C for 15 s and 59°C for 1 min. The primers and the probes used in the study were designed using Primer Express software (PerkinElmer Life Sciences) and synthesized by Life Technologies Inc. and Megabases. Clones with a threshold cycle (C T ) value of 30 or less were selected as positive for Pax3. Of the clone selected, the B9, E7, and H6 (16) were used in the study described in this paper because they showed the highest Pax3 expression. A parameter, threshold cycle (C T ), is defined for each PCR reaction that is the cycle number at which the reporter fluorescence generated by the cleavage of the sequence-specific probes passes above a fixed base line. This C T value has been shown to be a straight line when the log of the initial template is plotted versus C T (26).

RNA Isolation
Total RNA was isolated from parental (DAOY) cells, from pcDNA vector controls, and Pax3-transfected clones B9, E7, and H6 (16). When cells in T25 flasks were near confluence, total RNA was extracted using TRIzol (Life Technologies Inc.) following the manufacturer's instructions. After extraction, the RNA was treated with DNase I (Roche Molecular Biochemicals) at 1 unit/10 g of RNA, incubated at 25°C for 20 min, and heat-inactivated at 75°C for 10 min. The RNA was then cleaned using the RNeasy Mini columns (Qiagen) following the manufacture's instructions. Optical density was measured using a Beckman DU530 Life Science UV-visible spectrophotometer.
Oligonucleotide Microarray Analysis cDNA Synthesis-First strand synthesis was performed using Super-Script Choice system (Life Technologies Inc.). The manufacturer's protocol was slightly modified by Affymetrix including use of a high per-formance liquid chromatography-purified T7-(dT) 24 primer (Genset Corp.) and incubation at 42°C rather than 37°C. The final concentrations in the 20-l first-strand reaction are 100 pmol of T7-(dT) 24 primer, 10 g of total RNA, 1ϫ first-strand buffer, 10 mM dithiothreitol, 500 M of each dNTP, and 400 units of Superscript II reverse transcriptase. The final concentrations in a 150-l second-strand reaction are 1ϫ secondstrand reaction buffer, 200 M each dNTP, 10 units of DNA ligase, 40 units of DNA polymerase I, and 2 units of RNase H. A phenol/chloroform extraction using phase lock gels (Eppendorf-5 Prime, Inc.) and an ethanol precipitation are used to purify the double-stranded cDNA.
Synthesis of Biotin-labeled cRNA-Starting with 5 l of purified cDNA, an in vitro transcription reaction was performed using the Enzo BioArray high yield RNA transcript labeling kit (Affymetrix) according to the manufacture's instructions. The in vitro transcription product was purified in RNeasy spin columns (Qiagen) according to the manufacturer's protocol. This was followed by an overnight ethanol precipitation and resuspended in 15 l of diethyl pyrocarbonate-treated water (Ambion Inc). The cRNA was quantified using a Beckman DU530 Life Science UV-visible spectrophotometer. 20 g of cRNA was fragmented (0.5 g/l) according to Affymetrix instructions. The 5ϫ fragmentation buffer includes 200 mM Tris acetate, pH 8.1, 500 mM potassium acetate and 150 mM magnesium acetate. Individual reagents were received from Sigma.
Hybridization-Quantification of cRNA from total RNA was adjusted to reflect carryover of unlabeled total RNA with an equation given by Affymetrix: adjusted cRNA yield ϭ cRNA measured after in vitro transcription (g) Ϫ (starting amount of total RNA) (fraction of cDNA reaction used in in vitro transcription). 15 g of adjusted fragmented cRNA was added to a 300-l volume hybridization mixture that included final concentrations of 0.1 mg/ml herring sperm DNA (Promega/ Fisher), 0.5 mg/ml acetylated bovine serum albumin (Life Technologies Inc.), and 2ϫ MES hybridization buffer (Sigma). The mixture also contained the following hybridization controls: 50 pM oligonucleotide B2 (Genset Corp.) and 1.5, 5, 25, and 100 pM cRNA BioB, BioC, BioD, and Cre, respectively (ATCC). 200 l of this mixture was hybridized to the chips with 24-m x 24-m probe cells for 16 h according to Affymetrix procedures. The 50-m ϫ 50-m test chips are hybridized with 80 l of target for 16 h.
Data Analysis-Six data files (DAOY, PcDNA-1, PcDNA-2, B9, E7, and H6) were uploaded into Affymetrix MicroDB 1.0 software. This data base file was then sorted and studied with Affymetrix data mining tool 1.2. Using DAOY as a base line, genes showing fold changes between Ϫ1.5 and 1.5 in the vector controls were organized into a list. The data from the transfected cell lines were then queried using this gene list, an absolute call of "Present," and a difference call of "Increased" or "Decreased." The resulting list of genes was studied, and further information on genes of interest was compiled.

Hybridization of cDNA Probe to the CLONTECH Neurobiology Atlas cDNA Expression Arrays
Preparation of cDNA Probes-Preparation of cDNA probes was done with Moloney murine leukemia virus reverse transcriptase (Life Technologies, Inc.) in the presence of gene-specific primers (CLONTECH) and [␥-33 P]dATP according to manufacturer's protocol. For cDNA synthesis, we used equal amounts (3 g) of total RNA from control DAOY and B9 clone. The labeled cDNA was purified on ChromaSpin 200 diethyl pyrocarbonate-H 2 O columns (CLONTECH), and fractions 2-3 corresponding to cDNA were pooled and counted in a Beckman LS liquid scintillation spectrometer.
Hybridization-We use 20 ϫ 10 6 cpm of 33 P-labeled probes/membrane. After the test hybridization, the pre-hybridization and the hybridization of the Atlas human cDNA expression arrays-neurobiology (CLONTECH) was done at 68°C for 30 min and overnight, respectively. Washing of membranes was performed according to the manufacturer's protocols. The membranes were visualized on a Storm 860 PhosphorImager (Molecular Dynamics) after a 4-day exposure time. The screens were scanned and quantitated using the Storm 860 PhosphorImager.
Image and Image Analysis-ImageQuant Version 1.1 (Molecular Dynamics) was used to analyze the phosphorimage. Each gene spotted on the membrane was evaluated by the densitometry feature of this software. A grid was placed over each section of the membrane such that one box surrounded an individual, double-spotted gene. Densitometry analysis provided a volume intensity of each box. A volume of 900 was the lowest value visible by the naked eye. All genes exhibiting intensities less than 900 in both B9 and DAOY were eliminated. For each remaining gene, the B9/DAOY volume ratio was calculated. Ratios of 1 indicate equal volumes and, therefore, no change in gene expression between B9 and DAOY. Ratios below 1 indicate a greater volume in DAOY and down-regulation of gene expression in B9. Ratios above 1 indicate less volume in DAOY and up-regulation of gene expression in B9. From initial quantitative PCR results, it was determined that a ratio between 0.8120 and 1.1810 represents the lowest detectable difference, a 1-cycle difference in C T values. Due to detection limitations, all genes between 0.8120 and 1.1810 were considered to have equal expression and were excluded.

Real-time Quantitative RT-PCR
The 7700 sequence detector system chemistry is described in our previous paper (16). Each 50 l of RT-PCR reaction (PerkinElmer Life Sciences) volume included 25 ng of total RNA, 5 l of 10ϫ TaqMan buffer A (composed of 500 mM KCl, 100 mM Tris-HCl, 0.1 M EDTA, 600 nM passive reference dye, pH 8.3, at room temperature), 10 l of 25 mM MgCl 2 , 1.5 l of each dNTP (10 mM dATP, dCTP, dGTP, and 20 mM dUTP), 0.5 l of forward and reverse primers (10 M), 1 l of the corresponding TaqMan probe (5000 nM), 0.25 l of AmpliTaq Gold supplied at 5 units/l, and 0.25 l of murine leukemia virus reverse transcriptase. RT-PCR cycle parameters were 48°C for 30 min and 95°C for 15 min followed by 40 cycles at 95°C for 15 s and 59°C for 1 min. The primers and the probes used in the study were designed using Primer Express software (PerkinElmer Life Sciences) and synthesized by Life Technologies Inc. For murine Pax3, the forward and reverse primers were 5Ј-CCA ACC ATA TCC GCC ACA A-3Ј and 5Ј-TCT TAG AGA CGC AAC CAT GGG-3Ј respectively, and the TaqMan probe was 6 FAM-ATG GCA TTC GGC CTT GCG TCA TTT-TAMRA. For human ␤-actin, the forward and the reverse primers were 5Ј-TCA CCC ACA CTG TGC CCA TCT ACG A-3Ј and 5Ј-CAG CGG AAC CGC TCA TTG CCA ATG G-3Ј, respectively, and the TaqMan probe was 6 FAM-AT GCC CCC CCC ATG CCA TCC TGC GTT-TAMRA. For Rho GDI, the forward and the reverse primers were 5Ј-ACC CTG TCA CTC AAC GTG GTC-3Ј and 5Ј-TGG AAG ATC TGG CCC TGA TG-3Ј, respectively, and the TaqMan probe was 6 FAM-AAC AAG AGG CTT AAA ACC GGG CTT TCA CC-TAMRA. For PEA-15, the forward and the reverse primers were 5Ј-AAG GAA GAA GCC AGA CTG GTT AGA-3Ј and 5Ј-GGA GTC CTA GAG GTG TGT GTG TTA AG-3Ј, and the TaqMan probe was 6 FAM-CTG GCC ACT GCT GCA GAC ACC TG-TAMRA. For MyoD, the forward and the reverse primers were 5Ј-TAG GAG AGG CGG AGA ACT GAA G-3Ј and 5Ј-GAA GGG TGC TGC GTG GAA-3Ј, respectively, and the TaqMan probe was 6 FAM-AGG GCA AGG ACA CAG CGC GGT-TAMRA. For Tenascin, the forward and the reverse primers were 5Ј-GCA AAC GGG CAT AAA TTG GA-3Ј and 5Ј-GCT GGT TGT ATT GAT GCT TTG GTA-3Ј, respectively, and the TaqMan probe was 6 FAM-AGG AAT AAG GCG GCC CAG AGC GA-TAMRA. The primer and probe concentrations were optimized for MyoD, and the concentrations of reverse and forward primers used for MyoD were 300 and 900 nM, respectively. The MyoD probe concentration was 5000 nM.

SYBR Green RT-PCR
SYBR Green allows us to perform real-time RT-PCR using a 7700 sequence detector system (PerkinElmer Life Sciences). Measuring SYBR Green fluorescent dye that intercalates into the minor groove of the double-stranded DNA determines the amount of PCR product. Each 50 l of RT-PCR reaction (PerkinElmer Life Sciences) volume included 25 ng of total RNA, 5 l of 10ϫ SYBR Green PCR buffer (includes passive reference 1), 7 l of 25 mM MgCl 2 , 6 l of each dNTP (2.5 mM dATP, dCTP, dGTP, and 5 mM dUTP), 0.5 l of forward and reverse primers (5 M), 0.25 l of AmpliTaq Gold supplied at 5 units/l, and 0.25 l of murine leukemia virus reverse transcriptase. RT-PCR cycle parameters were 48°C for 30 min and 95°C for 15 min followed by 40 cycles at 95°C for 15 s and 59°C for 1 min. SYBR Green RT-PCR was performed with Versican V1, V2, and V3 splice variants. The forward and reverse primers for the Versican V1 splice variant was 5Ј-CCC AGT  GTG GAG GTG GTC TAC T-3Ј and 5Ј-CGC TCA AAT CAC TCA TTC  GAC GTT-3Ј, respectively, for the Versican V2 splice variant, the forward primer was 5Ј-GCA CAA AAT TTC ACC CTG ACA TT-3Ј, and the  reverse primer was 5Ј-TGC ATA CGT AGG AAG TTT CAG TAG GA-3Ј,  and for the Versican V3 splice variant, the forward primer was 5Ј-CCC  TCC CCC TGA TAG CAG AT-3Ј, and the reverse primer was 5Ј-GGC  ACG GGT TCA TTT TGC-3Ј.

Analysis of STX Promoter Activity
The plasmids pB01-NhN3.5 and pB01-SN0.45, containing STX promoter and luciferase reporter gene, and pPGBII, containing only the luciferase gene (37), were kindly provided by Dr. S. Tsuji (Institute of Physical and Chemical Research, RIKEN, Saitama, Japan). The DAOY cells, the pcDNA3 vector, and the Pax3-transfected stable cell line B9 were seeded at 5 ϫ 10 4 cells/60-mm-diameter dish in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum 24 h before transfection. The luciferase plasmids (5 g) used as the reporter, and Renilla luciferase plasmid (0.5 g) (Promega dual luciferase system), used as an insertional control for transfection efficiency, were transfected into the cells with DOTAP (N-[1-(2,3-dioleoyloxy)propyl]-N,N,N-trimethylammonium salts). After 42 h, the cells were washed three times with phosphate-buffered saline and then lysed with passive lysis buffer. The luciferase activity was measured using a Lumat LB 9501 luminometer (EG&G Berthold, Germany).

Preparation of Nuclear Extracts
The nuclear extracts from wild type DAOY and B9 Pax3 transfectant was prepared using the nuclear extraction kit (NE-PER nuclear and cytoplasmic extraction reagents from Pierce) as per the manufacturer's protocol.

GST/Pax3 Fusion Protein Production and Isolation
The GST/Pax3 fusion cDNA (25) was kindly provided by Dr. Mary R. Loeken (Harvard Medical School, Boston, MA). The Escherichia coli was transformed with GST/Pax3 fusion protein construct in pcDNA, and the cells were grown overnight in LB medium. The GST/Pax3 fusion protein production was induced with 0.1 mM isopropyl ␤-D-thiogalactopyranoside (Sigma) for 2 h. The cells were pelleted and sonicated to release the protein. The supernatant containing GST/Pax3 among other proteins was loaded onto glutathione-Sepharose 4B column (Amersham Pharmacia Biotech). The unbound proteins were washed with phosphate-buffered saline and eluted with 10 mM reduced glutathione in 50 mM Tris-HCl, pH 8.0. The eluted protein was concentrated using Centricon 10 (Amicon) as per the manufacturer's instruction.

Electromobility Shift Assays (EMSAs)
For EMSA experiments, 2 g of nuclear extracts from B9 or DAOY cells were incubated in a 15-l reaction mixture containing 15 mM Tris, pH 7.5, 6.5% glycerol, 90 mM KCl, 0.7 mM EDTA, 0.2 mM dithiothreitol, 1 mg/ml bovine serum albumin, 50 M pyrophosphate, 300 ng of sperm DNA as a competitor for nonspecific DNA binding, and 20,000 cpm of 32 P-labeled double-stranded oligonucleotide probes. Double-stranded probes covered the region Ϫ158 to Ϫ118 of the STX promoter (37) and designated as CSC1, 5Ј-TAGAGTTAGTGGGAGGAGACCAGGCAAGG-C-3Ј; CSC2, 5Ј-AGTGGGAGGAGACCAGGCAAGGCGCGGAGCA-3Ј; and CSC3, 5Ј-CAGGCAAGGCGCGGAGCAAACTGTCAAAC-3Ј. The suspected sequence of interaction with Pax3 is underlined. After a 20-min incubation at room temperature, free DNA and DNA-protein complexes were resolved in 4% (in case of nuclear extracts) and 8% (in case of GST/Pax3) polyacrylamide gels using 0.25ϫ TBE as running buffer. The electrophoresis was performed in cold buffer at 150 mV and 30 mA for 4 h. The gels were dried and subjected to PhosphorImager analysis to view the shifted bands.

RESULTS
Oligonucleotide Microarray-A human medulloblastoma cell line, DAOY, was stably transfected with murine Pax3 and pcDNA vector controls. Of the stable Pax3 transfectants, B9, E7, and H6 were chosen for the oligonucleotide microarray because they showed the same relative Pax3 expression (16). Biotin-labeled cRNA from DAOY cells, two vector-transfected controls (pcDNA-1 and pcDNA-2), and three Pax3 transfectants (B9, E7, and H6) were prepared, and microarray analysis was carried out as described under "Experimental Procedures."

Pax3 and Its Downstream Target Genes
After scanning, the six data files (DAOY, pcDNA-1, pcDNA-2, B9, E7, H6) were uploaded into Affymetrix MicroDB 1.0 software (Supplemental Table I). This data base file was then sorted and analyzed with Affymetrix Data Mining Tool 1.2. Using DAOY as a base line, genes showing fold changes greater than Ϫ1.5 and ϩ1.5 in the vector controls were excluded from further analysis. These represented genes that were possibly altered by the transfection and selection process. The remaining genes that varied by less than 1.5-fold were organized into a list and compared with the Pax3 transfectants. The expression of 1126 genes of a total of 7000 genes analyzed appeared to be altered by vector transfection (pcDNA-1 and pcDNA-2) and excluded from further analysis. The data from each of the three transfected cell lines (B9, E7, and H6) were then queried against the "unchanged" gene list. Genes that had an absolute call of Present and a difference call of Increased or Decreased in each of the transfected clones were selected. The resulting list of genes was examined and compared with known genes modulated by Pax3.
Upon analysis of this resulting list of genes that changed, it became apparent that the Present or Absent call had to be modified in our further calculations because known target genes such as MyoD and genes related to myogenesis were not identified. This appeared to be due to the method of "binning" to an absolute call of Present. For instance, MyoD and myogenic repressor 1 (MF1) were changed in the Pax3 transfectants but were missed because they were called Absent in DAOY or in transfectants. Thus, the genes that were Absent in DAOY were also included in subsequent analysis.
Based on a comparison with the known Pax3 targets, we selected candidates that were altered in all three (B9, E7, and H6) clones or, alternatively, in two of three clones. The modi- Pax3 and Its Downstream Target Genes fied criterion yielded a cluster of genes that included most of the known downstream targets of Pax3. In addition to the known downstream targets, we observed a total of 270 genes that were altered by Pax3 transfection. Of these 270 altered genes, 202 genes showed an increase, and 68 genes showed a decrease in their expression. More specifically, a direct comparison of the data sets for B9, E7, and H6 clones revealed that 38 genes displayed consistent changes (26 up-regulated and 12 down-regulated). In B9 and E7 clones, the expression of an additional 36 genes were altered, of which the expression of 27 genes were up-regulated, and 9 genes were down-regulated. E7 and H6 clones showed alterations in the expression of an additional 38 genes, of which 15 genes were up-regulated and 23 genes were down-regulated. Similarly, B9 and H6 showed changes in the expression of 56 genes, of which 38 were upregulated and 18 were down-regulated. Finally the expression of 102 genes was altered in any one of three clones. Of this, 96 genes were up-regulated, and 6 genes were down-regulated. The data and the relative expression levels of different selected genes in at least two of the three Pax3 transfectants as compared with DAOY controls are summarized and categorized according to their respective biological function(s) in Table I.
Validation of Microarray Results-To confirm whether the changes observed by Pax3 were authentic, two different independent methods were used. We performed a repeat hybridiza-tion experiment using a CLONTECH neurobiology cDNA expression array, which does not required any amplification step in contrast with the T7 amplification in the Affymetrix microarray protocol. Furthermore, real-time quantitative RT-PCR was used to corroborate both sets of array data.
cDNA Expression Arrays-The B9 clone was selected for CLONTECH neurobiology cDNA expression array analysis because it showed the highest Pax3 expression (16). Fig. 1 shows the autoradiographic image of 33 P-labeled cDNA obtained from control DAOY cells and the Pax3-transfected B9 clone. The spot intensities were quantitated using ImageQuant software (Molecular Dynamics) For each gene, DAOY was compared with B9, and the intensity ratio of B9/DAOY was calculated. Ratios of 1 indicate equal intensities and, therefore, no change in gene expression between the two cell lines. Ratios below 1 indicate down-regulation of gene expression in B9; on the other hand, ratios above 1 indicate an up-regulation of gene expression in B9 (Supplemental Table II). Using this criterion, we found that 494 genes of 588 genes were expressed both in DAOY and B9.
Of the 7000 genes on the Affymetrix chip and 588 genes on CLONTECH array, 107 genes were common to both. Of these 107 genes, 81 genes were eliminated from our Affymetrix data due to changes observed in the pcDNA3 vector transfection. This left 27 genes for direct comparison between the Affymetrix and CLONTECH arrays. Of the remaining 27 genes, 21 dis- Pax3 and Its Downstream Target Genes played the same relative expression change in both the Affymetrix and CLONTECH neurobiology arrays. Of the six that were not the same, three were at/or below our limits of detection. The three remaining, namely Versican, Tenascin, and protein-tyrosine phosphatase were called different.
Real-time Quantitative RT-PCR-In efforts to provide further experimental evidence that the gene changes observed on both arrays were valid, real-time quantitative RT-PCR was used to analyze a selection of representative genes. Throughout the real-time quantitative RT-PCR experiments, RNA was extracted from DAOY, vector, and the B9 clone only. First, we selected genes where there was a disagreement between Affymetrix and CLONTECH arrays; these included Versican and Tenascin. In addition, three genes (MyoD, RhoGDI, and Drebrin) that showed the same relative changes by "fold change" criteria (Affymetrix) and by "ratio" criteria (CLONTECH) were also analyzed by real-time quantitative RT-PCR (Fig. 2). A significant increase in the levels of MyoD and a decrease in RhoGDI and Drebrin E2 gene expression were observed. As shown in Table II, these data correlate with the values of B9 versus DAOY as determined by both Affymetrix and CLON-TECH methods. On the other hand, Tenascin was not changed in the Affymetrix arrays but was decreased on the CLON-TECH. In this case, real-time quantitative RT-PCR confirmed that there was a decrease. Interestingly, Versican was slightly increased in Affymetrix data and decreased in CLONTECH.
Real-time quantitative RT-PCR was attractive method for validation because it could detect different isoform changes that were not obvious in the Affymetrix chip or CLONTECH cDNA array experiments. For example, the Versican gene can be expressed in four known splice variants, V0, V1, V2, and V3, which share significant homology. The discrepancy between array data may be due to the sequences being detected. Realtime quantitative PCR made this distinction possible in that the expressions of splice variant V2 was increased 26.8-fold, and V3 was decreased more than 2-fold (B9/DAOY ratio of 0.4335), whereas V1 was not changed (Fig. 3 and Table II). In summary, there was very good agreement between data obtained from both arrays and real-time quantitative RT-PCR (Table II).
Comparison of Promoter Regions-After validation of the array data, the mechanisms by which the transcriptional regulation by Pax3 is brought about can now be explored. We hypothesized that there may be some similarity in the promoter regions of the altered genes. The promoter sequences for several known genes altered by Pax3 or PAX3-FKHR were examined for potential Pax3 binding motifs. The Pax3 downstream targets are MITF (23) The rectangles around the spots are representative genes that were changed significantly in B9 compared with DAOY. The white rectangle, corresponding to the coordinate D2e, is RhoGDI (GenBank TM accession number L20688); the black rectangle, corresponding to the coordinate E7f, is Drebrin E2 (Gen-Bank TM accession number U00802); the orange rectangle, corresponding to the coordinate F4a, is Tenascin (GenBank TM accession number X78565); the yellow rectangle, corresponding to the coordinate F2k, is Versican (GenBank TM accession number X15998). The B9/DAOY ratio data of the spot intensities for each of these individual spots were calculated and used for further analysis.

FIG. 2. Validation of representative Pax3 downstream target genes in Pax3 transfected clone B9 (green) and parental DAOY cells (red) by real-time quantitative RT-PCR. The expression of human MyoD (top panel), RhoGDI (middle panel)
, and Drebrin (lower panel) was analyzed by RT-PCR reaction using the 7700 sequence detection system (PerkinElmer Life Sciences) as described under "Experimental Procedures." The plot shows the change in the fluorescence intensity (⌬Rn) at each PCR cycle. ⌬Rn is the normalized reporter signal corrected for initial reporter signal. To calculate the ⌬Rn, the initial reporter signal is subtracted from the normalized reporter signal at each PCR cycle (26). A cycle threshold (C T ) is defined as the fractional cycle number at which the reporter fluorescence passes a fixed threshold above base line. Samples with higher message expression will have a lower C T value. The cycle threshold was set at 0.06 ⌬Rn. The cycle at which human ␤-actin fluorescence crosses the threshold is 22.05. The  (35). Despite numerous Pax3 targets described in the literature, a single consensus binding motif has not emerged (25). We therefore, scanned for known paired domain and homeobox domain binding motifs in the promoter region for those genes. Commonly described motifs for paired and homeodomain binding are ATTA(N)GTTCC or GTTAT. Similarly the paired domain recognition sites GTTCC, GTTAT, GT-TAC or the respective inverted sequences are reported Pax3 binding sequences (25,41). An additional Pax3 binding site, GTGTGA, has been described for the melanocyte-specific tyrosinase-related protein-1 promoter (30). This consensus sequence is also present in MITF (23) and c-RET (29). The presence of the complimentary sequences TAAT or CAAGG was observed in the STX. By searching for these Pax3 binding motifs, we assigned a score based on the number of times a Pax3 binding motif was present in the promoter region of a gene. Thus we came up with a score of 16, 40, 12, 23, and 9 for MyoD, MITF, NCAM, c-RET, and Tyrp1, respectively (Table  IIIA). The same criteria was applied to genes implied as Pax3 downstream targets but lacking a characterized promoter sequence for Pax3 binding. These genes included Versican (36), STX (37), Dep-1 (38), MBP (39), MARCKS (40), and PDGF-␣ receptor (33), which obtained a binding "motif score" of 7, 11, 4, 5, 1 and 17, respectively (Table IIIB). These criteria of scanning the promoter region for putative Pax3 binding motifs suggested that a score of 9 or 10 and above may have the potential for Pax3 binding and, therefore, could be a direct Pax3 downstream target. The list of changed genes with sequenced promoter regions was scanned for these putative Pax3 binding motifs. Of 270 genes that were changed in our Affymetrix analysis, there were 91 genes that had published promoter sequences. Of these 91 published promoter sequences, 71 gene promoters had potential Pax3 binding motifs. Of these 71 gene promoters, 17 of these published promoter sequences had a motif score of 10 or above (Table IIIC), 40 were between 3 and 9, and the remaining 21 promoter sequences had motif score of less than 3. Unfortunately, this list excluded those genes where the promoter sequence is either not known or not published. Therefore, in order not to miss out on some important unknown Pax3 downstream targets, we made a separate list of all the genes that showed a fold change of approximately Ϯ5-fold or greater. We found an additional 17 genes in this list that could be putative Pax3 targets (Table IV). As a control, genes that were not altered by Pax3 transfection, ␤-actin, pyruvate L, Recoverin, and delayed rectifier potassium channel, were also analyzed for Pax3 binding motifs and did not contain these features (Table IIID).
Identification of STX as a Pax3 Downstream Target-Based on microarray results and similarities in the promoter region, we identified 17 previously unidentified potential Pax3 downstream targets. Because there is no clear consensus binding motif for Pax3 (25), we wanted to demonstrate that our selection criteria could identify downstream targets. One candidate gene, STX, was selected to validate the selection criteria. This gene was chosen because the increased expression had been previously shown to alter the phenotype and aggregation properties of the DAOY cells (16). In addition, the motif score for STX was 11 and near the lower cutoff score of our putative Pax3 downstream targets. We reasoned that the authenticity of our prediction as putative Pax3 downstream target based on the motif score was best tested with a motif score that was toward the lower end of our cutoff score.
Analysis of Promoter Activity-To determine whether Pax3 could influence STX promoter activity, we used luciferase reporter plasmidspB01-NhN3.5and pB01-SN0.45 containing the STX promoter sequences (37). The promoterless luciferase gene in pPGBII was used as a control (37). The constructs were assayed for promoter activity by transient transfection into stable Pax3 transfectant B9, stable pcDNA3 vector transfected cells, and wild type DAOY cells. Fig. 4A shows the activation of luciferase expression by reporter plasmids containing putative Pax3 binding sites. The luciferase activity due to each reporter plasmid was normalized as to the Renilla luciferase activity by cotransfecting an internal control plasmid pRL-null carrying cDNA encoding Renilla luciferase gene. The pB01-SN0.45 construct showed the highest level of promoter activity. Both pB01-NhN3.5 and pPGBII control vector were essentially inactive. These data were consistent with the previous results of Yoshida et al. (37), indicating the pB01-SN0.45 construct to be minimal essential promoter for STX. This construct showed 10 -12-fold increases in activity over the pB01-NhN3.5 and pPGBII control constructs. These data suggested that the minimal essential region of the STX promoter contained elements that are responsive to Pax3 binding. Identification of a Pax3 Binding Sequence-The STX minimal essential promoter region did not appear to contain known consensus Pax3 binding motifs. We used EMSA with both nuclear extracts and purified GST/Pax3 fusion protein to identify sequences that may bind Pax3. Ten double-stranded oligonucleotides, 30 base pairs each, were synthesized. These covered the region from Ϫ298 to Ϫ118 of the STX promoter and overlapped each other by 10 base pairs at the 3Ј end. Of the 10 oligonucleotides tested, only two showed specific binding that was increased in B9 compared with DAOY (Fig. 4B). These were CSC2 and CSC3 covering the sequence between Ϫ151 to Ϫ118 (complete sequences of these oligonucleotides can be found under "Experimental Procedures"). Common to both of these oligonucleotides was the CAAGG motif suggested by our selection criteria described above and previously suggested as a Pax3 binding sequence (25,41). An additional oligonucleotide, CSC1, also contained the CAAGG motif but did not exhibit increased binding in the Pax3 transfectant B9 (Fig. 4B). When the CAAGG motif was in the middle of the oligonucleotide (CSC2), the nuclear extract from B9 bound with the greatest intensity.
The increased binding of CSC2 and CSC3 to B9 nuclear extracts suggested that increased Pax3 might be responsible for the increased binding. A purified GST/Pax3 fusion protein was used to determine whether CSC2 and CSC3 could interact directly with Pax3. EMSA studies using the GST/Pax3 fusion protein (Fig. 4C) indicated that Pax3 could bind directly to CSC2 oligonucleotide, which had the CAAGG motif in the middle, and not to CSC1 or CSC3, which had CAAGG motif at either the 5Ј end or 3Ј end. Thus, our studies indicate that Pax3 could bind directly to a sequence in the STX promoter and suggest that STX may be a previously unidentified biological downstream target of Pax3.

DISCUSSION
Using oligonucleotide arrays and RNA from stable Pax3transfected DAOY cells to identify potential Pax3 downstream targets, a total of 270 genes were altered by Pax3 transfection. Of these 270 altered genes, 202 genes showed an increase, whereas 68 genes showed a decrease in their expression. Before making any definitive conclusions about these changes, we determined how reproducible and representative these changes were. We employed two independent techniques to validate our Affymetrix data, namely the CLONTECH neurobiology cDNA expression array and real-time quantitative RT-PCR. The CLONTECH array was chosen to validate the Affymetrix results because it provided a repeat hybridization of 107 genes and did not require an amplification step. The consistency observed between the two types of arrays indicated that changes were reproducible and that the in vitro transcription step did not skew the message representation. Furthermore the sensitivity of real-time quantitative RT-PCR helped define the limits of significant changes and resolved any apparent inconsistencies in the array data. Real-time quantitative RT-PCR and cDNA expression array analysis confirmed the changes of RhoGDI, MyoD, Drebrin, Tenascin, and glutathione S-transferase. On the other hand, certain gene changes observed in Affymetrix were inconsistent with those observed with the CLONTECH array. Notably, an increase in astrocytic phosphoprotein PEA-15 was observed with the CLONTECH array, whereas the Affymetrix microarray gave us a No Change call. Real-time quantitative RT-PCR was instrumental in resolving this discrepancy, indicating that a relatively small magnitude of changes like those for PEA-15 is below the cut-off sensitivity of the array method used. Similarly, real-time quantitative RT-PCR resolved the discrepancy between the Affymetrix and CLONTECH data for Versican. An increase in the expression of Versican (V0), a common region in all the splice variants (V1, V2, and V3) observed in Affymetrix microarray, was confirmed to be due to the splice variant V2. This result suggested that the observed changes in Versican gene expression probably arose from the oligonucleotide sequences selected by Affymetrix.
Quantitative real-time RT-PCR validated the gene expression changes observed by the array analysis, but it remains to be seen whether these genes are direct downstream targets of Pax3. Despite numerous downstream targets identified in the literature, a consensus sequence for Pax3 binding motif does not exist (25). Analysis of known Pax3 binding motifs did, however, reveal similarities in their promoter regions. Of the genes that displayed significant alterations in expression, 17 previously undescribed downstream targets of Pax3 had promoter regions with similarity to known Pax3 targets. In addi-tion, another 17 genes showed reproducibly large changes, but their promoter sequences are unknown. These genes may also be Pax3 downstream targets.
Because there is no clear consensus binding motif for Pax3 (25), we used promoter-luciferase reporter transfection assays and EMSAs to demonstrate that our selection criteria could identify downstream targets. We chose a candidate gene STX to validate the selection criteria, because the increased expression of STX had been previously shown to alter the phenotype and aggregation properties of the DAOY cells (16). In addition, the motif score for STX was 11 and near the lower cutoff score of our putative Pax3 downstream targets. We reasoned that the authenticity of our selection criteria for predicting a gene as a potential Pax3 downstream target based on the motif score can be best tested with a motif score that was toward the lower end of our cutoff score. The demonstration that STX promoter activity was increased by Pax3 and the specific binding of Pax3 to STX promoter clearly suggests that the selection criteria we used for identifying downstream targets were appropriate and could be extended to identify common sequences in promoter regions of potential downstream targets of transcription factors. The Pax3 binding motif in the STX promoter appeared to be CAAGG sequence that is complimentary to the GTTCC motif previously identified as a paired domain recognition site (25,41). Although Pax3 is known to bind to GTTCC on the sense strand, it appears that Pax3 may also bind when the a Score is the number of times a paired or homeobox binding element appears in a promoter region. Paired and homeobox binding elements are described under "Experimental Procedures" and "Results." b Identified as a Pax3-FKHR fusion protein downstream target.

Pax3 and Its Downstream Target Genes
complimentary sequence, CAAGG, is present on the sense strand. Thus we showed at least one previously undescribed downstream target, STX, to be a biological downstream target of Pax3.
A limitation of the methodology used in this application is that the gene expression changes observed are dependent on many factors such as clonal selection and differential hybridization. In an effort to minimize the differences resulting from these factors, we used three Pax3-transfected clones (16) and performed repeat hybridizations. An additional possibility that may affect the expression of downstream target genes is the levels and the availability of Pax3. Chalepakis et al. (8) show that the transcriptional regulation by Pax3 is concentrationdependent and biphasic. Furthermore, Phelan and Loeken (25) demonstrate that low levels of Pax3 in the nucleus activated only the high affinity responsive promoters. Conversely, at concentrations above the minimum threshold, only the low affinity responsive promoters may show transcriptional activation. Although we do not know if Pax3 is above or below a threshold in the Pax3 transfectants used in our study, we selected these three clones because they showed the same relative levels of Pax3 expression.
It has been suggested that when Pax3 interacts with other proteins through the paired domain, it behaves more like a repressor. Examples of such Pax3-binding proteins are Msx1 (11), Ets (43), and hDaxx (44). On the other hand, if Pax3 interacts with proteins that leave the integrity of its homeodomain and paired domains intact, the resulting complex is a stronger transcriptional activator. An example of such strong transcriptional activator is a PAX3-FKHR fusion protein. Khan et al. (45,46) use microarrays to identify downstream targets of a fusion protein, PAX3-FKHR. A comparison of our data with that of these investigators showed an overlap in some of the genes that may be downstream targets of Pax3. An interesting example is the PDGF-␣ receptor (47). Although it is implied to be a Pax3 downstream target, promoter studies show that Pax3 by itself cannot mediate transcriptional activation of the PDGF-␣ receptor promoter, whereas PAX3-FKHR can (47). This observation underscores the importance of Pax3 binding partners involved in transcriptional activation by Pax3. In addition to PDGF-␣ receptor, Khan et al. (45,46) observe a significant increase in IGFBP5, ATF3, and PCNA expression levels in their microarray analysis, but we did not see any change in their expression levels by Pax3 alone. Even though these genes have high Pax3 binding motif score in their promoters, it appears that other factors in addition to Pax3 are essential in eliciting transcriptional activation.
Another way Pax3 may regulate downstream target genes is indirectly through regulation of transcription factor genes, which in turn regulate its downstream target genes. Among the genes identified as changed by Affymetrix analysis, 40 genes with known promoter sequences had a motif score of 9 or less. This implied that their promoter sequences were not that similar to known Pax3 targets and may be regulated by a transcription factor that may be a Pax3 target. c-MET was observed to be up-regulated by Pax3 in our Affymetrix analysis, although its Pax3 binding score was 2. It is very interesting to note that the c-MET promoter has a MyoD binding site. It is therefore possible that Pax3, which up-regulates MyoD (Pax3 binding score of 16), can up-regulate c-MET. In this respect c-MET is an indirect downstream target of Pax3 (Pax3 3 MyoD 3 c-MET).
If we view our data from the perspective of early embryonic development regardless of the combinatorial nature of Pax3 activity (48), its biological effect is to ensure that neural crest cells do not execute a differentiation program that is inappropriate for their dorsoventral position in the neural tube. It is interesting to note that certain sets of genes that that delay differentiation, namely Id-2, Id-4, and myogenic repressor-1, show an increase in expression in our Affymetrix analysis. Similarly several genes that are involved in cell migration show changes in expression in the Pax3 transfectants. These include: Versican, Tenascin, RhoGDI, Drebrin E, (HSPG2), type IV collagenase, and tissue inhibitor of matrix metalloproteinase-3 (TIMP-3). Up-regulation of specific genes and the delay in others is expected of a cell that is committed to migration (15). These events prepare the extracellular matrix as well as the intracellular cytoskeletal elements for efficient migration. Our previous study (16) showed that overexpression of Pax3 up-regulated STX, which resulted in an increase in the polysialylation of NCAM (PSA-NCAM). PSA-NCAM prefers heparin sulfate proteoglycan, which is migration-permissive, whereas Versican, a large chondroitin sulfate proteoglycan (CSPG), is a migration non-permissive substrate and promotes NCAM-NCAM-mediated homophilic adhesion (16, 49 -51). Migration of neural crest cells during early embryonic development would be facilitated if non-permissive substrates were down-regulated. Our data support this hypothesis. We observed that V2 splice variant of Versican is up-regulated and V3 splice variant is down-regulated. Henderson 45 and pB01-NhN3.5 and pPGBII (promoterless luciferase gene) (5 g) were transfected into wild type DAOY, pcDNA3 vector transfectants, and Pax3 transfectant B9 and assayed for luciferase activity 42 h after transfection as described under "Experimental Procedures." B, EMSAs of binding reactions performed with nuclear extracts from DAOY or Pax3 transfectant B9 and 32 P-labeled double-stranded oligonucleotides in the absence or presence of a 50-fold molar excess of unlabeled oligonucleotides CSC1, CSC2, or CSC3 as described under "Experimental Procedures." C, EMSAs of binding reactions performed with GST/Pax3 fusion protein and 32 P-labeled double-stranded oligonucleotides in the absence or presence of a 50-fold molar excess of unlabeled oligonucleotides CSC1, CSC2, and CSC3 as described under "Experimental Procedures." RLU, relative luminescence units.
serve an overexpression of Versican in Splotch mouse and suggested that Pax3 may serve to negatively regulate Versican expression that is associated with defective neural crest migration in Pax3 mutant mouse (52). It is therefore possible to speculate that it could be the V3 Versican splice variant that is affected in Splotch mouse.
Migration involves not only cell surface interaction with extracellular matrix components but also of cell intracellular cytoskeletal elements that prepares the cell for migration. Rho-GDI is one such gene that plays a critical role in the cytoskeletaldependent cell functions (53). It is required for transcription of muscle-specific gene, myogenin through MEF-2. Down-regulation or inhibition of RhoGDI function suppresses myogenesis (42). A significant down-regulation of RhoGDI and an up-regulation of Id-2, Id-4 and myogenic repressor-1 (MF-1) in Pax3 transfectants supports the role of Pax3 in migration and delaying differentiation until the cells reach their destination for differentiation to take place. Our previous results and these observations suggest that Pax3 not only modifies NCAM through polysialylation but also regulates extracellular matrix and intracellular cytoskeletal genes.
Thus in summary, we report several Pax3 downstream targets using Affymetrix gene chip analysis. To confirm the observed gene changes as representative of the actual changes in the cell, we validated our Affymetrix data and compared it with a CLONTECH array data and found that there was a good agreement between the two sets of data. Validation of the criteria used to establish stringency conditions to arrive at a given result in both the Affymetrix as well as CLONTECH array data was established by performing real-time quantitative RT-PCR. Based on the Pax3 binding motif score in the promoter regions of the genes that were changed by Pax3 transfection, we postulate that there may be at least 17 new genes that may be considered as direct Pax3 downstream targets and at least another 17 genes (whose promoter sequences are not known) that may also be putative Pax3 downstream targets.