In GE-HTS, a gene expression signature is used as a surrogate of a biological state. In the present context, we sought to define a signature of ERK activation mediated by PDGFR stimulation. Specifically, we treated SH-SY5Y neuroblastoma cells with the BB homodimer of PDGF (PDGF-BB), which resulted in PDGFRβ phosphorylation and subsequent ERK activation. We selected PDGFRβ over PDGFRα for our studies because of previous observations that PDGFRα might mediate functions of other PDGF isoforms in addition to PDGF-A 78. The activation state of the members of the PDGFβ pathway can be traced by increase in their phosphorylation levels shortly after introduction of the growth factor 9. In particular, ERK phosphorylation peaks at about 15-20 minutes after induction, and then decreases to background levels some 20-30 minutes later 10. Accordingly, we performed gene expression profiling using Affymetrix U133A arrays 30 minutes following PDGF stimulation, thereby identifying those genes whose expression is correlated with PDGFR activity. In order to identify the component of the gene expression signature that was attributable to ERK activation by PDGFR (as opposed to other pathways downstream of PDGFR), we also pretreated the cells with the MEK inhibitor U0126 and the ERK inhibitor apigenin, and repeated the gene expression profiling studies (Figure 1a).

To define the signature of ERK activation, we developed and applied a rank-pairwise comparison algorithm as described in Materials and methods. We note that the genes identified in this manner are chosen because of their ability to reflect the PDGF-stimulated state - not because of their necessarily being critical effectors of PDGFR signaling. The top three genes identified in this fashion were those for c-fos, early growth response 1 (EGR1), and activity-regulated cytoskeleton-associated protein (ARC). All three genes were previously shown to be upregulated by activation of ERK, and we further confirmed their regulation by reverse transcriptase (RT)-PCR (Figure 1b) 111213. Two additional genes, ribosomal protein RPL23A and ATP5B, were selected as internal controls, because their expression was not significantly altered by PDGFR activation.

Having defined a gene expression signature of PDGFR/ERK activation, we next sought to screen a library of small molecules to find compounds that would reverse the signature (for primary screen data, see Additional data file 1). We chose TIP5 fibroblast cells for the high-throughput screen instead of SH-SY5Y neuroblastoma cells used to define the gene expression signature. Both TIP5 and SH-SY5Y cells have wild-type PDGFR/ERK signaling, which makes it unnecessary to employ mutant and/or constitutively activated PDGFR cascades. TIP5 cells, however, were more adherent to 384-well plates, making them more amenable to the screening setting.

The screen was performed as follows. TIP5 cells were plated in 384-well plates, serum-starved overnight and compounds then added by pin transfer. The compound library, previously described in 2, consisted of 1,739 chemicals with previously established biological functions. Some of the compounds have been approved for use in humans by the Food and Drug Administration. After a 30 minute compound-incubation period, PDGF-BB was added. 45 minutes later, the growth medium was discarded, and cells were lysed. RNA was then extracted, the signature genes amplified by RT-PCR, and the PCR amplicons quantified by single-base extension mass spectrometry, as we previously described 2 (Figure 1c). Cells were treated in triplicate at two concentrations (approximately 10 μM and 50 μM). Compounds were defined as hits if the expression of two marker genes, c-fos and EGR1, normalized by expression of control genes was significantly (more than one standard deviation) lower than average expression in all positive control wells. Compounds that inhibited the signature of the activated PDGFR/ERK pathway in four out of six replicas were selected as hits for further characterization.

Three wells met the hit selection criteria: aurintricarboxylic acid (ATA; free acid), aurintricarboxylic acid triammonium salt (aluminon), and quinacrine dihydrochloride (mepacrine) (Figure 2a,b); all three were therefore selected for further studies. Western analysis of total lysates from cells treated with these compounds demonstrated that both ATA and its salt (which in solution is identical to ATA), but not quinacrine dihydrochloride, abrogated PDGF-mediated phosphorylation of ERK (Figure 3a), thereby identifying ATA as an inhibitor of the ERK pathway. Quinacrine dihydrochloride did not inhibit ERK phosphorylation, but it has been previously shown to be a non-specific inhibitor of phospholipase A2 14. Activated ERK phosphorylates phospholipase A2 15, and as a result stimulates transcription of the c-fos and EGR1 genes, two components of our ERK signature 16.

We then relaxed hit selection criteria, and identified nine more potential candidates. However, further study indicated that none of these nine additional compounds affected activation of the PDGFR/ERK pathway.

Disruption of phosphorylation of ERK by ATA was an indication that ATA inhibited the PDGFR/ERK pathway upstream of ERK. Subsequent analysis indicated that phosphorylation of both MEK (Figure 3b) and PDGFR (Figure 3c) was abrogated by ATA, thus pointing to PDGFR as a possible ATA target.

To address the possibility that ATA might in some fashion deplete PDGF ligand from the growth medium, TIP5 cells were first incubated with ATA for 30 minutes. Next, the cells were washed thrice with serum-free medium and then stimulated with PDGF. As shown in Figure 3d, PDGFR phosphorylation remained inhibited, suggesting that PDGF ligand was unlikely to be the target of ATA.

The experiments described so far indicated that ATA inhibits PDGF-mediated ERK phosphorylation by inhibiting PDGFR phosphorylation. To localize the portion of PDGFR targeted by ATA, we utilized a series of chimeric receptor constructs (Figure 4a). The first chimera, TEL/PDGFR, is a naturally occurring, leukemia-associated fusion of the oligomerization domain of the transcription factor TEL (ETV6) to the transmembrane and cytoplasmic domains of PDGFR, resulting in constitutive activation of PDGFR 17. As shown in Figure 4b, ATA was unable to inhibit TEL/PDGFR phosphorylation at concentrations as high as 100 μM, indicating that ATA does not target the transmembrane or cytoplasmic portions of PDGFR present in the TEL/PDGFR chimera.

The next chimera, termed PER, is composed of the extracellular domain of PDGFR and the transmembrane and cytoplasmic domains of epidermal growth factor receptor (EGFR) 18. ATA inhibited PER phosphorylation in PER-PC12 cells (Figure 4c), thus mapping the site of ATA action to the extracellular domain of PDGFR. To exclude the possibility of ATA inhibiting any receptor tyrosine kinase extracellular domain, we tested ATA against a third chimera, EKR, consisting of the extracellular domain of EGFR and the transmembrane and cytoplasmic domains of c-KIT 19. ATA failed to inhibit EKR (Figure 5a), indicating that ATA exhibits some specificity for the PDGFR extracellular domain. Similarly, ATA failed to inhibit insulin-like growth factor (IGF)-induced phosphorylation of IGF1 receptor (IGF1R; Figure 5b), or EGF-induced phosphorylation of EGFR (Figure 5c) 20. Interestingly, ATA did inhibit stem cell factor (SCF)-mediated activation of cKIT (Figure 5d). The cKIT and PDGFR extracellular domains have 41% sequence similarity (26% identity), whereas no significant homology is seen between the extracellular domains of PDGFR and EGFR or IGF1R.

We note that whereas phosphorylation of the PER chimera is PDGF-dependent (and ATA inhibitable) in PER-PC12 cells, PER is constitutively active in 501 MEL and MCF7 cells, and in those contexts PER phosphorylation is not fully abrogated by ATA (Figure 6a,b). These experiments further point to the possibility of ATA inhibiting PDGF binding to the extracellular domain of PDGFR and disrupting ligand-mediated activation of the receptor.

In order to characterize the features of the ATA molecule required for biological activity, we analyzed a diverse set of ATA structural analogs (Figure S1 in Additional data file 2) available from the Available Chemicals Directory 21. We split compounds into three groups to test three different hypotheses on the structure-activity relationship in the series. The activities of methylenedisalicylic acid, salicylic acid and 3-methylsalicylic acid (Figure S1a in Additional data file 2) were analyzed to examine if the skeletal-triphenylmethane structure of ATA was essential to its activity. Aurin, uranine and phenolphthalein sodium salt (Figure S1b in Additional data file 2) were tested to evaluate the roles the carboxyl and hydroxyl groups on the triphenylmethane scaffold play in the inhibitory potency of ATA. Compounds in the third group (Figure S1c in Additional data file 2) were evaluated to test the effect of various modifications of the phenyl rings on the inhibitory properties of ATA. No compounds in the series inhibited PDGFR at concentrations sufficient for ATA inhibition (less than 5 μM). In the first group, methylenedisalicylic acid (Figure 7a), but not methylsalicylic or salicylic acids inhibited PDGFR phosphorylation at 50 μM, suggesting that increasing the number of substituted salicylic acid moieties from one to three boosts the inhibitory potency of ATA. The positions and number of carboxyl and hydroxyl groups were essential for PDGFR inhibition, as indicated by the fact that no compounds in the second group inhibited PDGFR at 100 μM concentration. These results corroborate earlier reports that both the aurin triphenyl methane ring system and the carboxylic acid groups are necessary for ATA inhibitory properties 22.

In the third group, Basic Violet 3, Ethyl Violet and Victoria Pure Blue BO inhibited PDGFR in the 5-10 μM range (Figure 7b-d). Interestingly, these three compounds exhibited less specific patterns of receptor inhibition than ATA, inhibiting not only cKIT, but also EGFR and IGF1R at 10-100 μM (Figure 8). Moreover, different from ATA, Ethyl Violet and Victoria Pure Blue BO readily translocated across the cell membrane, as indicated by their inhibition of cytoplasmic TEL/PDGFR in Ba/F3 cells at 10 μM (Figure 9). Taken together, these results suggest that the inhibitory mechanism of Basic Violet 3, Ethyl Violet and Victoria Pure Blue BO is different from the extracellular receptor inhibition mechanism of ATA.

The EKR construct 19 was kindly provided by Dr. Ullrich, Department of Molecular Biology, Max-Planck-Institut fur Biochemie. PER chimera 18 was a gift of Dr. Tyson and Dr. Bradshaw, Department of Physiology and Biophysics, University of California, Irvine.

Chemical compounds apigenin, U0126, quinacrine dihydrochloride and ATA were obtained from Calbiochem 50; Methyl Violet B base, Rhodamine 6 G tetrafluoroborate, sulforhodamine, Ethyl Violet, Victoria Pure Blue BO, Rhodamine B, Lissamine Green B, Methyl Violet 2B, Rhodamine 6G, (L-Asp)2Rhodamine 110 TFA, Rhodamine 110 chloride, Eosin B, Rhodamine 123 hydrate, Rhodamine 19 perchlorate, Acid Fuchsin calcium salt, p-Rosolic acid, Basic Violet2, Gentian Violet, pararosaniline hydrochloride, and salicylic acid were purchased from Sigma 51; and 3-methylsalicylic acid, 5,5'-methylenedisalicylic acid, phenolphthalein sodium salt, and Uranine K were obtained from ABCR 52.

Growth factors PDGF, EGF, and SCF were obtained from Cell Signaling 53, R³IGF from Sigma, and interleukin (IL)3 from R@D Systems 54.

Cell culture reagents RPMI 1640, Dulbecco's modified Eagle's medium (DMEM), and HAM's F-10 were purchased from Mediatech 55, penicillin and streptomycin from Invitrogen 56, and fetal bovine serum from Sigma. p44/42 MAP kinase, phospho-p44/42 MAP kinase (Thr202/Tyr204), MEK1/2, phospho-MEK1/2 (Ser217/221), PDGF-BB, phospho-PDGFRβ (Tyr751), phospho-EGFR (Tyr1068), cKIT, Phospho-cKIT (Tyr719), IGF-IαR, and Phospho-IGF-IR (Tyr1131)/insulin receptor (Tyr1146) antibodies were obtained from Cell Signaling. EGFR and mouse cKIT antibodies were purchased from Santa Cruz Biotechnology 57. Alfa-tubulin antibody was obtained from Sigma.

SH-SY5Y neuroblastoma cells were purchased from American Type Culture Collection 58. The IL3-dependent pro-B lymphoid cell line Ba/F3 and Ba/F3 cells expressing TEL/PDGFRβ 1759 were obtained from Dr. Gary Gilliland. TIP5 primary fibroblasts 60 were a gift from Dr. Stephen Lessnick. We thank Dr. Ruth Halaban for 501 MEL human melanoma cells. PER-expressing PC12 cells were generously provided by Dr. Darren Tyson. SH-SY5Y, PC12, TIP5 and MCF7 cells were cultured in DMEM, BaF3 cells and BaF3 cells expressing TEL/PDGFRβ were maintained in RPMI 1640 medium, and 501 MEL cells were grown in Ham's 10 medium. Medium for IL3-dependent Baf3 cells was supplemented with 0.05 ng/ml IL3. Media for all cell lines except PC12 contained 10% fetal bovine serum, 10 U/ml penicillin, and 10 μg/ml streptomycin. PC12 cells were grown in DMEM with 15% horse serum, 5% fetal bovine serum, 10 U/ml penicillin, and 10 μg/ml streptomycin. All cells were grown at 37°C in 5% CO₂.

SH-SY5Y cells were grown to confluence and starved overnight in serum-free medium in order to silence any sustained effects from growth factor signaling. Prior to induction with 50 ng/ml PDGF, cells were treated with pathway inhibitors 74 μM apigenin or 50 μM U0126, or with dimethyl sulfoxide (DMSO) as carrier for 60 minutes. Total RNA was isolated 30 minutes after PDGF addition. Experiments were performed in duplicate. The RNA was processed and hybridized with Affymetrix U133A GeneChips as described in 61.

To define the ERK/PDGFR activation signature, a pair-ranking algorithm was used. Genes were ranked according to their raw expression values on each chip. Ten genes with maximum change in ranking were selected for each one of three pairs of conditions: cells with PDGF versus cells without PDGF, cells with PDGF versus cells with PDGF and apigenin, and cells with PDGF versus cells with PDGF and U0126. Three genes common to all three conditions were selected as a signature of the activated ERK/PDGFR pathway. The signature was then trimmed from three to two genes based on their relative strength of expression in TIP5 cells.

TIP5 cells were grown to confluence and starved overnight with 20 μl serum-free medium per well of 384-well plates. We added 20 μl of compounds diluted in media so that the final concentration of compounds would be approximately 10 μM in three out of six replicas, and 50 μM in three remaining replicas. Media containing carrier (DMSO) was added to control wells instead of compounds. The compound library was composed of 1,739 chemicals either approved for use in humans by the Food and Drug Administration or extensively biologically characterized 262 (the full list of tested compounds is available in these publications). After 30 minutes of compound treatment, cells were induced with 40 μl of PDGF diluted in media (final PDGF concentration 40 ng/ml). PDGF was added to half of control wells to measure PDGF response; only media was added to the remaining control wells. After 40 minutes of PDGF induction, media was discarded, cells were lysed and RNA was extracted and quantified as described in 2.

Briefly, 15 μl of lysis solution containing a hypotonic detergent, dithiothreitol and RNAse inhibitor were added to medium-free cells for 15 minutes. The lysates were transferred to a 384-well oligo-dT-coated plate and incubated with 6 μl of 2.5× binding buffer. After 30 minutes of incubation lysates were discarded and reverse transcription was carried out in a 5 μl Moloney murine leukemia virus (M-MuLV) RT reaction at 37°C for 2 h.

After incubation, the RT mixture was discarded and multiplex PCR was carried in a 5 μl volume. The resulting mixture was treated with shrimp alkaline phosphatase and the single-base extension reactions were carried out in 9 μl reaction volumes with 1× Thermosequenase buffer, 2.7 μM of each primer, 0.2 mM of each ddNTP and 0.58 units per reaction of Thermosequenase as described in 2.

The lysis buffers, 384-well custom-coated oligo-dT plates, and M-MuLV were purchased from Pierce 63 and were used according to a modified version of the Express Direct mRNA Capture and RT-PCR system. The shrimp alkaline phosphatase, Thermosequenase buffer, ddNTP and Thermosequenase were obtained from Sequenom 64.

The primers used for multiplex PCR reactions were: EGR1, 5'-AGC GGA TAA CAC CTC ATA CCC ATC CCC TGT-3' and 5'-AGC GGA TAA CTG TCC TGG GAG AAA AGG TTG-3'; c-fos, 5'-AGC GGA TAA CGC TTC CCT TGA TCT GAC TGG-3' and 5'-AGC GGA TAA CAT GAT GCT GGG AAC AGG AAG-3'; ATP5B, 5'-AGC GGA TAA CCA AAG CCC ATG GTG GTT ACT-3' and 5'-AGC GGA TAA CGC CCA ATA ATG CAG ACA CCT-3'; RPL23A, 5'-AGC GGA TAA CAA GAA GAA GAT CCG CAC GTC-3' and 5'-AGC GGA TAA CCG AAT CAG GGT GTT GAC CTT-3'.

The following primers were used for single-base extension reactions: EGR1, 5'-TTC CCC CTG CTT TCC CG-3'; c-fos, 5'-TGC CTC TCC TCA ATG ACC CT-3'; ATP5B, 5'-GAC TGT GGC TGA ATA CTT CA-3'; RPL23A, 5'-GTC TGC CAT GAA GAA GAT AGA A-3'.

To select compounds that inhibited expression of the pathway signature, the following procedure was performed. For each compound on each plate four ratios were determined: expression of EGR1 versus expression of ATP5B(V_EGR1/ATP5B), EGR1 versus RPL23A (V_EGR1/RPL23A), c-fos versus ATP5B (V_c-fos/ATP5B), and c-fos versus RPL23A (V_c-fos/RPL23A). For each plate a median (μ) and standard deviation (σ) were determined for each of four ratios. A compound was considered a plate hit if each of the four ratios for this compound were at least one standard deviation smaller than the median (V < (μ - σ): V_EGR1/ATP5B < μ_EGR1/ATP5B - σ_EGR1/ATP5B; V_EGR1/RPL23A < μ_EGR>1/RPL23A - σ_EGR1/RPL23A; V_c-fos/ATP5B < μ_c-fos/ATP5B - σ_c-fos/ATP5B; and V_c-fos/RPL23A < μ_c-fos/RPL23A - σ_c-fos/RPL23A). Compounds that were plate hits in four out of six replicas were selected for further consideration.

For transfection experiments, 501 MEL cells or TIP5 cells were grown overnight to 50% confluence and transfected using Fugene 6 transfection reagent (Roche 65) as recommended by the manufacturer. Then 24 h after transfection, medium was exchanged for a serum free one, and cells were serum starved overnight.

Otherwise, adherent cells (TIP5, MEL 501) were grown to confluence, serum starved overnight, and treated with compounds and growth factors as indicated. Cells growing in suspension (BaF3 cells and BaF3 cells expressing TEL/PDGFR protein) were grown to 10⁶ cells/ml and treated with compounds as indicated. After treatment media was removed, adherent cells were scraped with Sample Buffer from Cell Signaling, and suspension cells were pelleted and resuspended in Sample Buffer. The resulting lysates of approximately 1 × 10⁵ cells were boiled, chilled, run on 4-15% gradient gels from BioRad 66, transferred to a polyvinylidene difluoride membrane from Millipore 67, blocked, probed and visualized as recommended by the antibody manufacturers.

Comparative sequence analysis between PDGF (UniProtKB/Swiss-Prot entry P09619), cKIT (UniProtKB/Swiss-Prot entry P10721), EGFR (UniProtKB/Swiss-Prot entry P00533), and IGFR (UniProtKB/Swiss-Prot entry P08069) was performed with BLAST 2 SEQUENCES 68.

Each primary screen replica plate contained 16 wells with PDGFβ and carrier DMSO as a positive control for PDGF activation (called PDGF in Figure 2b), and 16 wells with carrier only as a negative control for PDGF activation (called No PDGF in Figure 2b). The expression levels of marker genes normalized by expression of control gene (ratios c-fos/ATP5B and EGR1/ATP5B) were averaged for 16 PDGF wells to have a single value for the positive PDGF control per plate, and for 16 No PDGF wells to have a single value for the negative No PDGF control per replica plate. Only one well was allocated for each hit compound on a single replica plate.

To compare data between replica plates in Figure 2b, the ratios c-fos/ATP5B and EGR1/ATP5B were adjusted to be equal to 1 for positive PDGF control. This means that on each replica plate the marker/control ratios in all wells were divided by the corresponding value for the positive PDGF control for this plate. The procedure was performed independently for both c-fos/ATP5B and EGR1/ATP5B ratios. As a result of this procedure, the c-fos/ATP5B and EGR1/ATP5B ratios for the hit compounds and for the No PDGF control on each plate were divided by PDGF control c-fos/ATP5B and EGR1/ATP5B ratios for this plate. The resulting adjusted values were then averaged between three replica plates.

The data have been deposited in the Gene Expression Omnibus 69 with accession number GSE7403.