Phospho-Proteomic Screen Identifies Potential Therapeutic Targets in Melanoma

Cell Culture

Yale University (YU) designated and WW165 melanoma cell strains were derived from primary and metastatic lesions as described (10). Melanoma tumors were excised as part of patient clinical care and were collected with the participants’ informed written consent according to Health Insurance Portability and Accountability Act (HIPAA) regulations with the approval of the Yale Human Investigation Committee. Primary and metastatic melanoma cell strains were used prior to passage 20. Most melanoma cells were cultured in OptiMEM (Invitrogen, Carlsbad, CA) supplemented with 5% fetal bovine serum (FBS) and 1% penicillin/streptomycin (basal medium). The WW165 primary melanoma and YUHEIK mucosal melanoma cells were grown in basal medium supplemented with 0.1 mM IBMX (3-isobutyl-1-methylxanthine, Sigma-Aldrich; St. Louis, MO). Normal human melanocytes from newborn foreskins (NBMEL) and discarded adult skin (RMP32F) were grown in basal medium supplemented with 16 nM 12-O-tetradecanoyl phorbol-13-acetate (TPA), 0.1 mM 3-isobutyl-1-methylxanthine, 2.5 nM cholera toxin, 1 µM Na3VO4, and 0.1 mM N6,2’-O-dibutyryladenosine 3:5-cyclic monophosphate (Sigma-Aldrich), termed TICVA (11). HEK293T cells from the American Type Culture Collection (ATCC, Rockville, MD) were grown in DMEM containing 10% FBS, 1% penicillin/streptomycin, and 1% HEPES. MDA-MB-231 cells, MDA-MB-453 cells, and BT474 cells (ATCC) were grown in RPMI containing 10% FBS and 1% penicillin/streptomycin.

Gene Expression Analysis

Whole genome gene expression analysis was based on data generated by the Yale SPORE in Skin Cancer (12). Briefly, NimbleGen human whole genome expression microarrays (array 2005-04-20_Human_60mer_and array 2006-08-03_HG18_60mer) were used for hybridization of cDNA from normal melanocytes and melanoma cells at NimbleGen Systems Iceland LLC. Vínlandsleið 2–4, 113 Reykjavik, Iceland (currently Roche Applied Science, Basel, Switzerland) and by the Yale W.M. Keck Foundation Biotechnology Resource as described (12). The microarray data from melanoma cell lines (YUCAS, YUCOT, YUDOSO, YUGOE, YUHEF, YUHUY, YUKIM, YUKSI, YULAC, YUPLA, YUROB, YUROL, YUSIV, YUTICA, YUZOR, WW165 and YUHEIK) and four independent normal melanocytes cell cultures (NBMEL and RMP32F) were used to determine a gene expression call (expressed/non expressed) and a mean RTK expression value. An expression call threshold (761.0) was determined using Stat4 library in R. Briefly, parameters of a bimodal normal distribution were estimated separating the gene expression intensity data into noise (μ_noise= 279.9) and signal (μ_signal= 2360.6). Genes were considered expressed if their observed intensity value (x) was greater than the μ_signal - 1SD_signal (i.e. x>761.0). To assess correlation between RTK activation and ligand expression, we obtained a list of 473 known receptors ligands pairs from the Database of Ligand-Receptor Partners (DLRP) (13). The list was manually augmented with additional literature-derived receptor-ligand pairs. Within our data set 118 receptor-ligands pairs were used to correlate ligand expression (determined by NimbleGen array) with RTK phosphorylation (determined phospho-RTK Array).

RNA from snap-frozen melanoma tumors was extracted directly from frozen tumor blocks (YUCOT, YUSIV, and YUSTE) or after micro-dissection of snap-frozen tumors (YUDOSO and YULAC). RNA was isolated using the RNeasy Mini Kit from Qiagen (Valencia, CA) and reverse transcribed with the iScript cDNA Synthesis Kit from BioRad (Hercules, CA) using 0.8 µg of RNA per reaction. Universal TaqMan Master Mix (Applied Biosystems; Foster City, CA) was used to perform quantitative real-time PCR (qRT-PCR) using a 1 to 10 dilution of the resulting cDNA. The following primers were used following the manufacturer’s protocols: AXL (Hs0024357_m1), HER3 (Hs00951455_m1), and GAPDH (Hs99999905_m1) (Applied Biosystems). Relative mRNA expression was determined with the delta Ct method using GAPDH as the reference gene. Gene expression values for tumor RNAs were normalized to a negative control. A single RNA preparation was analyzed for each micro-dissected tumor, and duplicates for each tumor block.

Cell Lysis, RTK Arrays, Immnuoblotting, and Hierarchical Clustering

Where indicated, cells were starved in 0.1% serum prior to lysis, and pervanadate (50 µM final) was added for 20 minutes before lysis. Cells were lysed in NP-40 lysis buffer (1% NP-40, 150 mM NaCl, 50 mM Tris pH 7.4, 5 mM EDTA, 10% glycerol with complete EDTA-free protease inhibitor tablets [Roche] and Phosphatase Inhibitor Cocktails 1 and 2 [Sigma-Aldrich] added immediately before cell lysis). Cell lysates (250 µg) were analyzed with the Human Phospho-RTK Array Kit [R&D Systems (Minneapolis, MI)]; array maps at http://www.rndsystems.com/pdf/ary001.pdf. Protein samples were prepared for electrophoresis by addition of 5× Laemmli sample buffer, and immunoblotting was performed on nitrocellulose membranes blocked in 5% milk in TBST (1.5 M NaCl, 0.2 M Tris-HCl, 0.05% Tween 20). Membranes were incubated with antibodies in 5% milk or 5% bovine serum albumin (BSA) overnight at 4°C: anti-pHER3, anti-pIGF1R, anti-IGF1R, anti-pKIT, anti-MER, anti-pMET, anti-MET, anti-pSTAT3, anti-STAT3, anti-pAKT, anti-AKT (Cell Signaling Technology; Boston, MA), anti-HER3, anti-GAPDH (Santa Cruz Biotechnology; Santa Cruz, CA), anti-TYRO3 (Abcam; Cambridge, MA), and anti-KIT (Dako; Carpinteria, CA). Secondary antibodies conjugated to horseradish peroxidase were used at a 1:10,000 dilution for one hour at room temperature prior to development (Thermo Scientific; Waltham, MA).

Non-supervised hierarchical clustering was performed using manually-determined phospho-array data values ranging from 0 (background) to 5 (maximum phosphorylation) (Tables 1–2). Clustering of RTK data was performed using Pearson correlation with complete linkage along sample and protein dimensions.

Table 1.

Survey of tyrosine phosphorylated receptors by RTK capture array analysis in untreated melanoma cell lines. Summary of receptor activations in breast cancer cell lines, melanoma cell lines, and newborn melanocytes (NBMEL). Signal intensity was visually scored on a scale from 0 (background) to 5 (most intense). Melanoma cells are grouped by NRAS/BRAF status.

Open in a new tab

^*

BRAF and NRAS not determined (ND) for breast cancer cell lines.

Table 2.

Survey of tyrosine phosphorylated receptors by RTK capture array analysis in pervanadate (PV)-treated melanoma cell lines. Summary of RTK activation in melanoma cell lines and newborn melanocytes (NBMELs) scored as described in legend to Table 1.

Open in a new tab

Immunoprecipitation

HER3 immunoprecipitation (IP) was performed with 1 mg of protein and 800 ng of HER3 antibody sc-7390 (Santa Cruz; Santa Cruz, CA); KIT IP with 300 µg protein and 1µg of A4502 (Dako);MET IP with 550 µg protein and 1µg of MET antibody sc-10 (Santa Cruz); IGF1R IP with 500 µg protein and 100ng of antibody 3027 (Cell Signaling Technology) at 4°C overnight. Then, 12 µL of a 50% slurry of protein A/G beads (Thermo Scientific; Waltham, MA) was added to each IP for 1 hour. Beads were washed twice with NP-40 lysis buffer and once with salt-Tris buffer (0.1 M NaCl, 10mM Tris-HCl pH 7.4), then incubated at 100°C for 8 minutes in 2.5X Laemmli sample buffer.

shRNA and lentiviral infection

shRNA targeting AXL (RHS3979-9568950, AXL 1) (14) and (RHS3979-9568949, AXL 2), HER3 (RHS3979-9630819) (15), IGF1R (TRCN0000121300) (16), empty vector pLKO control (RHS4080) (Open Biosystems; Huntsville, AL), and a scrambled shRNA control (plasmid 1864) (Addgene; Cambridge, MA) were used. Lentivirus stocks were produced by cotransfecting 293T cells with pMD2.G and psPAX2 (Addgene) using FuGENE6 (Roche), with supernatants collected daily for three days and pooled (Addgene protocol). Experimental infections were performed at a multiplicity of infection of five.

Cell Proliferation

Cells were plated in duplicate clear-bottom 96-well plates at a density of 1,000 cells per well in 100 µL of OptiMEM. Approximately six hours after plating, cells were incubated with lentivirus in OptiMEM containing 4 µg/ml polybrene for 20 hours. Immediately following infection, a T₀ time point was taken by incubating one plate with CellTiter-Glo® (CTG) per the manufacturer’s instructions (Promega; Madison, WI). Relative cell accumulation was calculated by dividing the value for the T₇₂ time-point by the value for the T₀ timepoint. Assays were repeated 4 times, with a representative plot shown for each.

Cell Migration

Cell lines were stably infected with scrambled shRNA or AXL shRNA and were seeded at 30,000 cells/well in 24-well plates with 8-micron filter inserts (BD Biosciences; Franklin Lakes, NJ) in OptiMEM with 0.1% FBS. OptiMEM with 5% FBS was used as a chemoattractant. Cell lines were analyzed in triplicate, with a representative graph chosen for each.

Transcriptional Profiling of Melanoma Cell Lines

We used transcription profiling to identify RTKs commonly expressed in melanoma cell lines and in normal melanocytes. Of the 58 human RTK genes, 43 RTKs were transcribed above a conservatively chosen expression threshold in at least one melanocyte culture, and 53 RTKs were above the threshold in at least one melanoma cell line (Supplementary Tables 1–2). Some RTKs, including IGF1R, MET, and TYRO3, were highly expressed in most melanomas and normal melanocytes, whereas fibroblast growth receptor 2 (FGFR2), EGFR, and ROS1 were rarely, if ever, detected above the expression threshold in either melanoma cells or melanocytes. AXL and HER3 were expressed at higher levels in melanoma cells than in melanocytes, but KIT and MERTK expression was lower in melanoma cells compared to melanocytes (Supplementary Tables 1–2).

Survey of Active RTKs in Melanoma

Generally, the abundance of RTK mRNA does not predict functional activity: although RTKs such as HER2 and MET are frequently activated via gene amplification and over-expression, mutationally-activated RTKs can be potent oncogenes when expressed at low levels (17–18). Since tyrosine (Tyr) phosphorylation is indicative of the signaling activity of RTKs (19), we surveyed receptor activation using phospho-arrays that capture 42 RTKs from cell lysates and are probed with anti-phospho-tyrosine (anti-PTyr). In order to enhance the endogenous PTyr signals, cells were also analyzed after treatment with pervanadate, a tyrosine phosphatase inhibitor. Pervanadate treatment significantly enhanced the cell lysate signals seen on the phospho-RTK arrays, but did not greatly alter the pattern of RTK activation (Figure 1A). Phospho-array controls included breast cancer BT474 and MDA-MB-453 cells, which are known to express active HER2, and MDA-MB-231 cells, which express little or no active HER3, but have substantial amounts of AXL (Table 1) (20–22).

As expected, RTK mRNA levels were only loosely correlated with receptor activation. For instance, Colony-Stimulating Factor 1 Receptor (CSF1R) mRNA levels were high in both melanoma cell lines and normal melanocytes, yet CSF1R was rarely phosphorylated (Tables 1–2, Supplementary Tables 1–2). The same trend was also observed for EPHA2 in melanoma cell lines and FGFR1 in normal melanocytes. Of the 42 RTKs analyzed via phospho-arrays, IGF1R, HER3, and MET were frequently activated and transcribed above the expression threshold in melanoma cell lines. Pervanadate treatment generally increased the number of RTKs activated and expressed above the threshold (Supplementary Tables 1–2). Phosphorylation of other RTKs was more variable in comparison to RNA expression, suggesting the importance of other modulators such as growth factors or active co-receptors.

RTK activation patterns were variable, even among cell lines with BRAF and NRAS activating mutations. Such mutations occur in 44% and 12%, respectively, of the melanoma cultures analyzed in this report. Examples of array variability are found in pervanadate-treated newborn melanocytes (NBMELs), which have higher levels of phosphorylated Macrophage Stimulating 1 protein Receptor (MST1R) than melanoma cell lines YUHEIK and YULAC (Figure 1B). Similarly, YUHEIK cells showed more activated EGFR than either newborn melanocytes or YULAC cells, while YULAC cells had lower activation of KIT and higher activation of IGF1R (Figure 1B). Despite these differences, the Insulin Receptor family (IR, IGF1-R), EGFR family (especially EGFR and HER3), MET family (MET and RON), and TAM family (TYRO3, AXL, and MERTK) were commonly activated in melanoma cultures (Tables 1–2). A ranking of receptor activation across all melanoma cell strains was determined by summing the phosphorylation intensities for each individual RTK (Supplementary Table 3). Interestingly, the 15 RTKs with the highest overall activation included receptors that had not been previously reported to be active in melanoma, including TYRO3, AXL, MERTK, and EPHB2 (Supplementary Table 3). RTKs such as EGFR, FGFR3, IR, IGF1R, and TYRO3 were frequently activated with and without pervanadate treatment while HER3, MET, and RON were more commonly detected after pervanadate treatment, perhaps reflecting faster turnover of phospho-tyrosine by endogenous phosphatases (Tables 1– 2). The frequency of detection of phosphorylated EGFR and TYRO3 decreased with pervanadate treatment, possibly indicating a biological change such as heterologous desensitization through activation of other RTKs.

In both pervanadate-treated and untreated cultures, it was common to find multiple phosphorylated RTKs (Tables 1–2). Non-supervised clustering was used to evaluate possible co-associations (Supplementary Tables 4–5). Without pervanadate, there was some coassociation of IGF1R and TYRO3 phosphorylation, with a subgroup showing common activation with IR, FGFR3, and EPHB2 (Supplementary Table 4). With pervanadate treatment, activation of multiple RTKs across cell lines was more common (Supplementary Table 5). Recently, a related set of primary array data analyzing fifteen separate melanoma lines was reported. This data set identified active RTKs such as HER3, IGF1R, and TYRO3 in non-pervanadate-treated samples, but these results were not validated (23).

Validation of array data

We focused on validating TYRO3, MERTK, AXL, HER3, MET, KIT, and IGF1R because they were either among the most common or most novel active RTKs identified and because they represent attractive cancer therapeutic targets (Figure 1C–D). With the immunoblotting reagents available, we were able to determine the abundance of TYRO3 and MERTK, but could not assess their phosphorylation status for comparison with the phospho-array data. While high levels of phosphorylation in pervanadate-treated cells permitted analysis of total cell lysates, immunoprecipitated phosphorylated RTKs could also be detected in untreated cells. The pattern of receptor activation in the melanoma cell lines analyzed in Figure 1 did not vary substantially with pervanadate treatment (Figure 1C–D). We considered the immunoblots more accurate in the instances where the arrays did not correlate with the immunoblots. Hence, the RTK arrays provided a useful, but imprecise, indication of which RTKs are active in a given cell line.

RTK activation in melanoma

To identify receptors that might be activated through autocrine circuits or over-expression, we compared the phosphorylation levels determined by phospho-array analysis to the transcriptional levels of the RTKs and their respective ligands. In untreated cell lines, a few RTK phosphorylations correlated with ligand mRNA expression including FGFR3::FGF13, AXL::GAS6, and HER4 with assorted ligands (Supplementary Table 6). Similarly, in pervanadate-treated cells, moderate correlations were found between RTK phosphorylation and ligand expression for AXL::GAS6, EPHA2::EFNA3, and FGFR3::FGF13. In both datasets, receptor phosphorylation was moderately associated with RTK mRNA expression for AXL, while ERBB4 and EGFR also showed correlations in untreated cells and KIT mRNA expression was correlated with KIT activation in pervanadate treated samples (Supplementary Table 6).

The small number of cases for which receptor phosphorylations were identified in association with high ligand expression represent candidates for autocrine loops. Here, the limited sensitivity of the RTK arrays coupled with the possible chronic downregulation of ligand-activated RTKs means that there may be many false negatives. Additionally, elevated expression can move the designated RTK above phospho-array threshold for detection, which may skew the results of our screen. The co-associations of RTK phosphorylation with high receptor mRNA levels may, however, identify cases for which RTK overexpression contributes to receptor activation.

The strongest correlations were found in comparison of receptor and ligand mRNA expression levels. For a subset of RTK/ligand pairs, including FGFR3 with several of its ligands, PDGFRβ::PDGFβ, AXL::GAS6, and ERBB4::NRG3, there were strong associations of mRNA levels (Supplementary Table 6). These RTK/ligand pairs represent candidates for autocrine activation, although for some pairs subcellular compartmentalization or the need for proteolytic activation of pro-hormones may limit RTK activity.

RTK knockdown influences melanoma cell proliferation, migration, and signaling

The phospho-array survey identified IGF1R as active in most of the melanoma cell lines (Tables 1–2). Since cells were maintained in medium supplemented with serum, which is rich in ligand IGF1, we explored whether IGF1R phosphorylation was affected by serum concentration. Three melanoma cell lines were serum-starved for a period of 24–48 hours and immunoblotting was used to measure the presence of total and phosphorylated IGF1R (Figure 2A). The persistence of IGF1R phosphorylation indicates that serum IGF1 is not the primary factor influencing IGF1R activation.

The high prevalence of IGF1R activation in our panel of melanoma cell lines suggested that it might be functionally important. Four melanoma cell lines characterized by phosphorylated IGF1R were infected with empty vector (pLKO), scrambled shRNA (SC), or shRNA against IGF1R (Tables 1–2, Figure 2B). IGF1R knockdown corresponded with decreased cell proliferation and a reduction in AKT signaling, in accordance with earlier studies (6) (Figure 2B–C).

AXL and HER3 were selected for further validation of the data generated by the phospho-array screen. We focused on these two receptors because they are relatively unexamined in the context of melanoma pathogenesis and because TAM family members are poorly characterized in cancers in general. Lentiviral vectors encoding shRNA directed against AXL or HER3 were used to knock down the respective receptors singly and in combination (Figure 3A). The growth of YURIF cells, which do not express AXL or HER3 (Tables 1–2, Figure 1C–D), was not affected by lentiviral-mediated knockdown of either RTK (Figure 3B). In contrast, knockdown of AXL or HER3 strongly inhibited proliferation of YUDOSO cells (Figure 3B), which express active AXL and HER3 (Figure 1C–D). A milder reduction in proliferation was obtained by targeting active HER3 in YUROL cells (Tables 1–2, Figure 3B). YUKSI cells, which express AXL but have low AXL activity (Tables 1–2, Figure 1C–D, Figure 3A), displayed only slight reductions in proliferation and migration upon AXL knockdown (Figure 3D–E). These alterations in proliferation and migration correlated with a slight reduction in phospho-STAT3 (Figure 3A). Knockdown of active AXL in YUSIV, YUSTE, and YUTICA melanoma cells evoked a more pronounced decrease in cell proliferation, cell migration, and STAT3 phosphorylation (Tables 1–2, Figure 3A, D–E). Sporadic reductions in phospho-AKT were also observed upon AXL knockdown (data not shown). Further validation of these effects was obtained by targeting AXL using a second shRNA in four melanoma cell lines (Supplementary Figure 1). These results support the biological relevance of AXL and HER3 in melanoma cells and substantiate the importance of targeting activated receptors.

Melanoma cell lines predict presence of RTKs in tumors

To verify that expression of the RTKs identified in our screen is not a cell culture artifact, snap-frozen tissues derived from the same original tumors were analyzed via qRT-PCR. AXL mRNA was identified by transcriptional profiling in YUDOSO, YUSIV, and YUSTE, but not in YUCOT melanoma cells (Table 3). AXL mRNA and protein were variably detected in YULAC cells (Figure 1C–D, Table 3). qRT-PCR analysis of frozen tumor tissues revealed that YUDOSO, YUSIV, YUSTE, and YULAC tumors all express AXL, but the YUCOT tumor did not (Figure 4A). Similarly, HER3 mRNA was detected by transcriptional profiling of the YUCOT, YULAC, YUDOSO, YUSIV, and YUSTE melanoma cell strains as well as by qRT-PCR analysis of tissue from the corresponding tumors (Table 3, Figure 4B).

As expected, there was no direct correlation between the activity of an RTK in a cell line (Tables 1–2) and the presence of mRNA in the corresponding tumor (Figure 4A–B). Nonetheless, activated AXL and HER3 were only detected in melanoma cell lines for which the tumors expressed the corresponding receptors. The differences in rank expression of the RTKs in tumors versus cell lines may reflect tumor to cell strain differences, tumor heterogeneity, or differential infiltration by non-tumor cells in the tumor micro-environment.