Integration of Genotypic and Phenotypic Screening Reveals Molecular Mediators of Melanoma–Stromal Interaction

Tumor metastasis is the cause of more than 90% of solid tumor deaths (1). Metastatic disease is particularly onerous in melanoma where the average life expectancy of patients with advanced disease is 6 to 9 months (2). In order for metastasis to occur, cells must dissociate from the primary tumor, locate and intravasate into a blood vessel, adhere to and extravasate from the vessel, adapt to a new microenvironment, and undergo mitosis to form a new tumor at the secondary site (3, 4). Recently, cell communication networks between a tumor and its surrounding microenvironment have garnered much attention, with mounting evidence supporting a critical role in the process of metastasis [reviewed in (5–7)]. Over the past decade in vivo and in vitro model systems have been developed to better assess these particular communication networks (8–10); however, we still lack complete knowledge of the molecular mechanisms underlying tumor–stromal interactions. These molecular events are particularly consequential to cancers known to be “aggressive” such as melanoma (11). A critical hindrance to defining these molecular pathways is the paucity of simple, precise, and scalable methodologies allowing analysis of cell–cell communication.

Recent advances in microfabrication-based techniques have dramatically enhanced the degree of control of the analysis of cell behavior, in general, and cell–cell communication, in particular. Intercellular signaling has been studied by patterning cell pairs (12) or heterotypic cell populations, using a variety of techniques, including micro-stencils, microcontact cell printing, dielectrophoretic cell capture, and reconfigurable micromechanic manipulation (13–15). These studies have led to interesting insights into the role of cellular communication in the control of a variety of biological processes, including angiogenesis, liver functioning, and differentiation of various cell types. However, this type of experimentation is rarely combined with more conventional high throughput analysis of biochemical or genomic states of the interacting cells, for example, microarray assay of alteration of gene expression programs.

In this report, we analyze the interaction between melanoma cells and endothelial cells in a series of well-controlled cell coculture methods. A distinctive feature of this analysis is the ability to assess both the detailed phenotypes resulting from cell–cell communication, such as formation of cellular networks and cell migration and the large-scale changes in gene expression engendered by tumor-endothelial cell–cell communication. This analysis has allowed us to identify a set of genes upregulated during in vitro melanoma-endothelial cell–cell communication, many of which have established roles in the control of angiogenesis. We furthermore provide evidence for the importance of neuropilin-2 (NRP2), one of the genes identified as upregulated in melanoma cells during coculture, in the regulation of melanoma-endothelial intercellular communication through the control of patterned cell movement and the rate of melanoma cell division. These findings, as well as our ability to perform similar experiments with other tumor cell lines, support the general utility of this method for the analysis of varied instances of cell–cell communication involved in the formation of normal or malignant tissues, and may suggest new candidates for biomarkers of aggressive cancers.

Melanoma cell lines were provided from the laboratory in which they originated by Meenhard Herlyn (Wistar Institute, Philadelphia, PA) and have been stored frozen as early-passage stocks for use in individual experiments. Cell lines have been well-characterized previously (16) and were tested for specific genetic mutations and authenticated as previously described (17) and cultured in DMEM with 10% FBS. Human umbilical vein endothelial cells (HUVEC) were purchased from Cambrex Bio Science Walkersville, Inc and cultured in EGM-2 (Lonza). HUVECs were discarded following passage 10. Additional tumor cell lines were obtained as gifts from various laboratories at Johns Hopkins University and are described in the Supplementary Materials and Methods. GFP and RFP labeling of cells was performed as previously described (18). All cell lines were passaged less than 6 months in culture after which time cells were reinitiated in culture from frozen stocks. Detailed cell culture protocols are provided in the Supplementary Materials and Methods.

The changes in the positions of individual HUVEC cells were manually tracked over time and the coordinates were recorded using custom-made software package implemented in Mathematica (version 6). The spatial correlation factor of the velocity vectors was calculated as a function of cell–cell distance r (not exceeding 200 μm in our analysis) for all cell pairs, as described previously (19):

where v_i denotes the velocity vector of cell i, at position r_i, calculated from the difference between the cell positions in thw 2 images taken at an interval of 1 hour. When movement of a pair of cells is highly correlated, the value of C(r) can approach 1. Lower values of C(r) indicate lower correlation in the directional movement.

Cell proliferation XTT assays were performed using standard reagents (Cell Proliferation Kit II, Roche Applied Science). BrdU incorporation was measured using the BrdU Labeling and Detection Kit I (Roche Applied Science), following the manufacturer's instructions. Terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) staining was performed using the In situ Cell Death Detection Kit, TMR Red (Roche Applied Science). Please refer to Supplementary Materials and Methods for detailed protocols.

Microarray analysis was performed as previously described (17). MIAME Compliance: All samples were run in commercial arrays from Affymetrix, using Affymetrix GeneChip human U133Plus 2.0 arrays as described by the manufacturer's instructions. The JHMI Microarray Core Facility abides in all its procedures by current MIAME guidelines. Microarray data have been submitted to the NCBI's Gene Expression Omnibus (GEO) repository under the series record GSE8699. Please refer to Supplementary Materials and Methods for detailed protocols.

The relative expression levels of VEGFR1, VEGFR2, and VEGFR3 in human melanoma cell lines and RFP-HUVECs were determined by quantitative real-time PCR (qRT-PCR) analysis. Detailed protocols are provided in the Supplementary Materials and Methods.

Antibodies used included: NRP2 (sc-5542 and sc-13117;Santa Cruz), VEGFR2 (sc-504, Santa Cruz) and actin (sc-1616-R; Santa Cruz). For immunoprecipitations, protein extract was prepared using RIPA buffer with protease inhibitor cocktail (Sigma) and PMSF (Sigma). Please refer to Supplementary Materials and Methods for detailed protocols.

The lentiviral shRNA vectors pLKO.1 empty, pLKO.1 scrambled, and 5 pLKO.1 vectors targeting NRP2 were obtained from The RNA Interference Consortium shRNA library. Lipofectamine 2000 (Invitrogen) was used to transfect lentiviral shRNA as well as relevant helper plasmids into 293T cells. Viral supernatant collected from the transfection in addition with 8 μg/mL polybrene was used to infect 1205Lu cells. Cells were selected with 1 μg/mL puromycin.

To define the putative molecular determinants of melanoma–endothelial cell communication, we designed systems for concurrent analysis of genotypic and phenotypic changes during cell coculture. We explored three distinct in vitro 2-dimensional coculture systems of melanoma and endothelial cells (Fig. 1A) which were designed to increase the degree of control of the coculture and to easily distinguish constituent cells. First, we incubated adjacent colonies of melanoma and endothelial cells separated by a considerable distance (150–200 μm). The cells were then allowed to migrate and interact at the interface between colonies for 48 hours, thus creating self-organizing cellular networks (method I, Fig. 1A, left). In the second method, cells were premixed at a 50/50 ratio, plated in this randomly mixed state and allowed to interact for the subsequent 48 hours (method II, Fig. 1A, center). Finally, using stencil-based methodology, endothelial cells were patterned into circular or triangular shaped colonies of predefined size, with melanoma cells plated into the surrounding open spaces (method III, Fig. 1A, right). The number of cells in each of the patterned colonies was controlled precisely (88 ± 14 cells for 1-mm diameter circular colony; 72 ± 20 cells for 1-mm-sided triangular colony). To distinguish melanoma and endothelial cells during and following coincubation, we used the metastatic melanoma cell line, 1205Lu, carrying stably-integrated green fluorescent protein (GFP-1205Lu) and human umbilical vein endothelial cells stably transfected with red fluorescent protein (RFP-HUVECs). The 3 different coincubation techniques employed primarily differ in terms of the nature and extent of the interface between heterotypic cells: the interface is initially absent but is emerging in the first method, it is very extensive but poorly controlled in the second method, and it is well defined and controlled but not very extensive in the third method. These differentially defined interfaces allow us to vary the extent of heterotypic cell interactions that might preferentially occur as a result of modulating their spatial relationship.

Using all 3 methods, we then examined the patterns of cell movement and colony organization over time. In method I, we observed striking emergence of endothelial cell networks as the tumor-endothelial cell interface gradually emerged and cells progressively intermingled (Fig. 1B, bottom). Notably, these structures included continuous loops composed of endothelial cells that enclosed tumor cells and were of shape and form analogous to those seen in tumor parenchyma (20, 21). This structured assembly of endothelial cells differed from the random assembly observed with endothelial cells cultured alone (Fig. 1F). Evaluation of random cocultures of melanoma and endothelial cells, (method II), showed minimal HUVEC network formation at 6 hours of coculture (Fig. 1C, top), but pronounced cell network formation at 48 hours (Fig. 1C, bottom). The networks formed by a gradual accumulation of endothelial cells into looped and branched structures distinguishable against the background of melanoma cells, with cell network morphology similar to that observed using method I. Using method III, we found that endothelial cells moved in collective migratory “streams” out of the prepatterned islands into the surrounding large scale colonies of melanoma cells, forming the initial branched network (Fig. 1D; Supplementary Movie 1). Thus, all 3 cell coincubation methods resulted in an extensive reorganization of endothelial cells into branched and looped networks at the interfaces between heterotypic cell colonies. Importantly, this type of reorganization, suggesting formation of capillary-like cell networks, is usually observed in appropriate gel structures, rather than on plastic or glass, further hinting at a strong effect of coculture.

We further sought to determine whether the observed cell patterning required direct cell–cell communication at the interfaces between heterotypic cell colonies, or could be mediated through secreted factors. We found that incubation of HUVECs in either basal medium alone or HUVEC-conditioned media (CM) for 48 hours failed to elicit HUVEC patterning in vitro (Figs. 1E, 1F). However, incubation of HUVECs in either melanoma cell–CM or coculture-CM induced cell reorganization into networks similar to those found using coculture methods I and II at 48 hours of incubation (Figs. 1G, 1H), suggesting that melanoma–HUVEC communication is mediated, at least in part, by one or more tumor-secreted soluble factor(s).

As all 3 coincubation methods used in the analysis resulted in progressively organizing networks of endothelial cells, we questioned whether these morphological changes were correlated with changes in the gene expression profiles within the cells. Among the coculture methods used, method II is particularly advantageous for a large-scale evaluation in changes of genomic composition of cells participating in coculture due to similar relative numbers of and an extensive interface between heterotypic cells. We took advantage of the nature of this system and the ability to sort heterotypic cells following coculture using cell type-specific fluorescent labels to examine the gene expression profiles of sorted cells following coculture. To define the molecular pathways governing melanoma–HUVEC communication, we performed gene expression profiling of GFP-1205Lu and RFP-HUVECs cultured alone or in mixture for 48 hours (Fig. 2A). Analysis of gene expression signatures associated with the coculture system demonstrated specific influences of tumor cells on endothelial cell expression profiles and vice versa. Our initial analysis focused on the expression profile differences in melanoma cells. Evaluation of the Gene Ontology biological classifications of the genes upregulated in GFP-1205Lu cells following coculture with HUVECs versus melanoma cells grown alone demonstrated alterations in genes controlling cell adhesion, cell migration, extracellular matrix organization, and angiogenesis, that is, signatures frequently associated with tumor progression and metastasis (Table 1). Among the top 30 genes upregulated in melanoma cells under coculture conditions were thymosin beta 4, a gene previously associated with tumor angiogenesis and melanoma metastasis (22–24), multimerin 1, a gene involved in endothelial cell adhesion (25), and NRP2; (Table 2). NRPs are transmembrane glycoproteins that modulate the development of the nervous and vascular systems (26–28). They function as coreceptors interacting with the vascular endothelial growth factor (VEGF) receptors and the plexins, and bind to 2 known ligands with distinct functions: class 3 semaphorins, involved in axonal guidance; and VEGF family members known to promote angiogenesis. Blocking NRP2 function has recently been shown to inhibit tumor metastasis through effects on lymphendothelial cell migration and tumor-associated lymphangiogenesis (29). NRPs have also been implicated in tumorigenesis as they are expressed in a variety of cancers (26, 30–36). In addition, NRP2 has recently been shown to regulate processes essential for melanoma metastasis and angiogenesis in vivo (34); however, the mechanism involved in this regulation is unclear. We therefore sought to evaluate the potential role of NRP2 in regulating the phenotypic cell responses to coculture conditions observed using the various methods described previously.

NRP2 protein expression was evaluated in GFP-1205Lu melanoma cells during homotypic cell culture or following coculture with RFP-HUVECs by immunoblotting, confirming coculture increased expression of NRP2 in GFP-1205Lu (Fig. 2B). As we noted that HUVEC patterning was dependent on tumor-associated secreted soluble factors, and because NRP2 may exist in both a secreted form and as a cell surface receptor (28), we evaluated the expression of secretable NRP2 in CM derived from either RFP-HUVECs alone, GFP-1205Lu cells alone, or from cells cocultured using method II (Fig. 2C). We found RFP-HUVECs alone can secret considerable amount of secreted NRP2, and furthermore, that GFP-1205Lu alone can produce low but detectable amounts of secreted NRP2. This raises the possibility that NRP2 could be produced by melanoma cells, at least partially, in soluble form, at levels that may be modulated by coculture, thus being a potential putative mediator of paracrine melanoma-endothelial cell–cell communication. Indeed, the coculture CM showed significantly higher concentrations of NRP2 than either monolithic culture. Because our coculture methods were developed to model melanoma metastasis, we sought to ensure that NRP2would indeed be expressed in metastatic melanomas in vivo. Examination of NRP2 expression in primary human melanoma tissues demonstrated specific, high-level expression of NRP2 within the tumor parenchyma in 5 of 5 metastatic melanomas evaluated (Fig. 2D-F); furthermore, such staining was consistent with tumor staining by the melanocyte marker, Melan-A (2G).

Because NRP2 can interact with VEGF signaling known to regulate both cell proliferation and migration, we investigated the functional significance of NRP2 overexpression in melanoma-endothelial cocultures using an NRP2-neutralizing antibody (37, 38). Interestingly, even in the absence of coculture, we found that an NRP2-neutralizing antibody severely decreased melanoma cell growth in vitro (Fig. 3A), suggesting that NRP2 is a critical mediator of melanoma cell proliferation. Studies with an alternative NRP2-neutralizing antibody also confirmed a growth inhibitory role (Supplementary Fig. S1) and BrdU incorporation assays demonstrated significant growth inhibition at 48 hours following antibody treatment (Fig. 3D). Growth inhibition was titratable, as decreasing amounts of antibody had less of an inhibitory effect (Fig. 3A-C). NRP2 knockdown in 1205Lu cells also showed growth inhibition of melanoma cells (Fig. 3E-F). We note that previously, a modeling analysis indicated that the effects of neuropilin knockdown and Ab-mediated inhibition of neuropilin-VEGFR coupling might be very different, favoring Ab-mediated perturbation for therapeutic purposes (39). Evaluation of cellular apoptosis by TUNEL staining demonstrated no notable increase in melanoma cell death (Supplementary Fig. S2A – C). In addition, tumor cell morphology was not significantly altered following treatment with neutralizing antibody (Supplementary Fig. S2D, E). Furthermore, antibody neutralization of NRP2 did not significantly alter cell migration in a scratch assay when proliferation was taken into account (Supplementary Fig. S2F). These results suggest that NRP2 expressed by melanoma cells in homotypic cell culture specifically supports cell proliferation, possibly by facilitating autocrine signaling by VEGF (40).

We next turned to investigate whether NRP2 could also control directed endothelial cell migration into branched and looped networks observed with all coculture methods using temporally resolved time-lapse imaging of HUVEC cell movement. For this purpose, the most controlled method of cell coculture, method III, was employed due to the high degree of reproducibility of the initial conditions. Collective movement of endothelial cells was quantified by estimating the correlation of cell velocities for various cell pairs C(r), as a function of the distance between the cells (r, as described in 19; see Methods for more details) for the control (HUVEC colony alone) and coculture experiments at early (5 hours) and late (40 hours) time points (Fig. 4, cf. Figs. 1A-III, 1D). Note that, although C(r) is expected to be 0 for sparse pure HUVEC colonies, it is commonly much higher for dense colonies, even in the absence of coculture. This is in part due to correlated cell movement away from the center of the colony into available spaces for the distances measured (<200 μm). We found that melanoma cells significantly enhanced directed HUVEC migration at the early but not at the late time point. Strikingly, NRP2-neutralizing antibody completely abolished this early enhancement effect by melanoma cells (Fig. 4B, E), suggesting that NRP2 plays a critical role in mediating endothelial cell organization during the initial stages of the cellular network formation. Furthermore, the effect of the NRP2-neutralizing antibody was most pronounced at relatively large cell–cell distances (>120 μm). These observations suggest that NRP2 effects on endothelial cell migration during coculture are likely to be long range and may influence cell behavior through a diffusion-based process rather than contact-based cell–cell interactions, in agreement with the results of the experiments with CM. These results further imply that the branching patterns observed in heterotypic cocultures (Figs. 1B-D) may, at least in part, be dependent on the initial (over the first 5 hrs.), NRP2-dependent correlation in cell movement, direction, and speed.

To define the pathways associated with the NRP2 function in melanomas, we evaluated the expression of NRP2 ligands and coreceptors in a panel of melanoma cell lines. Previous gene expression studies of melanoma cell lines from varying stages of malignant progression allowed us to define the molecular signatures associated with melanoma progression (17). These data were mined to investigate expression of NRP2, its homologue NRP1, and its binding partners: VEGFR1, plexinA4A, plexinA3, VEGF-A, VEGF-C, and Sema3F. We detected NRP2 expression in all stages of melanoma with lower expression noted in 2 of 3 radial growth phases (Supplementary Fig. S3A). The expression of VEGF-A was elevated in early versus late stage melanomas (Supplementary Fig. S3A), while low-level expression of Plexins, Sema3F, and VEGF-C was seen in all melanoma cell lines evaluated and Nrp1 expression was virtually absent (Supplementary Fig. S3A). VEGF receptor expression was further evaluated by quantitative RT-PCR (Supplementary Fig. S3B). All cell lines examined expressed high levels of VEGFR1, with variable low-level VEGFR2 and VEGFR3 expression. NRP2 protein expression was also at the highest level in vertical growth phase melanomas, with little detectable protein in 2 of 3 of early (radial) growth phase melanomas (Supplementary Fig. S3C). These results suggested that NRP2 could exercise its effects through its natural binding partners, including, most notably, VEGF receptors.

We next evaluated the ability of various tumor cells to promote HUVEC patterning using coculture method I (Fig. 5). Patterning was assessed by quantifying the circular areas formed by networks of HUVEC cells using an automated image-analysis system (Fig. 5A–C, bottom). Interestingly, we found a range of HUVEC pattern induction by various tumor cell lines with mild to moderate patterning induced by ovarian, colon, and pancreatic cancer cells, and the strongest patterning induced by non—small-cell lung cancer, prostate cancer, breast cancer, glioblastoma, and melanoma cells (Fig. 5D). Although there was not a strict correlation between tumor cell expression of NRP2 and patterning (Fig. 5E), 3 of the top 5 patterning-associated tumor cell lines expressed significant levels of NRP2.

Tumor growth and metastasis are complex processes that require intricate interactions with the surrounding stroma. One important consequence of these interactions is the extensive restructuring of endothelium adjacent to a developing tumor leading to tumor-associated angiogenesis. Another important consequence of such interactions is the ability of a tumor cell to breach an endothelial cell barrier to enter the vasculature, potentially resulting in metastasis. The mechanisms underlying angiogenesis and tumor metastasis are poorly understood due to their multifaceted and complex nature. To address these challenges, we took a reductionist approach to defining these systems by developing simple heterotypic cell coculture methods to serve as in vitro models of tumor–stromal interactions that would allow detailed genotypic and phenotypic analysis and would facilitate tracking and sorting of participating cells. These analyses led to identification of multiple genes altered in melanoma cells during coculture with HUVECs that are associated with the angiogenic phenotype, cellular migration, extracellular matrix remodeling, and cellular adhesion, suggesting that the described approach may be informative about the molecular alterations occurring during complex metastatic events in vivo. Indeed, previous investigations have remarked on the vasculogenic phenotype of advanced melanomas (41–44) which have been variously attributed to cell-autonomous angiogenic properties of melanoma cells themselves. Our current study suggests a more complex picture, according to which melanoma cells can be induced by neighboring endothelial cells to produce factors that can in turn influence endothelial cells, and their propensity to form functional vasculature.

A closer examination of the highly upregulated genes suggests that many of them code for secreted components that might be responsible for ensuring fast and robust formation of de novo vascular beds (12). Several of the highly upregulated genes identified have also been identified in reactive oxygen-driven tumors (44). This implicates the possibility of upregulation of reactive oxygen signaling in the coculture system.

We have chosen to further evaluate NRP2 as a mediator of melanoma–endothelial cell communication and investigate its role in melanoma development and progression. Although NRP2 has been shown to be expressed in melanoma cell lines (31, 33, 45), its expression in primary human melanomas and its precise role in melanoma development and progression has not been delineated previously. We found that NRP2 knockdown inhibited melanoma cell proliferation in addition to the specific and titratable inhibition of melanoma cell proliferation by an NRP2-neutralizing antibody, suggesting an important role of NRP2 in melanoma cell proliferation. It is notable that recently, semaphorins and VEGFs were shown to bind to distinct extracellular domains of NRP2 and that targeted antibodies can inhibit the binding of a specific ligand to NRP2 without influencing the binding of other ligands at other antigenic sites (46). The antibody evaluated in our studies was generated against amino acids 560 to 858 of NRP2 and would therefore block binding of both semaphorin and VEGF ligands. These results support a potentially critical role for VEGF and/or semaphorins in regulating melanoma cell growth in pure cultures or coculture experiments; however, the large numbers of interactions between NRPs, their coreceptors and numerous ligands suggest that the mechanisms of action of these antibodies is likely to be complex.

We also found a significant effect of NRP2-neutralizing antibody on HUVEC patterning induced by melanoma cells in coculture due, at least in part, to reduced long-range coordination of cell movement at the initial stages of network formation. This suggests that NRP2, likely in its secreted form, assists melanoma cells in recruitment of endothelial cells towards the development of a functional new vasculature, thus enhancing melanoma survival and providing potential routes for metastasis. The CM-induced cell patterning experiments further suggests that the interactions between NRP2 and its associated ligands and coreceptor are complex, and a delicate balance of their levels in the microenvironment is likely to be critical to the effects associated with cell patterning. The results also raises the possibility that posttranslational modification of NRP2 within 1205Lu cells may be different from that within HUVECs, thus causing the variable effects of NRP2 in CM from these 2 sources. NRP2 is notably polysialylated in dendritic cells (47); however, additional posttranslational modifications in other cell types have not been further elucidated to date. Clarification of the effects of NRP2 within the context of 1205Lu CM would be interesting for further studies. Given the in vivo (29) and this in vitro evidence regarding NRP2 functions in melanoma growth and tumor metastasis, we conclude that NRP2 is an important mediator of melanoma–endothelial cell communication.

Our studies suggest that in addition to metastatic melanoma, glioblastoma cells express high levels of NRP2, which in turn may promote communication with and patterning of endothelial cells. It is striking that NRP2 functions at an interface of neural cell and endothelial cell fates, and that melanoma cells elicit such a strong response to communication with endothelial cells through this coreceptor. Much recent work has focused on this particular interface between neural and vascular circuitry (48–50). As melanoma is a malignant growth of the neural crest-derived melanocyte, the strong communication network of these tumor cells with their associated vasculature may display features of the developmental interactions between neural and endothelial cells. Indeed, it may be that the striking ability for melanomas to metastasize at early stages of development relates to their genetic memory of developmental cues associated with neural crest migration.

Finally, it is notable that the analyses of cell–cell communication used in this study may prove useful in exploring interactions between endothelial cells and other cancer cell types (Fig. 5), as well as other instances of intercellular signaling occurring in vivo. Indeed the correlation of phenotypic and genotypic changes accompanying cell–cell communication on the scale of interacting heterotypic cell populations remains poorly explored. The coculture methods used here can help address whether the phenotypic changes engendered by cell–cell communication might depend on the nature and extent of the interface between interacting cell types. This analysis can be further extended to cover alterations in biochemical networks other than gene expression profiles. We thus anticipate that this method may become widely adopted as a means of discovery of novel factors controlling cell communication. We further suggest that such discovery efforts will lay the groundwork for identification of novel biomarkers and therapeutic targets for a variety of malignancies for which such communication networks are critical in tumor cell growth and metastasis.

No potential conflicts of interest were disclosed.

The author(s) have made the following declarations about their contributions: Conceived and designed the experiments: M.J. Stine, C.J. Wang, W.F. Moriarty, A. Levchenko, R.M. Alani. Performed the experiments: M.J. Stine, C.J. Wang, W.F. Moriarty, B. Ryu, and W.H. Westra. Analyzed the data: M.J. Stine, C.J. Wang, W.F. Moriarty, B. Ryu, W.H. Westra, R. Cheong, A. Levchenko, and R.M. Alani. Prepared the figures: M.J. S, C.J. Wang, W.F. Moriarty, W.H. Westra, B. Ryu, and R. Cheong. Wrote the paper: M.J. Stine, C.J. Wang, A. Levchenko, and R.M. Alani.

We thank D. Ginty, J. Flook, K. Fritz, F. Murillo, C.C. Talbot, A. DeMarzo, J. Hicks, and C. Herrera for technical assistance and helpful advice; P. Cole, D. Ginty, and members of the Alani Lab for careful review of this manuscript and suggestions.

This work was supported by grants from the National Cancer Institute CA107017 (R. Alani), the Flight Attendant Medical Research Institute (R. Alani), The American Skin Association (R. Alani), the Institute for Cell Engineering (A. Levchenko), The Joanna M. Nicolay Melanoma Foundation, The Murren Family Foundation, and The Henry and Elaine Kaufman Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.