Fibroblast-derived Neuregulin 1 Promotes Compensatory ErbB3 Receptor Signaling in Mutant BRAF Melanoma*

Background: Mutant BRAF melanomas respond to RAF inhibitors by up-regulating ErbB3 signaling. Results: NRG1 derived from fibroblasts activates ErbB3/ErbB2 signaling in vemurafenib-treated mutant BRAF melanoma cells, and NRG1 signaling is inhibited by ErbB3-targeting antibodies. Conclusion: Targeting ErbB3/ErbB2 enhances vemurafenib effects in mutant BRAF melanoma. Significance: Our data provide the preclinical basis to improve targeted therapies for mutant BRAF melanoma. Rapidly accelerated fibrosarcoma (RAF) inhibitors are first-line treatments for patients harboring V600E/K mutant BRAF melanoma. Although RAF inhibitors produce high response rates, the degree of tumor regression is heterogeneous. Compensatory/adaptive responses to targeted inhibitors are frequently initiated by the activation of growth factor receptor tyrosine kinases, including ErbB3, and factors from the tumor microenvironment may play an important role. We have shown previously that mutant v-raf murine sarcoma viral oncogene homolog B1 (BRAF) melanoma cells have enhanced activation of ErbB3 following RAF inhibition. However, the source of neuregulin 1 (NRG1), the ligand for ErbB3, is unknown. In this study, we demonstrate that NRG1 is highly expressed by dermal fibroblasts and cancer-associated fibroblasts (CAFs) isolated from mutant BRAF melanomas. Conditioned medium from fibroblasts and CAFs enhanced ErbB3 pathway activation and limited RAF inhibitor cytotoxicity in V600 mutant BRAF-harboring melanomas. Targeting the ErbB3/ErbB2 pathway partially reversed the protective effects of fibroblast/CAF-derived NRG1 on cell growth properties of RAF inhibitor-treated melanoma cells. These findings support the idea that NRG1, acting in a paracrine manner, promotes resistance to RAF inhibitors and emphasize that targeting the ErbB3/ErbB2 pathway will likely improve the efficacy of RAF inhibitors for mutant BRAF melanoma patients.

The RAF 2 inhibitors vemurafenib and dabrafenib and the RAF/MEK inhibitor combination dabrafenib plus trametinib are first-line treatments for melanomas that harbor BRAF V600E/K mutations (1)(2)(3). These targeted therapies are limited because patients nearly always progress because of drug resistance (4). Durable resistance is frequently mediated by stable alterations intrinsic to the tumor cells that lead to ERK1/2 pathway reactivation (5)(6)(7)(8). However, the initial response is likely determined by the degree of compensatory/adaptive activation of growth factor receptor tyrosine kinases (9 -12). The tumor microenvironment plays a key role in these adaptive responses to targeted inhibitors. For example, stromal fibroblasts may be subverted to phenotypically distinct CAFs, which promote supportive conditions for cancer cell growth through the production of soluble growth factors and the extracellular matrix (13)(14)(15)(16), induction of angiogenesis, and recruitment of inflammatory cells (14). The extent to which stromally derived factors are required for resistance to targeted inhibitors remains to be fully understood. Defining the molecular signals that control the cross-talk between malignant cells and the surrounding stroma may open avenues for new targeted therapy strategies to enhance the cytotoxic effects of RAF and MEK inhibitors.
Compensatory resistance mechanisms to mutant BRAF inhibition are rapid and reversible responses that counteract the actions of RAF inhibitors. We have demonstrated previously that up-regulation of ErbB3 signaling in V600E/D BRAF melanoma cells promotes survival of cells targeted with RAF inhibitors (10). Levels of phospho-ErbB3 were elevated following RAF inhibitor treatment of melanomas cells in culture, melanoma xenografts, and patient samples (10). Furthermore, application of an ErbB3-neutralizing antibody enhanced the efficacy of the RAF inhibitor PLX4720 in vivo (15). ErbB3 is a member of the EGF receptor family of receptor tyrosine kinases. Unlike the other members, ErbB3 shows low intrinsic kinase activity (17). Nevertheless, it still plays a role in the progression of several cancer types and is implicated in driving resistance to targeted therapies (10, 18 -21). Following the binding of its ligand, NRG1, ErbB3 het-erodimerizes with other ErbB family receptors, including ErbB2 and EGF receptor, to promote the activation of the AKT and ERK1/2 pathways (22). However, the cellular source of NRG1 remains unidentified.
In this study, we demonstrate that fibroblasts express high levels of NRG1 compared with mutant BRAF melanoma cells and that conditioned medium from fibroblasts and CAFs limits RAF inhibitor cytotoxicity. Additionally, ErbB3-and ErbB2targeting antibodies partially reverse the protective effects of fibroblast-and CAF-derived medium. Together, these data suggest a functional role for fibroblast-derived NRG1 in promoting resistance to RAF inhibitors in mutant BRAF melanoma.

Experimental Procedures
Growth Factors and Inhibitors-Recombinant human NRG1, insulin, and vemurafenib (PLX4032) were purchased from Cell Signaling Technology (Danvers, MA), Sigma-Aldrich (St. Louis, MO) and Selleck Chemicals LLC (Houston, TX), respectively. Seribantumab/MM121 was a gift from Merrimack Pharmaceuticals, and pertuzumab was obtained from the pharmacy at Thomas Jefferson University.
Isolation of CAFs-Human melanoma cancer biopsies (TJU-MEL25 and TJUMEL41) were obtained from Thomas Jefferson Hospital with patient consent. Following tumor excision, small pieces were digested with collagenase (Sigma) in complete medium at 37°C for 2-4 h. For the TJUMEL41 sample, pieces derived from different sections of the tumor were digested to generate CAF41A and CAF41B. Samples were then centrifuged at 4000 rpm for 4 min, the pellet was washed with complete medium, and then a second centrifugation was performed. The subsequent pellet was resuspended, and cells were cultured in DMEM supplemented with 10% FBS containing 5 g/ml insulin. CAFs were maintained in culture until passage 10. Cells were authenticated by morphology and by the expression level of ␣-smooth muscle actin and fibroblast activation protein.
Genomic DNA Sequencing-DNA was extracted from a portion of tumor samples and sequenced at the BRAF V600 loci.
Hematoxylin and Eosin Staining-The human melanoma sample TJUMEL25 was formalin-fixed, embedded in paraffin, and stained with hematoxylin and eosin.
ELISA-HFF, HTERT BJ1, and CAF25 cells were cultured in serum-free DMEM with or without 1 M vemurafenib for 24 h. Medium was collected and spun down to remove floating cells. Collected medium samples were analyzed using the NRG1-␤1 human ELISA kit (Abcam, Cambridge, MA) according to the instructions of the manufacturer. NRG1-␤1 concentrations were calculated from standard curves completed at the time of each assay. Data are representative of three independent experiments.
Western Blotting-Cells were washed twice in cold PBS and lysed with Laemmli sample buffer. For secreted NRG1 detection, medium was collected and centrifuged at 4000 rpm for 5 min to eliminate cellular debris and concentrated by centrifugation for 30 min at 4000 rpm using Amicon ultraconical tubes. Proteins were resolved by SDS-PAGE, and proteins were transferred to PVDF membranes. After blocking in 5% BSA, membranes were incubated with the indicated primary antibodies overnight at 4°C, followed by incubation with peroxidase-coupled secondary antibodies. Immunoreactivity was detected using HRP-conjugated secondary antibodies (CalBioTech, Spring Valley, CA) and chemiluminescence substrate (Ther-moScientific, Rockford, IL) on the Versadoc imaging system (Bio-Rad Coculture System-Cancer cells were plated in complete medium in individual wells of a 6-well plate. Following the indicated pretreatments, fibroblasts were plated in the transwell insert set on top of the cancer cells. The medium was replaced with fibroblast complete medium, and cells were cocultured for 3 days. Conditioned Medium Preparation-For Western blot analysis, fibroblasts were plated at 60 -70% confluence in single wells of a 6-well plate. Fibroblasts and cancer cells were treated similarly in DMEM containing 10% FBS. Conditioned medium from fibroblasts was then collected and transferred onto the melanoma cells. For longer-term growth and apoptosis assays, fibroblasts were plated in 15-cm plates at 50% confluence in 18 ml of complete medium. After 48 h, the medium was collected and concentrated to 450 l. 150 l of concentrated medium was added to 2 ml of melanoma cell medium in 6-well plates on the basis of calculations to give equivalent NRG1 levels to unconcentrated conditioned medium used for Western blot experiments. To promote repro-ducibility, after conditioned medium collection, fibroblasts were counted with the Scepter handheld automated cell counter (EMD Millipore, Billerica, MA). Counted cells per plate were 3.63 Ϯ
Alamar Blue Assay-Cells were plated in complete medium in individual wells of 6-well plates. The next day, cells were treated with vemurafenib (1 M) for 24 h or left untreated. Cells were either cocultured with fibroblasts (HFF or HTERT BJ1) using transwell inserts for 3 days or treated for 6 days with fibroblast conditioned medium with fresh vemurafenib. Treatments were renewed every 72 h. After 3 and 6 days, the medium was replaced with complete medium containing Alamar Blue (Invitrogen), and cells were allowed to reduce Alamar Blue for 2-3 h. The medium was collected in triplicate from each condition, and the absorbance of oxidized and reduced Alamar Blue was measured at wavelengths of 600 and 570 nm, respectively, in a Multiskan Spectrum spectrophotometer (Thermo Fisher Scientific). Changes in viability were calculated from the resulting absorbance reading using the guidelines of the manufacturer. All conditions were normalized to the DMSO control that was set to a value of 100.
Annexin V Staining-Cells were plated in complete medium in individual wells of 6-well plates. The next day, cells were treated with vemurafenib (1 M) or left untreated. After 24 h, vemurafenib treatment was renewed, and cells were treated for 72 h with fibroblast conditioned medium concentrated previously as described above. Cells were collected and washed twice with PBS and incubated with annexin V-allophycocyanin (BD Biosciences) for 15 min at room temperature protected from light. Samples were analyzed using the FACSCalibur flow cytometer.
Cell Growth Assay-Cells were treated for 6 days with fibroblast Conditioned medium containing vemurafenib and seribantumab/MM121 or pertuzumab, as indicated. Cells were then washed with PBS and stained with 0.2% crystal violet in 10% buffered formalin for 20 min. Subsequently, wells were washed and air-dried. Plates were scanned, and a quantitative analysis was performed using ImageJ software. The intensity mean was calculated for each well. Pictures were taken with a Nikon Eclipse Ti inverted microscope with NIS-Elements AR 3.00 software (Nikon, Melville, NY).

NRG1 Is Highly Expressed in Dermal Fibroblasts and Cancerassociated
Fibroblasts-To investigate stromally versus tumorderived NRG1, we first analyzed NRG1 expression across a panel of mutant BRAF melanoma cells compared with primary human dermal fibroblasts (HFF) and immortalized dermal fibroblasts (HTERT BJ1). Expression of NRG1 was low or unde-tectable in mutant BRAF melanoma cells but was readily detected in HFF and HTERT BJ1 cells (Fig. 1A). Because RAF inhibitors elicit paradoxical signaling effects in wild-type BRAF cells (23)(24)(25)(26), we tested the effect of the RAF inhibitor vemurafenib on NRG1 expression in fibroblasts. In contrast to keratinocytes harboring Harvey rat sarcoma viral oncogene homolog mutations (27), we observed neither paradoxical activation of ERK1/2 nor NRG1 up-regulation following vemurafenib treatment throughout a 24 -72 h time course in either HFF or HTERT cells (Fig. 1B). To further focus our studies on the tumor-associated stroma, we isolated CAFs from two independent BRAF V600E melanoma biopsy samples (Fig. 1, C and D). CAFs were characterized by their elevated expression of fibroblast activation protein and ␣-smooth muscle actin compared with normal fibroblasts (28,29) and by their lack of BRAF mutation (Fig. 1, E and F). NRG1 was poorly detectable in tumor lysates but was expressed readily in CAFs generated from the corresponding tumor, supporting our view that NRG1 is provided to mutant BRAF melanomas by the stroma (Fig. 1, E and  F). Furthermore, vemurafenib treatment did not elicit paradoxical activation of ERK1/2 or enhance NRG1 expression in CAFs (Fig. 1G). To analyze the effects on secreted levels of NRG1, ELISAs were performed and showed that NRG1␤1 secretion from fibroblasts and CAFs was not affected dramatically following vemurafenib treatment (Fig. 1H). Of note, the ELISA detects NRG1␤1, but multiple isoforms of NRG1 may be generated through alternative promoter use and splicing (30). Therefore, the total NRG1 concentration in the fibroblast conditioned medium is likely higher than the 0.5-1.2 ng/ml detected in this assay.
Because tumor cells can influence the expression profile of proximal cells (29), we also tested whether mutant BRAF melanoma-derived factors alter NRG1 expression in fibroblasts. Conditioned medium from untreated and vemurafenib-treated melanoma cells, WM115 and M238, did not alter NRG1 expression in fibroblasts (Fig. 1I). Also, in coculture transwell assays, vemurafenib-treated A375 and WM266-4 melanoma cells did not alter NRG1 expression in fibroblasts (Fig. 1L). Together, these data suggest that NRG1 is highly expressed in fibroblasts but that its expression is not enhanced by the paradoxical actions of RAF inhibitors.
Fibroblast-derived NRG1 Promotes ErbB3 Signaling and Increases AKT Phosphorylation in Melanoma Cells-Next we investigated whether fibroblast-derived NRG1 enhanced signaling in RAF-inhibited melanoma cells. To this end, we treated RAF-inhibited BRAF V600E/D melanoma cells (WM115, WM239-A, WM266-4, A375, and M238) with conditioned medium from vemurafenib-treated dermal fibroblasts. Exogenous NRG1 was added to cultures as a positive control (10). FBS. The next day, cancer cells were treated with fibroblast-derived conditioned medium or with exogenous NRG1 (10 ng/ml) for 1 h. Cell lysates were Western-blotted as indicated. B, WM115 cells were treated similarly as in A, but cells were treated with conditioned medium from CAF25. Cell lysates were Western-blotted as indicated. C, HFF and HTERT cells were treated with reagent alone (Ϫ) or transfected with a non-targeting control siRNA (siCTL) or NRG1-targeting siRNA (siNRG1) for 72 h. During the last 24 h of the knockdown, cells were cultured in serum-free medium. The medium was collected, concentrated, and analyzed for secreted NRG1 by Western blotting. Cell lysates were analyzed by Western blotting for cellular NRG1 and actin. D, WM115, WM239-A and WM266-4 cells were treated with vemurafenib for 24 h in DMEM supplemented with 10% FBS or left untreated. The next day, cells were treated for 1 h with conditioned (cond) medium from knockdown fibroblasts as indicated in C, and then cells were lysed and analyzed by Western blotting.

The Tumor Microenvironment and Targeted Inhibitor Resistance
Consistently, conditioned medium from HFF and HTERT BJ-1 cells induced phosphorylation of ErbB3 and downstream signaling to AKT and ERK1/2 to a level comparable with activation by exogenous NRG1 (Fig. 2A). Similar effects were obtained with CAF-derived conditioned medium (Fig. 2B). To test whether the protective effects of conditioned medium were dependent on NRG1 expression, we depleted NRG1 from fibroblasts using siRNAs (Fig. 2C). Notably, conditioned medium from NRG1 knockdown fibroblasts only weakly induced ErbB3 and AKT phosphorylation in all three RAF-inhibited mutant BRAF melanoma cell lines analyzed (Fig. 2D). These data show that the ErbB3 pathway in RAF-inhibited melanoma cells is activated by fibroblast-derived NRG1.
Fibroblast Conditioned Medium Drives Resistance in RAFinhibited Mutant BRAF Melanoma Cells-Exogenous NRG1 enhances growth and viability in RAF-inhibited melanoma cells (10). Therefore, we evaluated whether fibroblast-derived conditioned medium elicited similar effects. Initially, we analyzed the ability of fibroblast conditioned medium to rescue the effects of a range of vemurafenib doses on the BRAF V600D melanoma cell line WM115. Alamar Blue analysis showed that cell viability was effectively inhibited between dosages of 0.5-1 M of vemurafenib and that fibroblast-derived conditioned medium enhanced cell viability (Fig. 3A). On the basis of these findings, we continued our studies utilizing 1 M vemurafenib. Next we explored the effects of fibroblast conditioned medium on a panel of RAF-inhibited BRAF mutant melanoma cell lines. Vemurafenib-treated BRAF V600E/D melanoma cells were either cocultured with fibroblasts for 3 days in transwell chambers or treated with fibroblast-conditioned medium for 6 days. In both cases, fibroblast conditioned medium was able to partially prevent vemurafenib-mediated inhibition of melanoma cell viability (Fig. 3, B and C). Furthermore, fibroblast conditioned medium reduced the percentage of vemurafenib-induced apoptotic cells (Fig. 3D) and promoted cell growth in vemurafenib-treated WM115, W266 -4, WM239-A, and M238 cells (Fig. 3E). In a similar way, conditioned medium from CAFs partially rescued cell growth in RAF-inhibited WM115 melanoma cells (Fig. 3F). These data show that fibroblast and CAF conditioned medium counteracts RAF inhibitor effects on the growth of mutant BRAF melanoma cells.
Targeting ErbB3/ErbB2 Combines with RAF Inhibitors to Inhibit V600E/D BRAF Melanoma Cell Growth-To address the translational potential of our studies, we tested the effect of ErbB3-and ErbB2-neutralizing antibodies in blocking the protective effects of stromally derived NRG1. We utilized seribantumab/MM121, an ErbB3-neutralizing antibody that binds to the ectodomain 1 of ErbB3 (31), and pertuzumab, a clinicalgrade antibody that binds to ErbB2 subdomain II to block the ligand-dependent dimerization of ErbB2-ErbB3 (32). Administration of either seribantumab or pertuzumab strongly reduced the phosphorylation of ErbB3 and AKT in RAF-inhibited melanoma cells induced by normal fibroblast-derived or CAF-derived conditioned medium (Fig. 4, A and B). Consistent with previous findings (10), pertuzumab-mediated inhibition of ErbB3 phosphorylation indicates that ErbB3 utilizes ErbB2 as its main heterodimer partner in melanoma cells (Fig. 4A). Consistently, seribantumab and pertuzumab partially blocked the protective effects induced by fibroblast-derived conditioned medium on melanoma cell viability (Fig. 4C), apoptosis (Fig.  4D), and growth (Fig. 4E). Seribantumab and pertuzumab also inhibited the increase in melanoma cell growth induced by CAF conditioned medium (Fig. 4F). These data demonstrate that the protective effect of fibroblast-derived factors is targetable by ErbB3-and ErbB2-neutralizing agents.

Discussion
The NRG1-ErbB3 pathway is implicated in melanocyte homeostasis (33) and is related to the progression of several cancer types (18). Additionally, NRG1-ErbB3 signaling is implicated in driving resistance to targeted inhibitors (10, 19 -21). This knowledge prompted us to investigate the role and source of NRG1 in melanoma tumors and to better understand the ability of NRG1 to drive resistance to targeted therapies. We demonstrate that NRG1 secreted by fibroblasts is likely to provide the initiating signal for the compensatory/adaptive ErbB3 response in mutant BRAF melanomas treated with vemurafenib. These studies demonstrate the role of paracrine NRG1 signaling in regulating the response to targeted therapies in V600 mutant BRAF-harboring melanoma cells (Fig. 5). Additionally, our findings highlight the need to understand the mechanisms of NRG1 expression in normal and cancer-associ- . The next day, vemurafenib treatment was renewed at the indicated dose, and cells were treated with HFF or HTERT BJ1 concentrated conditioned medium (CM). As a control, concentrated DMEM with 2% FBS was added to DMSO-and vemurafenib-treated cells. Treatments were renewed after 72 h, and Alamar Blue assays were performed after 6 days. Conditions were normalized to the DMSO-treated condition. The results shown are the mean Ϯ S.D. of three independent experiments. ##, p Ͻ 0.01 compared with DMSO treatment. *, p Ͻ 0.05; **, p Ͻ 0.01 compared with the respective vemurafenib dosage treatment. B, WM115, WM239A and WM266-4 cells were treated with 1 M vemurafenib or left untreated. After 24 h, vemurafenib treatment was renewed, and cells were cocultured with or without fibroblasts (HFF or HTERT BJ1) using transwell inserts for 72 h in DMEM containing 10% FBS. Alamar Blue assays were performed to determine cell viability. Conditions were normalized to the DMSO-treated cell condition. The results are expressed as the mean Ϯ S.D. of three independent experiments. *, p Ͻ 0.05; **, p Ͻ 0.01. C, WM115, WM239-A, WM266-4, and M238 cells were treated with vemurafenib (1 M) for 24 h or left untreated. The next day, vemurafenib treatment was renewed, and cells were treated with concentrated fibroblast-conditioned medium. As a control, DMEM with 2% FBS was concentrated, and the concentrate was added to DMSO and vemurafenibtreated cells. Treatments were renewed every 72 h. After 6 days, Alamar Blue assays were performed. Results shown are the mean Ϯ S.D. of three independent experiments. *, p Ͻ 0.05; **, p Ͻ 0.01. D, WM115 and M238 cells were treated similarly as in C. After 72 h, apoptosis assays were performed. The results are expressed as the mean Ϯ S.D. of three independent experiments. *, p Ͻ 0.05; **, p Ͻ 0.01. E, WM115, WM266-4, M238, and WM239A cells were treated as in C. After 6 days, cells were fixed and stained with crystal violet. Pictures were taken at ϫ20 magnification (mag). Scale bars ϭ 50 m. Quantification was performed using ImageJ software as described under "Experimental Procedures." **, p Ͻ 0.01 compared with the control. #, p Ͻ 0.05; ##, p Ͻ 0.01 compared with vemurafenib treatment. F, WM115 cells were treated similarly as in B, but cells were treated with CAF25-dervied concentrated conditioned medium. After 6 days, cells were fixed and stained with crystal violet, and pictures were taken with ϫ20 magnification. Scale bars ϭ 50 m. Quantification was performed using ImageJ software as described under "Experimental Procedures." **, p Ͻ 0.01 compared with the control. #, p Ͻ 0.05 compared with vemurafenib treatment. OCTOBER  First, we show that NRG1 expression is low/undetectable in mutant BRAF melanoma cell lines and melanoma patient samples but highly expressed in fibroblasts and CAFs isolated from mutant BRAF melanomas. Our data are consistent with previous reports of low NRG1 expression in melanoma cells (20) and with studies demonstrating the presence of NRG1 in skin fibroblasts (34) and CAFs from pancreatic ductal adenocarcinoma (35). However, concurrent expression of ErbB3 and NRG1 may exist in a subset of human metastatic melanomas (36), and, in these tumors, direct targeting of ErbB3/ErbB2 may be efficacious (37). In other studies, we have identified a subset of wildtype BRAF/wild-type NRAS (neuroblastoma rat sarcoma viral oncogene homolog) melanomas that express both NRG1 and ErbB3 and are dependent on NRG1-ErbB3 signaling for growth (38). Despite the well characterized paradoxical signaling effects of first-generation RAF inhibitors (25,39), RAF inhibi-tors do not further promote NRG1 expression in fibroblasts. This contrasts with paradoxical effects of vemurafenib enhancing the production of VEGF from macrophages (40). Next-generation paradox breaker RAF inhibitors have been generated in an attempt to lessen these side effects (27,41). Additionally, we demonstrate the ability of fibroblast-and CAF-derived NRG1 to mediate the activation of the ErbB3 pathway in RAF-inhibited mutant BRAF melanoma cells. We show that fibroblastderived NRG1 partially blocks the effects of vemurafenib on cell growth and death in V600 mutant BRAF-harboring melanomas. Previous studies have reported the role of growth factors such as HGF (hepatocyte growth factor) produced from fibroblasts and exogenous NRG1 in driving resistance to vemurafenib (10,21,42,43). Importantly, our data demonstrate, through depletion experiments, the requirement for a fibroblast-derived growth factor, NRG1, in mediating resistance to RAF inhibitors. NRG1 does not promote the growth of mutant BRAF melanoma cells under basal conditions (10). Therefore, it supplemented with 10% FBS. The next day, cells were pretreated with seribantumab (1 M) or pertuzumab (10 g/ml) for 45 min before addition of conditioned (cond) medium derived from HFF and HTERT cells. Following 1 h of treatment, cells were lysed, and lysates were Western blotted as indicated. CTL, control. B, WM115 cells were treated similarly as in A, but CAFs were used instead of normal fibroblasts. Cells were lysed, and lysates were Western-blotted as indicated. C, WM115 and M238 melanoma cells were treated with vemurafenib for 24 h. The next day, vemurafenib treatment was renewed, and cells were treated for 45 min with seribantumab (Seriban, 1M) or pertuzumab (Pertuz, 10 g/ml). Cells were then treated with fibroblast-concentrated conditioned medium, and treatments were renewed every 72 h. As a control, concentrated DMEM supplemented with 2% FBS was added to DMSO-and vemurafenib-treated cell conditions. After 6 days, Alamar Blue assays were performed. DMSO-treated cell groups were set to 100%, and other conditions were normalized to this group. The results are expressed as the mean Ϯ S.D. of four independent experiments. *, p Ͻ 0.05; **, p Ͻ 0.01. D, WM115 cells were treated similarly as in C. After 72 h, annexin V apoptosis assays were performed. The results shown are the mean Ϯ S.D. of four independent experiments. *, p Ͻ 0.05; **, p Ͻ 0.01. E, WM115 and WM239-A cells were treated similarly as in C. After 6 days, cells were analyzed by crystal violet staining. Pictures were taken with ϫ20 magnification. Scale bars ϭ 50 m. Quantification was performed using ImageJ software as described under "Experimental Procedures." **, p Ͻ 0.01 compared with the control. #, p Ͻ 0.05; ##, p Ͻ 0.01 compared with vemurafenib treatment. †, p Ͻ 0.05; † †, p Ͻ 0.01 compared with vemurafenib ϩ HFF conditioned medium (CM). Ⅺ, p Ͻ 0.05; ⅪⅪ, p Ͻ 0.01 compared with vemurafenib ϩ HTERT conditioned medium. F, WM115 cells were treated similarly as in C, but concentrated conditioned medium from CAF25 was utilized. Cells were cultured for 6 days before fixation. Fixed cells were stained with crystal violet, and pictures were taken at ϫ20 magnification (mag). Scale bars ϭ 50 m. Quantification was performed using ImageJ software as described under "Experimental Procedures." **, p Ͻ 0.01 compared with the control. #, p Ͻ 0.05 compared with vemurafenib treatment. †, p Ͻ 0.05; † †, p Ͻ 0.01 compared with vemurafenib ϩ CAF25 CM.

The Tumor Microenvironment and Targeted Inhibitor Resistance
is likely that the presence of NRG1 in the tumor microenvironment becomes critical following treatment with RAF inhibitors.
To support the translational potential of our studies, we tested the role of ErbB3-and ErbB2-neutralizing antibodies in counteracting the effects of stromal NRG1. We used two different antibodies: seribantumab, which is currently being tested in several clinical trials (ClinicalTrials.gov identifier NCT02387216), and pertuzumab, a Food and Drug Administration-approved agent for combination therapy with trastuzumab and docetaxel for metastatic breast cancer (44). Both antibodies were able to partially reverse the effect of fibroblastderived conditioned medium on ErbB3 pathway activation and mutant BRAF melanoma cell growth properties. These data are consistent with the ability of ErbB3-targeting agents to increase the efficacy and duration of response of RAF inhibitors in vivo (15) and underscore the use of ErbB3-targeting agents to reduce the protective effects of fibroblast-derived NRG1 on RAF-inhibited melanoma cells. Although pertuzumab and seribantumab target ErbB3/ErbB2 signaling by preventing ErbB3/ErbB2 dimerization or the binding of NRG1 to ErbB3, respectively, other agents may be developed to directly sequester NRG1 (45) or to directly target CAFs (46). Our findings also have relevance to other tumor types in which ErbB3 has been implicated in driving resistance to targeted inhibitors (19,20). ErbB2-targeting antibodies are already Food and Drug Administration-approved for breast cancer, and there is considerable research activity with ErbB3-neutralizing antibodies for several cancer types (47,48).
ErbB3 is overexpressed in a subset of melanoma patients and associated with a poor prognosis (49). However, our findings emphasize the importance of stratification on the basis of NRG1 and ErbB3 levels as well as genetic status. Recent studies report that NRG1 expression is a predictor of response to seribantumab (31), providing the basis for using NRG1 levels as a biomarker. Therefore, further studies investigating NRG1 and ErbB3 as biomarkers for the combined therapy with RAF or MEK inhibitors and ErbB3/ErbB2-targeting agents in melanoma could prove to be beneficial. Such studies may identify BRAF V600 melanoma patients in whom stromal NRG1 contributes to compensatory/adaptive ErbB3 signaling and in whom targeting the ErbB3/ErbB2 signaling axis may improve the efficacy of RAF inhibitors.
Author Contributions-C. C. and A. E. A. conceived and coordinated the study and wrote the paper. C. C. performed most of the experiments. S. R. performed and analyzed the experiments shown in Fig. 1, A, B, H, and L, and aided in the writing of the manuscript. A. C. B. provided patient tissue samples. All authors reviewed the results and approved the final version of the manuscript.