Cooperative induction of receptor tyrosine kinases contributes to adaptive MAPK drug resistance in melanoma through the PI3K pathway

Abstract Vemurafenib‐induced drug resistance in melanoma has been linked to receptor tyrosine kinase (RTK) upregulation. The MITF and SOX10 genes play roles as master regulators of melanocyte and melanoma development. Here, we aimed to explore the complex mechanisms behind the MITF/SOX10‐controlled RTK‐induced drug resistance in melanoma. To achieve this, we used a number of molecular techniques, including melanoma patient data from TCGA, vemurafenib‐resistant melanoma cell lines, and knock‐down studies. The melanoma cell lines were classified as proliferative or invasive based upon their MITF/AXL expression activity. We measured the change of expression activity for MITF/SOX10 and their receptor (AXL/ERBB3) and ligand (NRG1/GAS6) targets known to be involved in RTK‐induced drug resistance after vemurafenib treatment. We find that melanoma cell lines characterized as proliferative (high MITF low AXL), transform into an invasive (low MITF, high AXL) cell state after vemurafenib resistance, indicating novel feedback loops and advanced compensatory regulation mechanisms between the master regulators, receptors, and ligands involved in vemurafenib‐induced resistance. Together, our data disclose fine‐tuned mechanisms involved in RTK‐facilitated vemurafenib resistance that will be challenging to overcome by using single drug targeting strategies against melanoma.


| INTRODUCTION
Malignant melanoma is an aggressive cancer with poor survival for patients with advanced disease. Besides immunotherapy, small molecule inhibitors selectively targeting the BRAF kinase such as vemurafenib, dabrafenib, and encorafenib have been successful in treating BRAF mutant melanoma. However, reactivation of the MAPK pathway as well as an increase in PI3K signaling is a major challenge for therapy response, due to both intrinsic and acquired resistance. [1][2][3][4][5][6][7] In an effort to circumvent resistance mechanisms, targeting the downstream kinase MEK in combination with BRAF inhibitors has been introduced into the clinic. However, even though responses have improved when using combination treatments, resistance continues to be a major obstacle for efficient therapy responses. 6,7 In response to MAPK pathway inhibition, melanoma cells may undergo a transcriptional reprogramming event, where proliferative melanoma cells switch into a phenotypically distinct invasive cell population. 8 The mechanism behind this process is incompletely understood. However, a study on the reprogramming phase in melanoma suggests MITF/SOX10 and AP1/TEAD as being the master regulators of the proliferative and the invasive transcriptome, respectively. 9 Moreover, it has been shown that during BRAF inhibitor treatment melanoma cells can lose MITF expression and de-differentiate, marking the transition to an invasive subpopulation of treatment-resistant cells. 5 Moreover, SOX 10 has been shown to have a paradoxical role in adaptive resistance. 10 Increased transcriptional levels of SOX10 are suggested to desensitize BRAF-mutant melanoma to MAPK inhibition, however, loss of SOX10 has also been shown to drive acquired resistance. 2,11 The lack of MITF and its upstream regulator SOX10 has also been shown to coincide with an upregulation of receptor tyrosine kinases (RTKs), including EGFR, ERBB3, and AXL, and in this way contributing to acquired resistance. 5,[12][13][14] Analyses of the mechanisms of drug resistance have revealed redundancy among the many surface receptors. Understanding the role of SOX10 and MITF in RTK regulation may disclose how treatment-induced transcriptional reprogramming could be utilized in the optimization of melanoma treatment strategies.
We therefore set out to investigate the role of the SOX10/MITF axis upon RTK regulation during the development of vemurafenib resistance. In a panel of melanoma cell lines, we monitored gene expression levels of ligands and receptors previously linked to the SOX10/MITF axis. We identified SOX10, MITF, ERBB3, and GAS6 as upregulated markers during early vemurafenib treatment, while EGFR, AXL, and NRG1 were upregulated after the establishment of vemurafenib resistance. Interestingly, we also found evidence for AXL/ERBB3 receptor redundancy, further demonstrating the challenges of various treatment strategies. Our study opens up for further elucidation of transcriptional reprogramming events during MAPK pathway inhibition that may contribute to stratifying the melanomas and serve as a tool for appropriate therapy selection.

| RESULTS AND DISCUSSION
We set out to investigate whether the MITF/SOX10 axis affects ERBB3 and AXL receptors throughout melanoma progression and treatment, as RTKs are frequently identified as targets of negative feedback loops and involved in resistance mechanisms in melanoma. [14][15][16][17][18] We chose to explore the ERBB3 and AXL receptors, as they have been shown to be involved with MITF in resistance towards small molecule inhibitors in melanoma. 3,5

| Correlations in patient samples
We have previously examined 470 melanoma patient samples from the Cancer Genome Atlas (TCGA SKCM) to investigate correlations between MITF/SOX10, SOX10/ERBB3, and MITF/NRG1. 13 We here expanded this examination by the inclusion of AXL and GAS6 correlations, and we observed an inverse correlation between AXL and MITF (R = À0.55, p = <.001), which is in agreement with results from previously published studies, in that melanomas lacking MITF exhibit high levels of AXL. 5, 19 We also observed an inverse correlation between AXL and SOX10 (R = À0.3, p = <.001), and finally, a moderate inverse correlation was found between AXL and ERBB3 in patient samples (R = À0.21, p = <.001) ( Figure 1A).  In an attempt to further elucidate the interplay between MITF/ SOX10 and ERBB3/AXL, we depleted SOX10 and MITF alone or in combination in the WM115 cell line, this cell line contains high levels of MITF, SOX10, and ERBB3, and low AXL expression. We found that MITF depletion alone increased both RNA and protein levels of AXL, ERBB3 and SOX10. Moreover, we found reduced RNA and protein levels of both MITF and ERBB3 following SOX10 depletion. AXL levels were not elevated following SOX10 depletion, while it did result   Figure 2C).

| ERBB3 and AXL receptor redundancy
We next asked whether this AXL high, MITF/SOX10/ERBB3 low, and NRAS-mutated cell line could be treated with AXL inhibitor monotherapy. Using MTS and IncuCyte measures as readout, we found that the WM852 cell line stopped proliferating at relatively low doses (1 and 2 μM), suggesting that the AXL inhibitor R428 alone may abrogate growth in these cells ( Figure 2D). These results are in line with a previous study, where it was demonstrated that the AXL receptor inhibitor amuvatinib had a cytotoxic effect against NRAS mutated melanoma. 22

| Receptor redundancy during vemurafenib treatment
To further investigate the consequences of AXL and ERBB3 receptor redundancy, we wanted to study the dynamics of the receptors in our cell lines in relation to the acquisition of resistance against BRAF inhibition. We treated the BRAFV600E mutated melanoma cell lines WM983, WM239, SKMEL28, WM9, and A375 with an increasing dose of the BRAF inhibitor vemurafenib. We considered the cells resistant when they proliferated at a 3 μM vemurafenib concentration, and we collected mRNA after 72 h, 1 week, and after the acquisition of resistance. RT-PCR results showed MITF and SOX10 upregulation at 72 h and 1 week following treatment in all five cell lines, with a loss of both MITF and SOX10 expression after resistance had been attained. Moreover, we found that ERBB3 was upregulated at both 72 h and 1 week following vemurafenib treatment in WM983B, WM239, WM9, and A375, in agreement with previously published work. 3,13 This is also supported by our previous work, where we measured ERBB3 levels after 2 weeks of vemurafenib treatment in WM983B and SKMEL28 at the protein level. 13 However, ERBB3 expression was comparable to normal levels in resistant cells. This seems to be compensated for by an increase of the ERBB3 ligand NRG1 in the resistant cells, which implies sustained signaling through the NRG1-ERBB3-PI3K pathway. NRG1 has previously been suggested to promote compensatory signaling through ERBB3 signaling in melanoma and colorectal cancer after BRAF inhibitor treatment. 23,24 Furthermore, the AXL ligand GAS6 was slightly upregulated after 72 h, and after 1 week, while the levels were reduced at the time of resistance with the exception of in the SKMEL28 cell line. Interestingly, AXL receptor levels were upregulated following resistance establishment in the AXL low/medium cell lines WM983B, WM239, and SKMEL28, while the AXL high and MITF low cell lines WM9 and A375 retain about the same transcription level of the factors after resistance as those of untreated controls. In contrast to NRG1-ERBB3 signaling, AXL upregulation may occur without an apparent GAS6 dysregulation in patient samples. In addition, AXL bypass signaling acts independently of GAS6 in approximately half of drug resistant lung cancer cell lines examined. 25 Furthermore, we included the RTK receptor EGFR, as it has been reported to confer resistance to MAPK inhibitors, 12 and has also been shown to regulate cell invasion signaling via AXL in glioblastoma cells. 26 In agreement with these studies, our results show that EGFR follows AXL expression during treatment, and eventually resistance, in our melanoma cell lines ( Figure 3A). Figure 3B illustrates the adaptive behavior of the F I G U R E 4 (A) Stratification of cell lines based on their transcriptional signature. The WM1366 cell line is thought to be classified as primary melanoma. However, stratifying it according to the transcriptional signature indicated in our results would suggest that this cell line is on the verge of becoming metastatic/invasive. Moreover, WM9, A375, LOXIMVI, and WM852 were stratified as invasive. MeWo and WM266.4 were stratified as being intermediate as they display medium-to-high levels of both invasive and proliferative signatures. Classified as of the Proliferative phenotype were WM983, SKMEL28, WM45.1, WM239, WM35, WM115, WM1341, and WM1382 cell lines. Finally, we included the transcriptional signature in the three treatment-resistant cell lines WM983BR, SKMEL28R, and WM239R resembling the invasive phenotype displaying Low levels of MITF, SOX10 and ERBB3 and high levels of AXL. B. Illustration of proposed signaling pathways involved in vemurafenib-induced resistance mechanisms in melanoma. Potential targets are illustrated as well as inhibitors. Vemurafenib targets BRAFV600 and trametinib targets MEK in the MAPK pathway, everolimus inhibits mTOR which is downstream of AKT and PKB, MK-2206 is a PI3K pathway inhibitor targeting all three AKT isoforms, cetuximab and erlotinib targets the EGF receptor, sapatinib is a pan ERBB inhibitor and lastly bemcentinib selectively targets the AXL receptor. transcription factors, RTKs, and their ligands during vemurafenib treatment in SKMEL28, WM983B, and WM239 cell lines.
Interestingly, Western blots of the AXL high/MITF low cell line A375 show a small pAKT increase, reduced AXL levels, and an extensive pERK level increase, suggesting that the A375 cell line mainly obtains resistance through increased MAPK pathway signaling. By contrast, the AXL low cell line SKMEL28 displays unchanged pERK levels, increased AXL levels, and increased pAKT levels, implying PI3K-induced resistance ( Figure 3C).
Adaptive cellular behavior involving RTK upregulation in response to MAPK treatment is of great interest, and patient trials combining MAPK inhibitors with RTK inhibitors are ongoing (ClinicalTrials.gov; NCT02872259). However, it is important to consider the tumor heterogeneity, as well as the complexity of the melanoma signaling network.
Receptor redundancy and pathway crosstalk is a major hindrance that complicates treatment tremendously. 3,15,16,27,28 A recent study proposes that human melanoma cells display a profound transcriptional variability 8 and that the addition of vemurafenib will induce epigenetic reprogramming in a subset of cells. During the first week of vemurafenib treatment, loss of SOX10 binding sites was observed, while 1-4 weeks of treatment revealed gain of binding sites mostly attributed to TEAD and AP-1 activation, suggesting dedifferentiation followed by activation of novel signaling pathways conferring to stable resistance. 8 Genomic cell bank. 30 All melanoma cell lines and the Hermes 3C cell line were cultured as previously described. 13 The cells were maintained at 37 C in a humidified atmosphere containing 5% CO 2 . Cell line identities were verified by short tandem repeat analysis and were routinely tested for Mycoplasma infections (VenorGeM, Minerva Biolabs, Berlin, Germany).

| Transfection and RNA interference
Cells were seeded on six-well plates and grown to 60% confluence.

| Western immunoblotting
Melanoma protein cell lysates were separated using SDS page using 4%-12% NuPAGE ® Novex Bis-Tris Midi-Gels (Invitrogen, Carlsbad, CA), and then transferred to a nitrocellulose membrane using iBot2 dry blotting system (Invitrogen, Carlsbad, CA). The membranes were then blocked using 5% BSA for 1 h, before incubation with primary antibody at 4 C overnight. To remove residual primary antibodies the membranes were washed for 3 Â 10 min in TBS/T (20 mM Tris-HCl pH 7.5, 137 mM NaCl, and 0.1% tween20). Next, a horseradish perox-

| Statistical analysis
Statistical significance of differences between control group and RTK levels was performed using student's t-test in GraphPad Prism 6. p < .05 were considered statistically significant. Experiments were performed in three biological replicates. For the construction of the heatmap logarithmized values (matrix)(log2 + 1) were used.

| TCGA data analysis
TCGA Melanoma (TCGA-SKCM) gene expression RNAseq files were extracted from UCSC Xena (https://xenabrowser.net/) (n = 472). 31 All data processing was done using R software, Correlation plot was generated using the function pairs.panels( ) in the psych package.
Pearson correlation analysis was used to calculate the correlation.

| RNA expression profiling
RNA was isolated using the protocol described above. The concentration of the RNA samples was measured using the NanoDrop ND1000 spectrophotometer (Nanodrop Technologies, Delaware, USA). RNA quality was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies Inc., California, USA) and mRNA expression profiling was performed using the Illumina HumanHT-12 v4 Expression BeadChip according to the manufacturer's protocol. Extraction of the data and quality control of the raw data was performed using Illumina's Genome studio software V2011.1. Heatmaps and clustering were performed in R. 32

ACKNOWLEDGMENTS
We gratefully acknowledge the funding by Helse Sør-Øst (project number 2014044). We also thank Vegard Nygaard, Bioinformatics Core Facility, OUS, Norway for TCGA analysis and Patrick Wernhoff, Department of Pathology, Norwegian Radium Hospital, OUS for heatmap generation.

CONFLICT OF INTEREST
The authors declare that they have no competing interests.

DATA AVAILABILITY STATEMENT
Gene expression microarray dataset can be found in the link below: http://ous-research.no/hovig/docs/Tine/Raw_microarray_ data_Melanoma_cells.zip

ETHICS STATEMENT
Not applicable.