Microenvironment-Driven Dynamic Heterogeneity and Phenotypic Plasticity as a Mechanism of Melanoma Therapy Resistance

Drug resistance constitutes a major challenge in designing melanoma therapies. Microenvironment-driven tumor heterogeneity and plasticity play a key role in this phenomenon. Melanoma is highly heterogeneous with diverse genomic alterations and expression of different biological markers. In addition, melanoma cells are highly plastic and capable of adapting quickly to changing microenvironmental conditions. These contribute to variations in therapy response and durability between individual melanoma patients. In response to changing microenvironmental conditions, like hypoxia and nutrient starvation, proliferative melanoma cells can switch to an invasive slow-cycling state. Cells in this state are more aggressive and metastatic, and show increased intrinsic drug resistance. During continuous treatment, slow-cycling cells are enriched within the tumor and give rise to a new proliferative subpopulation with increased drug resistance, by exerting their stem cell-like behavior and phenotypic plasticity. In melanoma, the proliferative and invasive states are defined by high and low microphthalmia-associated transcription factor (MITF) expression, respectively. It has been observed that in MITFhigh melanomas, inhibition of MITF increases the efficacy of targeted therapies and delays the acquisition of drug resistance. Contrarily, MITF is downregulated in melanomas with acquired drug resistance. According to the phenotype switching theory, the gene expression profile of the MITFlow state is predominantly regulated by WNT5A, AXL, and NF-κB signaling. Thus, different combinations of therapies should be effective in treating different phases of melanoma, such as the combination of targeted therapies with inhibitors of MITF expression during the initial treatment phase, but with inhibitors of WNT5A/AXL/NF-κB signaling during relapse.

drug or combination, and many of these are durable effects (3). Yet, drug resistance constitutes a major challenge for effective cancer treatment with melanoma being no exception. Rapid resistance to MAPKi is common and has also been reported for ICi (4-9). Although the molecular mechanisms leading to inher ent and acquired drug resistance have been discussed extensively in the literature, the dynamics leading to resistance are poorly understood but yet critical to designing better treatments. Besides genetic and epigenetic factors, other contributors to drug resistance are microenvironmentdriven tumor heterogeneity and plasticity (10-16).

MeCHAniSMS OF inTRinSiC AnD ACQUiReD DRUG ReSiSTAnCe in MeLAnOMA
Intrinsic refers to a preexistent drug resistance of the entire population or a subpopulation of cancer cells before exposure to the drug. For example, intrinsically resistant cancer cells do not harbor the targeted mutation or are not dependent on the path way inhibited by the drug. In the case of acquired drug resistance, the tumor responds initially to the treatment but relapses and progresses later. However, it is difficult to distinguish between intrinsic and acquired resistance as a small subpopulation of intrinsically resistant cancer cells subsequently enriched, may also explain initial response and later relapse (17)(18)(19)(20). Causative factors that contribute to MAPKi resistance can be broadly classi fied into three categories: mutational events, nonmutational events, and changes in the surrounding microenvironment (21). Mutational and nonmutational events that contribute to the development of drug resistance have been discussed previously (21, 22) and are not the focus of this review. In brief, the mecha nisms linked with these events predominantly lead to MAPK pathway reactivation and/or activation of parallel signaling pathways (e.g., PI3K/AKT/mTOR) (21, 23). Besides mutational and nonmutational events which are intrinsic to tumor cells, the tumor microenvironment contributes to the development of drug resistance by influencing the crosstalk between distinct cellular compartments. Solid tumors are comprised of tumor cells and stromal cells (e.g., fibroblasts, endothelial cells, and lymphocytes) that form an organlike structure which is embedded within the extracellular matrix (ECM) and nourished by a vascular network. Each of these components show varying distribution within the tumor resulting in a highly complex and heterogeneous tumor microenvironment (24). In melanoma, secretion of tumor necro sis factorα (25, 26), hepatocyte growth factor (HGF) (27), Wnt antagonist, sFRP2 (28), and increased production of ECM (29) by stromal cells in the tumor can cause resistance to MAPKi. Thus, the density of stromal cells in different parts of the tumor plays a key role in determining response and resistance to MAPKi. In addition, the distribution of the vasculature plays a crucial role in the acquisition of varying drug resistance mechanisms in different parts of the tumor, due to differences in the levels of nutrients and oxygen. Hypoxia can induce resistance to MAPKi by mediating upregulation of HGF/MET signaling (30), increas ing SNAIL, and decreasing Ecadherin expression (31).
There are three tumor heterogeneity models (37). The well accepted clonal evolution model (38) refers to acquired additional genetic mutations in cancer cells that contribute to their altered phenotype and malignant potential. This results in a Darwinian style selection of clones during disease progression (38). The stem cell model suggests that only a small fraction of tumor cells have the potential for maintaining the tumor and drive progression (39). These cancer stem cells have selfrenewal capability and can be differentiated into "nonstem cancer cells" that lose their tumorigenic potential by acquiring stable epigenetic changes and occupy the largest fraction of the tumor (37,39,40). These two models are complementary to each other, rather than mutu ally exclusive (41). Their common feature is the unidirectional, irreversible nature of the molecular changes that lead to tumor hetero geneity (37). An alternative model is "phenotypic plasticity" or "phenotype switching. " This model suggests that tumor cells with different phenotypic and functional behavior can dynami cally shift between different transcriptional programs (42)(43)(44). The different phenotypic states, described in terms of differential gene expression patterns, have been termed "proliferative" and "invasive" signatures (45). In this model, molecular changes resulting in tumor heterogeneity are reversible, unlike the clonal evolution and stem cell models. These changes are predominantly regulated by cues from the surrounding microenviron ment, e.g., hypoxia, stromaderived factors like HGF, TGFβ. For example, in response to hypoxia, proliferative melanoma cells can switch to the invasive phenotype by altering their gene expression profile (10, 46).

MiCROenviROnMenT-DRiven DYnAMiC HeTeROGeneiTY in MeLAnOMA
"Tumor microenvironment" is a broad term, which includes (1) the tumor stroma composed of fibroblasts, endothelial cells, immune cells, soluble molecules, and the ECM, (2) the epider mal microenvironment where the tumor had originated from, FiGURe 1 | Schematic representation of microenvironment-driven dynamic heterogeneity and phenotypic plasticity as a mechanism of melanoma therapy resistance. Tumor cells close to the blood vessels proliferate, while those away from blood vessels experience hypoxia and nutrient starvation that contribute to their slow-cycling phenotype. While treatment readily targets proliferating cells, slow-cycling cells can evade drug action and survive. Upon continuous treatment, this slow-cycling subpopulation is enriched within the tumor by clonal expansion. Due to their inherent cancer stem cell-like property, they are capable of self-renewal or differentiation into a proliferative tumor cells with increased drug resistance. In addition to this, the slow-cycling cells can switch their phenotype to fast proliferating cells upon exposure to oxygen and nutrients after replacing the original peripheral fast proliferating cells that had been killed by the therapy. These phenotypeswitched cells might be more drug resistant too, as they might have acquired resistance during their slow-cycling phase. and (3) different subcompartments within the tumor itself (47). Interactions between tumor cells and the microenvironment contribute to the malignant behavior of tumor cells, e.g., progres sion, metastasis, angiogenesis, migration, and invasion (48,49). In addition, microenvironmental stress signals in response to nutrient starvation and inflammation drive phenotypic plasti city and invasion and determine therapeutic outcome (16, 50). Similarly, a preexisting immuneactive tumor microenviron ment is necessary for a favorable response to ipilimumab, and potentially other ICi (51)(52)(53).
We have developed a 3D melanoma spheroid model, which recapitulates the in vivo tumor microenvironment and architecture (54,55), that combined with the fluorescent ubiquitinationbased cell cycle indicator (56) is a useful tool to study the microenvi ronment in vitro (57,58). This model is being complemented constantly, e.g., by including DRAQ7 as a realtime cell death marker (59) or by applying mathematical algorithms to predict spatial and temporal patterns of cell density and cell cycle (60,61). Due to an oxygen and nutrient gradient, melanoma sphe roids segregate into a continuously proliferating subpopulation in the periphery and a G1arrested subpopulation in the center (12). A similar phenomenon is observed in human melanoma xenografts in mice, where clusters of cycling cells are located near blood vessels and quiescent cells in central tumor zones (12). After isolating these two subpopulations from spheroids and plating them in 2D culture separately, within 24 h G1arrested central cells recommence their cell cycle and become indistin guishable from the proliferating peripheral subpopulation (12). This supports the phenotypic plasticity model (10, 23). The cell cycle phase can also contribute to drug sensitivity (13, 62, 63) and can be targeted for cell cycletailored melanoma therapy (64). For example, bortezomib preferentially kills melanoma cells in the S/G2/M phase of the cell cycle (15). By contrast, cell cycle arrest can confer tolerance to drugs (14, 64, 65).

THe ROLe OF A SLOw-CYCLinG SUBPOPULATiOn in MeLAnOMA THeRAPY ReSiSTAnCe
Although dysregulated proliferation is a hallmark of cancer (66, 67), a quiescent or slowcycling cell subpopulation is reported in many solid cancers, including melanoma. This slowcycling subpopulation is a major determinant of treatment resistance to targeted therapies (68)(69)(70). Increased level of oxidative phos phorylation in slow cycling compared to normal cells (69,71) contributes to drug resistance in many cancers including mela noma (72)(73)(74). MAPKi are predominantly effective in targeting rapidly proliferating cells, while the slowcycling cells are not readily responsive to MAPKi (69,75,76). Thus, cells in the slow cycling state or cells that switch to this state due to therapeutic stress, can evade the action of MAPKi.
Various mechanisms are utilized by this slowcycling sub population to contribute to drug resistance. First, clonal expan sion of the residual slowcycling cells, that have survived initial treatment, results in their enrichment within the tumor. A recent study suggested that these slowcycling cells are highly aggressive with increased metastatic potential (77). Second, the slowcycling subpopulation also displays increased cancer stem celllike behavior (78). Consistent with the stem cell theory, in melanoma, these slowcycling cells comprise only 0.5-5% of all tumor cells with selfrenewal potential and are defined by the expression of the H3K4 demethylase JARID1B (23). In addition, JARID1B positive cells are essential for maintaining tumor growth (23). During continuous treatment, slowcycling cells can gain the potential to differentiate into other cell types with an increased proliferation rate and drug resistance, subsequently resulting in relapse. The cells experience a high level of "therapeutic stress, " forcing them to employ several drug resistance mechanisms. Thus, overtime highly resistant drug tolerant cells are enriched within the tumor and contribute to the highly aggressive and drug resistant nature of metastatic melanoma after relapse. JARID1Bpositive cells can give rise to JARID1Bnegative cells and also vice versa (23). This supports the phenotype switching theory which indicates the plastic nature of tumor cells that is predominantly influenced by changing microenvironmental conditions (Figure 1). In addition to JARID1B, PGC1α defines another distinct slowcycling state in melanoma with increased treatment resistance (71,73). Taken together, slowcycling cells play a pivotal role in deve loping therapy resistance and cancer progression. Thus, it is cru cial to understand the underlying biology of the slowcycling phenotype to improve the current therapy regimens in melanoma.

THe ROLe OF MiCROPHTHALMiA-ASSOCiATeD TRAnSCRiPTiOn FACTOR (MiTF) in MeLAnOMA PLASTiCiTY AnD THeRAPY ReSiSTAnCe
Microphthalmiaassociated transcription factor is the master regulator of both normal melanocyte and melanoma biology (79,80). In melanoma, MITF acts as a molecular switch that determines whether the cell will differentiate, proliferate, or become quiescent with increased migratory behavior (44,(81)(82)(83)(84). The proliferative and invasive phenotypes of melanoma cells are defined by high and low levels of MITF, respectively, and melanoma cells are capable of switching between these two states, influenced by changing microenvironmental con ditions (10, 45).
Depletion of MITF can reduce proliferation through G1arrest (42,68,81,85) with increased expression of cancer stem cell markers (68,86). In response to hypoxia, MITF expression is downregulated (87). These properties are attributes of slow cycling JARID1Bpositive melanoma cells (20), supported by a negative correlation of MITF and JARID1B/SerpinE2 (77). Thus, in response to stress, e.g., hypoxia and/or nutrient starvation, melanoma cells switch from a proliferative MITF high to an invasive MITF low slowcycling phenotype. However, these subpopula tions are not mutually exclusive, as within a tumor there can be MITF high and MITF low cells, reflecting tumor heterogeneity as discussed above. In contrast to the proliferative MITF high pheno type, the invasive MITF low phenotype is mainly governed by receptor tyrosine kinases (e.g., AXL, EGFR, and ERB3), WNT5A or NFκB signaling, and the BRN2-NFIB-EZH2 axis (46,(88)(89)(90). Single cell expression analysis revealed that some MITF high cells also express the gene signature of the invasive MITF low pheno type (91,92). These and other studies indicate the presence of a third subpopulation in melanoma that expresses MITF, AXL, and WNT5A simultaneously (88,93,94). Consistent with this, we showed by using a 3D melanoma spheroid model that indeed melanoma cells can proliferate and invade simultaneously (12). In addition, another study has shown that invasive MITF low and poorly invasive MITF high cells cooperate to invade into the sur rounding matrix (95).
The role of MITF in drug resistance is controversial and the underlying mechanisms are yet to be understood. For instance, the presence of MITF is a marker for responsiveness to MAPKi treatment, but when MITF expression is upregulated, it can confer resistance to MAPKi (96). This might reflect the extreme end of the MITF rheostat model defined by differentiation, slow cycling (42), high PGC1α expression, and therapy resistance (20). Augmenting MITF levels in melanoma cells should switch the invasive slowcycling phenotype to a proliferative phenotype. This would increase drug sensitivity because MAPKi predominantly act on rapidly proliferating cells. In addition, overexpression of MITF will inhibit the switching of proliferative cells to the inva sive slowcycling phenotype in response to stress by maintaining MITF levels constant. However, MITF is also reported as a driver of melanoma progression (97)(98)(99) and longterm MITF depletion induces senescence in melanoma cells and/or promotes apoptosis (81,100,101). Melanoma cells upregulate MITF expression to recover the loss of MAPK signaling upon exposure to MAPKi, enabling the cells to tolerate MAPKi (102). Downregulation of MITF increases the cytotoxic effects of MAPKi on melanoma cells and also reduces the acquisition of drug resistance (101,103,104). Upregulation of MITF has also been seen in several MAPKi acquired resistant cell lines (89). However, the same study reports that another population of resistant cell lines has lost MITF expression. MITF is downregulated in the acquired drug resist ant phase and makes the cells more invasive (89). Thus, further investigation of these signaling pathways is required to determine in which combination these signaling pathways can be targeted along with the inhibition of MAPK signaling, to improve the outcomes of melanoma patients with disease relapse.
However, the situation appears to be even more complex, as in heterogeneous tumors MITF high and AXL high populations can coexist (33,102). Nevertheless, it has been shown that these subpopulations benefit from endothelin1 in the presence of MAPKi, as inhibiting endothelin1 signaling can effectively inhibit the growth of such heterogeneous tumors (105). More comprehensive studies are required to determine how MITF expression levels are altered in relation to the tumor's response to MAPKi during ongoing treatment. Combination of MITF inhibitors with MAPKi should improve the efficacy of MAPKi in treating phases with high MITF expression. On the contrary, inhibitors of WNT5A/AXL/NFκB in combination with MAPKi should improve the efficacy of MAPKi in treating phases with low MITF expression (Figure 2). Indeed, targeting AXL and BRAF/MEK simultaneously in a patientderived xenograft model confers an increased survival advantage to the mice compared to monotherapy with either AXL or combination therapy with BRAF/MEK inhibitors (106).
at various phases of melanomagenesis, MITF levels can be con sidered as a predictive marker for a suitable therapy regimen for treating a particular melanoma phase. We have developed an in vitro 3D melanoma spheroid model that mimics dynamic tumor heterogeneity to study the biology of microenvironment driven tumor heterogeneity and plasticity and as these dynamic changes are difficult, timeconsuming, and expensive to study in vivo.

AUTHOR COnTRiBUTiOnS
NH and FA wrote the manuscript together.

COnCLUSiOn
Tumor microenvironmentdriven dynamic heterogeneity is a major determinant of drug resistance in melanoma. This is mainly exerted by regulating the level of the master regulator MITF which is the major determinant of the dynamic pheno typic states in melanoma. A moderate MITF level determines the proliferative state of melanoma which is readily targetable with MAPKi. Both low and extremely high MITF levels give rise to two distinct slowcycling states of melanoma (e.g., MITF low /JARID1Bpositive and MITF high /PGC1αpositive) with increased oxidative phosphorylation that results in treatment resistance. Thus, targeting this slowcycling subpopulation by modulating MITF levels can be a potential strategy to overcome drug resistance in melanoma. However, MITF biology is highly complex and the downstream effects of MITF are extremely diverse (107). In addition, mechanisms that regulate MITF expression and activity are also numerous (80). Thus, modula tion of MITF expression and activity can have diverse effects on melanoma cell biology. Considering the dynamic expression of MITF in response to changing microenvironmental conditions