Immunotherapies blocking negative immune checkpoints were first approved for the treatment of metastatic melanoma and are now widely used for the treatment of a growing number of cancers, albeit with variable efficacy (1). Despite spectacular efficacy of immune checkpoint blockade (ICB) (anti-CTLA-4 and anti-PD-1 antibodies) in a subset of metastatic melanoma patients (2), primary (i.e. at the onset of treatment) or acquired resistance (i.e. after initial response) still occurs in 60% of patients (3).

The mechanisms of resistance to immunotherapies targeting inhibitory immune checkpoints, include both host extrinsic and intrinsic factors, the mechanisms of which are becoming increasingly clear (4). Predictive features of response by the tumor immune microenvironment have gained notoriety (5), particularly CD8⁺ T cell infiltration (6–8) and the presence of tertiary lymphoid structures (9–11), emerging as promising factors of response. Conversely, the presence of regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSC), M2 macrophages, and other inhibitory immune checkpoints, contribute to inhibiting anti-tumor immune responses (3, 12–14). However, the emergence of additional tumor parameters has improved our predictive capacity. Indeed, tumor cell-intrinsic alterations have been associated with resistance to immunotherapy in melanoma. Namely, JAK1/2 mutations were associated with decreased Interferon-γ (IFN-γ) sensitivity of melanoma cells (15), and mutations in the gene encoding beta-2-microglobulin (B2M) to loss of surface expression of MHC-I and decreased antigen presentation (16). Oncogenic alterations of melanoma cells, including activation of the WNT-β-catenin (17, 18), MAPK (19), CDK4-CDK6 (20) pathways, or loss of PTEN expression (21–23) have also been associated with T cell exclusion and immune resistance in mouse models and human melanoma samples. Altogether, these tumor-intrinsic genetic alterations define the tumor mutational burden (TMB), the value of which to predict response to ICB was confirmed recently in a pan-cancer meta-analysis of over 1,000 patients, including 5 melanoma datasets (24).

However, growing evidence suggests the contribution of tumor cell-intrinsic non-genetic mechanisms, akin to the epithelial-to-mesenchymal transition (EMT), in the acquisition of resistance to immunotherapy in melanoma (25–29). EMT is an embryonic reversible cell plasticity process, by which an epithelial cell loses its polarity and adhesion to other cells, while gaining motility and mesenchymal features (30). EMT is required during the delamination of the embryonic neural crest, from which melanoblasts, the progenitors of melanocytes, originate (31, 32). This process is driven by intense transcriptional, epigenetic and translational reprogramming, involving EMT-associated transcription factors (EMT-TFs) (33, 34). Aberrant reactivation of EMT has been thoroughly characterized in carcinomas, as a multi-step dedifferentiation process driving metastasis, drug resistance and disease recurrence (35, 36). More recently, emerging evidence showed that this process is not limited to epithelial cancers, since EMT-like transitions have been described in non-epithelial malignancies, such as glioblastoma and neuroblastoma (37–40). The term of phenotype switching is therefore used to refer to EMT-resembling plasticity in non-epithelial cancers.

Aberrant activation of EMT factors is increasingly reported to contribute to immune evasion in various carcinoma models (41–44). In contrast, their contribution in the context of melanoma has been poorly studied, while they display cell-type specific roles in this neural crest-derived cancer (45, 46). In this review, we will focus on the impact of non-genetic mechanisms related to melanoma cell phenotype switching during immune escape and resistance to immunotherapies. After a brief overview of EMT-like transcriptional, epigenetic and translational mechanisms, we will review recent data highlighting how melanoma cell plasticity may impact anti-tumor immunity, including the loss of melanocytic antigens and other cell autonomous mechanisms that enable melanoma cells to evade cytotoxic CD8⁺ T cell lysis. In addition, we will discuss how melanoma cell dedifferentiation can reprogram the immune tumor microenvironment, mainly through the secretion of inflammatory cytokines and chemokines. Metabolic rewiring of melanoma cells, which also plays a major role in their interactions with the tumor microenvironment [already reviewed in Avagliano et al. (47)] will not be addressed. A better understanding of the mechanisms by which phenotypic adaptations of melanoma cells contribute to the acquisition of resistance to immunotherapies, will provide the basis for the development of new combination therapies.

Melanoma cells undergo reversible phenotype switching between a proliferative/differentiated phenotype and an invasive/stem-like state. Microphtalmia-associated transcription factor (MITF), the master regulator of melanocytic differentiation (48) was described as the major driver of phenotype switching. Indeed, decreased MITF expression promotes the transition towards a more invasive state, with stem-like features (49–51). Up-regulated expression of the AXL receptor tyrosine kinase, led to the definition of the MITF^low/AXL^high phenotype (52). Recent analyses of tumor heterogeneity at the single cell level (20, 53, 54) further refined this phenotype switching model, with the description of intermediate states, including a neural crest stem cell-like (NCSC) phenotype [reviewed in (55, 56)]. The deciphering of the transcriptional and epigenetic mechanisms regulating these transitions is ongoing (57, 58). Indeed, analyses of the H3K27 acetylation pattern highlighted a major role for MITF and SOX10 transcription factors in the regulation of melanocytic enhancers, and AP-1 was described as the major regulator of mesenchymal-like state enhancers (59, 60).

Strikingly, EMT-TFs were shown to participate in the regulation of melanoma cell plasticity, with a specific expression pattern compared to that of carcinoma (61) [reviewed in (45)]. Our team demonstrated that ZEB2 and SNAIL2 are highly expressed in MITF^high proliferative melanoma cells, whereas decreased expression of ZEB2 and SNAIL2 and aberrant activation of ZEB1 and TWIST1 promoted phenotype switching towards an invasive, dedifferentiated and MITF^low phenotype (46). Notably, ZEB1 increases the expression of NCSC markers, such as NGFR (62), a major regulator of phenotype switching in melanoma (63).

Dynamic transcriptional regulation of EMT-TFs (ZEB1, TWIST1, SNAIL) and genes associated with invasive phenotypes (SOX9, POU3F2) is achieved through a bivalent promoter configuration, displaying both permissive H3K4me and repressive H3K27me histone marks; this “poised” state enables rapid induction of gene expression (64). In this context, the histone methyltransferase EZH2, the effector subunit of the PRC2 polycomb complex, was shown to promote EMT-like melanoma cell plasticity (65). An additional level of regulation of phenotype switching is provided at the translational level through the eIF4E/F translation initiation complex (66, 67). Dysregulated mRNA translation by aberrant activation of the MNK1/2-eIF4E axis plays a critical role in progression to an invasive state (68). Phosphorylation of eIF4E selectively increases the translation of a subset of mRNAs encoding proteins including NGFR or SNAIL (69). In addition, micro-environmental cues were shown to induce phenotype switching in melanoma cells in vitro, such as TNFα (25), TGF-β (70) and hypoxia (71), all known inducers of EMT in carcinoma models (30, 35).

Overall, our understanding of the EMT-like phenotype switching in melanoma has improved over the last decade, supporting its role in sustaining intra-tumoral heterogeneity [reviewed in (72, 73)]. Concomitantly, several studies gathered evidence for the role of melanoma cell-intrinsic plasticity in the resistance to BRAF^V600E targeted therapy (53, 57, 59, 62, 74), and reprogramming towards a NCSC phenotype was proposed as an adaptive response to targeted therapy [reviewed in (75)]. Emerging evidence sustaining the role of melanoma cell phenotype switching in immune resistance will be detailed below.

In addition to escaping recognition and lysis by immune cells, tumor cells may also modify the composition and functionality of the immune microenvironment. CD8⁺ T cell recruitment and activation following treatment is a major parameter associated with the response to immunotherapy (83, 123) ( Figure 1C ). TGF-β, a well-known inducer of EMT (124), has been associated with the exclusion of cytotoxic T cells in carcinoma models (125, 126). In accordance, combined treatment with antibodies targeting PD-L1 and TGF-β was shown to induce CD8⁺ T cell infiltration and tumor regression. However, few studies are available in melanoma.

Alterations in various oncogenic pathways have also been associated with T cell exclusion or dysfunction in melanoma (21). Indeed, activation of the β-catenin pathway and loss of PTEN are two major independent and non-redundant mechanisms mediating T cell exclusion in metastatic melanomas (18, 22). A longitudinal follow-up of a patient with metastatic melanoma who relapsed upon interleukin-2 treatment revealed an activation of the β-catenin pathway, associated with the absence of infiltrating CD8⁺ T cells and decreased chemokine expression in the recurrent tumor (23). These findings provide evidence that tumor cell-intrinsic β-catenin activation induces an immune-deprived environment, that impairs immune control even in the setting of IL-2 immune stimulation. More recently, another study showed that serine/threonine­protein kinase PAK4, a WNT signaling mediator, was enriched in immunologically cold tumors from patients with melanoma resistant to anti-PD-1 immune checkpoint blockade. In multiple mouse models, genetic deletion or pharmacological inhibition of PAK4 resulted in reversal of resistance to anti-PD-1 therapy (17).

In addition, high NGFR expression in melanoma cells has been linked to immune exclusion in human melanoma samples (84). Indeed, NGFR expression by melanoma cells, as assessed by immunohistochemistry, was inversely correlated with the presence of CD8⁺ T cells. This was also evidenced at the intra-tumoral level in melanoma heterogeneous cases. These findings suggest that NGFR^high melanoma cells are not only resistant to T cell targeting but are also associated with poor T cell infiltration.

More recently, we investigated the role of the EMT-inducing transcription factor ZEB1 in immune evasion, and demonstrated its role in preventing T cell infiltration in melanoma (127). Multi-immunofluorescence spatial analyses of the immune infiltrates in human melanoma samples highlighted that high ZEB1 expression in tumor cells was associated with decreased CD8⁺ T lymphocyte infiltration. Gain- or loss-of-function experiments in BRAF or NRAS-mutated melanoma mouse models demonstrated that ZEB1 prevents the recruitment and the activation of CD8⁺ T cells. Mechanistically, ZEB1^high melanoma cells showed a defective secretion of T cell-attracting chemokines, including CXCL10, suggesting the intrinsic role of ZEB1 in regulating the secretome and subsequent immune cell attraction. ZEB1 directly binds to the promoters of T cell-attracting chemokines, including CXCL10 to repress their transcription. ZEB1-mediated T cell exclusion promotes immune evasion, as we showed that ZEB1 overexpression promotes resistance, whereas Zeb1 knock-out improves the efficacy of anti-PD-1 immunotherapy in melanoma mouse models. Overall, our data indicate that CXCL10 partially accounts for ZEB1-mediated CD8⁺ T cell deficiency, and suggest additional mechanisms may contribute to immune escape. Moreover, ZEB1 may regulate CD8⁺ T cell-dependent tumor growth at least in part independently of MITF-mediated phenotype switching, since similar phenotypes were observed in MITF^high and MITF^low backgrounds.

Other melanoma-intrinsic pathways have also been associated with T cell deficiency. Knocking-out the epigenetic regulators Arid2 or Pbrm1 in B16F10 melanoma cells significantly increased CXCL9/CXCL10 production and CD8⁺ infiltration in vivo, resulting in decreased tumor growth (117). SOX2 overexpression in B16F10 melanoma cells, in addition to mediating resistance to T cell killing, decreased infiltration of CD8⁺ T cells, albeit the underlying mechanisms remain unclear (116). Although not related to a phenotype switch in the immune-competent melanoma mouse models used, Sox10 knock-out was also shown to reduce melanoma tumor growth in a T cell-dependent manner (128).

The link between the melanoma differentiation state and T cell exclusion/dysfunction was recently nicely addressed in novel mouse models, that will be useful tools to further dissect melanoma cell-intrinsic mechanisms of immune escape. Four immunocompetent melanoma mouse models differing in their differentiation status were generated and extensively characterized in relation with their differential sensitivity to immunotherapy (129). These models, based on different genetic backgrounds, showed similarities with previously described differentiation signatures (57); the M1 (NCSC-like) and M2 (undifferentiated) models were resistant to anti-CTLA-4 and anti-PD-1, while more differentiated models, M3 (melanocytic) and M4 (transitory), displayed partial response to ICB. The capacity to present antigens through MHC-I and to activate CD8⁺ T cells was maintained in all models. Importantly, resistant M2 tumors which displayed an undifferentiated signature, harbored only few T cells with a high exclusion score, except for Treg (130). However, TIL abundance in untreated tumors was not sufficient to explain ICB response, since M1 tumors were resistant despite a high CD8⁺ T cell/Treg ratio. CD8⁺ T cells in M1 tumors were shown to display a high dysfunctional score as determined by TIDE (T cell dysfunction and exclusion score), with increased expression of PD-1, LAG-3 and TIM3 exhaustion markers. Moreover, these studies highlighted the importance of assessing the myeloid cell compartment in addition to T cells: pro-tumor macrophages (especially PD-L1⁺) were enriched in M2 and M1 tumors, while dendritic cells (DC) and NK cells were further reduced in M2 tumors. Overall, melanoma cell dedifferentiation is associated with decreased T cell abundance and function, but may also impact the recruitment or function of other immune cell populations.

Previous studies in carcinoma have shown that expression of EMT-TFs (notably ZEB1, SNAIL and TWIST1) is associated with the recruitment of immunosuppressive cells, including regulatory T cells, myeloid-derived suppressor cells and tumor-associated macrophages (TAMs) (41–43, 131–133).

Dedifferentiation of melanoma cells has been associated with an immunosuppressive microenvironment ( Figure 1D ). Namely, the dedifferentiated MITF^low/c-Jun^high melanoma phenotype in human melanoma samples, was associated with increased infiltration by myeloid immune cells, as assessed by CD14 immunohistochemistry (134). Increased Gr-1⁺ myeloid cell recruitment was confirmed in dedifferentiated Hgf-Cdk4^R24C mouse melanoma models obtained after escape from ACT against the melanocytic antigen gp100. In an ACT model directed against the melanosomal protein RAB38, recurrent melanomas with dedifferentiated features also showed increased myeloid cell recruitment, related to increased secretion of myeloid cell-attracting chemokines (CCL2, 3, 5) (135). Overall, several studies reported that the MITF^low phenotype is associated with elevated NF-kB activity, and increased expression/secretion of pro-inflammatory cytokines IL-1β, IL-6 and CCL2 chemokine that may contribute to myeloid cell recruitment (136).

SNAIL was shown to favor an immunosuppressive microenvironment in melanoma models (137). SNAIL overexpression in melanoma cells in vivo promotes the recruitment of Tregs and impairs dendritic cell maturation in a process involving the immunosuppressive cytokines thrombospondin (TSP1) and TGF-β. Furthermore, blocking SNAIL with siRNA or anti-TSP1 monoclonal antibodies inhibited tumor invasion by relieving this immunosuppressive mechanism. In addition to promoting CD8⁺ T cell deficiency, ZEB1 overexpression in melanoma mouse models was also associated with an increased frequency of Treg, although at a later stage. No significant modifications in the overall frequency of macrophages and DCs were observed in these models upon ZEB1 expression, although more precise phenotyping of these immune populations will be required (127).

Moreover, CD133⁺ melanoma cancer stem cells (CSC) in the B16F10 model were associated with increased abundance of immunosuppressive cells, including Tregs, MDSCs, and M2 macrophages (138). Interestingly CD133⁺ CSC harbored an increased TGF-β signature that was regulated by miR-92 through an integrin-dependent TGF-β activation.

Recently, the MNK1/2-eIF4E axis was also involved in the regulation of melanoma plasticity and anti-tumor immune response by promoting a suppressive tumor microenvironment (68). In a BRAF^V600 ; PTEN KO mouse melanoma model in which eIF4E cannot be phosphorylated (eIF4EKI), they showed an increase in the secretion of many cytokines linked with the expansion, recruitment (such as CCL2, CCL12, and CCL5), and function (such as MMP-9) of immunosuppressive cells such as MDSCs. Immune phenotyping showed a significant infiltration of cytotoxic CD8⁺ T cells linked to a decrease in monocytic MDSCs in eIF4EKI compared with eIF4EWT melanomas (68). Combined treatment with MNK1/2 inhibitors and anti-PD-1/PD-L1 demonstrated its efficacy in several mouse melanoma models, arguing in favor of defining a new strategy to inhibit melanoma plasticity and improve response to anti-PD-1 immunotherapy.

Altogether cancer cells use several ways to promote immune escape notably by inducing an immunosuppressive microenvironment through phenotype switching. Strategies aiming at targeting key players of melanoma cell plasticity could thus improve response to immune checkpoint therapy ( Table 1 ). Such combination strategies may not only be addressed in preclinical mouse models, but their efficacy may be further evaluated ex vivo on fresh human melanoma slices (142), as nicely reported recently. This will also allow the validation of predictive biomarkers.

Previous reports have suggested that EMT or mesenchymal signatures were associated with a lack of response to ICB in carcinoma (36, 42, 43). As described in this review, preclinical studies in mouse models highlighted the contribution of many non-genetic mechanisms in the resistance to ICB in melanoma, but only few studies demonstrated the predictive value of these parameters in large human cohorts of ICB-treated patients. Difficulties reside in the inability to identify robust signatures/biomarkers of response to ICB from bulk RNA-Seq data of tumors at baseline, before treatment.

By comparing responding versus non-responding pre-treatment tumors Hugo et al. first described an innate PD-1 resistance (IPRES) gene signature, including upregulation of mesenchymal markers (such as AXL, WNT5A, or TWIST2) and immunosuppressive cytokines (27). Another ICB pre-treated melanoma cohort showed that TGFB1^low and SOX10^high expression was associated high cytotoxic T lymphocyte levels and a better overall survival (130). RNA-Seq data of patients treated with ICB revealed that the tumor-intrinsic NGFR signature predicts resistance to anti-PD-1 therapy before treatment and is further increased upon treatment (84). However, transcriptomic signatures such as IPRES have not been able to consistently predict response to anti-PD-1 in several independent melanoma cohorts (83, 89, 123, 143). One limitation resides in the fact that part of these melanoma samples received BRAF inhibitors (BRAFi) prior to ICB. Indeed, previous treatment with BRAFi may be a confounding parameter, as BRAFi is known to induce ZEB1 expression and promote a higher mesenchymal score (62). In line with these observations, a recent study demonstrated in melanoma patients and mouse models, that tumors relapsing after targeted therapy with MAPK pathway inhibitors are cross-resistant to immunotherapies (144). Resistance to targeted therapy in melanoma leads to activation of a cancer cell-intrinsic signaling program with enhanced and altered transcriptional output associated with an immunosuppressive tumor microenvironment, characterized by a lack of functional CD103⁺ DCs and T cells. Inhibition of the MAPK pathway in BRAFi-resistant tumors restored tumor infiltration and maturation of CD103⁺ DCs, reduced suppressive myeloid cells, increased T cell infiltration, and re-sensitize cross-resistant tumors to immunotherapy.

More recently, Pérez-Guijarro et al. defined a 45-gene melanocytic plasticity signature (MPS), consisting of 33 upregulated and 12 downregulated genes, in undifferentiated/NCSC resistant melanoma mouse models and validated its efficacy in predicting response in human cohorts. While the precise molecular mechanisms involved remain to be functionally addressed, they demonstrated that combination of MPS and TIDE (T cell dysfunction and exhaustion) signatures was a stronger predictor (129). Overall, this highlights the interest of combining cancer cell-intrinsic and immune parameters in order to better predict patient survival in response to ICB.

However, there may still exist limitations/biases of using bulk RNA-Seq approaches, since EMT-associated genes are not only expressed by cancer cells but also by cells from the tumor micro-environment, including cancer-associated fibroblasts (CAFs), endothelial cells, and immune cells. This constitutes an important confounding parameter in bulk RNA-Seq analyses, especially in samples from lymph nodes metastases. This problem can be overcome through single-cell RNA-seq analyses (20, 54) or through spatial analyses of tumors (127). By using scRNA-seq data, Tyler and Tirosh recently defined cancer or stromal cell-specific mesenchymal signatures. Such a deconvolution tool would allow to circumvent the confounding effect of stromal cells in bulk RNA-Seq data (145). In parallel, we already demonstrated the power of multiplexed immunofluorescence analyses to specifically analyze ZEB1 expression in melanoma cells, by excluding stromal and immune cells. Whether ZEB1/EMT-TF specific expression by melanoma cells may predict response to ICB in patient cohorts remains to be addressed.

Furthermore, since plasticity is a dynamic event, longitudinal follow-up of patients during immunotherapy, and not just baseline analyses, is required to comprehend the dynamic evolution of cancer cells and their microenvironment. In a recent study conducted in the group of G. Boland, a patient was followed for 9 years, collecting tumors from primary tumor, metastatic recurrence, pre-treatment, on-treatment, post-progression and at autopsy. This allowed them to establish an evolutionary dynamic map of resistance to ICB. Deconvolution of bulk RNA-Seq and highly multiplexed immunofluorescence characterized a dedifferentiated neural-crest (NGFR^high) tumor population during resistance to immunotherapy. Moreover, they described distinct NGFR^high tumor cell distribution patterns, highlighting site-specific heterogeneity in tumor-immune interactions that will require further investigation (29).

Throughout this review, we described several complementary mechanisms by which melanoma cell-intrinsic pathways can impact the crosstalk with the immune tumor microenvironment. Transcriptomic analyses in patient samples contributed to pinpointing that melanoma dedifferentiation markers may prove useful in combination with immune parameters to design a composite score for better predicting patients that may or may not respond to immunotherapy. In the future, the development of single cell spatial transcriptomic analyses, performed on samples biopsied in a longitudinal way upon treatment, will be crucial to precisely decipher the crosstalk between cancer cells and immune cells. A better understanding of these complex, intertwined interactions, will pave the way to novel combinatory treatments, to rescue resistance to ICB and improve metastatic melanoma patient care.

VB and FB contributed equally to writing and figure preparation. SD provided grant support and critical reading. JC wrote and supervised the whole process. All authors contributed to the article and approved the submitted version.

The team is supported by the Lyon Integrated Research Institute in Cancer (SIRIC LYriCAN INCa-DGOS-Inserm_12563), the Institut Convergence PLAsCAN (ANR-17-CONV-0002), the ERiCAN program of Fondation MSD-Avenir (Reference DS-2018-0015), the Institut National contre le Cancer (INCA-DGOS PRTK Melpredict), ARC (Sign’it grant Birdman), the Ligue Nationale contre le Cancer, the Société Française de Dermatologie (SFD), the association Melarnaud and Vaincre le Mélanome. FB is supported by ITMO Cancer of Aviesan within the framework of the 2021-2030 Cancer Control Strategy, on funds administered by Inserm.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.