Immune checkpoint inhibitors (ICIs) are revolutionizing the treatment of melanoma and other types of cancers, with 7 agents now approved in the US and a rich pipeline of new agents and mechanisms in development ^[1][2][3][4][1,2,3,4]. Notwithstanding the excitement around these developments, there is a significant unmet medical need in the form of patient stratification and therapy resistance. Melanoma affects more than 1 million Americans, and there is an increasing incidence of melanoma worldwide. Approx. 300,000 new cases are diagnosed in the US each year [5], with the average annual cost for treatment estimated at $3.3 billion [6].

Numerous successes have been achieved with anti-PD1, anti-CTLA4, or combination therapies ^[5][7][5,7] in treating melanoma and other types of cancers. The groundbreaking finding by Leach et al. [8] showed that antibodies blocking the T cell co-inhibitory receptor CTLA-4 can augment immune responses against tumor cells in mice. This finding gave rise to ipilimumab, the first ICI to increase the survival of patients with metastatic melanoma ^[9][10][9,10] which was granted FDA approval in 2011 for the treatment of metastatic melanoma.

In 2014, additional T cell immune checkpoint blocking antibodies, anti-PD-1 (pembrolizumab) and anti-PD-L1 (nivolumab) [11], received FDA approval. The combination of anti-PD-1 (pembrolizumab) and anti-PD-L1 (nivolumab) resulted in the augmented survival of patients with metastatic melanoma compared to patients treated with ipilimumab alone or chemotherapy alone ^[12][13][14][12,13,14].

In 2022, relatlimab, which targets lymphocyte-activation gene 3 (LAG-3), was approved by the FDA for adult and pediatric use with metastatic melanoma [15].

Following those developments, high levels of PD-L1 expression in cancer cells and tumor mutational burden (TMB) have been shown in melanoma to correlate with clinical responses to ICI [13].

Despite these data, a substantial number of melanoma patients fail to respond to ICI, leading to premature death ^[16][17][18][16,17,18]. Considerable effort has been made to identify biomarkers that predict clinical response/resistance to ICI [19]. Despite the successes of ICI, even with the combination of ipilimumab and nivolumab, the five-year survival for the intention-to-treat (ITT) population was only 53%. This means that 47% of patients do not reach long-lasting benefits and succumb to the disease [20], suggesting the need for predictive biomarkers of response to overcome ICI resistance. Presently, physicians have no idea which patients will or will not respond to ICI in the absence of predictive biomarkers of response.

2. Melanoma Biomarkers of Response to ICI

2.1. Established Clinical Biomarkers

The earliest approved clinical biomarkers that helped to inform the prognosis of metastatic melanoma relied on baseline clinical characteristics, such as serum levels of lactate dehydrogenase (LDH), a marker of tumor burden ^[21][41]. Elevated serum LDH levels have been demonstrated to be a negative prognostic marker, irrespective of the given treatment ^{[22][23][24][25]}[42,43,44,45]. Generally, high LDH levels are associated with poor overall survival (OS) compared with normal LDH levels. Elevated levels of LDH have been used as a biomarker to assess patient staging ^[26][46]. For example, it has been shown that among ca. 30% of patients with 4–5-year OS following treatment with BRAF and MEK inhibitors, only a few of those patients had high LDH levels before the initiation of therapy ^[27][47]. Likewise, elevated serum levels of S100B have demonstrated to have a prognostic value in both metastatic ^[28][29][48,49] and high-risk resected melanoma settings ^[30][50]. Gene expression profiling has shown to have a prognostic value that complements existing biomarkers in patients with melanoma ^[27][31][32][47,51,52]. In metastatic melanoma, a well-characterized predictive biomarker of response guiding the therapeutic decision process is the BRAF V600 mutation, which is somehow predictive of response to BRAF ± MEK inhibition with low rates of primary resistance ^[33][34][35][53,54,55]. The response rate to BRAF and MEK inhibitors in metastatic melanoma patients with the BRAF V600 mutation is ca. 70% in selected patients, with less than 10% of patients having the highest response to progressive disease ^{[35][36][37][38]}[55,56,57,58]. Several oncogenic driver mutations have been identified as predictive biomarkers of response from targeted agents, including NRAS, NF-1, and c-KIT, that provide insight into the probability of therapeutic response to a specific treatment ^{[39][40][41][42][43]}[59,60,61,62,63]. Presently, the presence of a BRAF V600 mutation is the only validated predictive biomarker for melanoma patients.

2.2. Emerging Predictive Biomarkers

Predictive biomarkers of response include TMB, neoantigen load ^{[7][44][45][46][47][48]}[7,65,66,67,68,69], HLA-I genotype ^[49][50][70,71], cytolytic activity ^[51][72], aneuploidy ^[52][73], and T cell repertoire ^[26][46], which exhibit high predictability to ICI response. Other predictive biomarkers of response are PDL-1 expression, LAG-3 expression, CD8+ T cells at tumor invasive margin, IL-17 expression, immune-related gene expression signatures, and T-cell receptor (TCR) signature. Additional biomarkers that are used to assess ICI response include ctDNA profiles, absolute lymphocyte count, proliferating CD8+ T cells, increase of T-cell subsets and checkpoint molecules (PDL-1, LAG-3), granzyme B expression, and TCR signature ^{[53][54][55][56]}[37,38,39,40]. Unfortunately, many of those potential biomarkers of ICI response have not yet been validated ^{[57][58][59][60][61]}[74,75,76,77,78]. A recent study by Carter et al. ^[57][74] questioned the validity of the immuno-predictive score (IMPRES), a predictor of ICI response in melanoma consisting of 15 pairwise transcriptomic signatures that analyze the relationship between immune checkpoint genes reported by Auslander et al. ^[59][76]. The IMPRES is context-dependent and could not reproducibly predict ICI response in the context of metastatic melanoma ^[59][76]. Moreover, Xiao et al. ^[60][77] questioned the reproducibility of the Immune Cells.Sig ^[62][79] signature in melanoma, demonstrating inconsistencies in the prediction capability of ImmuneCells.Sig across different RNA-seq datasets ^[60][77]. The performance of the ImmuneCells.Sig signature in predicting ICI outcomes in four melanoma patient datasets, using the same implementation scheme as Xiong et al. ^[62][79], showed that there were inconsistencies across different datasets ^[60][77].

2.2.1. Gene Expression Signatures

Gene expression signatures (GES) have also been identified as predictive biomarkers of response to ICI and have been validated in several independent datasets (e.g., immune-predictive score, IMPRES consisting of 15 immune genes). IFN-γ-responsive genes were also used to predict ICI response in metastatic melanoma ^{[26][51][52][59][63][64][65][66][67][68][69][70]}[30,46,72,73,76,80,81,82,83,84,85,86]. A panel of pan-tumor T cell-inflamed GES consisting of 18 IFN-γ-responsive genes was validated and confirmed to predict the response to ICI in pre-treatment tumor specimens from nine types of cancers, including melanoma ^[67][83]. MHC-I/II gene signatures have also been explored as predictive biomarkers of ICI response in melanoma ^[69][71][72][85,87,88]. Carter et al. ^[57][74] reported that immuno-predictive score (IMPRES), a predictor of ICI response in melanoma encompassing 15 pairwise transcriptomics relations between immune checkpoint gene ^[59][76], did not reproducibly predict the response to ICI in metastatic melanoma. It was argued that many factors may contribute to the limited successes of those biomarkers, such as: (1) the predictive biomarkers have been derived from pre-clinical studies; (2) evaluation of the biomarkers in clinical specimens only included baseline biopsies and peripheral blood samples; (3) batch effect, lack of reproducibility might have contributed to the failure of the biomarkers for ICI response ^{[57][58][60][61]}[74,75,77,78]. To address these issues, numerous researchers have developed predictive biomarkers to reduce batch noise and other technical issues. Expanded predictive biomarker panels have resulted in higher reproducibility as opposed to predictive signatures based on individual biomarkers ^{[73][74][75][76]}[89,90,91,92]. Tian et al. ^[75][91] reported that the combined BRAF, KRAS, and PI3KCA mutation signature resulted in a favorable predictive response to cetuximab for patients with colorectal cancer ^[75][91].

2.2.2. Gene Expression Signatures at Baseline and On-Treatment Tumor Specimens

Genome-wide analysis of transcriptomic and genomic profiles of baseline and on-therapy tumor specimens from patients treated with ICI provides a comprehensive view into the mechanisms underlying tumor response and resistance to ICI ^[77][93]. Grasso et al. ^[77][93] reported that the mechanism of action of ICI is based on the interaction between immune effector cells and cancer cell targets. Tumor studies conducting comprehensive analyses of transcriptomic and genomic profiling have focused not only on the genetic alterations and gene expression profiles of cancer cells ^{[63][69][72][78]}[25,30,85,88], but also on the composition of immune infiltrates and expression of immune-activating gene programs ^{[26][46][58][64][66][67][69][70][72][79][80][81][82][83][84]}[36,46,67,75,80,82,83,85,86,88,94,95,96,97,98]. Du et al. ^[85][64] reported that pathway-based signatures derived from on-treatment tumor specimens were predictive of the response to anti-PD1 blockade in patients with metastatic melanoma. Other studies of breast cancer suggested that post-treatment tumor samples were more informative than pre-treatment samples ^[86][87][88][99,100,101]. Conversely, Wallin et al. developed adaptive immune signatures based on tumor samples obtained during the early course of treatment, showing that the signatures were highly predictive of the response to ICI in patients with metastatic melanoma ^[89][102]. Auslander et al. built an immune-predictive score (IMPRES), which encompasses 15 pairwise transcriptomic relations between immune checkpoint genes, to predict the response of metastatic melanoma to ICB therapy ^[59][76]. The IMPRES signature produced better predictive scores with post-treatment samples than with pre-treatment samples in two independent datasets ^[59][76]. In support of these findings, a recent proteome profiling study of samples from patients with metastatic melanoma undergoing either tumor infiltrating lymphocyte based or anti-PD1 immunotherapy demonstrated that the fatty acid oxidation pathway was significantly enriched in responder patients. These results underlined the critical role of mitochondrial metabolism, including fatty acid metabolism, in conferring response to immunotherapy ^[90][103]. It is now widely accepted that post-treatment tumor specimens are generally much more informative than pre-treatment specimens and may provide more valuable insight into dynamic changes at the transcriptional level that correlate with clinical response, resulting in a higher predictive score. In conclusion, although pathway signatures derived from post-treatment samples are highly predictive of therapeutic response to anti-PD1 in patients with metastatic melanoma, further studies are warranted to confirm the predictive value of those signatures in larger cohorts of patients with metastatic melanoma.

2.2.3. Pathway Signatures

Du et al. ^[85][64] developed pathway-specific signatures in pre-treatment (PASS-PRE) and on-treatment (PASS-ON) tumor specimens based on transcriptomic data and clinical information from a large dataset of metastatic melanoma patients treated with anti-PD1. Both PASS-PRE and PASS-ON signatures were validated in three independent datasets of metastatic melanoma. Compared to existing molecular signatures, it was concluded that the on-treatment (PASS-ON) tumor specimen signature exhibited a robust and better predictive value for metastatic melanoma patients who responded to anti-PD1 across all four datasets. The pre-treatment pathway signatures included six pathways for predicting the response to anti-PD1 treatment, including: (1) complement cascade; (2) regulation of insulin-like growth factor IGF transport and uptake by insulin-like growth factor binding proteins IGFBPS; (3) binding and uptake of ligands by scavenger receptors; (4) plasma lipoprotein remodeling; (5) IL2 family signaling; and (6) retinoic acid (RA) biosynthesis pathways. Complement cascade, binding, and uptake of ligands by scavenger receptors and IL2 family signaling pathways are related to immune and inflammation, whereas plasma lipoprotein remodeling and the RA biosynthesis pathways are related to metabolism. In contrast to the pathway-based signature analysis of on-treatment samples, Du et al. ^[85][64] identified four pathways, including: (1) peroxisomal lipid metabolism; (2) generation of second messenger molecules; (3) fatty acid metabolism; and (4) PD1 signaling. Of note, peroxisomal lipid metabolism and fatty acid metabolism are related to fatty acid and lipid metabolism ^[90][103]. Generation of second messenger molecules is a central signaling pathway in T-cell receptor (TCR) stimulation. Likewise, PD1 signaling plays an important role in immunoregulation as an immunoregulatory signaling pathway. To further validate the predictive performance of a pathway-based super signature for on-treatment samples (PASS-ON), Du et al. ^[85][64] tested three independent datasets with RNA-seq data available for on-treatment samples, including those of Gide et al. and Lee et al. ^[82][91][96,104], and the MGH cohort demonstrated the effectiveness of PASS-ON in predicting patient response to anti-PD1. Patients with high PASS-ON signature scores were associated with significantly improved PFS compared to those with low signature scores in all tested patients. Furthermore, Du et al. ^[85][64] demonstrated that the time-response interaction pathway-based super signature for pre- and on-treatment samples had reasonable predictive power. The study suggested that pathway-based biomarker signatures derived from on-treatment tumor specimens compared to pretreatment tumor specimens were better predictors of response to anti-PD1 therapies in metastatic melanoma patients.

2.2.4. Tumor Antigens

Huang et al. ^[92][22] investigated several melanoma-relevant tumor-specific antigens, cancer germline genes, melanocyte differentiation antigens, overexpressed antigens, neoantigens, neuropeptides, and other sources of immunogenic antigens, such as immunogenic epitopes as novel predictive biomarkers of ICI response to melanoma. Neoantigens are derived from tumor-specific somatic mutations and are exclusively expressed in cancer cells and absent in normal human tissue. The majority (95%) of somatic mutations are single-nucleotide variants (SNVs), which lead to aberrant protein and peptide expression with single amino acid substitutions ^[93][105]. Neopeptides also arise from nucleotide insertions or deletions (indels), leading to the expression of aberrant proteins and peptides with frameshift or non-frameshift sequences depending on the number of nucleotides added and deleted. While the minority of mutations are indels (<5% for melanoma) ^[94][95][106,107], frameshift mutations can generate a number of immunogenic neoepitopes that are highly distinct from the self. Other sources of immunogenic antigens, including immunogenic epitopes, can also derive from mutations associated with gene fusion, aberrant messenger-RNA splicing with retained introns, or aberrant translation resulting in cryptic antigens, and genomically integrated endogenous retroviral sequences as a result of previous retroviral infections, although they are epigenetically silenced, can be reactivated in tumors ^[94][106], as in the case of cancer germline antigens. Furthermore, tumors often present aberrant patterns of DNA methylation, resulting in the demethylation, ectopic expression, and presentation of cancer germline genes to T cells relevant in immune recognition ^[94][96][106,108]. For example, cancer germline genes such as MAGEA1 and NY-ESO-1 are silenced epigenetically through methylation in human tissue, with the exception of male germ cells and trophoblastic cells, which lack MHC-I molecules. PRAME (preferentially expressed antigen in melanoma), a member of the cancer-testis antigen family, has been reported to be frequently overexpressed in many cancers, including melanoma, which indicates advanced cancer stages and poor clinical prognosis ^[97][98][109,110]. As such, overexpressed PRAME is a potential immunotherapy target. PRAME-specific immunotherapies are currently in development for many cancers, including melanoma. For example, a recent study demonstrated that uterine carcinosarcoma, synovial sarcoma, and leiomyosarcoma patients would potentially benefit from PRAME-specific immunotherapies ^[97][109].

2.2.5. Genomic Alterations

Considerable effort has been made to identify genomic alterations and transcriptome profiles as predictive biomarkers of ICI response. Numerous studies have identified distinct stages of CD8+ T cells linked to positive response or failure to ICI treatment ^[84][98]. Moreover, tumors from patients responding to ICI showed a higher number of cancer-associated somatic mutations (i.e., mutated antigens or neoantigens) targeted by T cells ^[44][65]. The IFN-γ signature has also been shown to predict the response to ICI (e.g., anti-PD-1) in melanoma ^[99][111] and in other types of cancers ^[6][77][6,93]. Gene expression signatures obtained from bulk melanoma tumor or single-cell profiling and the TME have been shown to be correlated with sensitivity and resistance to several ICIs ^{[70][72][84][100][101][102]}[86,88,98,112,113,114]. Collectively, the gene expression signatures associated with response to ICI in metastatic melanoma represent distinct characteristics and play an important function in different signaling pathways, including the inflammatory response, type I interferon signaling pathway, cytokines, and others.

3. Who Is Responding to ICI?

Recent data have demonstrated that immunotherapies against immune checkpoints (e.g., CTLA-4 or PD-1) downregulate two main negative regulators of the anti-tumor immune response ^{[77][103][104][105]}[93,115,116,117], resulting in durable anti-tumor responses in a subset of cancer patients, including those with melanoma ^[2][106][2,118]. Another key factor contributing to anti-tumor immune response following ICI treatment ^[77][93] is the pre-existing level of T cell infiltration of the tumor ^{[107][108][109]}[119,120,121], representing the immunogenicity of the cancer cells. Analysis of tumor biopsies from ICI-treated patients showed that clinical responses associated with ICI were mediated by tumor-infiltrating T cells reactivated following ICI treatment ^[109][110][121,122]. A combination of immunohistochemical (IHC) analysis with RNA-seq performed on cancer biopsies from patients treated with anti-CTLA-4 antibody (ipilimumab) before or after treatment with anti-PD-1 antibody (nivolumab) demonstrated that a major response to anti-CTLA-4 requires cancer cells with high levels of MHC-I expression at baseline, whereas the response to anti-PD-1 was more strongly associated with a pre-existing interferon-γ gene expression signature ^[72][88]. Liu et al. demonstrated that the response to anti-PD-1 therapy (with or without prior anti-CTLA-4 treatment) was associated with increased MHC-I and MHC-II expression ^[69][85]. Biopsies from patients with metastatic melanoma treated with anti-PD-1 monotherapy (nivolumab) in part 1 of the CheckMate 038 study showed an increase in immune cell subtypes with elevated immune activation gene signatures seen in responders to therapy ^[78][25]. The transcriptome analysis of tumor biopsies from patients treated with anti-PD-1 monotherapy (nivolumab) or in combination anti-PD1 plus anti-CTLA-4 therapy (ipilimumab) correlated well with the in vitro analysis of gene expression signatures of melanoma cell lines following exposure to interferon-γ ^[77][93]. It appears that cancer cells become enablers of the immune response via the expression of IFN-γ-response genes, triggering the upregulation of antigen presentation, amplification of the interferon response, and induction of chemokines (i.e., CXCL9 and CXCL10) to entice immune cells to the TME. Thus, T cell-induced IFN-γ correlates with ICI therapy response ^[77][93]. Collectively, the degree of the anti-tumor T cell response and downstream IFN-γ signaling are the main drivers of response or resistance to ICI therapy ^[77][93]. However, little is known about how tumor-intrinsic loss of IFN-γ signaling impacts TILs. The question remains whether tumor-intrinsic IFN-γ signaling actively regulates the infiltration or function of TILs? Shen et al. ^[111][123] demonstrated that IFN-γR1 knockout melanomas and IFN-γR1 knockout melanomas in B6 mice had reduced infiltration and function of tumor-infiltrating lymphocytes (TILs). Furthermore, long-distance effects of IFN-γ on tumor cells also play a crucial role in anti-tumor immunity ^[111][123]. These recent findings revealed an important role of tumor-intrinsic IFN-γ signaling and IFN-γ-response genes in shaping TILs.

4. Therapeutic Strategies to Turn “Cold Tumors” into “Hot Tumors”

Several strategies have been investigated to elucidate the mechanisms underlying how T cells are driven into “hot tumors” in order to improve the efficacy of ICI therapy ^[112][124]. Several clinical trials have tested these novel therapeutic modalities as interventions in combination with ICI to overcome ICI monotherapy resistance and attempt to turn “cold tumors’ into “hot tumors”. Grasso et al. ^[77][93] showed that a robust anti-tumor immune response relies on the interplay of key factors that can be modulated with innovative interventions. Recent preclinical and clinical findings have provided insight into the immunological implications of canonical cancer signaling pathways (e.g., WNT-beta-catenin signaling, cell cycle regulatory signaling, mitogen-activated protein kinase signaling, and pathways activated by loss of the tumor suppressor phosphoinositide phosphatase PTEN), thus providing new opportunities for the development of new treatments for those patients who do not respond to ICI monotherapies ^[113][218]. Combined therapeutic strategies from preclinical ^{[103][105][114]}[115,117,219] and clinical studies [14] have shown that anti-PD-1 plus anti-CTLA-4 (nivolumab and ipilimumab) treatments elicit stronger immune stimulation in stage IV melanoma than monotherapy alone, resulting in a favorable anti-tumor immune response. Response to ICI, either following anti-CTLA-4 monotherapy or in combination with anti-PD-1 therapy, triggers a robust T cell response that generated an appreciable antitumor response ^[113][218]. Another approach to turn “cold tumors” into “hot tumors” is through the intra-tumoral delivery of oncolytic viruses or Toll-like receptor agonists capable of inducing intra-tumoral interferon production, which triggers the pattern recognition pathways with consequent boosting of the anti-PD1 immune response rate ^{[115][116][117]}[187,190,220]. Recently, different approaches have been used to boost the response to ICI. Activation of the STING pathway ^[118][221], inhibition of immune suppressive factors (e.g., WNT signaling or the adenosine pathway) ^{[99][119][120][121]}[111,209,222,223], as well as the release of other immune checkpoints (e.g., LAG-3, TIM-3, or TIGIT, etc.) in T cells ^[99][111] have been explored, but no favorable clinical outcomes have yet been reported in patients. Altogether, these new therapeutic strategies provide new opportunities for cancer immunotherapy for patients who do not respond to ICI ^[99][111].

5. Lessons Learned: ICI Therapies in Melanoma

Huang et al. ^[92][22] reported that several factors, including the immune TME, tumor-associated immune cells, and different host factors, contribute to the ICI resistance. Melanoma resistance has armed people with a handful of information that can be applied to other types of cancers, such as: (1) the ability to present cancer antigens through MHC-I and elevated TMB; (2) tumor-antigen-specific T cells play a crucial role in the response to ICI; (3) reactivation of terminally exhausted T cells could be considered a biomarker for PD-1 blockade, which is detectable as early as one week after ICI dosing; (4) melanoma immunosuppressive mechanisms are complicated and need additional research to remodel their interaction, cooperation, and dynamics during tumor progression and in immunotherapy resistance; (5) Treg cells are emerging as a key mechanism of resistance to PD-1 blockade, but not necessarily CTLA-4 blockade; (6) emerging neoadjuvant immunotherapy trials are anticipated to provide new insight into pharmacodynamic immune responses and advance the development of rational immunotherapy and neoadjuvant combination regimens while avoiding toxicity and significantly improving patient management; (7) longitudinal assessment of pre- and on-treatment patient specimens is required to determine prognostic vs. predictive use of immune and other parameters, including genomic parameters correlating with patient outcomes, and deduce their biologic role in response to ICI therapy based on their modulation during treatment; and last but not least (8) melanoma-specific oncogenic programs supporting metabolic plasticity and fitness, together with clinical and preclinical evidence of differential activity of ICI therapy depending on the tumor metabolic state, should provide new research opportunities to evaluate these relationships as potential biomarkers for patient stratification and treatment allocation, and formulate novel precision-medicine combinations depending on metabolic and immune therapies.