The Role of Immune Checkpoint Blockade in Uveal Melanoma

Uveal melanoma (UM) represents the most common intraocular malignancy in adults and accounts for about 5% of all melanomas. Primary disease can be effectively controlled by several local therapy options, but UM has a high potential for metastatic spread, especially to the liver. Despite its clinical and genetic heterogeneity, therapy of metastatic UM has largely been adopted from cutaneous melanoma (CM) with discouraging results until now. The introduction of antibodies targeting CTLA-4 and PD-1 for immune checkpoint blockade (ICB) has revolutionized the field of cancer therapy and has achieved pioneering results in metastatic CM. Thus, expectations were high that patients with metastatic UM would also benefit from these new therapy options. This review provides a comprehensive and up-to-date overview on the role of ICB in UM. We give a summary of UM biology, its clinical features, and how it differs from CM. The results of several studies that have been investigating ICB in metastatic UM are presented. We discuss possible reasons for the lack of efficacy of ICB in UM compared to CM, highlight the pitfalls of ICB in this cancer entity, and explain why other immune-modulating therapies could still be an option for future UM therapies.


Introduction
Uveal melanoma (UM) represents the most common ocular malignancy in adults. About 85% to 90% of the primary tumors arise from melanocytes residing in the choroid, the pigmented layer of the eyeball that also contains blood vessels [1,2], and less frequently from the ciliary body or iris of the eye where melanocytes are also present [3]. With an incidence of 4-7 cases per million in Europe, it is much rarer than cutaneous melanoma (CM) [4]. Several factors have been associated with an increased risk of developing UM. These include the presence of choroidal or cutaneous nevi [5], fair skin, light eye color [6], oculodermal melanocytosis [7,8], inactivating mutations of the tumor-suppressor BRCA1-associated protein 1 (BAP1) [9,10], and exposure to ultraviolet (UV) radiation [11].

Cutaneous vs. Uveal Melanoma: Genetic Differences and Implications for Prognosis
Although UM and CM both arise from melanocytes and share similar risk factors, these melanoma subtypes substantially differ biologically and clinically [5]. Similar to other cancer entities that are associated with the exposure to environmental carcinogens such as nonsmall cell lung cancer (NSCLC),

Immune Checkpoints in Cancer Therapy
The immune response is a delicate process that needs to be balanced carefully in order to fulfill its function in the defense against pathogens on the one hand and to avoid massive damage of healthy tissue on the other hand. Several immune checkpoints help to control autoimmunity and maintain self-tolerance. Cytotoxic CD8+ T lymphocytes (CTLs) are able to identify and destroy infected cells as well as tumor cells by recognizing epitopes of antigens that are presented by human leukocyte antigen (HLA) class I molecules on the cell surface and thereby play an essential role in maintaining the integrity of the body [81].
The anti-CTLA-4 monoclonal antibody ipilimumab was the first immune checkpoint inhibitor approved for therapy of metastatic CM by the US Food and Drug Administration (FDA) in 2011 [82]. CTLA-4 is a critical negative regulator of the T cell response. In general, T cell activation requires the transduction of two signals [83]: Signal 1 involves the interaction of a foreign peptide or mutated self-antigen bound to the major histocompatibility complex (MHC) on antigen-presenting cells (APCs) and the T cell receptor (TCR) on T cells. Additionally, a second antigen-independent signal has to be transduced by costimulatory B7 proteins CD80 (B7.1) and CD86 (B7.2) located on APCs that interact with the co-receptor CD28 on T cells [83]. Only both signals can fully activate T cells, which includes their clonal expansion and the expression of effector cytokines. CTLA-4 is expressed on both activated T cells and regulatory T cells (T regs ). It is constitutively internalized; thus, about 90% of all CTLA-4 is found in intracellular vesicles unless its surface expression is upregulated due to stimulatory signals [84]. CTLA-4 binds to the costimulatory ligands CD80 and CD86 with a higher affinity than CD28 and thereby prevents their interaction with CD28 and subsequent T cell activation [85]. Additionally, this interaction also increases the expression of indoleamine 2,3-dioxygenase (IDO) in dendritic cells (DCs) [86]. IDO mediates immunosuppressive effects by converting the amino acid tryptophan to kynurenine, a metabolite that directly induces apoptosis in T cells. Furthermore, IDO-mediated tryptophan depletion impairs T cell function by leading to cell cycle arrest and anergy [87]. CTLA-4 binding to its ligands also leads to the removal of CD80/86 from the APC surface by a mechanism termed trans-endocytosis and further impairs T cell activation [88]. Thus, the number of activated T cells is controlled. CTLA-4 plays an important role in the maintenance of self-tolerance and the control of autoreactive T cells, as shown in studies with knockout mice [89]. The inhibitory effect of CTLA-4 can be overcome either by upregulating CD80/86 on APCs-which occurs physiologically in response to inflammatory stimuli in vivo-or therapeutically by administering antibodies targeting CTLA-4. These prevent the interaction with CD80/86 on APCs and additionally lead to the elimination of immunosuppressive T regs via both direct cytotoxicity and antibody-dependent cell cytotoxicity in the tumor [85] and thereby boost T cell activation and the initiation of effector T cells ( Figure 1A).
A second immune checkpoint that is exploited for cancer therapy involves the PD-1-PD-L1 axis. The interaction of the receptor PD-1 and its ligands programmed death ligand 1 (PD-L1) and programmed death ligand 2 (PD-L2) is an important mechanism to avoid tissue damage from autoreactive T cells and maintain the peripheral tolerance [90]. PD-L1 is constitutively expressed in low levels on the cell surface of nonhematopoietic cells as well as on T cells, B cells, macrophages, and DCs [90,91]. Mediated by proinflammatory cytokines such as interferon gamma (IFN-γ) and tumor necrosis factor-α (TNF-α) that are released by activated T cells, its expression is upregulated during inflammatory processes. PD-L1 binds to its receptor PD-1, which is widely expressed on T cells, B cells, and some myeloid cells [90]. When the TCR is simultaneously bound, the interaction of PD-L1 and PD-1 leads to the transduction of an inhibitory signal that blocks TCR signaling [91] by reducing PTEN phosphorylation, phosphoinositide 3-kinase (PI3K) inhibition [92], and MEK-ERK pathway activation [93]. This results in diminished T cell function due to reduced production of the T cell effector cytokines IFN-γ, TNF-α, and interleukin-2 (IL-2) and a decreased T cell proliferation and survival [94][95][96]. Thus, the PD-1-PD-L1 axis limits an extensive and too-long-lasting immune response that may damage healthy tissue. The second PD-1 ligand PD-L2 has been less extensively investigated yet. Its expression is limited to macrophages, DCs, mast cells, and B cells and can be induced by IL-4; however, the transcriptional regulation has not been fully elucidated yet [97]. In contrast to PD-L1, binding of PD-1 to PD-L2 suppresses the activation of T cells in lymphoid organs [98]. been detected in various tumor types on tumor cells, invading immune cells and endothelial cells [103]. Antibodies against PD-1 or its ligand PD-L1 are able to inhibit the PD-1-mediated transduction of inhibitory signals and prevent the inactivation of tumor-reactive immune cells. This strategy has been used successfully for the treatment of a variety of tumors including CM [77] ( Figure 1B).  (a) CTLA-4 is a critical negative regulator of the T cell response in the early activation phase of the adaptive immune response. It binds to the costimulatory ligands CD80 and CD86 on antigen-presenting cells with a higher affinity than CD28 and thereby prevents their interaction with CD28 and subsequent T cell activation. Anti-CTLA-4 antibodies block the interaction of CTLA-4 and CD80/86 and boost T cell activation and the anti-tumor response.
(b) The PD-1-PD-L1 axis is an important mechanism to avoid tissue damage from autoreactive T cells and maintains the peripheral tolerance. Binding of PD-L1 to its receptor PD-1 blocks T cell receptor (TCR) signaling, resulting in limited T cell function. Antibodies targeting PD-1 or its ligand PD-L1 are able to inhibit their interaction and prevent the inactivation of tumor-reactive immune cells.
Tumor cells are capable of escaping T cells by exploiting the PD-1-PD-L1/PD-L2 axis to dampen the antitumoral immune response. PD-L1 can be expressed in high levels and on the majority of both tumor cells and tumor-infiltrating immune cells in several cancer entities like CM [77,99], NSCLC [100], epithelial ovarian cancer [101], and breast cancer [102]. Similarly, PD-L2 expression has also been detected in various tumor types on tumor cells, invading immune cells and endothelial cells [103]. Antibodies against PD-1 or its ligand PD-L1 are able to inhibit the PD-1-mediated transduction of inhibitory signals and prevent the inactivation of tumor-reactive immune cells. This strategy has been used successfully for the treatment of a variety of tumors including CM [77] ( Figure 1B).

Studies Investigating Immune Checkpoint Blockade for UM Treatment
In the past, UM patients have been excluded from most trials investigating ICB for CM treatment. However, some prospective phase I and II trials have also included UM patients. An up-to-date summary of the most important pro-and retrospective analyses, dosage schemes, and study results are presented in the following overview (Table 1). All studies investigated ICB in patients with metastatic melanoma; exception: Fountain 2019/ NCT01585194 (adjuvant ICB in patients with a high risk for developing metastases). *: Authors did not distinguish between grade 3 and 4 when reporting the severity of AEs. Abbreviations: CI = confidence interval; DCR = disease control rate; EAP = expanded access program; ICB = immune checkpoint blockade; ipi = ipilimumab; mo. = months; nivo = nivolumab; NPP = named patient program; n.r. = not reported; pembro = pembrolizumab; RFA = radiofrequency ablation; TAC = transarterial chemotherapy; TACE = transarterial chemoembolization; wks. = weeks.

Ipilimumab
The anti-CTLA-4 monoclonal antibody ipilimumab was the first immune checkpoint inhibitor approved for therapy of metastatic CM by the FDA in 2011 [82].
The prospective investigator-initiated SECIRA-UM trial investigated different dosages of ipilimumab combined with radiofrequency ablation (RFA) in 41 patients with UM and at least two liver metastases in a phase Ib/II trial [104]. No complete responses (CR) or partial responses (PR) were observed. Severe adverse events (AEs) were reported in 52% of the patients who had received the highest dosage (10 mg/kg), most frequently immune-related colitis. The frequency of grade 3 and 4 AEs was associated with a higher dosage of ipilimumab [104].
Five additional prospective studies with 185 patients in total investigated ipilimumab at a dose of 3 mg/kg body weight [105][106][107][108][109]. In the study published by Jung et al. [109], 10 UM patients (total n = 106 melanoma patients) were treated. The median PFS was 2.8 months (graphically determined based on the published time-to-event analyses), and it was unclear whether the median OS had been reached at the time of publication. Secondary outcomes were not reported; furthermore, the UM subgroup was poorly described [109]. In the other prospective studies that investigated this dosage scheme (3 mg/kg), overall response rates (ORR) ranged from 0% [108] to a sobering 4.8% [107]. The median PFS and OS were quite short, with a median PFS ranging from 2.8 months [108] to 3.6 months [107] and a median OS between 5.2 months [106] and 6.8 months [108]. The frequency of severe treatment-related AEs differed between the studies and ranged between 6% [107] and 36% [108].
Higher ipilimumab dosages of 10 mg/kg were investigated in three pro-and one retrospective analysis [110][111][112][113]. Interestingly, a prolonged median OS of up to 9.8 months was observed; however, the ORR did not differ from other trials investigating lower dosages. Median PFS was not determined in any of these four studies. Similar to the SECIRA-UM trial [104], the high dosage of ipilimumab was also associated with a higher number of severe AEs. Nine of the 10 study participants developed grade 3 AEs, and in one patient who participated in an adjuvant study, temporal arteritis resulted in persistent blindness [112].
Itchins et al. [114] reported the results of a single-center cohort analysis on 37 metastatic UM patients who received a sequential therapy with transarterial chemotherapy (TAC), ICB (predominantly ipilimumab), and a systemic chemotherapy. Although the ORR was comparably poor, this combined approach led to the longest median PFS (9 months) and OS (17 months) among all therapies investigating ipilimumab. However, interpreting these results may be problematic, as the authors did not distinguish between different ICB antibodies when reporting the outcomes [114].
Bol et al. [115] compared the efficacy of ipilimumab, pembrolizumab, and a combined therapy with ipilimumab and nivolumab in a retrospective population-based study including 86 patients. None of the 24 patients treated with ipilimumab responded to the therapy. The median PFS was 3.0 months and the median OS was 9.9 months [115].

Tremelimumab
Tremelimumab (CP 675206) is another monoclonal CTLA-4-targeting antibody. In a multicenter phase II trial published by Joshua et al. [116], 11 UM patients received 15 mg/kg tremelimumab. None of the patients achieved an objective response (CR or PR). The median PFS was 2.9 and the median OS was 12.8 months. Secondary outcomes were not clearly reported. The poor results regarding PFS and radiological response led to the withdrawal of the trial at the first interim stage [116].
Taken together, the anti-CTLA-4 antibodies ipilimumab and tremelimumab showed a poor efficacy on the one hand with poor response rates and short PFS and OS, and its use in an adjuvant setting also achieved only discouraging results. On the other hand, a significant number of patients developed severe AEs. These results highlight the low efficacy and high toxicity of these antibodies for the treatment of metastatic UM.
In the prospective CheckMate 172 trial [117], a subgroup of 75 UM patients with progressive disease after anti-CTLA-4 therapy was included, 34 of whom were evaluable for outcome analysis. Two PR were achieved in this subgroup(ORR 6%), and 15 patients (44%) had stable disease (SD) after a minimum follow-up of 1 year. The median OS was 11 months, thus better than the median OS for ipilimumab in this group of patients. Data on the UM subgroup regarding PFS or severe AEs were not reported [117].
Three other reports retrospectively evaluated the efficacy of nivolumab in a rather low number of patients [118][119][120]. ORR ranged from 0% [118] up to 25% [119]. The median PFS was not reported in [119] and was 2.3 months in the other studies, and the median OS ranged from 9.6 months [118] to about 14 months [120]. Remarkably, the studies reported by Tian et al. [119] and van der Kooij et al. [118] also included patients receiving pembrolizumab; however, the authors did not distinguish between both interventions when reporting the outcomes. In general, the patients tolerated the treatment well. Severe AEs were only reported for one patient who developed a grade 4 hyperglycemia due to ICB-related diabetes [120].
Altogether, in comparison with ipilimumab, patients receiving nivolumab achieved a similar median PFS but a longer median OS. The response rates varied from 0% up to a promising 25% in the retrospective analyses, whereas the ORR in the phase II trial was 5.8%, which is comparable to the results achieved with ipilimumab. However, the safety of nivolumab seems to be better than that of ipilimumab.

Pembrolizumab
The efficacy of pembrolizumab for the treatment of metastatic UM was evaluated in several expanded access programs (EAPs), retrospective analyses, and one phase II trial [115,[121][122][123][124][125]. Kottschade [123]. One patient had a CR after receiving one dose, and two patients had a prolonged SD (20% response rate, 60% clinical benefit rate). However, these three patients had either no liver metastases or only small metastases. The median PFS was 11.0 months, which was almost three times longer than those achieved in the aforementioned EAPs. The median OS was not reached. Serious AEs > grade 3 were reported for one patient [123]. Rossi et al. [126] published the results of a prospective observational single-arm study evaluating the efficacy of pembrolizumab in 17 patients with metastatic UM. ORR was 11.7%, but no CR was achieved. The median PFS was 3.8 months, which is comparable to the results reported by Kottschade and Karydis [121,122]. Interestingly, patients with both liver and extrahepatic metastases had a longer median PFS than those with hepatic metastases only (8.4 vs. 3.1 months). No severe AEs > grade 3 were observed [126].
Additionally, three retrospective studies analyzed the efficacy of pembrolizumab and other anti-PD-1 and anti-PD-L1 antibodies [115,124,125]. In the publication of Bol et al. [115], the outcomes for each antibody were separately reported, whereas the other two publications did not clearly distinguish between the interventions. Bol et al. [115] compared the efficacy of pembrolizumab, ipilimumab, and a combined therapy with ipilimumab and nivolumab in a retrospective population-based study including 86 patients. ORR for pembrolizumab was 7%. The median PFS was 4.8 months and the median OS was 10.3 months [115]. Algazi et al. [124] published the results of a multicenter analysis including 56 metastatic UM patients who received either pembrolizumab, nivolumab or the anti-PD-L1 antibody atezolizumab. The median PFS was 2.6 months and the median OS was 7.7 months; however, the authors did not clearly distinguish between the interventions when reporting the outcomes. A PR was observed only in two patients who had received either nivolumab or pembrolizumab [124]. A second retrospective analysis published by Piperno-Neumann et al. [125] described the results of 21 metastatic UM patients who were treated with pembrolizumab or nivolumab. The median PFS was 3 months and thus similar to the one reported by Algazi et al. [125]. In general, the therapy was well tolerated by the patients as no serious AEs > grade 3 were reported in any of the three retrospective trials.
In summary, PD-1 blocking antibodies showed a better clinical efficacy in patients with metastatic UM regarding the median OS and a more favorable toxicity profile compared to CTLA-4 blocking antibodies. The ORR achieved with nivolumab and pembrolizumab varied extremely in different trials, which makes it difficult to evaluate the efficacy.

Combination Therapies/Combined ICB: CTLA-4 and PD-1 Blockade
Several studies have evaluated the effect of combined ICB in UM. A combination of nivolumab and ipilimumab was assessed in an EAP reported by Shoushtari et al. [127]. Sixty-four patients with metastatic melanoma, including a UM subgroup of six patients, were included. No objective responses were achieved in the UM subgroup. The median PFS for UM patients was 2.8 months. No data on OS or serious AEs were specifically reported for the UM subgroup [127]. Piulats et al. reported data of an interim analysis of the prospective phase II GEM1402 trial (NCT02626962) [128]. The ORR was 12%, the median PFS was 3.3 months, and median OS was 12.7 months. A high number of patients (54%) developed grade 3 and 4 AEs, and nine patients discontinued the treatment [128].
Besides, four retrospective analyses investigated the efficacy of combined treatment approaches. One analysis published by Heppt et al. [129] compared the efficacy of PD-1 monotherapy using pembrolizumab or nivolumab with a combined anti-PD-1 and anti-CTLA-4 therapy in 101 UM patients. The median PFS were 3.1, 2.8, and 2.8 months and the median OS were 14 months, 10 months and not reached for pembrolizumab, nivolumab, and combined ICB, respectively. The ORRs were 3% and 5.7% for nivolumab and pembrolizumab monotherapy, respectively, in contrast to 16.7% for the combinational approach; however, no CR was observed in any treatment subgroup. Severe AEs ≥ grade 3 were more frequent in the treatment arm receiving the combination therapy compared to those receiving anti-PD-1 monotherapy (26.7% vs. 7.4% or 12.5%) [129]. Another multicenter analysis evaluated the efficacy of two ICB treatment combinations in 64 metastatic UM patients [130]. The patients received nivolumab or pembrolizumab plus ipilimumab. The authors did not distinguish between both combinations when reporting the results. The ORR was 15.6%, the median PFS was 3.0 months, and the median OS was estimated to be 16.1 months. Similar response rates were reported by Piulats et al. 2018 [128] and Heppt et al. 2017 [129] for combined ICB in this cancer entity as mentioned above. A high number of patients developed severe ICB-related AEs of grade 3 (24 patients, 37.5%) and grade 4 (1 patient, 1.6%) [130]. Furthermore, Karivedu et al. [131] published a retrospective analysis of eight metastatic UM patients who were treated with transarterial chemoembolization (TACE) and a combined ICB with nivolumab and ipilimumab. The median OS was 14 months and the median PFS was not reported. Two patients (25%) achieved a PR. Four patients developed treatment-related colitis; however, the severity of the AEs was not clearly reported. [131]. In a retrospective population-based study, Bol et al. [115] evaluated the efficacy of a combined therapy with ipilimumab and nivolumab in 19 patients. The combined approach led to an ORR of 21.1%. The median PFS was 3.7 months and the median OS was 18.9 months [115].
Taken together, the longest median OS was achieved with combined ICB. The median PFS was around 3 months and comparable among the different studies and antibodies. The longest median PFS of 9 months was achieved in a study investigating a treatment approach combining ICB, TAC, and systemic chemotherapy [114]. Throughout the studies, most of the responding patients achieved only a PR. Regarding safety, ipilimumab showed the most unfavorable toxicity profile as a high number of patients developed severe AEs > grade 3 but did not clinically benefit from this treatment. Thus, a combined ICB using PD-1-and CTLA-4-blocking antibodies seems to be the best option that is currently available for patients with metastatic UM.

ICB in Uveal Melanoma-Doomed to Fail?
The enormous success of ICB for the treatment of metastatic CM raised the expectations that metastatic UM patients would also benefit from the new therapy options. However, as presented above, the results were rather disappointing. Several reasons might explain the poor response rates of ICB in metastatic UM as discussed in the following part.

Low Mutational Burden, Few Neoantigens, and Low PD-L1 Expression
Several studies have shown that ICB seems to be most effective in tumors with a high mutational burden, such as CM, cutaneous squamous cell carcinoma, or NSCLC [132,133], as these tumors are more likely to express a broad range of neoantigens that can be recognized by CTLs [134]. The exposure to environmental noxae, such as tobacco smoke, UV radiation, and other carcinogens facilitate the accumulation of dozens of mutations and result in the formation of tumors with a high mutational burden [12]. Currently, the tumor mutational burden is the best biomarker to predict response to ICB across tumor entities. In contrast to CM, UM is a tumor with a remarkably low mutational burden of around 0.5 per Mb sequence [20]. Environmental mutagens like UV radiation have been discussed as risk factors in UM; however, their impact on UM initiation and progression seems to be quite low and is still controversial [135]. Therefore, the probability of expressing neoantigens that are recognizable for T cells is relatively low. Additionally, the fact that only about 10% of UM primary tumors [99] and 5% of the UM cells in metastatic UM sites express PD-L1 [136] may also contribute to the low susceptibility of UM to ICB mediated by anti-PD-1 or anti-PD-L1 antibodies. Nevertheless, a small subset of patients appears to benefit from ICB, and durable responses have been achieved with ICB. Rodrigues et al. report the case of a UM patient with metastases in liver, lungs, and bones who had a CR after ICB with pembrolizumab [137]. Another UM patient, who had received ipilimumab plus pembrolizumab had an SD for 10 months and a prolonged survival of 2 years in the metastatic stage [138]. The tumors of both patients were characterized by an extraordinary high mutational burden and the presence of germline mutations of methyl-CpG-binding domain protein 4 (MBD4) [137,138] that are found in about 2% of UM patients [137]. Like the frequently in UM mutated BAP1, the MBD4 gene is located on chromosome 3 (3q21.3) [139]. MBD4 mutations are also present in up to 43% of colorectal, endometrial, gastric, and pancreatic cancer with microsatellite instability [139]. The protein acts as a DNA glycosylase and is involved in the base excision repair mechanism where it initiates the removal of deaminated cytosines and 5-methylcytosines and, in particular, thymine and uracil incorrectly pairing with guanine within CpG dinucleotides [140]. Altogether, these findings indicate that the low mutational burden of UM significantly contributes to the weak ICB susceptibility.

Development of the Primary Tumor in an Immune-Privileged Organ
UM arises in the eye in an immune-privileged environment that possesses inhibitory properties against both the innate and the adaptive immune system [141]. This may protect cancer cells not only at the site of the primary tumor but also may hamper a successful anti-tumor immune response in other sites of the body and thus contribute to ICB failure in UM. In contrast, autoimmune-mediated diseases such as uveitis can severely affect the eye. Nevertheless, several mechanisms in the eye, including physical barriers, the presence of immunosuppressive cells, and soluble as well as membrane-bound immune modulators, protect the sensitive retinal tissue from excessive immune responses that may cause vision loss and blindness [141]. The special anatomical structure of the eye limits the access of immune cells to the surrounding tissue through tight junctions between vascular endothelial cells forming the blood-ocular barrier and a lack of lymphatic vessels [142,143]. Additionally, soluble factors in the aqueous humor, including transforming growth factor-β (TGF-β), macrophage migration-inhibitory factor (MIF), vasointestinal polypeptide (VIP), and α-melanocyte-stimulating hormone (α-MSH) [144] dampen the immune response by inhibiting the IFN-γ secretion of activated CD4+ T cells [145]. TGF-β also inhibits the activity of proinflammatory macrophages [144] and upregulates CTLA-4 on CTLs, which promotes the formation of immunosuppressive T regs [146]. T regs are found in up to 24% of UM primary tumors, and their presence is associated with a poor prognosis [147]. T regs formation is also stimulated by α-MSH, another soluble factor that is physiologically present in the aqueous humor [148], and can dampen neutrophil activation [142]. The soluble factor MIF is able to inhibit cell lysis mediated by natural killer (NK) cells [149]. As a part of the innate immune system, NK cells are able to recognize and destroy cells with a decreased or lacking MHC class I expression that would evade CTLs otherwise [150]. NK cells are able to destroy UM cells in vitro [151]. However, in contrast to most other tumors including CM [152], loss of MHC class I expression in UM is associated with a better prognosis [153,154].
UM cells not only evolve in an immunosuppressive environment, but they are also able to create or maintain this favorable environment also in other sites of the body by fostering mostly tumor-promoting immune cells and blocking tumor-killing T cells and NK cells. This is achieved by producing soluble mediators, such as TGF-β [155], α-MSH, and MIF [156], that are also present in the healthy eye. The expression of the membrane-bound PD-L1 was found in the retina, cornea, and iris [157][158][159], but also on UM cells [160]. However, unlike CM, only about 10% of UM primary tumors [99] and 5% of the UM cells in metastatic UM sites express PD-L1, but about 50% of tumor-infiltrating lymphocytes (TILs) express its receptor PD-1 [136]. Tumor-infiltrating immune cells in UM are mostly T cells and smaller numbers of NK cells, B cells, and macrophages [161]. Interestingly, a higher number of TILs has been associated with a worse prognosis in UM [162]. Macrophages in UM are mostly of the M2 phenotype, which facilitates angiogenesis and immunosuppression by producing anti-inflammatory cytokines such as TGF-β and IL-10, which dampen the activation and function of T cells, NK cells, and DCs [163]. Chemokines like CCL2 and CCL22, which are released by UM cells, promote the polarization towards this phenotype [164]. The presence of M2 macrophages in UM may also explain why a high number of TAMs in UM is also associated with a poorer prognosis [165].

Liver as an Immune-Modulating Organ
The liver is the primary organ where UM metastases develop. At time of death, hepatic metastases are present in more than 90% of UM patients [48]. Interestingly, the liver also acts as an immune-modulating organ. It is the first organ which the blood from the gastrointestinal tract passes and therefore needs a tightly balanced immune response in order to defeat pathogens like microbes and viruses quickly but maintain tolerance against harmless dietary antigens. This response is predominantly mediated via the components and resident cells of the innate immune system including the complement system and NK cells [166].
NK cells seem to play an important role in UM tumor growth and the establishment of hepatic metastases. Lymphocyte distribution in the liver significantly differs from other sites of the human body. About 30% to 50% of all lymphocytes in the liver are NK cells compared to 5% to 20% in the peripheral blood [167]. Disrupting the NK cell function in mouse models increased the number of UM liver metastases, whereas promoting NK cell function by interferon-β application resulted in a decreased formation of liver metastases [168,169]. These data indicate that UM cells are generally susceptible to NK-cell-mediated killing in the liver but may be protected in the immunosuppressive environment in the eye. Metastatic UM cells residing in the liver have developed several strategies to circumvent NK cell-mediated lysis. UM cell lines derived from hepatic metastases show a stronger production of MIF, a cytokine that inhibits NK cell function [156]. NK cells are activated via the natural-killer group 2 member D (NKG2D) receptor on their cell surface, which interacts with MHC class I related chain (MIC)-A/B molecules [170]. MIC-A/Bs are nonclassical MHC molecules and are expressed on the cell surface of various tumor entities, including about 50% of primary UMs [171]. Interestingly, it is absent in UM liver metastases, indicating that a decreased MIC-A/B expression also contributes to immune escape from NK-cell-mediated tumor cell lysis in the liver [171].
Besides this, the adaptive immune response against hepatic metastases is rather weak. Only small numbers of CTLs are found in UM liver metastases, and these are mostly observed peritumorally [172]. Thus, the infiltration of lymphocytes into the tissue is weak, and cell lysis mediated by CTLs seems to play a minor role in this process, indicating that ICB, which aims to boost the T cell-mediated anti-tumor response, is likely to fail. Besides this low accessibility to CTLs, only about 5% of the tumor cells in metastatic UM sites and 50% of TILs express PD-1 [136]. These observations are in line with several studies that showed poorer response rates of anti-PD-1 therapy also in CM with liver metastasis, also indicating that the T cell response in this organ plays only a minor role in tumor growth control [173,174]. Additionally, in the phase II study reported by Johnson et al. [123], the only patients who responded were those without or with only small hepatic metastases, and in the study presented by Rossi et al. [126], patients with liver metastases had a much lower median PFS compared to patients with extrahepatic metastases.

Making Cold Tumors Hot for Immune Cells: Considerations for Future Immunotherapies
A special immune-modulating environment at the sites of the primary tumor, the liver as the preferred place for metastatic disease, and acquiring immunosuppressive properties help UM cells to evade from an anti-tumoral immune response. This partly explains why UM is less responsive to ICB than other cancer entities. However, despite the low number of neoantigens, UM expresses a couple of immunogenic antigens as melanoma antigen recognized by T cells (MART-1/Melan-A), glycoprotein 100 (gp100), tyrosinase, tyrosinase-related protein-1 (TRP-1) [175], or melanoma-associated antigen (MAGE) [176]. Furthermore, CTLs isolated from UM primary tumor tissue [177] or the peripheral blood of UM patients [178] as well as NK cells [151] are able to lyse UM cells in vitro. These findings demonstrate that other therapy options may be able to enhance an antitumoral immune response in UM despite the poor ICB results. Thus, is there a chance to develop an immunotherapy that can effectively mediate a successful antitumor response?
Highly expressed antigens such as MART1, gp100, or TRP-1 may be suitable targets for, e.g., dendritic cell (DC) vaccination approaches. DCs are professional APCs and are required for the induction of an antigen-specific immune reaction by presenting captured antigens via MHC class I and II molecules to naïve CD8+ and CD4+ T cells [179]. Therapeutic DC vaccination uses this mechanism to provoke a specific antitumor immune response by generating tumor-specific CD8+ T cells and efficient CD4+ T helper cells. Monocytes or other hematopoietic progenitors are isolated from the tumor patient via leukapheresis and cultured in the presence of stimulatory cytokines such as IL-4 and granulocyte-macrophage colony-stimulating factor (GM-CSF) [179,180]. These monocyte-derived DCs are then loaded with tumor-specific peptides (i.e., gp100, tyrosinase) or mRNA encoding these antigens and retransferred into the patient [180] (Figure 2). In the last years, DC vaccination showed several promising results. A phase I/II trial investigating autologous DCs loaded with gp100 and tyrosinase in 14 metastatic UM patients achieved an antitumor immune response in four patients (29%), one SD, and a median OS with metastatic disease of 19.2 months. No serious AEs of grades 3 or 4 were reported, and the most frequently reported AEs were flu-like symptoms (eight patients, 57%), erythema at the site of the DC injection (six patients, 43%), and fatigue (five patients, 36%) [180]. Another open-label, nonrandomized phase II trial (NCT00929019) investigated DC vaccination in an adjuvant setting compared to observation only. Twenty-three UM patients with a high risk for developing metastases as characterized by monosomy 3 received autologous DCs loaded with tumor mRNA after resection of the primary tumor. DC vaccination achieved a 3-year OS of 79% and a median disease-free survival of 34.5 months. Similar to the study mentioned above, no grade 3 or 4 toxicities were observed, and the most common AEs were flu-like symptoms (21 patients, 91%) and erythema at the site of the DC injection (20 patients, 87%) [181]. Currently, a multicenter, open-label, randomized phase III trial investigating adjuvant DC vaccination versus observation only is ongoing (NCT01983748) [182]. Patients receive autologous DCs loaded with tumor mRNA intravenously over a period of 2 years or standard care with observation only and staging every three months.

Conclusions
UM is a rare malignancy and by far less susceptible to ICB than CM. Its genetic characteristics and the ability to adopt immunosuppressive strategies from the sites where it originates help UM cells to evade from the immune system and lead to checkpoint inhibitor resistance. However, some promising therapy approaches that aim to modulate and enhance the antitumoral immune response are in the pipeline and might prove successful in the next few years. As a subset of patients shows durable responses and clinical benefit to combined ICB, future studies are urgently warranted to better characterize and understand the underlying mechanisms of response.  Monocytes or other hematopoietic progenitors are isolated from the tumor patient via leukapheresis and develop in vitro in the presence of stimulatory cytokines to mature DCs. These are then loaded with tumor-specific peptides (i.e., gp100, tyrosinase) or mRNA encoding these antigens and retransferred into the patient in order to boost the anti-tumor response. (B) Adoptive T cell transfer: TILs are isolated from hepatic metastases and expanded in vitro. After a lymphodepleting chemotherapy, the patient receives the expanded TILs followed by IL-2 administration. (C) Bispecific proteins, i.e., IMCgp100 (tebentafusp): The bispecific protein IMCgp100 combines a TCR against gp100 and CD3 scFv. The engineered TCR of the molecule binds to gp100 on UM cells presented by the MHC class I protein HLA-A*02:01, and the anti-CD3 antibody fragment binds to and activates CD3+ T cells. DC = dendritic cell, TILs: tumor-infiltrating lymphocytes, CD3 scFv = anti-CD3 single-chain antibody fragment, MHC I = major histocompatibility complex I, TCR = T cell receptor, IL-2 = interleukin-2.
The adoptive transfer of tumor-infiltrating lymphocytes (TILs) involves the isolation of TILs from hepatic metastases and their expansion in vitro. Before the expanded cells are retransferred, followed by a T-cell-stimulating IL-2 administration, the patients receive a lymphodepleting chemotherapy. This therapy option has shown promising results in a phase II trial (NCT01814046). Twenty-one HLA-A2-positive UM patients with hepatic metastases were included, and 20 patients were evaluable. One patient had a CR and six patients a PR (30%). Severe AEs of grade 3 and higher, including lymphopenia, neutropenia, and thrombocytopenia, were observed in all 21 patients, and 14 patients (67%) had a grade 3 anemia. These were most likely caused by the lymphodepleting chemotherapy but not by the adoptive T cell transfer itself. [151,183]. Currently, an interventional, open-label phase II trial (NCT03467516) investigating adoptive TIL transfer in metastatic UM patients has finished recruiting and will evaluate the ORR, CR rate, duration of response, disease control rate, PFS, and OS.
Another attempt includes IMCgp100 (tebentafusp), a bispecific protein that combines a TCR recognizing gp100 and an anti-CD3 single-chain antibody fragment (scFv). The engineered TCR binds to gp100 on UM cells presented by the MHC class I protein HLA-A*02:01 and the anti-CD3 antibody fragment binds to and activates CD3+ T cells [184]. IMCgp100 has achieved encouraging results in phase I/II trials. In a phase I first-in-human trial (NCT01211262) including 14 UM patients, two PR (14%), 8 SD (57%) and 4 PD (29%) were observed. [185]. Another phase I study (NCT02570308) investigated intra-patient escalating doses of IMCgp100 in 19 heavily pretreated metastatic UM patients. Two PR (11%) were observed in this trial, the median PFS was 5.6 months and the 1-year PFS was 62%, the 1-year OS 79.5% (95% CI 55-93), median OS was not yet reached. Rash (90%), pruritus (90%) and edema (63%) were the most frequently observed AEs. Similar to the first-in-human study, hypotension (16%) was the most frequent severe AE of grade 3 and higher [186]. Currently, a randomized controlled trial comparing IMCgp100 vs. physician's choice chemotherapy or immunotherapy is recruiting patients with advanced UM (NCT03070392).
In highly prevalent tumor types, the use of high-throughput data (HTD) technologies for molecular profiling like DNA and RNA sequencing generates currently valuable knowledge to elucidate the molecular relationships behind pathogenesis, improve diagnostics, find druggable targets, and assess the mechanisms of resistance to therapy. The application of these technologies to UM is in its infancy [30]. However, in general, parallel and hierarchical integration analyses used for CM should also be applicable to other cancer entities as well [187]. These new techniques could significantly contribute to the understanding of UM pathogenesis and metastatic spread but also to the adaptation of therapies successful in other tumor entities like CM. As indicated, the liver is an immune-regulating organ which UM tumors benefit from developing. HTD techniques can be used to dissect the molecular features that make the liver microenvironment advantageous for UM. The use of single-cell RNA sequencing can provide detailed information about the populations of stromal and immune cells interacting in the tumor niche (see [188] for an example in CM). When combined with network biology methodologies, the RNA sequencing data can provide clues about the specific transcriptional programs that tumor and immune cells have activated [189,190]. The deep understanding of these programs, which govern key tumor and immune cell phenotypes, can help in finding molecular targets for coadjuvant therapy making UM cold tumors hot [191]. We also mentioned that UMs often have a low number of neoantigens, but still a number of overexpressed proteins holding immunogenic antigens. RNA sequencing analysis of tumor samples, differential expression, and computational algorithms for epitope detection can be combined to select in a personalized manner T cell epitopes from overly expressed proteins, which could be used in DC-or epitope-based immunotherapy [192].

Conclusions
UM is a rare malignancy and by far less susceptible to ICB than CM. Its genetic characteristics and the ability to adopt immunosuppressive strategies from the sites where it originates help UM cells to evade from the immune system and lead to checkpoint inhibitor resistance. However, some promising therapy approaches that aim to modulate and enhance the antitumoral immune response are in the pipeline and might prove successful in the next few years. As a subset of patients shows durable responses and clinical benefit to combined ICB, future studies are urgently warranted to better characterize and understand the underlying mechanisms of response. Conflicts of Interest: L.H. has received fees as consultant/member of advisory boards by Amgen, MSD, BMS, Roche, Curevac, Pierre-Fabre, Novartis, and Sanofi-Aventis and research grants to institution from Novartis. M.S. received honoraria for lectures from MSD, BMS, Roche, and Pierre-Fabre and served on advisory boards for Novartis, Roche, MSD, and Pierre-Fabre. C.B. has received speaker´s honoraria or fees as member of advisory boards by Amgen, BMS, Immunocore, Merck, MSD, Novartis, Pierre Fabre, Roche, and Sanofi-Aventis. C.B. has been investigator of clinical trials sponsored by Amgen, Array Pharma, AstraZeneca, BMS, MSD, Novartis, Regeneron, and Roche. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; nor in the decision to publish the results. All other authors declare no conflict of interest.