Uveal melanoma (UM) is a malignant tumor of the eye that originates from the pigment cells of the choroid layer or the ciliary body which is clinically and biologically distinct from cutaneous melanoma. Although the incidence is much lower than that of cutaneous melanoma, UM belongs to the most common malignant intraocular tumors in adults [1]. In approximately 50% of all cases, patients develop distant metastasis during the course of the disease, which affects predominantly the liver. Clinical risk factors for metastases are posterior localization in the eye, tumor size of more than 10 mm, and presence of vascular loops. Molecular biomarkers associated with a higher risk of metastasis are monosomy 3 or genomic alterations of BAP-1 [2]. Once distant metastases have occurred, the prognosis is dismal with an average survival time of approximately 1 year across all therapeutic regimens [3].

Patients with metastatic UM have so far benefited little or not at all from the treatment innovations achieved in cutaneous melanoma in recent years. Neither targeted therapy with MEK inhibitors nor checkpoint blockade with ipilimumab or PD-1 inhibitors as monotherapy was able to significantly improve the prognosis of patients with UM [4, 5]. The response rates were consistently in the single-digit percentage range in a panel of previous studies [6–9]. In cutaneous melanoma, combined checkpoint blockade with ipilimumab and nivolumab revealed response rates and survival outcomes superior to PD-1 inhibitor monotherapy, albeit at the cost of high immune-related toxicity [10]. However, the significance of combined checkpoint blockade in UM is unclear and has only been investigated in case reports and small case series [6, 11, 12]. In this study, we evaluate the clinical course of 64 patients with metastatic UM who received combined checkpoint blockade. We report clinical outcomes with respect to response, survival, and adverse events (AE). Furthermore, clinical and laboratory parameters were investigated which may have prognostic value in UM patients treated with checkpoint blockade.

A total of 64 (100%) patients with metastatic UM were included. Fifty patients (78.1%) were naïve to systemic treatment and received combined checkpoint blockade as first-line systemic therapy. Regarding genotype, the presence of monosomy 3 as risk factor was specifically investigated in 7 patients and identified in 2 of them. BRAF, NRAS and KIT were analyzed and reportedly wildtype as expected in 30, 22, and 20 patients, respectively. Mutations and inactivations of MBD4 which were previously linked to a hypermutator profile with high sensitivity to PD-1 inhibition were not investigated in any case [18, 19].

Previous ipilimumab and PD-1 inhibitor monotherapy were applied in 2 (3.1%) and 12 (18.8%) cases, respectively. Both patients treated with ipilimumab before showed PD. Specifically, 4 patients (6.3%) had received nivolumab and 8 (12.5%) pembrolizumab before. In 4 cases, SD was achieved while 8 patients showed PD upon PD-1 inhibitor monotherapy. The median duration of the clinical benefit was 6.5 months in the 4 patients with SD. Liver-directed therapies were reported in 31 patients (48.4%). Most patients had an ECOG status of 0 (n = 49, 76.6%). Serum LDH was elevated in 33 cases (51.6%) at baseline. Other baseline characteristics are listed in detail in Table 1. Ipilimumab plus nivolumab was given in 59 patients (92.2%), while 5 patients (7.8%) received ipilimumab plus pembrolizumab. The median number of treatment cycles was 3 (range 1–4) for the combination of ipilimumab with a PD-1 inhibitor in the induction phase, and 0 (range 0–27) for PD-1 inhibitor maintenance therapy in the overall population. A total of 19 patients (29.7%) received a PD-1 inhibitor maintenance therapy. Among these, the median number of PD-1 inhibitor cycles was 3 (range 1–27).

The best ORR to combined checkpoint blockade was 15.6% (n = 10) relating to the entire population (4 patients were not evaluable for a radiologic response). Two patients achieved a complete response (3.1%) and 8 (12.5%) a partial response. The median duration of response was 25.5 months (range 9.0–65.0). Stable disease was achieved in further 14 cases (21.9%), resulting in a disease control rate of 37.5% with a median duration of the clinical benefit of 28.0 months (range 7.0–65.0) (Table 2). The median PFS was 3.0 months (95% CI 2.4–3.6). The median OS was estimated to 16.1 months (95% CI 12.9–19.3) with a median follow-up period of 9.2 months (95% CI 7.8–10.6) (Fig. 1).Tab2

Best response rates to combined checkpoint blockade

The median time to response in patients with CR or PR after treatment initiation was 12 weeks (range 5–31). For the patients with SD, the median duration until the benefit was observed also amounted to 12 weeks (range 9–30). Interestingly, all 4 patients with SD after previous single PD-1 inhibitor blockade had PD to combined checkpoint blockade. Among the remaining 8 patients with PD after previous single PD-1 inhibitor blockade, one achieved a PR to combined checkpoint blockade. Thus, these data suggest that the effects of single and combined checkpoint blockade were observed independently from each other.

A total of 78 AE were reported in 39 patients. Thus, the majority of patients developed any treatment-related AE (60.9%). Of all events, 37 AE were graded as severe (grade 3 + 4). They were observed in 25 patients (39.1%; grade 3: 37.5%; grade 4: 1.6%). The treatment was discontinued in 25 cases (39.1%) due to unacceptable toxicity. However, no treatment-related deaths occurred during treatment or the observation period. The most common events were colitis (20.3%), hepatitis (20.3%), thyreoiditis (15.6%), hypophysitis (7.8%), fever (4.7%), and myalgia with myositis (4.7%). In all 5 cases with hypophysitis, the individual hormone axes including ACTH, cortisol, FSH, LH, TSH, and testosterone were investigated but not specifically graded. In 3 cases, the pituitary gland was enlarged in MRI examinations. All patients received systemic replacement of hydrocortisone. All AE are listed in Additional file 1.

In univariate Cox regression, ECOG status (p = 0.000096), the presence of bone metastasis (p = 0.011), and the best response to checkpoint blockade (p = 0.002) were significantly associated with OS (Additional file 2). The risk factors ECOG status, serum LDH, serum levels of CRP, and presence of bone metastasis were further integrated into a multivariate Cox regression model. Of these factors, a significant association with OS was confirmed for ECOG status (p = 0.007) only (Table 3, Fig. 2a).Tab3

Multivariate Cox regression analysis of clinical parameters and serum biomarkers

We recently identified a prognostic score of the serum biomarkers LDH, CRP, and relative eosinophil count (REC) in a cohort of 94 UM patients receiving PD-1 inhibitors [6]. The score assigns one risk point for each unfavorable factor, i.e., elevated LDH, elevated CRP, and a REC < 1.5%, defining four distinct prognostic groups (low, intermediate, high, and very high risk). Each patient receiving combined checkpoint blockade was assigned to a risk group and the score was validated with Kaplan-Meier estimates. Due to a small sample size, patients with low and intermediate risk were pooled. The risk groups showed significantly different survival probabilities (p = 0.000005). The median survival times were superior for the low plus intermediate group (17.7 months, 95% CI 14.7–20.8) compared to the high (15.4 months, 95% CI 12.7–18.2) and very high risk group (7.1 months, 95% CI 0.0–16.2) (Fig. 2b). However, the score neither correlated with the response rate (p = 0.609) nor with the DCR (p = 0.446), suggesting that it was generally prognostic but not specifically predictive for the response to combined checkpoint blockade.

Subgroup analysis were performed for patients with metastasis to the central nervous system (CNS) at treatment initiation and for the treatment responders. Four patients showed an involvement of the CNS. Two of them had neurological symptoms. Two patients achieved SD, 2 showed PD. The median PFS for the CNS subgroup was 3.0 months (95% CI 0.0–6.1) while the median OS was not reached. In contrast, none of the treatment responders (CR or PR) had CNS involvement when the treatment was initiated (Table 4). The median time from detection of the primary tumor to metastatic disease was 43 months among the responders. Data on the assessment of the risk of metastasis formation of the primary tumors were sparse, as e.g. the presence of monosomy 3 or the MBD4 status was not investigated in any of the responders.Tab4

Characterization of the responders to combined checkpoint blockade (n = 10)

Here, we present a comparatively large cohort of patients with metastatic UM who were treated with combined checkpoint blockade. We detected a 15.6% ORR, with a 3.1% complete and 12.5% partial response rate. This response rate is in line with our previous report showing 16% ORR, although only 12 patients were evaluable for their radiologic response and the follow-up time was short [6]. Another case series was recently published from a single-center experience where 2 out of 8 patients treated with nivolumab and ipilimumab had a partial response [11]. Other preliminary data on the efficacy of the combined checkpoint blockade have been proposed as conference abstracts, but appear preliminary to date. Najjar et al. reported results from a multi-center, retrospective analysis in 66 patients from 11 U.S. centers, revealing an ORR of 13% and a DCR of 31% [20]. In addition to these estimates in a real-world setting, prospective trials are currently underway. A preliminary analysis of the Spanish phase II trial GEM1402 (NCT02626962) showed an ORR of 12% and disease stabilization in 52% of cases [21]. Another phase II trial is currently ongoing in the U.S. in 30 patients with UM (NCT01585194). A recently presented interim analysis revealed an ORR of 17% and disease control in 50% [22]. Thus, we conclude that the ORR of 15.6% identified in this population is a solid estimate for the efficacy of combined checkpoint blockade in UM and a good indicator of what we can expect from the final analyses of the prospective trials. This regimen appears to be significantly superior compared to the sobering efficacy values observed with ipilimumab and PD-1 inhibitor monotherapy [6–9, 23–26]. Considering the data available so far, we conclude that the increase of ORR of the combined blockade versus PD-1 inhibition alone amounts to approximately 10%. Further evidence for a better efficacy of the combined regimen is supported by the observation of complete responders, albeit to a small extent. This is notable as UM is considered a “cold” tumor due to a low mutational burden and a unique immunosuppressive tumor microenvironment [27–29]. Further research is urgently needed to identify the radiologic, immunologic, and molecular determinants for treatment response in this small subset of patients. Regarding safety, the rate of severe AE was lower compared to the events reported in the pivotal trial in cutaneous melanoma (CheckMate-067) [30]. In particular, the occurrence of potentially life-threatening grade 4 AE was surprisingly low, suggesting that the regimen may be better tolerated in UM. However, it is also conceivable that the retrospective design and the small number of cases of this study causes an underreporting of AE.

Among clinical parameters and serum biomarkers, only the ECOG performance status was a consistent prognostic factor in multivariate analysis. Other parameters such as serum LDH, CRP, and the REC showed a significant association neither with OS nor with the treatment response when they were considered as single factors. However, when integrated into a prognostic score, they were useful for risk stratification and discriminated groups with distinct survival probabilities. Thus, the risk score identified previously in a distinct cohort was successfully validated in this population [6]. As there was a significant association neither with the ORR nor the DCR, we conclude that the score is generally prognostic but not specifically predictive for the response to checkpoint blockade.

The major limitations of this study are its retrospective design and the lack of a control group. When compared to historical controls, the median OS of 16.1 months is superior to survival estimates from other studies. Recently, the median OS benchmark for metastatic UM was identified as 10.2 months in a meta-analysis on individual data from 912 patients pooled from 29 trials [31]. Another analysis on individual-level data from 2494 patients proposed a median OS of 1.07 years across all treatment modalities. In this context, the OS observed in our cohort treated with combined checkpoint blockade appears more favorable, although external cohorts should be interpreted with caution and the comparison may be subject to significant confounding. A further limitation comes from the paucity of molecular and genetic analysis on the primary and metastatic tumors which are urgently needed to better characterize and understand the pattern of treatment response in UM.

Altogether, our study implies that combined checkpoint blockade represents the hitherto most effective treatment option available for metastatic UM available in routine care outside of clinical trials. Based on our analysis and preliminary data from others, we hypothesize that the ORR achieved with combined checkpoint blockade will be 15–17%. Future trials are warranted to identify specific biomarkers for treatment response.