Long-term Survival and Clinical Benefit from Adoptive T-cell Transfer in Stage IV Melanoma Patients Is Determined by a Four-Parameter Tumor Immune Signature

Melanoma is the most aggressive form of skin cancer and has long been recognized as a highly immunogenic tumor and a good target for immunotherapy (1). In different types of cancer, including melanoma, the presence of type I cytokine–oriented tumor-infiltrating lymphocytes (TIL) has been associated with improved survival (2). Indeed, a strong ongoing immune response was linked to spontaneous regression in about half of the primary melanomas (3) and longer survival of patients with stage I–III primary and regionally metastasized melanoma (4–6). More recently, a large study in patients with stage IV (distant metastases) melanoma revealed that even at this stage, intratumoral T-cell content was associated with improved survival (7). However, the predictive value for survival was not so strong, indicating that other immune-related factors previously studied in primary melanoma may also play a role (8–12); this involvement of other immune parameters was also suggested by studies at the gene expression level (13, 14). In parallel, studies showing that a strong intratumoral T-cell infiltrate fosters a better response to PD-1 checkpoint therapy (15) and autologous tumor cell vaccination (11), but also that intratumoral macrophages can hamper CTLA-4 checkpoint therapy (16), suggest that the tumor's immune contexture may also influence the response to immunotherapy.

In this study, we have expanded on earlier studies (4, 7) by assessing the influence of a series of immune factors in the metastatic tumor microenvironment in a large group of stage IV melanoma patients with up to 10 years of follow-up since metastasis. We identified four parameters, each of which was associated with better survival. These parameters comprised CD8 T cells, the presence of galectin-9⁺ dendritic cells (DC)/DC-like macrophages, a higher M1/M2 macrophage ratio, and galectin-3 expression by tumor cells. The presence of at least three parameters was an independent prognostic factor for survival, which was validated by analysis of these parameters in stage IV melanoma patients in The Cancer Genome Atlas (TCGA) database. Furthermore, with the introduction of targeted therapies and checkpoint inhibitors, adoptive cell transfer (ACT) has mostly become a salvage therapy (17) for treatment of stage IV melanoma patients. Analysis of the predictive value of this signature for the response to ACT revealed that the pretreatment tumors of patients without clinical benefit (CB) predominantly display two or fewer of the beneficial immune parameters, whereas the presence of three or four of these parameters was most frequent in patients showing sustained CB after ACT.

Formalin-fixed, paraffin-embedded tissue blocks from 73 stage IV metastatic melanoma patients undergoing surgery were collected at Leiden University Medical Center (LUMC, Leiden, the Netherlands) and at the Netherlands Cancer Institute (NCI, Amsterdam, the Netherlands). The patients were included in clinical studies that were approved by a local ethical committee (LUMC study P04.085, NCI study EudraCT 2010-021885-31), and all patients gave written informed consent. All specimens were from metastases, and biopsies were taken before any immunotherapeutic treatment. Classification of metastases was done by tumor–node–metastasis (TNM) staging criteria (18), and information on the concentration of lactate dehydrogenase (LDH) at the moment of sampling was collected. Included in the cohort of 73 patients were 17 patients that were treated with ACT in ongoing clinical studies in the LUMC and NCI. Of these patients, 7 were classified as patients without CB [progressive disease (PD)] and 10 patients with CB (stable disease (SD), partial response (PR), and complete response (CR)] according to RECIST1.1 criteria.

The presence of T-cell and macrophage infiltrate in the tumor area and the expression of galectin-1, galectin-3, and galectin-9 by the tumor was analyzed using previously determined optimal antibody concentrations and immunofluorescence staining protocols as described before (19, 20). Briefly, T-cell infiltrates were stained with antibodies to CD3, CD8, and FoxP3. Macrophages were identified using antibodies to CD14 and CD163. To examine which cells expressed galectin-9, a small part of the cohort received triple immunofluorescence staining with antibodies to galectin-9, CD68, and CD11c. All secondary antibodies were isotype-specific antibodies labeled with the fluorochromes Alexa Fluor 488, 546, or 647. The expression of Tbet was analyzed by IHC as described before (19) with the exception that after incubation with the primary Tbet antibody and incubation with BrightVision poly-HRP anti-mouse/rabbit/rat IgG, the antigen-antibody reactions were visualized using the NovaRED Substrate Kit for peroxidase (Vector Laboratories). For all antibody labeling, negative controls and controls omitting the primary or secondary antibody were performed. Positive control tissue slides were included for all antibody labeling, using tonsil for T cells, colon for galectin, and placenta for macrophage controls. Images were captured using a confocal microscope (LSM15, Zeiss) for the immunofluorescence labeling and a spectral microscope (Leica DM4000 B, Leica Microsystems) for the immunohistochemical stains. Random images (five per slide) were taken for analysis. Analysis of the images was done using ImageJ. Intratumoral T cells, macrophages, galectin-9⁺ cells, and Tbet⁺ cells were manually counted using the “cell counter” plugin of ImageJ and presented as number of cells/mm² (average of five images). Galectin-1 and galectin-3 expression by tumor cells were analyzed using an immunoreactive score (IRS; ref. 21), taking into account the percentage of positive cells and the intensity of the staining (Supplementary Table S1A).

For validation of our results of the IHC and immunofluorescence experiments, we analyzed data from the publicly available TCGA database (5). To reconstruct our parameters, we used gene expression profiles of the subset of patients with stage IV melanoma. CD8⁺ T cells were identified by taking the average of CD8A and CD8B expression, galectin-9⁺ DC-like macrophages by GALS9 expression, M1/M2 macrophage ratio by the ratio of CD86 and CD163 expression, and expression of galectin-3 by tumor cells by LGALS3 expression. This approach assumes that the genes we used are expressed preferably by the target cells for the parameter, and not in the other cell types in the sample. We used the z-scores as available in the TCGA data. This will yield a more comparable weight of all parameters. For each parameter, a high and a low group was defined by splitting the samples on the median value for the parameter.

The differences between different patient groups were analyzed using the nonparametric Mann–Whitney U test for comparison of continuous variables between two groups and the one-way ANOVA or Kruskal–Wallis test for the comparison of three groups. For comparison of categorical data, the χ² test or the Fisher exact test was used. Correlation analysis was done using Spearman ρ correlation. Correlation between immune parameters and overall survival (OS) since metastasis was calculated by the Kaplan–Meier method and statistically analyzed by the log-rank test. Univariate and multivariate Cox proportional hazards models were used to determine the HR that represents the relative risk of death among patients in the different indicated groups. In the multivariate Cox regression models, analysis was corrected for age, serum LDH, and the pattern of visceral metastases (TNM staging), the latter two being established prognostic factors for stage IV melanoma. Interaction analyses were performed on the parameters that were identified as prognostic for survival. For all tests, P values <0.05 were considered statistically significant. For statistical analysis, the software package SPSS statistics 20.0.0 was used (SPSS Inc.).

To study the immune markers that relate to survival in stage IV metastatic melanoma, tumor biopsies from a group of 73 patients were investigated (Supplementary Table S1B). The mean age at the moment of sampling was 52 (range, 25–74; 41 males and 32 females). The patients were divided in tertiles (<9 months, 9–20 months, >20 months) based on the survival of stage IV disease. The mean age of the group of patients with the longest survival cohort was a bit lower, but this difference was not significant (one-way ANOVA, P = 0.09). Maximal follow-up of the patients since collection of the metastatic sample was 120 months, and the median survival since metastasis was 13 months (range, 1–120). Within the cohort of 73 patients, 17 patients were included that were treated with ACT. Of these 17 patients, 6 patients received T cells that were generated by mixed lymphocyte tumor cultures of autologous tumor cells and peripheral blood mononuclear cells (22). The other 11 patients received T cells that were generated by rapid expansion of TILs (23). Within the ACT-treated group of 17 patients, 10 were classified as having CB (3× SD, 4× PR, and 3× CR). The CB patients showed a significantly better survival compared with the patients with no CB when calculated by the Kaplan–Meier method and analyzed by the log-rank test (PD; P < 0.001; Supplementary Fig. S1). The different methods used to generate T-cell batches for ACT did not influence clinical outcome (Fisher exact test, P = 0.64).

To study the immune signature in each tumor, the numbers of T cells, macrophages, and galectin-9–positive immune cells were quantified per square millimeter of tumor area, whereas the expression of galectin-1 and galectin-3 by tumor cells was determined via the IRS (Fig. 1; Supplementary Table S2A). CD8⁺ T cells were defined as CD3⁺CD8⁺FoxP3^–, CD4⁺ T cells were defined as CD3⁺CD8⁻FoxP3^–, and regulatory T cells (Treg) were defined as CD3⁺CD8⁻FoxP3⁺ (Supplementary Fig. S2A). Two types of macrophages were defined: CD14⁺CD163^– (M1) and CD14⁺CD163⁺ (M2) macrophages (Fig. 2B). Galectin-9 was, based on morphology, expressed by cells of myeloid origin (Fig. 2C). Additional costaining with CD68 and CD11c revealed that a majority (70%) of these galectin-9–expressing cells coexpressed CD11c, and part of these cells also expressed CD68, indicating that the galectin-9⁺ myeloid cells were predominantly DCs or DC-like macrophages (Fig. 3A and B).

Galectin-1 and galectin-3 were expressed by the tumor cells, but with different intensities between samples (Fig. 2C). To determine the type I orientation of the TIL, part of the cohort was stained for Tbet (Fig. 3C). A strong positive correlation between the number of T cells and the number of Tbet⁺ cells was observed (Spearman ρ correlation coefficient 0.720, P < 0.001; Supplementary Table S2), suggesting a type I immune contexture in strongly T-cell–infiltrated tumors (24). In addition, the number of galectin-9⁺ DCs/DC-like macrophages was also strongly related to that of intratumoral T cells and Tbet⁺ cells (Supplementary Table S2B), fitting with the notion that galectin-9 is expressed on immune cells upon exposure to proinflammatory mediators (25). No correlation was found between OS and any of the investigated markers (Supplementary Table S2B).

Quantification of the T cells revealed huge variability in the number of CD8, CD4, and Treg cells, as well as tumor-infiltrating M1 and M2 macrophages, within each survival group. Patients with high T-cell counts and a higher ratio of M1/M2 macrophages were more often found in the long- and medium-term survival groups (Fig. 1A and B). The number of patients with dense galectin-9⁺ DCs/DC-like macrophages was higher in the groups of patients with medium- and long-term survival (Fig. 1C). The expression of galectin-1 and -3 by tumor cells was not so different between the patient groups, albeit that in the long-term survival group, galectin-1 expression was somewhat lower and galectin-3 higher (Fig. 1C). Notably, none of the observed differences were statistically significant between the three groups (Supplementary Table S2).

To investigate the potential influence of all these parameters on survival since metastasis, a univariate and multivariate Cox analysis was performed, and this revealed that high numbers of CD8⁺ T cells and galectin-9⁺ DCs/DC-like macrophages, a high M1/M2 ratio, and tumors expressing galectin-3 were prognostic factors for survival (Table 1). For calculation of the correlation between these parameters and survival using the Kaplan–Meier method, the patients were stratified on the basis of the median cell counts or IRS. The presence of these four parameters was also related in this analysis to longer survival (Fig. 2A). Because these four different elements in the tumor immune microenvironment were prognostic for survival, we performed interaction analyses of each combination of two parameters. Patients with a combination of two of any of these parameters (CD8⁺ high/M1/M2 high/galectin-9⁺ high/galectin-3 high) showed better survival compared with the other patients, resulting in decreased HRs in the multivariate Cox regression analysis (Table 1) as well as higher survival rates when the Kaplan–Meier analysis was used (Fig. 2B). Each component positively contributed to the other (Fig. 4).

To investigate whether it would be possible to define a certain immune signature for each of the three survival cohorts, the fraction of tumors displaying combinations of immune parameters was plotted (Figs. 3A and 5). This revealed that the fraction of tumors displaying a certain immune signature differed significantly between the three survival cohorts (χ² test, P < 0.001). The group with the longest survival contained the largest fraction of patients with a tumor that was positive for four beneficial parameters, whereas this immune phenotype was not present in the short survival cohort. We then questioned how many beneficial parameters were required to have a major impact on survival. The group of patients with metastases displaying two or less of the beneficial immune parameters showed a low survival rate. The survival rate increased when patients displayed three or four of the immune parameters, with a 100% survival for patients with a tumor that displayed all four immune parameters (Fig. 3B). In addition, a multivariate Cox analysis was performed with age, gender, LDH level, and TNM stage as covariates. This showed that an immune signature consisting of more than three parameters was an independent prognostic factor for survival (Table 1).

To validate our results, this immune signature was analyzed in a set of stage IV melanoma patients in the TCGA database, which contained 23 patients. In this cohort, patients with a tumor that displayed three or four of the immune parameters had better long-term survival (P < 0.001) compared with the rest of the patients (Fig. 3C).

We analyzed 17 pre-ACT treatment biopsies of metastases. Their immune microenvironment showed that the tumors from patients with CB showed higher numbers of intratumoral T cells compared with the patients showing PD (Fig. 4; Supplementary Table S2A). CB patients showed a significantly higher ratio of M1/M2 macrophages, which was mostly the result of a low M2 infiltration (Fig. 4). In addition, the number of intratumoral galectin-9⁺ DCs/DC-like macrophages and tumors expressing galectin-1 was lower in the CB patients. Next, the fraction of tumors displaying the combination of immune parameters linked to better survival was plotted (Fig. 5A). Thirty percent of the CB patients displayed an immune signature positive for four of our identified beneficial immune parameters, whereas none of the PD patients showed this immune phenotype. Of the 5 CB patients that showed an immune signature of two or less beneficial immune parameters, 4 showed a low infiltrate of CD8⁺ T cells (Supplementary Table S3), which potentially was corrected by the ACT. Analysis of the effect of this immune signature on the survival after ACT revealed that sustainable CB was only found in those patients whose tumor immune signature included three to four of the identified parameters before treatment with ACT (Fig. 5B).

In this study, we investigated the tumor immune contexture in stage IV metastatic melanoma and analyzed its association to survival. We found that an infiltrate of CD8⁺ T cells, galectin-9⁺ DCs or DC-like macrophages, a high M1/M2 ratio, and a high expression of galectin-3 by the tumor were associated with survival. The groups of patients with an immune signature consisting of a combination of three or four immune contexture–defining elements displayed the longest survival. Notably, most of the patients in our cohort were treated in the preimmunotherapy era when the OS generally was low (26). Since the introduction of immunotherapies, the OS of stage IV melanoma patients has increased considerably. Independent validation of our results on a more recent cohort of 23 stage IV melanoma patients with a higher median OS in the TCGA database (5) sustained our observation that patients with tumors that expressed at least three of the defined immune parameters survived longer than patients with tumors not displaying three to four of these immune parameters. Notably, the immune signature is also prognostic for longer survival after ACT treatment and may be predictive for response after ACT, as treatment of stage IV melanoma patients by adoptive T-cell transfer was more likely to be successful and durable when the metastases of these patients displayed at least three of the identified four beneficial immune parameters prior to treatment. However, a firm statement about the predictive value of the immune signature requires validation in a prospectively ACT-treated patient cohort.

Our study clearly showed that not only T cells but multiple elements defining the tumor immune contexture are of importance for the prognosis of stage IV metastatic melanoma patients and their response to ACT. Rather than the numbers of several types of infiltrating immune cells, which did not differ among the short-, medium, and long-term survival groups, it is their combined presence and the processes they reflect that bears an impact on outcome. We showed that there were strong correlations between the Tbet⁺ lymphocyte counts and CD4 and CD8 T-cell counts, suggesting an ongoing type I cytokine intratumoral T-cell response. The notion of such a proinflammatory signature is sustained by the presence of galectin-9⁺ myeloid cells, the number of which was strongly correlated to the number of Tbet⁺ cells and T cells. Galectin-9 is known to be expressed by myeloid cells after exposure to proinflammatory cues in the microenvironment (25, 27). On the basis of the coexpression of CD68 and CD11c, we conclude that these galectin-9⁺ cells predominantly are DCs and DC-like macrophages that can be crucial for the induction of strong antitumor immunity similar to their published role as crucial mediators of anticancer immune responses after anthracycline-based chemotherapy (28). It is not clear whether galectin-9 is also secreted by these myeloid cells. If so, this could also contribute to the antitumor response, as galectin-9 was shown to increase the numbers and cytolytic capacity of CD8 T cells and natural killer (NK) cells in the Meth-A and B16F10 mouse tumor models (29, 30). The presence and activity of NK cells was not assessed in our study but may form an additional parameter associated with survival, as highly activated NK cells have been found in tumor-infiltrating lymph nodes of melanoma patients (31). In addition, the presence of a proinflammatory environment was reflected by the higher M1 to M2 macrophage ratio found in long-term survivors. Previous studies in melanoma did not find associations between macrophage infiltrate and survival, most likely because these studies did not distinguish between functionally different macrophages and did not relate the presence of macrophages to other immune cells (4, 8, 32). In other types of cancer, however, the presence of proinflammatory (M1) macrophages and, in particular, a high M1 to M2 ratio was shown to be related to improved survival (33, 34).

In our study, a high expression of galectin-3 by the tumor was associated with long-term survival, which was also reported for primary melanoma (12). This is counterintuitive as existing literature mostly describes an immunosuppressive role for galectin-3 (35). Galectin-3 can hamper T-cell activation by interfering with T-cell receptor signaling (36) and by binding to the checkpoint inhibitor LAG-3 (37). In addition, it has been shown to increase apoptosis in effector T cells (38, 39). The positive effect of galectin-3 expression on survival is possibly explained by its direct effect on the tumor cell and might be a result of a higher susceptibility to oxidative damage and apoptotic death of galectin-3–expressing melanoma cells (40).

The presence of an at least three-parameter beneficial immune signature was also found in the patients displaying sustained CB after ACT. The other 5 CB patients, lacking this signature, displayed only short-term benefit from ACT. None of the patients that were unresponsive to ACT showed an immune signature with the presence of all four parameters. We hypothesize that the quality of the ACT T-cell product determines whether a patient responds to therapy. Indeed, T-cell batches that were infused into CB patients comprised tumor-specific T cells with predominantly a Th1 cytokine profile, whereas the T-cell batches that were infused into the patients with PD showed a non-Th1 cytokine profile (22). We speculate that even if good-quality ACT is given to patients whose tumor displays a less desirable immune signature, in particular those showing a failure to recruit large numbers of T cells, only a short period of CB can be achieved. It can be envisaged that an initial failure to recruit T cells, potentially due to a lack of T-cell–attracting chemokines secreted by these tumors (13), will also hamper ACT. Unfortunately, we do not have the tumor material after ACT treatment to test our hypothesis.

Our study is limited by the relatively low number of patients that we could analyze, in particular the patients treated with tumor-reactive T cells and the stage IV melanoma patients available for analysis in the TCGA cohort. Validation of our results in other cohorts will be important to understand the true value of our findings.

We conclude that a tumor immune contexture of at least three of the four parameters described in our study is prognostic to long-term survival of stage IV melanoma patients. Moreover, this same immune signature is required for sustainable responsiveness to ACT of these patients. It would be of interest to validate the predictive value of this immune signature in a prospective cohort of patients treated with ACT as well as to study whether this signature may predict the response to immune checkpoint blockers. If so, the use of this immune signature will be important for personalized therapy, that is, development of tailor-made therapy with higher predicted treatment benefit.

No potential conflicts of interest were disclosed.

Conception and design: S.M. Melief, V.T.H.B.M. Smit, S.H. van der Burg, E.M.E. Verdegaal

Development of methodology: S.M. Melief, V.V. Visconti, E.M.E. Verdegaal

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.M. Melief, E.H.W. Kapiteijn, E.M.E. Verdegaal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.M. Melief, M. Visser, M. van Diepen, J. Oosting, S.H. van der Burg, E.M.E. Verdegaal

Writing, review, and/or revision of the manuscript: S.M. Melief, M. van Diepen, E.H.W. Kapiteijn, J.H. van den Berg, J.B.A.G. Haanen, V.T.H.B.M. Smit, S.H. van der Burg, E.M.E. Verdegaal

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.M. Melief, M. Visser, J.H. van den Berg

Study supervision: S.H. van der Burg, E.M.E. Verdegaal

We would like to acknowledge the NKI-AVL Core Facility Molecular Pathology & Biobanking (CFMPB) for supplying NKI-AVL Biobank material and/or laboratory support.

This study was supported by the Landsteiner Foundation for Blood Transfusion Research (project number 1207) and the Dutch Cancer Society (KWF UL 2012-5544).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.