Accumulation of Circulating CCR7⁺ Natural Killer Cells Marks Melanoma Evolution and Reveals a CCL19-Dependent Metastatic Pathway

Immune checkpoint therapy has changed prognoses for many melanoma patients, increasing their overall survival (1, 2). These successes in melanoma therapy have emphasized the role of the immune response in controlling tumor burden. Temporal cascades of gene mutations characterize the natural history of melanoma from the appearance of the early malignant nevus to its metastatic systemic spread (3). However, less is known regarding changes in the immune response as the disease progresses. Analysis of circulating T and NK cells during melanoma progression may identify prognostic biomarkers that could be assayed through patient liquid biopsies and mark disease progression.

Most melanoma patients at diagnosis have already developed local or disseminated lymph node metastases, but do not have visceral metastases. Prognoses for melanoma patients with early-stage disease (II, III) are more favorable than prognoses for patients with stage IV disease.

Self-renewing cancer stem cells (CSC) are thought to be responsible for metastatic spread (4–6). CSCs express chemokine receptors that are involved in CSC phenotype promotion and/or maintenance (7).

NK cells, members of the innate lymphoid cell group 1 (ILC1) family (8), are the most abundant ILC population found in circulating blood. NK cells are cytotoxic lymphocytes that have little effect on primary tumor lesions (9–11) but can control solid tumor metastatic spread (12). NK cell recognition of melanoma cells has been reported both in vitro and in vivo (13–15). Human NK cells are regulated by activating and inhibitory receptors. Activating receptors include Natural Cytotoxicity Receptors (NCRs) (NKp30, NKp44, and NKp46), NKG2D, and DNAM-1, which recognize stress-inducible molecules expressed on the tumor cell surface. Inhibitory receptors are mainly HLA-class I recognizing KIRs that induce NK cell tolerance (16). In humans, circulating NK cells fall into two subsets, CD56^bright and CD56^dim NK cells (17). The proportion of the two subsets varies with anatomical localization (16). CD56^bright NK cells prevail in secondary lymphoid organs and uterus, produce cytokines, and have a low cytotoxic potential. CD56^dim NK cells are effector cells that lyse cancer and virus-infected cells. Moreover, CD56^dim NK cells derived from individuals previously exposed to pathogens, such as human cytomegalovirus (HCMV), may include “memory-like” NK cells (18). Unknown cofactors associated with HCMV infection may induce the onset of a fully mature NK cell subset that is characterized by expression of the inhibitory checkpoint protein PD-1 (19).

Much is known about the regulation of NK cell–mediated cytotoxicity (20). NK cells selectively recognize melanoma metastatic cell lines (14) derived from tumor-infiltrated lymph nodes. We showed that the presence of CD57⁺KIR⁺CCR7⁺CD3⁻ NK cells in melanoma-infiltrated lymph nodes exerts an autologous antimelanoma cytotoxicity and frequency of such cells may predict patient survival (21). The NK cell subset repertoire provides a number of variables that are associated with melanoma patients' response to anti-immune checkpoint treatment with ipilimumab (22). We and others demonstrated that NK cells are able to control CSC-mediated lung metastasis (23–25). To better elucidate the immune pathogenesis of melanoma, we here analyze circulating NK and T cells subpopulations, as defined by receptor expression, in the context of melanoma. We focused particularly on cell subsets expressing chemokine receptors that control cell migration and the cytochemokine serum environment. Our objective was to understand changes in the immune response associated with melanoma clinical evolution. Our investigations led to insights concerning the chemokine biology of melanoma cells, particularly a subpopulation of melanoma cells that are CSC. Altogether, our results suggest a role for the CCR7–CCL19 pathway during melanoma progression, in that a subpopulation of tumor cells may exploit the immune system through chemokine signaling.

For the characterization of changes happening in T and NK cell compartments in melanoma patients during disease progression, 166 melanoma patients (42 stage III and 124 stage IV) were enrolled in Italy at the NCI Fondazione “G. Pascale” of Naples, and 24 stage IV melanoma patients were recruited at the Oncology clinic Karolinska University Hospital, Stockholm, Sweden. For each patient cohort, Ethical Committees associated with NCI of Naples and Karolinska University Hospital, Stockholm, granted ethical permission. Written informed consent was obtained from all patients in accordance with the Declaration of Helsinki for the use of human biological samples for research purposes. Stage III melanoma patients did not receive any treatment at the time of the enrollment, whereas stage IV were naïve or had been subjected to different types of chemotherapy. For lymphocyte compartment analysis and functional assays, peripheral blood mononuclear cells (PBMC) were isolated from 80 and 29 patients, respectively. Cytokine quantification was performed on 88 patients.

PMBCs from 42 persons and sera from 9 persons, all sex- and age-matched healthy donors, were also isolated as controls at the Pugliese-Ciaccio Hospital and University Magna Graecia of Catanzaro, Catanzaro, Italy. The experiments were performed once per patient.

PBMCs from 80 melanoma patients and 42 healthy donors were isolated by Biocoll separating solution (Biochrom AG) density gradient centrifugation. For functional experiments requiring NK cells, these cells were isolated from healthy donor PBMCs using human NK Cell Isolation Kit (Miltenyi Biotec) according to the manufacturer's instructions. The purity of isolated NK cells was >95%, as assayed by flow cytometry.

The K562, Huh7, a2780, RKO, and HDFa cell lines were obtained from ATCC in 2013 (RKO and K562), 2014 (Huh7), 2015 (HDFa), and 2016 (a2780). Melanoma cell lines were obtained, based on informed consent, from surgical melanoma specimens of patients admitted at the Fondazione IRCCS Istituto Nazionale dei Tumori, Milan (2009), Istituto Nazionale Tumori – IRCCS “Fondazione G. Pascale,” Naples (2018), and San Raffaele University Hospital, Milan (2013; Supplementary Table S1). K562, a2780, and all the melanoma cell lines were cultured in RPMI-1640 medium (Life Technology) supplemented with penicillin (100 IU/mL) and streptomycin (100 mg/mL) and 10% FBS. Huh7 and RKO cell lines were cultured in DMEM (Life Technology) supplemented with penicillin (100 IU/mL) and streptomycin (100 mg/mL) and 10% FBS. The HDFa cell line was cultured in DMEM (Life Technology) supplemented with penicillin (100 IU/mL) and streptomycin (100 mg/mL) and 20% FBS. For all experiments, cells were used within 2 weeks after thawing.

Melanoma stem cell isolation from metastatic melanoma specimens and propagation was performed as previously described (26), in accordance with the ethical standards on human experimentation. Melanoma stem cells were constantly authenticated by using the short tandem repeat (STR) system (GlobalFilter STR Kit; Applied Biosystems) followed by DNA sequencing (ABIPRISM 3130 genetic analyzer; Applied Biosystems). Cells, growing as multicellular spheres, were cultured in serum-free medium supplemented with epidermal growth factor (EGF; 20 ng/mL; PeproTech) and basic fibroblast growth factor (bFGF; 10 ng/mL, PeproTech), using ultra-low attachment flasks (Corning Incorporated). Melanoma stem cells were washed twice with PBS and cultured in attachment conditions in 10% FBS DMEM for at least 21 days to obtain sphere-derived adherent cells (SDAC).

To exclude mycoplasma infection, cells were routinely analyzed by the MycoAlert PLUS Mycoplasma Detection Kit (Lonza). In order to assess the sphere-forming capacity of CSCs, 1,000 cells/mL were plated in ultra-low attachment flasks. After 10 days, spheres that reached ≥ 50 μm of diameter were counted. In order to assess the tumorigenic capacity of CSCs, a total of 25 × 10³ CSCs were suspended in serum-free medium 1:1 matrigel (BD Biosciences Pharmingen) and subcutaneously injected into NOD/SCID mice.

All the used cell lines were maintained in culture for no longer than 3 weeks and were not authenticated in the past year.

Thawed PBMCs from metastatic melanoma patients and healthy donors, as well as fresh melanoma cells and CSCs, were subjected to immunofluorescence staining (Supplementary Table S2). The isotype-specific goat anti-mouse was from Southern Biotechnology. For NCR ligand detection, NKp30-Fc and NKp46-Fc were obtained as previously described (27). After incubation, cells were washed twice with PBS 1X, resuspended in FACS Flow, and acquired by FACS CANTO II, FACS ARIA I, Accuri C6, or a FACS Verse flow cytometer (BD Biosciences). 7 AAD Staining Solution (BD Biosciences) was added before each acquisition to distinguish between dead and live cells. In all the experiments, isotype-matched controls were used to set up the negative values. Data were analyzed using FlowJo version 10, version 9.3.1 software analysis (Treestar US) or FAC Suite software (BD Biosciences).

To quantify the cell surface expression of CD107a, degranulation assays were performed. Thawed lymphocytes derived from 27 melanoma patients (4 stage III and 25 stage IV) and 18 healthy donors were cultured at 37°C in 5% CO₂ in the presence of 8 μL of PE-conjugated CD107a/IgG1 antibody (BD Biosciences) in U-bottom 96-well plates. After 1 hour, Brefeldin A (10 μg/mL; Sigma Aldrich) was added for 3 hours of incubation. Cells were then collected, washed with PBS 1×, and stained with anti-CD56APC and anti-CD3FITC (Miltenyi Biotec) and acquired as described above.

Resting NK cells were cocultured at 37°C in 5% CO₂ with melanoma cells at 5:1 ratio in flat-bottom 12-well plates (Corning Incorporated) supplied with 1640-RPMI (Life Technology) supplemented with penicillin (100 IU/mL) and streptomycin (100 mg/mL) and 10% FBS. After 4 hours, cells were collected, stained, and acquired as described above.

Based on a protocol described elsewhere (28), cytotoxic assays were performed using fluorescent 5,6-carboxy-fluorescein-diacetate (CFDA). In brief, target cells were labeled with 150 μg/mL of CFDA-mixed isomers (Invitrogen) for 30 minutes and then incubated in 96-well U-bottom plates at 37°C in a humidified 5% CO₂ incubator for 4 hours with freshly purified allogenic NK effector cells at different effector:target (E:T) ratios. Target cell–specific lysis was analyzed by flow cytometry (FACS CANTO II and FACS ARIA I; BD Biosciences) and calculated as: % of specific lysis = (CT − TE)/CT × 100, where CT indicates target cells' mean fluorescence in control tubes and TE indicates target cells' mean fluorescence in tubes containing effector cells. Data were collected and analyzed as described above.

Hematoxylin and eosin (H&E) staining was performed on paraffin-embedded sections of patient-derived primary melanoma, relative metastasis, and xenograft tumor tissues, according to the manufacturer's instructions.

Immunofluorescence staining was performed on 5-μm-thick embedded sections of human melanoma tissues and on melanoma CSCs cytospinned on glass slides. Cells were fixed in 2% paraformaldehyde for 20 minutes at 37°C. Slides were exposed overnight at 4°C to CD44 (BU75, mouse IgG2a,k Ancell), CD271 (C40-1457, mouse IgG1a,k BD), ABCB5 (N13, goat IgG, SantaCruz), CD166 (MOG/07, mouse IgG2a,k, Novocastra), CCR7 (goat, Abcam), PD-L1 (mouse IgG1, R&D Systems), Gal-9 (rabbit IgG1, Thermo Scientific), or isotype-matched controls IgG2a (mouse monoclonal Ancell), IgG1k (mouse BD), IgG (goat Thermo Fisher Scientific). Primary antibodies were revealed using Alexa Fluor-488 anti-rabbit, mouse, or goat secondary antibodies. Nuclei were counterstained with Toto-3 iodide (Molecular Probes). In patient-derived primary melanoma and metastasis tissues, melanoma cells were distinguished from lymphocytes on the basis of cell morphology and size. H&E staining was performed by incubating sections with hematoxylin for 2 minutes and eosin for 1 minute.

To quantify serum cytokines, samples of thawed sera from 112 patients and 9 healthy donors were analyzed using the biochip analyzer Evidence Investigator (Randox Labs) and the “Cytokine Array I” kit (Randox), for the simultaneous quantification of the following cytokines: IL2, IL4, IL6, IL8, IL10, IL1a, IL1b, vascular endothelial growth factor, interferon-g (IFNg), monocyte chemotactic protein-1 (MCP1), tumor necrosis factor-a (TNF-a), and EGF. ELISA kits following manufacturer's instructions evaluated the concentrations of IL15, IL21, CCL19, and CCL21 in patients' serum and supernatants obtained from melanoma cell lines and melanoma CSC cultures; R&D Systems for IL15, BioVendor for IL21, and Aviva Systems Biology for CCL19 and CCL21. Chemokine secretion from melanoma cells was assessed by measuring CCL19 and CCL21 concentration in 10 different melanoma cell lines and 7 melanoma CSCs using ELISA kits following the manufacturer's instructions (Aviva Systems Biology).

Data obtained from multiple experiments were analyzed for statistical significance. Data from two related groups were analyzed using the paired Student t test or Wilcoxon signed rank test for samples that were or were not normally distributed, respectively. Data from two unrelated groups were analyzed using an unpaired Student t test or Mann–Whitney test for samples that were or were not normally distributed, respectively. Data from three related groups were analyzed using a two-way analysis of variance (ANOVA) followed by Bonferroni correction. Data from three unrelated groups were analyzed using one-way ANOVA followed by Bonferroni correction or Kruskal–Wallis test followed by Dunn correction for samples that were or were not normally distributed, respectively. P values < 0.05 were considered statistically significant. Kaplan–Meier (KM) curves were used to compare the survival of patients below or above median CCL19 serum concentration. The log-rank test was used to compare the KM curves and calculate the P value. Statistical computations were performed using the GraphPad Prism 5.0 software.

SIMCA, version 14 (MKS Data Analytics Solutions), was applied for multivariate analysis (29). A total of 85 variables, including immune profile and biographical variables, were applied in the analysis.

Orthogonal Projections to Latent Structures and Discriminant Analysis (OPLS-DA) was used to distinguish groups and identify parameters characterizing each group. As a development of classical principal component analysis, in OPLS-DA the systematic variation in the data is summarized into scores (T) that represent the systematic variation in the N-dimensional variable space related to a Y-variable outcome. In this projection method, all the variations related to the separation between groups are present in the predictive component(s) t[x], while the variation unrelated to separation between groups is visualized as “orthogonal components” to [x]. The contribution of each variable to the phenotype of each group of samples is indicated by the regression coefficient, that ranges from −1 (perfect negative correlation) to +1 (perfect positive correlation). OPLS-DA model quality is assessed by internal cross-validation and presented as the percentage of data explained and predicted by the models. A good biological model is defined as having > 40% predictive capacity.

The model was built on the basis of excluding biologically irrelevant and/or redundant variables (i.e., subsets with no signal) and screening for outliers (samples deviating significantly both from their own group and the global OPLS-DA model).

To investigate the correlation between immune variables and melanoma progression, the immune profile of PBLs from 42 healthy donors and 80 melanoma patients, grouped according to TNM classification of the American Joint Committee on Cancer (stage III, n = 15; stage IV, n = 65; ref. 30), was analyzed by flow cytometry. Patients at stage IV were analyzed in a previous study (22); those profiles were used here to compare with immune profiles obtained from patients at stage III.

Data from the immune profiles, together with biographical variables (Supplementary Table S3), were used to create a multivariate model. Based on the principal component analysis, Orthogonal Projections to Latent Structures Discriminant Analysis (OPLS-DA) evaluates all the variables simultaneously, giving them the same importance independent of the value range. This approach allows the assessment and display of both the covariation between variables and the correlation to a specific group of samples. Because the components are orthogonal, the difference between groups is represented by the first principal component (horizontal axis), whereas orthogonal components (vertical axis) stand for unrelated variation. Compared with traditional methods for statistical analysis, the advantage of using OPLS-DA models for biomarker identification is that a smaller cohort of patients is needed for the analysis.

The OPLS-DA model was built including only samples characterized for at least 50% of the variables considered. Thus, 37 healthy donors and 11 stage III and 42 stage IV melanoma patients were included. The OPLS-DA model gave a good separation between healthy donors and the two groups of patients (Fig. 1A), explaining 79% of the difference between them with a cross-validated predictive capacity of 69%. Overall, 66 variables contributed to distinguishing the groups. The ten most significant variables characterizing each group as well as the relative contribution of each characterizing variable are reported in Fig. 1B–D.

The immune profile characteristic of healthy donor featured high frequency and expression of NKp46 on the CD56^dim NK cell compartment. The stage III immune profile was characterized by high frequencies of chemokine receptors (CXCR2, CCR2) and NK activating receptors (NKG2D, NKp30, NKG2C). Stage IV melanoma patients were characterized by elevated percentages of CCR7 and CXCR2 on NK cells and high expression of inhibitory receptors (Tim-3, PD-1) on both NK and T cells.

Overall, the immune variables applied in the multivariate analysis identified separate groups and identified the variables best correlating with disease staging. Because the OPLS-DA identifies even those variables that are significant only when taken together with other parameters, variables reported by the model were also analyzed in univariate mode to select biomarkers that remain significant even when considered on their own. The complete list of variables characterizing the immune profile of each group, together with the univariate validation, is reported in Supplementary Table S4.

The functional features of circulating NK and T cells at different disease stages were analyzed to understand their possible role in the pathophysiology of melanoma. Lymphocyte activation was investigated in a subset of patients. Representative plots for NK and T cells degranulation are showed in Fig. 2A and B. NK cells from melanoma patients showed higher spontaneous degranulation compared with healthy donors (Fig. 2C). On the other hand, T-cell degranulation was higher in stage IV melanoma patients compared with healthy donors (Fig. 2D).

To characterize cytokine profiles, sera from stage III and stage IV melanoma patients were analyzed and compared. Most of the stage IV melanoma patients had been previously characterized (22). The univariate analysis showed significant differences between stages III and IV for 5 of 15 cytokines: CCL2, IL6, CXCL8, IL15, and CCL19. All the cytokines showed a steady longitudinal serum concentration increase paralleling melanoma disease clinical evolution. Because CCL2, CXCL8, and CCL19 reached the highest concentration at stage IV, we analyzed expression of their cognate receptors on both NK and T-cell subpopulations (Supplementary Table S4). Among them, only the percentage of CCR7 on the CD56^bright NK cell subset displayed a pattern similar to the respective chemokine (Fig. 3A and B). Indeed, the frequency of CCR7⁺CD56^bright NK cells reached its peak in stage IV melanoma patients, as did CCL19 concentration, suggesting an ectopic recruitment of this NK subset in the blood. Thus, this pathway may feature in melanoma progression.

The increase of CCL19 in sera may depend on its production by melanoma cells. Indeed, melanoma can produce IL8 and MCP-1 (31–33). Thus, we speculated that the high CCL19 observed in stage IV melanoma patients could be due to ectopic production from melanoma cells. To test this hypothesis, we measured CCL19 concentrations in supernatants from primary and metastatic melanoma cells, melanoma CSCs, fibroblasts, and other solid tumors (hepatic, ovarian, and colon carcinoma) cells. We observed CCL19 secretion by melanoma cells (Fig. 3C).

Patient serum concentrations of the other four cytokines are listed in Supplementary Table S5. Serum concentrations of CCL2 exceeded the physiological range in all melanoma stages, whereas concentrations of the other three cytokines (IL6, CXCL8, and IL15) reached pathological levels only in stage IV melanoma. We previously demonstrated that melanoma cells produced CCL2, IL6, and CXCL8 (21). These soluble factors were found in culture supernatants of infiltrated lymph nodes from melanoma patients and were able to induce CCR7 expression on CD56^bright NK cells (21). The IL15 cytokine induced phenotypic changes in NK cells from melanoma patients (22).

CCR7 is a receptor involved in lymph node homing, and lymph nodes are the first anatomical sites vulnerable to melanoma metastatic spread. Therefore, we hypothesized that CCR7 could be ectopically expressed in melanoma metastatic cell lines, driving their lymph node metastatic colonization. Thus, we analyzed a panel of patient-derived melanoma cell lines for CCR7 expression. We also analyzed other surface molecules key to NK-melanoma cell cytotoxic synapse formation, such as MICA/B, PVR, HLA-class I, PD-L1, and Galectin-9. We analyzed ten melanoma cell lines of different anatomical origins. CCR7 was always expressed by a small fraction (1%–5%) of melanoma cells (Fig. 4A) that appears to be a specific subpopulation. Indeed, such CCR7⁺ melanoma cells coexpressed CCR7 with two immune checkpoints ligands, PD-L1 and Galectin-9 (Fig. 4A and B), recognized by PD-1 and Tim-3 on lymphocytes, respectively.

Thus, in each of ten different melanoma lines, a small fraction of cells expresses a chemokine receptor known for lymph node homing and also expresses PD-L1 and Galectin-9 that could protect from T and NK cell cytotoxic attack. Other molecules involved in NK cell cytotoxic synapses were measured. We found that CCR7⁺ melanoma cells displayed low expression of MHC class I molecules and PVR, but high expression of NCR ligands (Fig. 4C). On the other hand, the activating ligands MICA/B and ULBP2, recognized by NKG2D, showed variable expression, being expressed on only four and six of the ten cell lines analyzed, respectively (Supplementary Fig. S1A and S1B). The low MHC class I and the high NCR ligands expression on CCR7⁺ melanoma cell surface suggested an increased susceptibility to NK cell cytotoxicity.

To test whether NK cells could target CCR7⁺ melanoma cells, we performed cell cocultures using melanoma cell lines and NK cells. After coculture with fresh NK cells, a reduction in the frequency of CCR7⁺PD-L1⁺Galectin-9⁺ (referred to as CCR7⁺ cells) subpopulation was observed (Fig. 4D, left), whereas the CCR7⁻ PD-L1⁻ Gal-9⁻ (referred to as CCR7⁻ cells) melanoma cell population was not affected by NK cell exposure (Fig. 4D, right). Furthermore, CCR7⁺ melanoma cells that survived after the NK cells coculture were found to express higher HLA-class I molecule levels (Supplementary Fig. S1C). PD-L1 expression was not affected by NK cell exposure, thus ruling out the possible effect of IFNγ produced by NK cells (34) in the induction of MHC class I molecules on the resistant CCR7⁺ melanoma cells (Supplementary Fig. S1D).

CCR7⁺ melanoma cells are characterized by potentially higher migration capability, lower expression of membrane-associated MHC class I molecules, low frequency within the bulk melanoma cell population, and increased susceptibility to NK cell–mediated killing. These features resemble those of CSCs (24, 35). To test the hypothesis that CCR7⁺ melanoma cells are CSCs, we compared CCR7⁺ cells to CSCs derived from melanoma generated with previously established methods (Table 1 and Fig. 5). First, we checked melanoma patient-derived CSCs for their ability to grow as spheres and to generate highly proliferating differentiated cells. We also tested for their aptitude to initiate tumor growth in immunocompromised mice (Fig. 5A and B). We showed that melanoma CSCs generated xenografts with histology typical of human melanomas (Fig. 5B). CSC stemness was confirmed by the expression of known stemness markers such as CD44, CD271, ABCB5, and CD166 (Fig. 5C; Supplementary Fig. S2A; refs. 36–38). Finally, we measured the percentage of CSCs expressing CCR7 and observed that, in the four lines tested, CSCs homogenously expressed CCR7 (Fig. 5D). PD-L1 and Galectin-9 were also expressed on CSCs but to a different extent (Supplementary Fig. S2B). Thus, CCR7 expression seems to identify CSCs.

Because we have previously demonstrated that human colon adenocarcinoma–derived CSCs (25) and human and mouse breast adenocarcinoma CSCs (20) are susceptible to NK cell cytotoxic recognition, we evaluated melanoma CSC susceptibility to NK cell–mediated lysis. Indeed, freshly purified NK cells showed an enhanced capability to recognize and kill melanoma-derived CSCs (Fig. 5E–G).

To summarize, well-defined melanoma CSC share a number of characteristics with the melanoma subpopulations explored in this paper: expression of CCR7, PD-L1, and Gal-9 as well as high susceptibility to NK cells. Thus, we hypothesize that this melanoma cell population plays a role in metastasis.

We then evaluated the clinical relevance of CCR7, Gal-9, and PD-L1 in the process of metastasis dissemination and outgrowth. Immunohistochemical analysis on tissues derived from patients who have melanoma metastases revealed the enrichment of CCR7⁺ cells in melanoma metastatic lesions of the lymph node and parotid gland as compared with primary melanoma (Fig. 6A–C). Gal-9 and PD-L1⁺ cells concomitantly expressed CCR7, suggesting that CCR7 is required for the metastatic outgrowth of aggressive melanoma cells (Fig. 6D).

We also analyzed the potential correlation between CCL19 serum concentrations and survival in a small cohort of patients. Our preliminary data (Supplementary Fig. S3) showed a tendency for a better overall survival in those patients who had less CCL19 in their blood; this observation needs to be validated in a larger patient cohort.

Immune profile analysis of melanoma patients at different stages of disease progression suggests that melanoma progression leads to changes in peripheral blood NK and T-cell subsets. In stage III melanoma, the NK cell compartment is dominated by subpopulations expressing CXCR2 and CCR2, both of which are receptors for the melanoma-produced chemokines IL8 and CCL2 (MCP-1), possibly indicating a more robust migration of NK cells in early-stage melanoma disease. Indeed, both IL8 and CCL2 act as tumor autocrine growth factors increasing melanoma cell migration in metastatic lesions (39, 40). Thus, it is conceivable to speculate that these NK cell subsets may be able to migrate to the early melanoma metastatic foci. At this stage, circulating NK cell subsets expressing activating receptors (NKG2D, NKp30, and NKG2C) prevail, whereas the T-cell compartment is characterized by a reduced frequency of mature CD57⁺CD8⁺ cells associated with low levels of Tim-3 and CCR7. Circulating CD56^dim NK cells in stage III melanoma patients appear to be activated and display basal degranulation without any stimulation in vitro. Later in the disease's evolution (stage IV), a similar feature is evident for T cells. This sequential activation, involving first the innate and then the adaptive immunity, recapitulates the physiological dynamics of the immune response.

On the other hand, stage IV melanoma is characterized by an increased frequency and accumulation of CCR7⁺ CD56^bright NK cells in patients' blood. This phenomenon could be due to the cumulative effects exerted by CCL2, IL6, and IL8, as previously demonstrated (21), whose serum concentrations are increased in the latest stage of the disease. Serum at stage IV has higher concentrations of the CCR7-ligand CCL19 than serum at stage III. The ectopic presence of CCL19, otherwise characteristic of lymph nodes, could be part of a pathogenic mechanism interfering with NK cell migration in melanoma-infiltrated lymph nodes. T cells expressing CCR7, in contrast with CD56^bright NK cells, did not increase in the blood of stage IV melanoma patients, suggesting that T cells may overcome the increased CCL19 sera concentration and redistribute to peripheral tissues. This could be explained by at least two, not mutually exclusive, mechanisms: (i) other lymph node homing receptors expressed by T cells, such as CD62L, could drive them to the lymph nodes independently of the CCR7–CCL19 axes, and (ii) the cytokine milieu associated with stage IV melanoma (which includes IL6, CCL2, and IL8) does not induce CCR7 expression on T cells.

Blood NK cells from stage III and IV melanoma patients showed increased expression of CD107a when explanted to in vitro culture without further stimulus. This spontaneous degranulation may reflect the recent in vivo activation of the cells (that the cells have been in action recently in vivo). CD107a expression might thus mimic a “smoking gun” following the NK cell attack on circulating melanoma cells. However, there are other possible explanations, such as generalized activation of NK cells by cytokines and other factors associated with advanced disease.

The scenario emerging from our study is that melanoma cells and related CSCs may produce CCL19, as described for cervical cancer (41). The increased serum concentration of CCL19 found in stage IV melanoma patients could be explained by the higher disease burden, where an increased number of CSCs actively secrete the chemokine. This could promote melanoma metastatic dissemination through at least two mechanisms: (i) CSCs migration from skin to blood circulation through the molecular combination between their CCR7 expression and the high patient's CCL19 blood concentration, (ii) the high blood concentration of this chemokine, normally present only in the lymph nodes, retains CD56^bright NK cells in the blood and prevents their progression to the lymph nodes, reducing their frequencies in the melanoma-infiltrated lymph nodes as we have previously demonstrated (21). However, it is also possible that CD56^bright NK cells resident in lymph nodes may be rerouted to the blood. Indeed, studies indicate that cross-talk between NK cells and DCs in the lymph nodes affects the CD8⁺ T-cell antitumor immune response, as reviewed elsewhere (42, 43). NK cells can edit the DC population (i) by physical elimination of immature DC or (ii) by IFNγ production that induces full maturation of the DC skewing toward a Th1 immune response (44, 45). Thus, we speculate that the perturbation in CD56^bright NK cell migration in the lymph nodes may contribute to the failure of the antitumor immune response in the late stage of melanoma disease. Our data complement findings demonstrating that NK cell frequencies in melanoma tissues positively correlate with anti–PD-1 immunotherapy and overall survival (46).

Regardless of the pathological meaning of CCL19 concentration and the accumulation of CCR7⁺CD56^bright NK cells in the blood circulation, our study demonstrates concomitant expression of CCR7 on melanoma CSCs and NK cells and that NK cells can recognize and eliminate melanoma CSCs that drive the disease's metastatic spread.

Ex vivo analysis showed that CCR7 is expressed more in metastatic melanoma cells than in melanoma cells at the primary tumor site. These data corroborate and expand previously reported findings (47, 48), suggesting that melanoma cells rely on CCR7 during metastasis.

Moreover, Gal-9⁺ and PD-L1⁺ melanoma cells concurrently expressed CCR7 showing the presence of an aggressive subpopulation of CCR7⁺ cells endowed with immune evasion capabilities. Our results provide a foundation for developing melanoma therapies that could interfere with this metastatic pathway by the use of monoclonal antibodies targeting CCL19.

A.M. Grimaldi has received speakers bureau honoraria from Bristol, Myers Squibb, Novartis, Roche, and MSD and is a consultant/advisory board member for Novartis and MSD. P.A. Ascierto reports receiving a commercial research grant from Bristol-Myers Squibb, Roche-Genentech, and Array and is a consultant/advisory board member for Bristol-Myers Squibb, Roche-Genentech, Medimmune, AstraZeneca, Syndax, SunPharma, Idera, Sanofi, Ultimovacs, Sandoz, Immunocore, MSD, Array, Novartis, Merck Serono, Incyte, Pierre Fabre, Genmab, and Newlink Genetics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K. Karre, E. Carbone

Development of methodology: P.A. Ascierto, E. Carbone

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.M. Cristiani, A. Turdo, T. Apuzzo, M. Capone, G. Madonna, D. Mallardo, C. Garofalo, A.M. Grimaldi, R. Tallerico, E. Marcenaro, S. Pesce, V. Agosti, F.S. Costanzo, A. Rizzo, E. Carbone

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.M. Cristiani, A. Turdo, V. Ventura, R. Tallerico, E. Marcenaro, S. Pesce, P.A. Ascierto, M. Todaro, E. Carbone

Writing, review, and/or revision of the manuscript: C.M. Cristiani, A. Turdo, K. Karre, P.A. Ascierto, M. Todaro, E. Carbone

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Ventura, M. Capone, G. Madonna, D. Mallardo, E.D. Giovannone

Study supervision: M. Todaro, E. Carbone

Other specify (cytometry consultancy): G.D. Zotto

S. Pesce is the recipient of a fellowship awarded by Fondazione Umberto Veronesi. A. Turdo is the recipient of an AIRC fellowship. We thank Claudia Cantoni (University of Genoa and Istituto Giannina Gaslini, Genoa) for supplying the NKp30-Fc and NKp46-Fc fusion protein. This work was supported by grants to E. Carbone from the Italian Association for Cancer Research (IG15521); Wenner-Gren Stiftelserna, Sweden; Italian Ministry of Health grant “Progetto Ricerca Finalizzata 2011–2012”; grant CO-2011–02348049 co-funded by Bristol Myers Squibb, Fondazione Melanoma Onlus, Naples; E. Marcenaro from AIRC-Special Program Metastatic disease: the key unmet need in oncology 5 per mille 2018 (21147) and Progetto Roche Ricerca 2017. M. Todaro received funding from the Italian Association for Cancer Research (IG14415) and (PSN) 2015-Linea Progettuale 6, Azione 6.2,CUP I76J17000470001. P.A. Ascierto was supported by the Italian Ministry of Health “Ricerca Corrente.” G. Madonna was supported by the Institutional Project “Ricerca Corrente” Fondazione “G. Pascale.”

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.