Single‐Cell Immune Mapping of Melanoma Sentinel Lymph Nodes Reveals an Actionable Immunotolerant Microenvironment

Understanding the cellular response of the immune system to cancer cells in the draining regional lymph nodes may unveil new insights into how to stimulate clinically meaningful immunity against cancer. The aptly named “sentinel lymph node” (SLN) is the first lymph node encountered by a metastatic cancer cell and serves as a key immune recognition site that shapes subsequent antitumor innate and adaptive immune responses (1). SLNs are readily identified by surgeons through radioisotope and dye tracing, and the presence of cancer cells in the SLN is a major prognostic indicator predicting cancer recurrence and survival for patients with clinically localized cutaneous melanoma.

Among all cancer types, melanomas have the highest frequency of somatic mutations, induce robust and diverse CD8⁺ T-cell responses, and are highly responsive to immune checkpoint inhibitors (ICI) that target PD-1 and/or CTLA-4 (2). Furthermore, SLNs are routinely evaluated for the presence of melanoma cells in patients who have >1-mm-thick melanomas; the presence of tumor cells in the SLN confers a higher risk of recurrence and death. Given the highly mutagenic nature of melanoma cells and the established clinical significance of their migration to SLNs, a comprehensive examination of their immune repertoire may identify new approaches to stimulate immunity against melanoma and other cancer types.

Primary melanoma-derived factors convert draining SLN(s) into an immune-privileged site (3). Previous studies have shown that adaptive immunity in the SLN is biased towards tolerance in response to the primary melanoma (1). These alterations include suppression of migratory dendritic cells (DC; ref. 4) and increased infiltration of T-regulatory cells (5)—both cellular events that contribute to the induction of regional cross-tolerance and suppression of effective antimelanoma-antigen T-cell responses. Moreover, there is growing evidence that immune cell-mediated tolerogenic changes in tumor-draining SLNs can precede local disease spread and melanoma growth (6–8), setting the stage for recurrence and metastatic spread to distant sites. The potential clinical impact of these changes is significant, because increased T_regs in melanoma⁺ SLNs have been found to correlate with reduced overall survival (1, 9), whereas increased mature DC-LAMP⁺ dendritic cells in SLNs have been found to correlate with increased overall survival (10). Importantly, our current understanding of the impact of metastatic melanoma cells on the immune repertoire in SLNs remains rudimentary, and an unbiased survey of the SLN immune microenvironment in patients with melanoma is anticipated to open new avenues for the development of novel immunotherapeutic approaches.

We have developed a robust immune profiling strategy to map the immune microenvironment of SLNs resected from patients with clinical stage I to III cutaneous melanoma. We used mass cytometry by time-of-flight (CyTOF), flow cytometry, and T-cell receptor immunosequencing to allow for the single cell analysis of the distinct immune cell compartments that reside in the SLN of each patient with melanoma. We identified unique tumor-driven T, NK, and innate immune cell signatures that are present in stage III melanoma-bearing SLNs, but absent in stage I/II non–melanoma-bearing SLNs. These immune cell changes are expected to compromise antimelanoma immunity, resulting in an increased risk of locoregional and distant metastasis. Finally, on the basis of the observed high expression of the PD-1 and CTLA-4 checkpoint molecules on T cells, along with multiple marked changes in innate immune cell signatures, this study provides the key rationale for the clinical testing of neoadjuvant approaches to block checkpoint molecules and to activate innate cell immune responses.

Human sentinel lymph node (SLN) tissue samples were obtained from patients with cutaneous melanoma who underwent SLN biopsy at the University of Louisville. Informed written consent was obtained from all patients on a protocol that was approved by the University of Louisville Institutional Review Board (IRB number: 08.0491). These studies were conducted in accordance with U.S. Common Rule. All patients underwent SLN biopsy using a combination of radioactive tracer and isosulfan blue dye. A random portion of the most radioactive SLN (2 mm in greatest dimension or one-quarter of the SLN, whichever was smaller) from each patient was obtained fresh at the time of surgery. All patients with a SLN biopsy were either positive (MEL⁺ SLN; pathologic stage III) or negative (MEL⁻ SLN; pathologic stage I/II) for metastatic melanoma by pathologic analysis of serial sections by hematoxylin and eosin (H&E) staining and IHC staining for MART-1 and SOX-10. A portion of the SLN was enzymatically digested and cryopreserved in cryovials. Cryovials were then transferred to −150°C for storage. Details of SLN tissue processing and cryopreservation are provided in Supplementary Materials and Methods.

The collected SLN samples were divided into three cohorts. Please see Scheme I and Supplementary Table S1 in Supplementary Materials and Methods section for details of the Cohorts and the groups within. Samples in Cohort I (11 stage I/II and 11 stage III) were exclusively used for mass cytometry (CyTOF) analysis. All unstimulated samples of Cohort I were included in Group I and analyzed. A subgroup of the samples in cohort I (subgroup 1: 10 stage I/II and 5 stage III) were stimulated with PMA + Ionomycin (P + I) and analyzed using mass spectrometry (CyTOF). Samples in the Cohort II (45 stage I/II and 20 stage III) were exclusively used for flow cytometry and ImmunoSeq studies. Group I (Cohort II) consisted of all the Cohort II samples and were used for surface marker analysis. A subgroup of the samples in Cohort II (subgroup 1: 7 stage I/II and 8 stage III) were further evaluated to measure TCR clonality and frequency using ImmunoSeq analysis. All samples in Cohort III (11 stage I/II and 10 stage III) were used for sorting naïve and effector-memory T cells, which were further expanded in vitro and used for functional assays.

Melanoma cell line A375 (ATCC CRL-1619) and T2 antigen-presenting cell lines were purchased from ATCC and maintained in DMEM (Corning) or RPMI1640 (Corning) media containing 10% (v/v) FBS. We do not culture these cell lines longer than 6 to 8 weeks and all of our stocks come from thawed vials that were frozen at passage two after receiving the cells from ATCC. A375 and T2 cell lines were authenticated by ATCC cell bank using the short tandem repeat (STR) profiling.

Antibodies were either purchased pre-conjugated from Fluidigm or purchased purified, carrier-free, and conjugated in-house using the Maxpar DN3 polymer kits from Fluidigm according to the manufacturer's instructions. Cryopreserved patient SLN samples were thawed and were either unstimulated (Cohort I, Group 1) or stimulated (Cohort I, subgroup 1) with phorbol 12-myristate 13-acetate (PMA) and Ionomycin (P+I) for 4 hours and stained as described previously (11) and data were acquired on a Helios mass cytometer (Fluidigm). Data obtained from the instrument were normalized using the Nolan Lab Matlab-based normalizer (https://github.com/nolanlab/bead-normalization). Details of CyTOF mass cytometry sample preparation, staining, and antibody panel is provided in Supplementary Materials and Methods section. Briefly, cells were stained with surface antibody cocktail, washed and stained with live/dead stain—115In-Maleimide DOTA. After surface and live/dead staining, the cells were fixed, permeabilized, and stained with the intracellular antibody cocktail. CyTOF data were analyzed through Cytobank (12) and the R package Cytofkit (13). Details of mass cytometry data processing and visualization are provided in Supplementary Materials and Methods section.

Single cell suspensions from MEL⁺ SLN and MEL⁻ SLN (Cohort II, group 1) were washed twice and resuspended in FACS buffer (2% FBS + PBS) and stained with multicolor antibody (Ab) cocktails according to the manufacturer's recommendations. Stained cells were analyzed with a FACSCanto II (BD Biosciences) and data were analyzed using FlowJo v10 (BD Biosciences). FACS dot plots are represented with log-scale axes. Histograms are represented on a log-scale x-axis and a linear y-axis. Details on flow cytometry staining, representative flow plots/gating strategy, and antibody panels are provided in the Supplementary Materials and Methods section.

Genomic DNA was extracted from SLN cells (Cohort II, subgroup 1). Samples were analyzed using Adaptive's ImmunoSEQ Immune-Profiling Kit and Illumina's NextSeq platform for “deep resolution” of the TCR V-β CDR3 region. This analysis results in a 10× sequence coverage for T cells from 3.6 μg of DNA, with an input of 200,000 T-cell genomes and an output of 1,000,000 sequences. Adaptive's bioinformatics pipeline processed the data, after a quality control check, to quantitate the T-cell repertoire. Analyses of T-cell receptor sequencing data are described in detail in Supplementary Materials and Methods section.

CD8⁺ T-cell reactivity against melanoma epitopes was tested in SLN samples obtained from human leukocyte antigen (HLA)-A2⁺ patients. On day 0, before SLN removal surgery, 18 mL of blood was drawn from each patient. Peripheral blood mononuclear cells (PBMC) were isolated from heparinized blood by Ficoll density gradient centrifugation. To determine the patients' HLA-A1, HLA-A2, or HLA-A3 status, PBMCs were stained with the following antibodies: (i) A1/A36 (HLA-A1, One Lambda); (ii) MA2.1 and BB7.2 (HLA-A2); and (iii) GAP A3 (HLA-A3; obtained form ATCC) and analyzed by flow cytometry.

Cryopreserved MEL⁻ and MEL⁺ SLN cells (Cohort III) were sorted into CD8⁺ T naïve (T_N) and T effector memory (T_EM) subsets on a BD FACSAria sorter (BD Biosciences). The details pertaining to the subsequent generation of effector cells, analysis of surface markers, and in vitro co-culture killing assays are provided in the Supplementary Materials and Methods section.

Total RNA and real-time analysis was using the following primers and probes: 18S endogenous control gene (Hs99999901.s1; VIC) and KLRG1 (Hs00195153_m1; FAM). Details of RNA extraction and PCR analysis are described in Supplementary Materials and Methods section.

GraphPad Prism 8.0 software (GraphPad Prism Software, Inc.) and SAS version 9.4 (SAS Institute, Inc.) were used for statistical analyses. Two-group comparisons between control and test samples (groups compared are indicated in the respective figures) were done by two-tailed Student t tests. Multiple data comparisons were derived by one-way ANOVA followed by Tukey post hoc test. Ulceration status and gender between stage I/II and stage III groups comparisons were analyzed using Chi-squared test. Correlations between T-cell clonality and cytometry data as well as between population frequencies and LN micro-metastatic burden/tumor Breslow thickness were assessed using Spearman rank correlation. For all tests, statistical significance was assumed where P < 0.05.

The data generated in this study are available within the article and its supplementary data files.

To characterize the immune microenvironment of melanoma SLNs, we collected one most radioactive SLN sample from each of 108 patients diagnosed with primary cutaneous melanoma. After pathologic staging based on the AJCC staging system, 67 patients were diagnosed with pathological stage I/II (melanoma-negative LNs or MEL⁻ SLN) and 41 patients with pathologic stage III disease (melanoma-positive LNs or MEL⁺ SLN). Clinical characteristics are summarized in Supplementary Table S1. No significant differences were found in gender or age of the MEL⁻ and MEL⁺ SLN groups. Breslow thickness was significantly different between groups and primary tumor ulceration was more common in the MEL⁺ SLN group compared with that in the MEL⁻ SLN group (P = 0.0001).

We performed high-dimensional single-cell immune profiling using mass cytometry (CyTOF; refs. 14, 15) to analyze the SLN microenvironments. We used viSNE and SPADE algorithms to visualize CyTOF data in two dimensions while preserving single-cell resolution (16, 17). This allowed us to analyze the different immune cell lineages present in SLNs across all patients (Fig. 1A and B; Supplementary Fig. S1A). The most abundant SLN-resident immune cell lineages were T and B cells. The other immune cell lineages present in the SLNs were NK cells, neutrophils, and dendritic cells. Strikingly, CD14⁺ and CD16⁺ monocytes as well as CD14⁺ macrophage subpopulations were not detected in both MEL⁻ and MEL⁺ SLNs.

To further evaluate the immune cell architecture in SLNs in an unbiased manner, we analyzed the CyTOF data using the PhenoGraph algorithm. PhenoGraph algorithm allows for automated partitioning of high-dimensional single-cell data into subpopulations to identify common cellular communities across all patients (18, 19). For PhenoGraph clustering of T-cell subsets, we initially clustered single cells based on shared protein expression in the SLN tissue from each patient and then merged clusters from each patient to identify metaclusters that are common across all patients (Fig. 1C). PhenoGraph clustering across all SLNs (both MEL⁻ and MEL⁺) identified twenty distinct T-cell metaclusters. The absolute cell numbers in each of the T-cell metaclusters are included in Supplementary Table S4A in Supplementary Materials and Methods section. To identify major T-cell populations, a heatmap that included all metaclusters and expression levels of T-cell markers was generated. The identified subsets are labeled (Fig. 1D).

Given that the immune checkpoint inhibitors (ICI; e.g., anti-PD-1 and anti-CTLA-4 antibodies) have resulted in enduring remissions and better clinical outcomes for patients with advanced melanoma (20, 21), we examined the expression of checkpoint molecules CTLA-4 and PD-1 on all T-cell metaclusters identified collectively in both MEL⁻ and MEL⁺ SLNs. Comparison of CTLA-4 expression on all T-cell metaclusters showed high levels of CTLA-4 on γδ T cells in the SLNs (Fig. 2A and B), whereas T_regs showed very low levels of surface CTLA-4 (Fig. 2B).

Analysis of PD-1 expression on collective T-cell metaclusters showed that highest levels of PD-1 expression was observed on a distinct subset of CD4⁺ and CD8⁺ T cells (Fig. 2C and D). Among the CD4⁺ and CD8⁺ T differentiation subsets—CD45RA⁺CCR7⁺ naïve (T_N), CD45RA⁻CCR7⁺ central-memory (T_CM), CD45RA⁻CCR7⁻ effector-memory (T_EM), and CD45RA⁺CCR⁻ terminally differentiated (TEM) subsets (22)—expression of PD-1 was higher on CD8⁺ T_EM cells when compared with that in other three CD8⁺ T-cell subsets; in CD4⁺ T subsets, higher expression of PD-1 was observed on both CD4⁺ T_EM and activated T_CM cells. Expression of PD-1 on naïve T cells was very low (Fig. 2D).

Next, we evaluated the expression of CD57 on all T-cell metaclusters identified in both MEL⁻ and MEL⁺ SLNs. CD57 has been shown as a marker expressed on end-stage senescent T cells (23). Our data show that in CD4⁺ T subsets, CD57 was expressed on CD4⁺ T_EM cells. In CD8⁺ T cell subsets, CD57 marker was solely expressed on CD8⁺PD-1⁺ subset (Fig. 2E and F). Evaluation of other surface markers (heatmap in Fig. 1D) showed that CD8⁺PD-1⁺CD57⁺ T-cell subset co-expressed CD27 and CD28 markers. It appears that this subset of CD8⁺ T cells closely resemble the CD8⁺CD57⁺ TILs that accumulate in metastatic tumors, including melanoma, that fail to downregulate CD27 and CD28 markers and possess limited ability to kill tumor cells (24, 25).

Increased frequencies of CD8⁺PD-1⁺CD57⁺ and CD8⁺ T_EM subsets and reduced activated CD8⁺ T cells/Treg ratios are distinct features of melanoma-bearing SLNs. We used flow cytometry on Cohort II, group 1 patient samples (n = 45 MEL⁻ and 20 MEL⁺ SLNs, SI Scheme 1) to further delineate the phenotypes of CD8⁺ and CD4⁺ T cells. Initially, we compared the relative frequencies of total CD8⁺ PD-1⁺ T cells and CD8⁺PD-1⁺CD57⁺ T-cell subsets in MEL⁺ and MEL⁻ SLNs. Our data showed that frequencies of total PD-1⁺CD8⁺ T and CD8⁺CD57⁺PD-1⁺ cells were significantly higher in MEL⁺ SLNs (Fig. 3A and B). We also evaluated if primary melanoma growth progressively affects CD8⁺CD57⁺PD-1⁺ phenotype in the SLNs. First, we used Breslow thickness as a measure of primary tumor burden (26, 27). The frequencies of CD8⁺CD57⁺PD-1⁺ cells in all SLNs significantly correlated with primary melanoma growth (Fig. 3C).

We next sought to assess the relative frequencies of CD8⁺ T-cell differentiation subtypes (T_N, T_CM, T_EM, and TEM subsets) in MEL⁻ and MEL⁺ SLNs (22). We observed an abundance of CD8⁺ T_EM subset and a significantly lower proportions of CD8⁺ T_N subset in the MEL⁺ SLNs when compared with their respective frequencies in MEL⁻ SLNs. No significant changes were observed in the frequencies of T_CM and TEM CD8⁺ T-cell subsets (Fig. 3D). Correlation analysis showed significant correlation between primary tumor burden and frequencies of CD8⁺ T_EM subsets (Fig. 3E). SLN tumor burden, expressed as the maximum diameter of SLN tumor deposits, has been reported to be an independent prognostic factor in patients with stage III melanoma (28). Our data indicate that micrometastatic tumor burden in the MEL⁺ SLN positively correlate with frequencies of PD-1-expressing CD8⁺ T_EM (Fig. 3F), suggesting that expansion of differentiated CD8⁺ T_EM increases with growing metastases in the MEL⁺ SLNs.

Further analysis of surface marker expression revealed a distinct phenotype of CD8⁺ T cells in MEL⁺ SLNs—frequencies of activated CD8⁺ CD69⁺ T cells and activated CD8⁺ T cell/T_reg ratio were significantly diminished in MEL⁺ SLNs (Fig. 3G and H). We also evaluated if primary melanoma growth progressively affects activated T cell/Treg phenotype in the MEL⁺ SLNs and our data showed that CD8⁺CD69⁺ T cell/T_reg ratio significantly correlated with primary melanoma growth (Fig. 3I).

Analysis of other T-cell subsets revealed that the proportions of γδ T cells and CD4⁺ T_regs cells were increased whereas activated CD4⁺ CD69⁺ T cells were significantly decreased in the MEL⁺ SLN tissues when compared with that in MEL⁻ SLN tissues (Fig. 3J).

We compared the phenotypes of CD8⁺ and CD4⁺ T cells and their correlations to primary Breslow thickness in the data obtained using CyTOF on samples in Cohort I, group 1 patient samples (n = 11 MEL⁻ and 11 MEL⁺ SLNs, SI Scheme 1). MEL⁺ SLNs displayed increased frequencies of total PD-1⁺CD8⁺ T cells, CD8⁺ T_EM cells, and CD4⁺ T_regs; a lower proportion of CD8⁺ T_N subset and activated CD8⁺ T cell/T_reg ratio (Supplementary Fig. S2). However, given the lower numbers of samples, these differences did not raise to the level of significance present in the data obtained using flow cytometry above.

We employed high-throughput deep sequencing of the human TCR V-β CDR3 sequences to better characterize the expansion and clonality of the T cell repertoire in the MEL⁻ and MEL⁺ SLNs (Cohort II, subgroup 1, SI Scheme 1; ref. 29). Our data indicate that there was an increased diversity of TCR clonotypes in MEL⁺ SLNs (Supplementary Fig. S3A). A small subset of the top 100 most frequent TCR clonotypes were expanded in these MEL⁺ SLNs (Supplementary Fig. S3B). Correlation studies indicate that the frequencies of CD8⁺ T_EM cell subset uniquely and significantly correlated with TCR clonality (Supplementary Fig. S3C), whereas the frequencies of the CD8⁺ T_N cells negatively correlated with TCR clonality in MEL⁺ SLNs (Supplementary Fig. S3D), but no such correlations were significant in MEL⁻ SLNs (Supplementary Figs. S3E and S3F). These results suggest that increased clonal expansion of CD8⁺ T_EM subset occurs in MEL⁺ SLNs.

Using PhenoGraph clustering, we evaluated the expression of cytolytic markers granzyme B and perforin on all identified T-cell metaclusters in both MEL⁻ and MEL⁺ SLNs (Cohort I, group 1, SI Scheme 1). Expression of these cytolytic markers were highest in the CD8⁺PD-1⁺CD57⁺ metacluster (Fig. 4A,–D). Intriguingly, comparison of expression levels of Granzyme B with perforin in CD8⁺PD-1⁺CD57⁺ subset showed that a higher relative expression of granzyme B (Supplementary Fig. S1C); such a cellular phenotype is associated with lower tumor killing capabilities (24, 25). Granzyme B and perforin were also expressed in CD4⁺ T_EM and CD8⁺ T_EM subsets (Fig. 4B and D) albeit at much lower levels than that expressed in CD8⁺PD-1⁺CD57⁺ subset. Comparison of relative frequencies of total CD8⁺ T cells expressing cytolytic markers in MEL⁻ and MEL⁺ SLNs show relatively higher frequencies of granzyme B⁺ CD8⁺ T cells and lower frequencies of perforin⁺CD8⁺ T cells in the MEL⁺ SLNs (Fig. 4E and F).

We evaluated the cytokine profile of the different T-cell subsets within the MEL⁻ and MEL⁺ SLNs after stimulation with PMA + Ionomycin (P+I) using PhenoGraph clustering (Cohort 1, subgroup 1, Scheme 1). Intracellular markers analyzed are listed in Supplementary Table S3 in Supplementary Materials and Methods section. We detected a CD8⁺ T-cell cluster expressing cytotoxic effector molecules—CD107a, IFNγ, TNFα, IL2, and MIP-1β—in stimulated T-cell subsets (Supplementary Figs. 4A and S4B). Other subsets identified include: IL2-producing CD4⁺ T_CM and T_EM, IL17-producing CD4⁺ T cells, Th1 cytokines-producing CD4⁺ T_EM, and CD107a-expressign CD4⁺T effectors (Supplementary Figs. 4A and S4B). The absolute cell numbers in each of the T-cell metaclusters are included in the Supplementary Table S4C in Supplementary Materials and Methods section.

We next compared the relative frequencies of CD4⁺ and CD8⁺ T cells expressing intracellular cytokine markers in MEL⁻ and MEL⁺ SLNs upon P+I stimulation. Similar frequencies of CD8⁺ T cells expressing effector cytokines (IFNγ, TNFα, and IL2) were observed in both MEL⁺ and MEL⁻ SLNs (Fig. 4G). Similarly, no differences were observed in the frequencies of CD8⁺ T cells expressing cytolytic markers (CD107a and MIP-1β; Supplementary Fig. S4C) either. However, evaluation of cytokine profile of CD4⁺ T cells revealed that MEL⁺ SLNs harbored higher frequencies of CD4⁺ T cells expressing both Th1 effector (TNFα, IL2) and Th2 (IL4) cytokines (Fig. 4G).

Because we observed an expansion of PD-1-expressing CD8⁺ T_EM cells in MEL⁺ SLNs, we further characterized the phenotypic and functional characteristics of the effector cells generated from both CD8⁺ T_EM and T_N subsets obtained from MEL⁻ and MEL⁺ SLNs (Cohort III, SI Scheme 1). CD8⁺ T_N and T_EM cells were purified from cryopreserved MEL⁻ and MEL⁺ SLN samples, activated with anti-CD3/CD28, expanded in the presence of IL2 for 16 days (SI Scheme 3), and assessed for the surface expression of T-cell differentiation markers CD45RA and CCR7. Our results indicate that majority of cells derived from T_N and T_EM subsets in both MEL⁻ and MEL⁺ SLNs acquired an effector-memory phenotype (CD45RA⁻CCR7⁻) postexpansion (Fig. 5A). However, a significantly lower in vitro expansion was obtained in effector cells derived from the T_EM (T_EFF EM) when compared with effector cells derived from T_N subsets (T_EFF N) in both MEL⁻ and MEL⁺ SLNs (Fig. 5B). Furthermore, T_EFF EM cells derived from MEL⁺ SLN display lower expansion than that observed in T_EFF EM cells from MEL⁻ SLN (Fig. 5B). To determine the phenotypic status of the expanded effector cells, expression of terminal differentiation markers CD57 and killer cell lectin-like receptor G1 (KLRG1) were analyzed in T_EFF EM and T_EFF N subsets from MEL⁻ and MEL⁺ SLNs. Senescent and terminally differentiated effector cells that exhibit diminished in vivo anti-tumor activity with short life span have previously been shown to express high levels of CD57 and KLRG1 markers (30, 31). Our data indicate that T_EFF EM cells derived from MEL⁺ SLN display higher levels of CD57 and KLRG1 expression than that observed in T_EFF EM cells from MEL⁻ SLN (Fig. 5C,–E). Furthermore, expression of CD27—a marker indicative of greater in vivo proliferative potential and tumor response (32)—was significantly lower in T_EFF EM cells from MEL⁺ SLN when compared with that in T_EFF EM cells from MEL⁻ SLN (Fig. 5F and G). No differences in the phenotypic status of T_EFF N was observed between MEL⁻ and MEL⁺ SLNs (Supplementary Figs. S5A–S5C).

We also assessed the expression of immune checkpoint molecules on T_EFF EM cells from MEL⁻ and MEL⁺ SLNs. In addition to PD-1, immune checkpoint molecules LAG-3, TIM-3, and TIGIT have been identified as coinhibitory regulators of effector functions of CD8⁺ T cells; their expression on T cells can signal an exhausted state (33). We found that the frequencies of T_EFF EM expressing PD-1, LAG-3, and TIGIT markers were higher in MEL⁺ SLNs (Fig. 5H).

Next, we tested the effector functions of T_EFF N and T_EFF EM cells against HLA-A2⁺ target melanoma epitopes. We evaluated the in vitro tumor killing capabilities and effector cytokine production of the effector cells from MEL⁻ and MEL⁺ SLNs obtained from HLA-A2⁺ patients (SI Scheme 3). T_EFF EM cells from MEL⁺ SLNs showed higher melanoma cell killing and IFNγ production when compared with that obtained with MEL⁻ SLNs (Fig. 5I and J). No such significant changes were observed when activities of T_EFF N cells from MEL⁻ and MEL⁺ SLNs were compared (Supplementary Figs. S5A–S5C). Taken together, these results indicate that CD8⁺ T_EFF EM from MEL⁺ SLNs are end-stage senescent CD8⁺ T cells with high cytotoxic potential and low proliferative capacity, which are progressing towards terminal differentiation.

We next analyzed the distribution of innate immune cells present in SLNs. viSNE cluster analysis revealed that NK cells were the least abundant lymphocyte lineage when compared with all other immune cell lineage frequencies present in SLNs (Supplementary Fig. S1D; Fig. 1A). PhenoGraph clustering of non–T-cell subsets (CD3^neg) in SLNs from all patients identified two predominant NK cell subsets—CD56^dimCD16^negCD57⁺ and CD56^brightCD16^neg (Fig. 6A and B; refs. 34, 35). Strikingly, NK-cell subsets expressing CD16 were excluded from the SLNs. PhenoGraph analysis of expression of cytolytic markers in the CD3^neg subsets revealed that CD56^brightCD16^neg NK cells expressed higher levels of perforin and granzyme B when compared with their levels in all other non–T-cell subsets (Fig. 6C,–F). We next compared the frequencies of CD56^brightCD16^neg NK cells expressing granzyme B and perforin in MEL⁻ and MEL⁺ SLNs. Clearly, the frequencies of granzyme B- and perforin-expressing cells were significantly lower in MEL⁺ SLNs (Fig. 6H and I), suggesting that MEL⁺ SLNs harbor NK cells with lower cytolytic potential. The absolute cell numbers of the NK cell clusters are shown in Supplementary Data-1 Excel file. The other innate immune cells present in the metaclusters were the CD14⁻CD16^high neutrophils (Fig. 6A and B).

PhenoGraph clustering of CD3^neg SLN cells identified four predominant B cells subsets present within the B cell compartment—naïve B cells, resting memory B cells, activated memory B cells, and CD38⁺ plasma cells (Fig. 6A and B). These subsets were identified on the basis of the expression of CD19, CD20, CD27, CD21, CD38, CD69, and HLA-DR markers. However, comparison of relative frequencies of naïve, memory, and plasma B cells present in MEL⁻ and MEL⁺ SLNs showed no significant differences (Supplementary Fig. S1B).

The data presented herein provide the first comprehensive, unbiased assessment of the immune microenvironment in melanoma MEL⁻ and MEL⁺ SLNs. We observed increased effector-memory αβ T cells, TCR clonality, and γδ T cells selectively in the MEL⁺ SLNs, suggestive of an anti-tumor immune response. Intriguingly, we also observed an overwhelmingly immunotolerant environment in the MEL⁺ SLNs as indicated by (i) reduced CD8⁺CD69⁺ and CD4⁺CD69⁺ activated T cells; (ii) high PD-1 expression on CD4⁺ and CD8⁺ T cells; (iii) high CTLA-4 expression on γδ T cells; (iv) increased proportions of T_reg cells; (v) increased proportions of CD8⁺CD57⁺CD27⁺PD-1⁺ effector cells (with low cancer cell killing capacity; refs. 24, 25); and (vi) high CD57 and KLRG1 and low CD27 expression on T_EFF EM cells indicating reduced proliferative potential and short life span. Taken together, these data demonstrate that the presence of melanoma cells in the SLN correlates with an immunotolerant microenvironment. Given that we observed increased TCR clonality in the MEL⁺ SLNs, one potential scenario is that the melanoma cells precede the immunotolerant microenvironment and, through unidentified mechanisms, tolerize the immune microenvironment and suppress the induction of effective melanoma-specific immunity.

In contrast to MEL⁻ SLNs, we found that MEL⁺ SLNs harbor higher percentages of T_regs and fewer activated T cells. On the basis of this data, we hypothesize that the reduced CD8⁺CD69⁺ T-cell/T_reg ratio can be a likely biomarker of disease recurrence for patients with stage III melanoma. Our future studies will evaluate the changes in CD8⁺CD69⁺ T-cell/T_reg ratio in the context of disease progression to provide conclusive evidence for the use of CD8⁺CD69⁺ T-cell/T_reg ratio as a prognostic biomarker. This altered T-cell composition at the melanoma-bearing SLN site was accompanied by a clonal expansion of CD8⁺ T that are in the effector-memory stage of differentiation with a significant proportion of them being PD-1-expressing CD8⁺CD57⁺ T-cell subtype. These CD8⁺CD57⁺PD-1⁺ T cells co-expressed CD27 and CD28 and were granzyme B⁺. A similar situation is observed in the case of metastatic melanoma lesions (24, 25) and chronic viral infections (CD8⁺ T cells encounter persistent antigenic stimulation; ref. 36), where CD8⁺CD57⁺PD-1⁺ T cells expressing granzyme B fail to downregulate CD27 and CD28 and transition to perforin^hi end-stage CTLs. Furthermore, effector cells differentiated from CD8⁺ T_EM cells obtained from MEL⁺ SLNs, while exhibiting increased ex vivo cytotoxic potential, display increased expression of multiple exhaustion markers (PD-1, LAG-3, and TIGIT). Collectively, these phenotypic changes to the T cells in the MEL⁺ SLNs work in tandem to promote an immunotolerant microenvironment and can potentially contribute to disease progression. There are at least two limitations to our study. First, we used effectors cells derived from CD8⁺ T_EM to evaluate the expression of exhaustion markers/immune checkpoint receptors (PD-1, TIM-3, LAG-3, and TIGIT) (33). Analysis of expression levels of these markers directly on CD4⁺ and CD8⁺ T-cell subsets in MEL⁺ SLNs will better represent the exhaustion status of T cells. Second, because there exists considerable overlap between markers indicating T-cell exhaustion and activation, the degree of co-expression between PD-1 and the alternative immune checkpoints (TIM-3, LAG-3, and TIGIT) will need to be assessed to better define the hierarchical loss of effector functions in the T cells present in MEL⁺ SLNs.

viSNE clustering analysis indicted that in SLNs, CD16⁻ NK cells were the least abundant population among all immune cell lineages. Intriguingly, NK cells from MEL⁺ SLNs expressed lower levels of granzyme B and perforin cytolytic molecules. On the basis of this observation, it is possible that NK cells present in SLNs in stage III patients may display poorly cytolytic activity and may fail to eliminate the melanoma cells. Our future studies will directly assess the melanoma killing activity of the NK cells. Therapeutic strategies that can induce the expansion or activation of NK cells in the SLNs can be beneficial for the cytotoxic clearance of melanoma cells and reduce the risk of disease recurrence in these patients.

The γδ T cells play a significant role in cancer (37); however, little is known about the role of SLN-resident γδ T cell subtypes in melanoma progression or the regulatory role of immune checkpoint receptors PD-1 and CTLA-4 in γδ T-cell biology. Earlier studies have shown that higher ratio of Vδ1 to total γδ T cells in patients with melanoma predict patient mortality and directly correlate with reduced therapeutic response to anti-CTLA-4 antibody (38) whereas higher frequencies of Vδ2 cells were associated with anti-CTLA-4 therapeutic response and improved survival in patients with melanoma (38, 39). Our CyTOF data indicate high levels of surface CTLA-4 expression on both MEL⁻ and Mel⁺ SLN-resident γδ T cells, indicating the possibility of beneficial checkpoint inhibitor intervention in patients with early-stage melanoma. Conversely, T_regs showed very low levels of surface CTLA-4. A key feature of CTLA-4 is its rapid and constitutive endocytosis from the plasma membrane and about 90% of CTLA-4 is present in intracellular organelles (40). It has been shown that activation of T_regs leads to transport of CTLA-4 to the T-cell surface (41). On the basis of these studies, it appears that the majority of CTLA-4 in T_regs present in the SLNs could be in the intracellular stores and that activation of these cells could lead to rapid transport of CTLA-4 to the cell surface. Our future studies will confirm this phenomenon of surface versus intracellular CTLA-4 expression in both nonactivated and activated T_regs.

CD14⁺ and CD16⁺ monocytes have diverse functions in lymph nodes (42, 43). CD16⁺ monocytes have been reported to participate in cancer surveillance as well as recruit and activate CD16⁺ NK cells to tissues that harbor tumor cells (44). Even in the SLNs from patients with early-stage melanoma, CD14⁺ and CD16⁺ monocytes were not detected in both MEL⁻ and MEL⁺ SLNs. The absence of CD16⁺ monocytes may contribute to lack of CD16⁺ NK cells in SLNs from stage I/II and III patients. Strikingly, macrophages were absent in both MEL⁻ and MEL⁺ SLNs. We also observed a significant decrease in frequencies of DCs in MEL⁺ SLNs patients, although the activation status of CD11c⁺CD14⁻ conventional DCs was not altered (Supplementary Fig. S5D). This is in agreement with recent work by van den Hout and colleagues (4), describing a dramatic decrease in the frequencies and activation status of skin-derived DC subsets in melanoma-bearing SLNs. It is possible that the migration of DC or its precursor is hampered in stage III patients (45, 46) or tumor-derived factors in the local milieu could directly hinder the maturation or viability of DCs in the tumor-bearing SLNs (47).

The first step in treatment of patients with T2–4 (>1 mm Breslow thickness) clinically node-negative melanoma is to perform wide local excision of the primary tumor along with SLN biopsy. In effect, the one immune organ most capable of initiating a specific antitumor immune response is removed prior to any consideration of adjuvant immune checkpoint inhibitor therapy. Although an excellent staging procedure to determine which patients have nodal metastases and beneficial in terms of reducing the risk of regional nodal melanoma recurrence, SLN biopsy (even with concomitant completion of lymphadenectomy for patients with tumor-positive SLN) does not improve overall survival, although subgroup analysis showed that melanoma-specific survival was better for SLN-positive patients compared with those who developed palpable nodal metastasis (48). Resecting the SLN at the outset removes not only metastatic melanoma cells, but also diverse and clonally expanded CD4⁺ and CD8⁺ T cells that, if activated, could promote antimelanoma systemic immunity. This argument is supported by a recent murine study of melanoma wherein removal of tumor-draining LNs (TDLN) abrogated response to immune checkpoint blockade (49). These suggest that an immune checkpoint blockade strategy prior to SLN biopsy may provide clinical benefit for patients with stage III melanoma (MEL⁺ SLNs). The rationale for such an approach is supported by a recent clinical trial wherein a neoadjuvant single-dose anti-PD-1 therapy induced potent pathologic responses in some stage III/IV patients with resectable melanoma (50).

In conclusion, this study combined single-cell analysis, flow cytometry, and TCR repertoire analysis to provide a comprehensive analysis of the immune microenvironment of SLNs. Our findings demonstrate that the presence of melanoma cells in a SLN correlates with an immunotolerant microenvironment and provide rationale for the development of neoadjuvant immunotherapeutic strategies to reshape the immune microenvironment in the SLNs and promote systemic melanoma-specific immunity.

M.E. Egger reports personal fees from Iovance Biotherapeutics and grants from SkylineDx outside the submitted work. H.T. Maecker reports personal fees from Caris Life Sciences and grants from NIH outside the submitted work. J.A. Chesney reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

K. Yaddanapudi: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. B.F. Stamp: Data curation, formal analysis, validation, investigation. P.B. Subrahmanyam: Data curation, formal analysis, methodology, writing–review and editing. A. Smolenkov: Resources. S.J. Waigel: Data curation, formal analysis. R. Gosain: Resources, data curation. M.E. Egger: Resources, writing–review and editing. R.C.G. Martin: Resources. R. Buscaglia: Formal analysis, validation. H.T. Maecker: Resources, writing–review and editing. K.M. McMasters: Conceptualization, resources, funding acquisition, writing–original draft, writing–review and editing. J.A. Chesney: Conceptualization, resources, funding acquisition, writing–original draft, writing–review and editing.

This work was supported by grants NIH/NIGMS P20GM135004 (to J.A. Chesney and K. Yaddanapudi) and NIH/NCI R21CA245560 (to K. Yaddanapudi). P.B. Subrahmanyam and H.T. Maecker were partially funded by grant 1U24CA224309 from the NIH. This project used an instrument funded by grant S10RR027582 from the NIH.

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.