Prevalence and Cellular Distribution of Novel Immune Checkpoint Targets Across Longitudinal Specimens in Treatment-naïve Melanoma Patients: Implications for Clinical Trials

Immunotherapy has now become an established pillar in the treatment of patients with many cancers, particularly melanoma and lung cancer, and is increasingly being used in patients with many other tumor types. With the advent of immune checkpoint inhibitors (ICI), targeting coinhibitory receptors on immune cells, the paradigm of translational cancer research has shifted to include not only the tumor itself, but also the immune system and tumor microenvironment. In particular, mAbs that target the receptors PD-1 and CTLA-4 have demonstrated unprecedented clinical efficacy and have revolutionized the treatment of melanoma in both the metastatic and adjuvant setting. Despite these incredible advances, the majority of patients treated with ICI therapy fails as a result of innate (primary) or acquired (secondary) resistance (1).

A number of resistance mechanisms for anti-PD-1 and anti-CTLA-4 therapy have been hypothesized. One such mechanism is the alternative immune checkpoint expression hypothesis, whereby other checkpoint inhibitory receptors dampen T-cell responses and contribute to severe T-cell exhaustion (1), such that anti-PD-1 alone is not enough to rescue T-cell functionality. It has been suggested that in such a state, the additional targeting of other coinhibitory receptors, such as TIM-3, TIGIT, LAG-3, VISTA, and others (1–5), may achieve better clinical outcomes. In addition, one well-described association with lack of response to immunotherapy occurs in tumors that lack adequate immune cell infiltration. Although such tumors are less responsive to ICI therapy, it has been postulated that mAbs targeting costimulatory receptors (OX40, GITR, ICOS, and others) on T cells might work synergistically with checkpoint inhibitors to increase functional immune infiltrates specific for tumor antigens (1).

There are a growing number of clinical trials now underway involving these novel targets as single-agent therapy or in combination with anti-PD-1. A major confounding factor for developing novel strategies, rational selection of targets for testing in clinical trials, and clinical decision making, however, is the lack of understanding of the expression profile of most of these targets in the tumors of patients with cancer. For these reasons, it is critically important to understand the prevalence and expression profile of both costimulating and coinhibitory receptors in human malignancies, how they change in the course of disease progression, and what immune cells (or tumor cells) these receptors are expressed on and enriched for. Such knowledge will provide the field with the necessary foundation to develop tailored therapies for individual patients in the future.

To address these questions, we analyzed the protein expression profile of coinhibitory receptors (PD-1, TIM-3, VISTA) and costimulatory receptors (GITR, ICOS, OX-40) by multiplex immunofluorescence staining in 96 tumor biopsies from 41 patients with melanoma, including matched biopsies for primary tumors, lymph node metastases, and distant metastases from the same patient. We then performed mass cytometry on leukocytes isolated from 18 fresh baseline melanoma tumor samples to evaluate the expression of checkpoint receptors (ICOS, GITR, OX40, PD-1, CTLA-4, TIGIT, TIM-3, PD-L1, and PD-L2) on innate and adaptive immune infiltrates.

Ninety-six tumor biopsies from 41 patients with melanoma, including matched biopsies for the primary tumors, lymph node, metastases, and distant metastases from the same patient were used to ascertain costimulatory and coinhibitory expression by multiplex IHC staining. The Melanoma Institute Australia database and archival files of the Department of Tissue Pathology and Diagnostic Oncology at the Royal Prince Alfred Hospital (Sydney, Australia) were used to identify patients with stage IIIC/IV melanoma with multiple formalin-fixed paraffin-embedded (FFPE) melanoma specimens available at various stages of disease progression, as described previously (6, 7). In patients with a history of multiple primary melanomas, the culprit primary melanoma was selected using a previously defined algorithm (8). An additional cohort of 18 fresh tumor biopsies taken from 18 patients with either stage III or stage IV melanoma were used to generate the mass cytometry data. Details of these patients and the site of biopsy are presented in Supplementary Table S3. The study was undertaken with Human Ethics Review Committee approval (protocol no X17-0312) and patients' informed consent. No patients received any prior systemic treatment and the site from which the biopsy specimens were obtained had not been previously treated at any time with radiotherapy or with topical or intralesional therapy.

All immunofluorescence staining was carried out on 4-μm–thick sections using an Autostainer Plus (Dako, Agilent Technologies) or on the bench with appropriate positive and negative controls. Opal Multiplex IHC Assay Kit (PerkinElmer) was used as per the manufacturer's protocol. Briefly, FFPE tumor specimens were cut at 4 μm and air dried overnight. Specimens were then baked at 65°C for 30 minutes, deparaffinized, and rehydrated by xylene and ethanol. Heat-induced antigen retrieval was performed in a Decloaking Chamber (Biocare), which heated samples at 95°C for 20 minutes in pH 9 antigen retrieval (AR) buffer (PerkinElmer). The primary antibody panels targeting (i) GITR (1/1,000; CST-D919D), ICOS (1/3,000; CST-D1K2T), OX40 (1/3,000; CST-E9U7O), CD3 (1/1,000; CM103R95); (ii) PD-1 (1/100; Cell Marque-NAT205), TIM-3 (1/500; CST-D5D5R), VISTA (1/2,000; CST-D1L2G), CD3 (1/1,000) (CM103R95); or (iii) CD14 (1/100; Sigma, polyclonal), CD68 (1/1,000; Kp-1), and CD8 (1/800; Ab4055) were incubated for 30–45 minutes. Primary antibodies were detected using either Opal Polymer HRP (GITR, ICOS, OX40, TIM-3, VISTA, CD3, CD14, CD68, CD8) (Perkin Elmer) or MACH 3 HRP-polymer (PD-1; Biocare) for 30 or 10 minutes, respectively, and then visualized using Tyramide Signal Amplification for 10 minutes (Opal 7-Colour IHC, PerkinElmer). Between subsequent staining runs, tissues were boiled in pH 9 AR buffer for 15 minutes to strip the primary antibody complex from the sample. On the last staining run, DAPI was added to the sample for 5 minutes. The samples were cover slipped using Vectashield (H-1400) and left overnight to the dry at 4°C.

Vectra 3 multispectral slide scanner was used in conjunction with Vectra 3.3 and Phenochart 1.0.4 software to image samples. Images were then unmixed and analyzed using inForm 2.3.0 software (PerkinElmer) to phenotype and quantify the expression of each of the markers on individual cells. Because endogenous microenvironment in lymph node metastases is formed by lymphoid tissue and is therefore unique compared with other sites of metastases, an expert pathologist reviewed the images and annotated any lymph node–associated structures/cells not associated with the tumor out of downstream analysis. Quantitative analysis was conducted using TIBCO Spotfire 3.3.1. Graphical and statistical analyses were performed using Prism version 6.0f (GraphPad Software) or TIBCO Spotfire 3.3.1. P values between matched primary and distant metastasis groups were determined using a Wilcoxon-paired matched test (nonparametric). All other P values were determined using a nonparametric Kruskal–Wallis test and Dunn multiple comparisons test, where appropriate. Correlation plots for each of the checkpoint receptors were performed using a linear regression analysis in TIBCO Spotfire 3.3.1. P values less than 0.05 were considered significant. All variability in the data is shown as the SEM.

Fresh tumor biopsies were collected and placed in RPMI1640 (Thermo Fisher Scientific) at 4°C before being subjected to a mechanical/enzymatic dissociation system (GentleMACS, Miltenyi Biotec). Dissociation was performed according to the manufacturer's protocol. Briefly, the tumor was cut into small fragments (<3 mm) and incubated in a C-tube (Miltenyi Biotec) with enzymes H, R, and A, made up to 5 mL with RPMI1640. The C tube and contents were then placed upside down onto the GentleMACS Dissociator with heating elements and then subjected to a mechanical disaggregation step followed by 30- to 60-minute incubation at 37°C (program 37C_h_TDK_1). After dissociation, tissue was passed through 70-μm strainer and washed with RPMI1640 supplemented with 100 U/mL penicillin, 100 μg/mL streptomycin, 25 mmol/L HEPES, and 50 μg/mL gentamicin at 1,800 rpm for 10 minutes before being resuspended in FCS supplemented with 10% dimethyl sulfoxide. Samples were slow cooled to −80°C and cryopreserved for future analysis.

A total of 2 × 10⁶ cells were stained for mass cytometry analyses, as described previously (9, 10). Briefly, cells were stained with 1.25 μmol/L Cell-IDTM Cisplatin in PBS (Fluidigm, catalog no. 201064) for 3 minutes at room temperature and quenched by rapid addition of FBS. Cells were then washed twice in FACS buffer, then stained with a fluorophore-conjugated antibody cocktail for 20 minutes at 4°C. Following washing with FACS buffer, cells were stained with a metal-conjugated surface stain antibody cocktail for 20 minutes at 4°C. Cells were then fixed and permeabilized using the FOXP3 Transcription Factor Staining Buffer Set, according to the manufacturer's protocol (eBiosciences, catalog no 00-5523-00). Cells were subsequently stained with a metal-conjugated intracellular antibody cocktail for 40 minutes at 4°C. Cells were then washed twice, once in Perm/Wash buffer and once in FACS buffer. Next, the cells were fixed overnight in 4% paraformaldehyde solution containing DNA Intercalator (0.125 μmol/L iridium-191/193; Fluidigm, catalog no. 201192B).

Prior to acquisition, cells were washed once in FACS buffer and twice in dH₂O. Cells were then diluted to 8 × 10⁵ cells/mL in dH₂O containing 10% EQ Four Element Calibration Beads (Fluidigm, catalog no. 201078) and filtered. Cells were acquired at a rate of 200–400 cells/second using a CYTOF 2 Helios Upgraded Mass Cytometer (Fluidigm). Flow cytometry standard (FCS) files were normalized to EQ bead signal and were then analyzed using FlowJo v10.2 (Tree Star). TD leukocytes were debarcoded manually in FlowJo. For manual gating of immune subsets, a minimum threshold of 50 cells was set to analyze the checkpoint expression on a particular immune subpopulation within patients. Subpopulations with fewer than 50 cells were excluded from the analysis.

Prior to staining, CD45⁺ leukocytes were isolated from TD samples by magnetic labeling with CD45 MicroBeads (Miltenyi Biotec, catalog no. 130-045-802) and subsequent separation using the autoMACS Pro Separator (Miltenyi Biotec, catalog no. 120-092-545), according to the manufacturer's instructions. To ensure that the CD45 MicroBeads did not prevent binding of the CD45 antibody, cells were simultaneously stained with metal-conjugated anti-CD45 (Pd104). The timing of anti-CD45 addition was determined by previous experiments. Isolated CD45⁺ cells were then counted and washed in FACS media.

Tumor dissociates with yield <2 × 10⁶ leukocytes after separation were spiked with donor peripheral blood mononuclear cells (PBMC) to increase pellet bulk. To differentiate between tumor dissociate leukocytes and PBMCs, a CD45-based barcoding approach was used prior to spike-in, as described previously (11). Tumor dissociate leukocytes were labeled with CD45-Pd104 during CD45⁺ isolation PBMCs were stained with CD45-Pd110 for 20 minutes at 4°C and then washed twice in FACS buffer. Labeled PBMCs were then added to tumor samples (if needed) up to a final count of 2 × 10⁶ cells.

Metal-conjugated antibodies used in CyTOF analysis are presented in Supplementary Table S5 (cell surface) and Supplementary Table S6 (intracellular). For some markers, fluorophore-conjugated antibodies were used as primary antibodies (Supplementary Table S7), followed by secondary labeling with anti-fluorophore metal-conjugated antibodies. Antibodies were either purchased from Fluidigm or conjugated in-house using MaxPar X8 reagent kits (Fluidigm), according to the manufacturer's protocol. The concentration of each antibody was assessed using a NanoDrop (Thermo Scientific) and was then adjusted to 200 mg/mL in BioStab Antibody Stabilizer (Sigma-Aldrich, catalog no. 55514). Conjugated antibodies were titrated for optimal concentration prior to use. Surface and intracellular antibody staining cocktail master mixes were prepared prior to each experiment. This protocol was carried out by the Ramaciotti Facility for Human Systems Biology, Sydney, Australia.

Cryopreserved single-cell isolates from tumor samples obtained from patients with metastatic melanoma (isolated as described previously) were thawed, washed, and counted. Tumor cells were stained at 4°C for 30 minutes with saturating concentrations of the following extracellular mAbs: anti-CD3, anti-CD4, anti-CD11c, anti-CD13, anti-CD14, anti-HLA-DR, (all from BD Biosciences), anti-CD8, anti-CD19, anti-CD45 (BioLegend), and anti-VISTA antibody (R&D Systems). Stained sample acquisition was performed on a 5-laser Fortessa flow cytometer (BD Biosciences) and acquisition was performed using FACS DIVA. Data were analyzed on FlowJo software.

t-Distributed Stochastic Neighbor Embedding (t-SNE) analysis was performed using Cytobank (www.cytobank.org) on 6 × 10⁴ T cells from a total of 18 tumor-dissociates. Analysis was performed using the following settings: Iterations 1000, Perplexity 30, Theta 0.5, and Seed “random.” Clustering was performed using the checkpoint receptor channels (ICOS, GITR, OX40, PD-1, CTLA-4, TIGIT, TIM-3, PD-L1, and PD-L2) so as to promote tight clustering of each checkpoint receptor and hence quick visualization of overlapping/distinct checkpoint-receptor positive T cells.

Ninety-six melanoma biopsies from 41 patients (Table 1) were stained for ICOS, GITR, OX40, PD-1, TIM-3, and VISTA using multiplex immunofluorescence-IHC. Representative staining for each of the markers are shown in Fig. 1A. The costimulatory receptor ICOS and the coinhibitory receptors PD-1, TIM-3, and VISTA were the most abundant (mean expression of 303 ± 94 cells/mm², 166 ± 47 cells/mm², 122 ± 36 cells/mm² and 286 ± 87 cells/mm², respectively), while GITR and OX40 were the least abundant (P < 0.001, mean expression of 13 ± 4 and 16 ± 6 cells/mm² respectively; Fig. 1B; Supplementary Table S1). When using a positive threshold of either >1 or >5 positive cells per 1 mm² to distinguish positive from negative biopsies for a particular marker, OX40 was the least common in patient tumors compared with other costimulatory and coinhibitory receptors (Supplementary Table S1).

We then examined the coexpression of CD3 and each of the various checkpoint receptors in the tumor environment. We first determined the proportion of T cells that were positive for each of the markers per patient (n = 41) in all samples. Melanoma-associated T cells in patients were ICOS⁺ CD3⁺ in 0% to 34% (average 12%) and PD-1⁺ CD3⁺ in 0% to 55% (average 10%; Fig. 1C). A smaller proportion of T cells expressed VISTA (average 7% ± 2%), TIM-3 (4% ± 1%), GITR (2% ± 1%), and OX40 (<1% ± 1%; Fig. 1C).

We next sought to determine the proportion of checkpoint receptor–positive cells that were T cells. A high proportion of ICOS, GITR, and PD-1–positive cells were T cells (88% ± 2%, 64% ± 5%, and 77% ± 4%, respectively), while significantly less TIM-3 (36% ± 4%) and VISTA (24% ± 3%) positive cells were T cells (Fig. 1D, P < 0.005). The cytomorphology characteristics (larger cell size, lower nucleus: cytoplasm ratio and irregular shape) of TIM-3- and VISTA-expressing non-T-cell populations suggested that these cells were myeloid-derived cells such as macrophages or dendritic cells. In line with this, we observed colocalization between VISTA, CD68, and CD14 (Supplementary Fig. S1A). OX-40 displayed a high degree of interpatient variability, with a mean of 42% ± 7% of OX40-positive cells being T cells across the patients, but 30% of patients expressed OX40⁺ cells that were all CD3⁻ (Fig. 1D). Nevertheless, 70% of OX40⁺ cells were found to be T cells when all positive cells from all patients and samples were combined (Supplementary Fig. S1B). In addition, VISTA was occasionally expressed on tumor cells (Supplementary Fig. S1C), although this was not observed for any of the other checkpoint receptors. Furthermore, the number of all checkpoint receptor–expressing cells was significantly correlated with the CD3 numbers (Supplementary Table S2).

Because tumors lacking T-cell infiltrate or PD-1/PD-L1 expression represent a broad class of patients generally less responsive to anti-PD-1 therapy, we examined PD-1–negative tumors (≤1 cell/mm²) to investigate the expression of alternative checkpoint markers. In these tumors, the abundance of alternative checkpoint receptors was low (generally fewer than 50 cells/mm²), with VISTA showing relatively higher expression compared with GITR (P < 0.05) and OX40 (P < 0.01; Supplementary Fig. S1D). The mean expression of ICOS, TIM-3, and VISTA checkpoint receptors in PD-1–negative tumors was negligible when compared with their expression in PD-1–positive tumors (P < 0.05; Supplementary Fig. S1E).

Together, these results show that only a small subset of T cells in the tumor express any given checkpoint receptor, and therefore only a select subpopulation of the total infiltrating lymphocytes are likely to be targeted by any given checkpoint therapy. Furthermore, some checkpoint-targeting therapies such as TIM-3 and VISTA are likely to function predominately through non-T-cell populations.

To determine whether the expression pattern of any of the checkpoint receptors changes during the progression of melanoma disease, we stratified the patient biopsies according to the stage of disease; primary (P), in-transit, metastasis (ITM), regional lymph node metastasis (LN), or distant metastasis (DM). GITR had a higher density of positive cells in primary biopsies compared to in-transit metastases (P = 0.016) and distant metastases (P = 0.003), but not regional lymph node metastases (Fig. 2A; Supplementary Fig. S2A). No pattern in the densities of any of the other checkpoint receptors between the various stages of disease was observed (one-way ANOVA per receptor).

We next sought to examine the proportion of T cells expressing a given checkpoint receptor during disease progression. Given that within our cohort of patient biopsies, there was a subgroup of patients (n = 6) with matched specimens for primary, regional lymph node, and distant metastases, we were able to explore this in a model that paralleled melanoma disease progression in patients. Interestingly, although a very small percentage of intratumoral T cells expressed GITR, there were strong trends for decrease in the proportion of GITR⁺ T cells between primary and regional lymph node (P = 0.03) and primary and distant metastases (P = 0.07) in individual patients (Fig. 2B). PD-1 and ICOS were expressed on a high proportion of T cells, relative to other checkpoint receptors at all stages of disease progression, with trends for higher percentages in regional lymph node metastases compared with primary (PD-1, P = 0.06) or distant metastases (ICOS, P = 0.07; Fig. 2B), possibly reflecting the unique microenvironment of the lymph node.

We examined checkpoint expression changes in a larger cohort (from within the 96 biopsies) of patients with matched biopsies between primary and lymph node metastases (n = 20) or primary and distant metastases (n = 14). This confirmed that the proportion of intratumoral GITR⁺ T cells were higher in primary tumors (∼5%) compared with matched regional lymph node (<1%, P = 0.04) and distant metastases (<0.5%, P = 0.0005; Fig. 2C), indicating that the proportion of intratumoral GITR⁺ T cells decreases during disease progression. No other statistically significant differences were observed for any of the other checkpoint receptors (Supplementary Fig. S2B and S2C). Our results suggest that GITR therapy may theoretically be more effective as an upfront therapy in patients with locally advanced but clinically localized primary melanoma rather than those with occult nodal or distant metastases.

To determine whether the site of tumor influenced the expression pattern of checkpoint receptors, we examined the percentages of T cells expressing a given checkpoint receptor in all distant metastatic biopsies by site of disease. Interestingly, no statistically significant differences were observed for most of the checkpoint receptors expressed on intratumoral T cells except higher ICOS expression in bone metastases relative to brain metastases (Fig. 2B).

As potential drug targets, it is imperative that there is an understanding of the distribution of checkpoint receptors in the tumor microenvironment, particularly, the various immune cells enriched for each of the costimulatory and coinhibitory receptors so as to anticipate the immune cells each checkpoint antibody may affect. Therefore, we performed 43 parameter mass cytometry (CyTOF–cytometry by time of flight), on-treatment–naïve melanoma tumor dissociates (n = 18) from 18 stage III or stage IV patients with melanoma (Supplementary Table S3) that had undergone CD45 isolation and enrichment. The mass cytometry antibody panel analyzed checkpoint expression on a broad range of immune cell populations, including central (Tcm), effector (Tem), and tissue resident (Trm) memory subsets of CD8⁺ and CD4⁺ T cells (defined by the expression of CD69 and CD103 in CD8 T cells; ref. 12), T-regulatory cells (Treg), natural killer (NK) cells, conventional dendritic cells (cDC1 and cDC2), monocytes, CD14±, macrophages (Mϕ CD68+ CD14±), and B cells (See Supplementary Fig. S3A for full gating strategy). The percentage of each immune population positive for each of the checkpoint receptors ICOS, GITR, OX40, PD-1, CTLA-4, TIGIT, TIM-3, PD-L1, and PD-L2 was determined by manual gating and the results are summarized in a heatmap (Fig. 3A) as well as in column graphs for each marker (Fig. 3B).

Our results revealed that costimulatory and coinhibitory receptor expression varied widely between different immune populations, and indeed within subsets of memory T cells. PD-1 and TIGIT were expressed largely on T-cell populations, with PD-1 being enriched on CD8⁺ Trm (mean >70% in all patients), and TIGIT being enriched on CD8⁺ Trm (>70%) and Tregs cells (>90%; Fig. 3A and B). Innate populations generally expressed relatively little TIGIT except for NK cells and monocytes, which expressed >15% and >10%, respectively (Fig. 3B). TIM-3 expression on T cells was largely restricted to the CD8⁺ Trm phenotype (>10%) and Tregs (∼9%) with minimal expression observed for other CD8 and CD4 memory populations (Fig. 3A and B). However, within the innate populations, TIM-3 was expressed highly on dendritic cells, most notably CD141⁺ cDC1 cells (∼20%) and to a lesser extent on NK cells (6%). These results are in line with our IHC data in which TIM-3 was largely expressed on non-T-cell populations. The CyTOF data also revealed ICOS, OX40, and CTLA-4 were expressed most highly on Treg cells (Fig. 3A and B; Supplementary Table S4). However, a small subset of non-Treg CD4⁺ T cells, and to a lesser extent, CD8⁺ T cells, also expressed ICOS and CTLA-4. In line with our IHC data, GITR expression was very low on all immune populations in these melanoma tumor dissociates. Interestingly, GITR expression was highest on B cells (∼2%) in these melanoma tumor dissociates relative to T-cell populations (<1%; P < 0.01) and was minimally expressed on other innate immune populations (Fig. 3A; Supplementary Table S4). As expected, PD-L1 was largely expressed on dendritic cells, macrophages, and monocytes, with the highest expression on conventional dendritic cells type-II (cDC2) and CD14⁺ CD68⁺ macrophages (both ∼10%). However, we found that PD-L2 was highly expressed on monocytes (∼25%). To identify the immune cells expressing VISTA, we assessed its expression on immune infiltrates in melanoma tumors on a separate cohort by flow cytometry. Our results showed that VISTA was predominately expressed on (CD3⁻CD19⁻CD45⁺ HLA-DR⁺) myeloid populations, such as CD11c⁺ CD14⁺ macrophages, CD11c⁺ CD14⁻ dendritic cells, monocytes, and also on B cells (Supplementary Fig. S3B), in line with what we observed for IHC.

In order to further assess the distribution and coexpression of checkpoint receptors on T cells at the single-cell level, we performed tSNE analysis on 60,000 T cells from our collective 18-melanoma tumor dissociates, clustering based on costimulatory/coinhibitory expression. The regions positive for each costimulatory and coinhibitory receptor are shown in Fig. 4A. In addition to this, we also explored cooccurrence and pairwise associations for each of the checkpoint receptors on T cells and myeloid cells by analyzing the correlation in the expression for each marker pair. The strength of the relationship between each checkpoint pair for each of the cell types is summarized in Fig. 4B and C, respectively, with the highest correlating pair in each cell type displayed as a scatter plot.

The results provided by tSNE analysis and cooccurrence heatmap (T cells) showed that the checkpoint receptors ICOS, PD-1, TIGIT, CTLA-4, and TIM-3 were highly coexpressed on cells (Fig. 4A and B). T cells expressing either PD-1 or TIGIT were largely seen to overlap with one another (Fig. 4A) and correlation of the expression between the two receptors was moderately strong (R = 0.58; P < 0.0001), suggesting that T-cell populations targeted by PD-1 or TIGIT agonists are likely to be very similar (Fig. 4A and B). These T cells were mostly EOMES⁺ and CD69⁺ (Supplementary Fig. S4A), which we previously demonstrated to represent a tissue-resident memory cell type associated with response to anti-PD-1 ± anti-CTLA-4 immunotherapy (12, 13). It is also interesting to note that ICOS, although a costimulatory receptor, had a degree of coexpression with many of the coinhibitory receptors on this EOMES⁺ CD69⁺ subpopulation, indicating that the expression of many inhibitory receptors does not necessarily preclude the expression of costimulatory receptors on the same population of cells (Fig. 4A; Supplementary Fig. S4A). In line with this, a weak correlation was observed between ICOS and many of the coinhibitory checkpoint receptors, including TIGIT (R = 0.38, P < 0.0001), TIM-3 (R = 0.29, P < 0.0001), CTLA-4 (R = 0.28, P < 0.0001), and PD-1 (R = 0.29, P < 0.0001; Fig. 4B). Interestingly, there was no correlation observed between ICOS and other costimulatory receptors, except for ICOS and OX40, which showed a weak correlation between the two (R = 0.27, P < 0.0001), likely representing Treg and CD4 populations of T cells based on the expression of FOXP3 and CD4 (Fig. 4A and B; Supplementary Fig. S4A).

On myeloid cells, pairwise association analysis revealed that the expression of costimulatory and coinhibitory receptors were largely unassociated with one another, except for PD-L2 and TIGIT, which showed a very strong correlative relationship (R = 0.86, P < 0.0001; Fig. 4C). Our enrichment analysis demonstrated that monocytes express relatively high levels of both TIGIT and PD-L2 in the myeloid compartment. Our pairwise association analysis indicates, however, that there was a relationship between these two receptors in the myeloid compartment generally. It is also interesting to note that PD-L1 and PD-L2 expression on myeloid cells lack any correlative relationship (Fig. 4C), suggesting divergent subsets, and therefore their inhibition may have distinct functions on this immune compartment.

The targeting of various novel costimulatory and coinhibitory receptors in tumors is focused on providing alternative treatment options to patients refractory to current anti-PD-1/anti-CTLA-4 therapies, and represents a step forward in the path toward personalization of immunotherapy treatments. Still, little is known about the actual mechanistic basis of these new therapies, which immune cells they are likely to target and modulate in the cancer setting, the relative prevalence of the targets in human malignancies, how they change during the course of the disease, and consequently, which patients are likely to benefit from them. Understanding these issues is critical to drug development, determining whether the target expression has any correlation with drug activity and ultimate clinical decision-making; to select the best immunomodulatory therapeutic regimen at the optimal time for individual patients based upon robust predictive biomarkers of response and resistance to therapy. For these reasons, we analyzed the prevalence and cellular distribution of novel immunotherapy targets in patients with treatment-naïve melanoma with primary melanomas and corresponding metastases, including longitudinal cohorts from individual patients that parallel melanoma disease progression. Our collective data demonstrate that only a small proportion of intratumoral lymphocytes express any given target and that some of these checkpoint receptors (TIM-3 and VISTA) are likely to be targeted on non-T-cell populations. In addition, we showed that while the majority of the checkpoint receptors demonstrate high interpatient variability with no significant patterns during disease progression (globally and matched), the proportion of GITR-expressing T lymphocytes consistently decreases from primary tumor to nodal and distant metastases. We have also demonstrated that while relative abundance profile of novel checkpoint receptors is similar in PD-1–positive and PD-1–negative tumors, their expression is drastically reduced in PD-1–negative tumors compared with PD-1–positive tumors, indicating that tumors lacking T-cell infiltration or PD-1 expression are less likely to have alternative novel targets in the tumor. Finally, we have detailed the immune populations enriched for each of the checkpoint receptors, including immune populations that are likely to be the same targets of multiple therapies.

Importantly, our data show that the costimulatory receptors OX40 and GITR are far less abundant in melanoma tumors at all stages of disease compared with other costimulatory (ICOS) and coinhibitory (PD-1, TIM-3, TIGIT, VISTA, CTLA-4) receptors, with overall less than 1%–2% of intratumoral T cells on average expressing either marker. While it is unknown whether these differences will translate into differences in the clinical efficacy of agents targeting these molecules, undoubtedly the data will have important implications for the field of personalized immunotherapy where quantitation of expression of each marker in treatment-naïve patients is necessary for the evaluation of their role as robust predictive biomarkers. Our results are in line with a study that demonstrated similar average counts of OX40⁺ cells in melanoma by IHC (14), and corroborates the finding of another study, which reported GITR expression to be drastically lower in patients with cancer compared with cancer in murine models (15). Concerning the immune subsets enriched for these receptors, the expression of OX40 on T cells was largely restricted to Tregs (∼21% expressing the OX40 receptor), while GITR was expressed on a higher proportion of B cells compared with T-cell populations, despite the majority of GITR⁺ cells being T cells in melanoma. This is not surprising as GITR is known to be expressed by activated B cells, NK cells, and other innate immune cell populations (16, 17); however, their exact function in melanoma remains unknown. Our data does suggest, however, that any benefit observed in patients from the use of GITR and OX40-targeting therapies is likely to incorporate immune subsets other than simply CD8⁺ effector T cells in the tumor, particularly Tregs in the case of OX40-targeting therapy in human melanoma.

Mechanistically, antibodies targeting OX40 and GITR have been shown to increase T-cell proliferation, effector function, and survival for effector populations expressing the target receptor, while depleting target-positive Tregs or dampening their function (18–22). However, given that such a small percentage of intratumoral T cells express GITR or OX40, this calls into question the likely effectiveness of these treatments and whether antibodies targeting these receptors are likely to have much mechanistic effect in nodal or distant metastases, as appears from early clinical trial data (23, 24). It is known that GITR and OX40 are intermediately expressed 24–72 hours after T-cell activation and then decrease days later (16, 18). It is possible that GITR and OX40 are expressed on a greater proportion of T cells, when T cells are initially primed against tumor antigen. In line with this, we found that the percentage of T cells expressing GITR is higher at the early stages of melanoma (primary) compared with later stages (nodal and distant metastases) in matched specimens from the same patient. One limitation to the interpretation of our data, however, is that the tissue microarrays were used. Because some tumors are heterogeneous in their cellular composition, it is possible that our analysis is not representative of other areas of tumor. Nevertheless, the consistency and strong statistical trends in matched and global analyses for GITR between primary and distant metastases, strongly argues against this being a spurious result. This would suggest that GITR-targeting antibodies could be biologically relevant at the very early stage of immune interaction with melanoma, but less effective for advanced disease. However, systemic immunotherapy is unlikely to be clinically relevant for patients with localized primary disease. In addition, our results indicate that comparative analysis for GITR expression between primary and distant metastases in predicting response to GITR therapies may be unsuitable and ineffective.

We also highlight that the majority of intratumoral TIM-3 and VISTA expression in melanoma is found on non-T-cell populations, including tumor cells (VISTA). Our cytometry data revealed that dendritic cells in the tumor, including cDC1 cells (defined by CD141 expression), and cDC2 cells, were enriched for the TIM-3 receptor. In murine models of breast cancer, cross presenting CD103⁺ dendritic cells express TIM-3 and upon cotherapy with a TIM-3–targeting antibody, they upregulate CXCL9, resulting in the activation of T cells, reduction in tumor burden, and increased survival compared with the single-agent paclitaxel therapy control arm (25). Such mechanisms may also exist in patients with melanoma treated with TIM-3–targeting antibodies, as cDC1 and cDC2 dendritic cells express TIM-3 and CD141⁺ cDC1 represent the human equivalent of cross-presenting CD103⁺ DCs in murine models (26). Although we were unable to explore VISTA expression by mass cytometry, our combined flow cytometry and IHC data showed that VISTA was predominately expressed on myeloid cells and B cells in melanoma tumors. Indeed, there are reports that VISTA can be expressed strongly on myeloid populations in addition to Tregs, and that mAb treatment targeting VISTA can decrease the number of myeloid-derived suppressive cells in melanoma tumors (27). Previously, we have shown that VISTA expression is increased in the melanoma biopsies of patients progressing on anti-PD-1 therapies (28). Therefore, the functional effects of targeting VISTA warrant further investigation and testing of its efficacy in clinical trials.

To the best of our knowledge, this is the first study to explore the expression of TIGIT extensively on immune subsets in melanoma patient tumor dissociates. We demonstrate that innate cells generally express minimal TIGIT compared with T cells, with natural killer cells and monocytes, showing a trend for higher TIGIT expression within the innate compartment. Most importantly, however, when we investigated TIGIT expression on T-cell subsets, TIGIT was highly expressed on CD69⁺CD103⁺CD8⁺ tissue resident memory T cells (>70%) and on Treg cells (>90%). This has important implications as we have previously shown that CD103⁺ CD8⁺ resident T cells are a critical population for melanoma control (that make up approximately 30% of tumor-infiltrating CD8⁺ T cells) and are the likely targets and responders to anti-PD-1 therapy (12, 13). Our study and others have shown that CD103⁺ CD8⁺ resident memory T cells are enriched for PD-1 and TIM-3 (12, 29), and together this suggests that TIGIT, PD-1, and TIM-3 inhibitors could all be primarily involved in reinvigorating the same important population of T cells in tumors. In line with this, our pairwise association analysis revealed that many of the coinhibitory receptors on T cells were correlated with one another, particularly PD-1 and TIGIT, further supporting this notion. Indeed, the expression profile of PD-1 and TIGIT on the immune subsets in melanoma was similar. Therefore, combining anti-PD-1 and anti-TIGIT may target the same T-cell subsets and it remains to be determined whether the addition of TIGIT inhibition in patients with advanced melanoma has an additive or redundant effect on disease control and its effect on the frequency of immune-related adverse events. Nevertheless, our data indicate that targeting TIGIT is likely to only benefit patients already responsive to anti-PD-1 therapy (i.e., patients with PD-1 expression on tumor-infiltrating lymphocytes in tumor) and therefore may have a purpose in a small percentage of patients who develop acquired resistance to anti-PD-1. Given the high expression of TIGIT on Treg cells, however, it is likely that anti-TIGIT therapy would have a dual role in depleting TIGIT⁺ Tregs, the more potent suppressive cells, as others have demonstrated (30, 31).

We have also shown that ICOS is a relatively abundant costimulatory receptor on T cells in melanoma, accounting for approximately 12% of tumor infiltrating T cells. ICOS was enriched on Treg cells, but also expressed on CD4, and to a lesser extent, CD8 memory populations, including on populations positive for coinhibitory receptors. We also demonstrated that ICOS expression weakly correlated with coinhibitory receptor expression on T cells. In the past, the ICOS/ICOSL pathway has been shown to be important in the efficacy of anti-CTLA-4 therapy (32, 33). Recently, a study in mice demonstrated that ICOS⁺ Th1-like CD4 effector populations are the likely targets of anti-CTLA-4 therapy (34). Given the abundance of ICOS in melanoma tumors, it is possible that targeting the ICOS/ICOSL pathway may have an effect on a significant subset of infiltrating T cells. Indeed, a recent study demonstrated that direct targeting of the ICOS/ICOSL pathway via an oncolytic virus helped increase activated T-cell infiltrates in the treated tumor and untreated tumors at distant sites (35). Future studies will need to explore the exact contribution of ICOS-positive CD4 and CD8 populations in melanoma disease.

The number of distant metastatic biopsies from specific sites in our study was limited. However, our results suggest that while the majority of checkpoint receptor expression did not change between sites, it is possible that site of disease may be an important factor in the expression of some checkpoint receptors. This should be validated and further examined in larger cohorts.

In conclusion, our study highlights significant differences in the prevalence and cellular distribution of expression of the major immune checkpoint receptors that are being targeted in current or soon-to-be commenced clinical trials in patients with melanoma. Furthermore, the interpatient expression of these potential drug targets was highly variable, suggesting that personalized treatment decisions based on predictive biomarker evaluation may be required to maximize treatment efficacy of novel drug combinations. In addition, the low abundance of some receptors (OX40 and GITR) and the coexpression of others (PD-1 and TIGIT) will need to be considered when designing clinical trials with these agents in melanoma.

R. P. M. Saw reports receiving speakers bureau honoraria from Bristol-Myers Squibb, and is a consultant/advisory board member for Novartis, MSD, and Amgen. J. F. Thompson is a consultant/advisory board member for GlaxoSmithKline, Bristol-Myers Squibb, Merck Sharp Dohme, and Provectus. A. M. Menzies is a consultant/advisory board member for Bristol-Myers Squibb, MSD, Novartis, Roche, and Pierre-Fabre. G. Long is a consultant/advisory board member for Bristol-Myers Squibb, Merck, Novartis, Aduor, Mass Array, and Pierre-Fabre. R. A. Scolyer is a consultant/advisory board member for Merck Sharp & Dohme, Novartis, Myriad, and NeraCare. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.J. Edwards, A. Tasker, U. Palendira, J.S. Wilmott, G. Long, R.A. Scolyer

Development of methodology: J.J. Edwards, A. Tasker, J.S. Wilmott, R.A. Scolyer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.J. Edwards, A. Tasker, M. Batten, A.L. Ferguson, R. Allen, R.P.M. Saw, J.F. Thompson, A.M. Menzies, J.S. Wilmott, G. Long, R.A. Scolyer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.J. Edwards, A. Tasker, I. Silva, C. Quek, M. Batten, A.L. Ferguson, R. Allen, R.P.M. Saw, J.S. Wilmott, G. Long, R.A. Scolyer

Writing, review, and/or revision of the manuscript: J.J. Edwards, I. Silva, B.M. Allanson, J.F. Thompson, A.M. Menzies, U. Palendira, J.S. Wilmott, G. Long, R.A. Scolyer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.M. Allanson, A.M. Menzies, J.S. Wilmott

Study supervision: J.J. Edwards, J.S. Wilmott, G. Long, R.A. Scolyer

This work was supported by Melanoma Institute Australia, the New South Wales Department of Health, NSW Health Pathology, the National Health and Medical Research Council of Australia (NHMRC) and Cancer Institute NSW. J.S. Wilmott, R.A. Scolyer and G. Long are supported by NHMRC Fellowships. U. Palendira and A. Ferguson were supported by the Cancer Council NSW (RG18-08). G. Long and J.F. Thompson are supported by the Melanoma Foundation of the University of Sydney through the University of Sydney Medical Foundation. J.J. Edwards is supported by the Research Training Program and Chen Family Scholarship at The University of Sydney. A.M. Menzies is supported by Cancer Institute NSW Fellowship. R.P.M. Saw is supported by Melanoma Institute Australia.

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.