Purpose: Understanding why some melanomas test negative for PD-L1 by IHC may have implications for the application of anti-PD-1 therapies in melanoma management. This study sought to determine somatic mutation and gene expression patterns associated with tumor cell PD-L1 expression, or lack thereof, in stage III metastatic melanoma to better define therapeutically relevant patient subgroups.

Experimental Design: IHC for PD-L1 was assessed in 52 American Joint Committee on Cancer stage III melanoma lymph node specimens and compared with specimen-matched comprehensive clinicopathologic, genomic, and transcriptomic data.

Results: PD-L1–negative status was associated with lower nonsynonymous mutation (NSM) burden (P = 0.017) and worse melanoma-specific survival [HR = 0.28 (0.12–0.66), P = 0.002] in stage III melanoma. Gene set enrichment analysis identified an immune-related gene expression signature in PD-L1–positive tumors. There was a marked increase in cytotoxic T-cell and macrophage-specific genes in PD-L1–positive melanomas. CD8Ahigh gene expression was associated with better melanoma-specific survival [HR = 0.2 (0.05–0.87), P = 0.017] and restricted to PD-L1–positive stage III specimens. NF1 mutations were restricted to PD-L1–positive tumors (P = 0.041).

Conclusions: Tumor negative PD-L1 status in stage III melanoma lymph node metastasis is a marker of worse patient survival and is associated with a poor immune response gene signature. Lower NSM levels were associated with PD-L1–negative status suggesting differences in somatic mutation profiles are a determinant of PD-L1–associated antitumor immunity in stage III melanoma. Clin Cancer Res; 22(15); 3915–23. ©2016 AACR.

This article is featured in Highlights of This Issue, p. 3705

Translational Relevance

Pharmacologic inhibition of the PD-1/PD-L1 axis has clinical efficacy in patients with advanced stage melanoma. Patients with PD-L1–negative locoregional metastasis have a worse prognosis and respond less frequently to checkpoint inhibition therapies. At present, little is known about the regulation of PD-L1 expression in vivo. Examining immunotherapy-naïve stage III melanomas, we observed that PD-L1–negative specimens had significantly lower NSM levels and exhibited a poor immune response gene expression signature when compared with PD-L1–positive specimens. This suggests that PD-L1 status may be dependent on somatic mutation profiles. Neither NSM level nor PD-L1 status were sufficient to predict clinically significant CD8+ T-cell tumor infiltration of stage III tumors. CD8Ahigh gene expression is a marker of good prognosis in stage III melanoma. Using IHC to assay PD-L1 status may be a robust method for determining the presence of clinically relevant immune responses in stage III melanoma patients.

Programmed cell death ligand-1 (PD-L1) regulates cellular and humoral immunity through the checkpoint receptor programmed cell death-1 (PD-1) on activated T cells and B cells (1). PD-L1 binding to PD-1 suppresses the duration and amplitude of immune responses through inhibition of CD8+ T cells and activation of CD4+ T-regulatory lymphocytes. Melanoma tumor cells aberrantly leverage this mechanism by expressing PD-L1 ligand affecting suppression of antitumor immunity (1).

Clinical trials of antagonists to the PD-1 immune checkpoint have demonstrated efficacy in the treatment of advanced stage melanoma, lung cancer, renal cancer, and colon cancer, among others (2–4). Two randomized, controlled, phase III clinical trials comparing monoclonal anti-PD1 (pembrolizumab or nivolumab) with standard-of-care ipilimumab (anti-CTLA4) demonstrated an improved response rate, progression-free survival (5, 6), and overall survival (5) in unresectable stage III/IV melanoma patients.

Several studies in advanced stage melanoma patients have observed that PD-L1–negative tumors have a worse prognosis than melanoma patients whose tumors express PD-L1 (6–9); however, other studies have reported opposite findings such that the prognostic significance of PD-L1 tumor cell expression remains unresolved (10–12). What is clear is that PD-L1–negative patients respond less frequently to checkpoint inhibition therapies (5, 6). Furthermore, PD-L1–negative patients respond better to combination anti-PD-1 and anti-CTLA-4 therapy than anti-PD-1 alone (6). Consequently, understanding why some melanomas do not develop a PD-L1 tumor response may help to define therapeutically relevant patient subsets, provide insights into enhancing the efficacy of immune checkpoint inhibition therapy, or uncover novel therapeutic targets in PD-L1–negative tumors.

The immunohistochemical pattern of PD-L1 expression by tumor cells is typically heterogeneous and contiguous to areas of tumor-infiltrating lymphocytes (TIL). IFNγ from lymphocytes can induce tumor PD-L1 expression in vitro (1) and this mechanism likely accounts for the majority of PD-L1 expression by tumor cells. However, not all TILs are associated with a PD-L1 tumor response and discordant PD-L1 status is observed in longitudinal melanoma samples from 50% of melanoma patients (13). Furthermore, in 25% of melanoma patients, PD-L1 was not detected across multiple longitudinal intrapatient specimens taken at various stages of disease progression, even among specimens with measurable lymphocytic infiltration (13). Uncovering the mechanisms responsible for this complex and heterogeneous PD-L1 tumor response is critical to a better understanding of anti-PD-1 therapy.

The objective of this study was to define the biologic and clinical relevance of PD-L1 status in stage III melanomas. To achieve this, PD-L1 IHC was assessed in lymph node metastases, which show the highest rate of PD-L1 expression compared with matched primary or distant metastases (13), from immunotherapy-naïve melanoma patients. PD-L1 IHC results were then compared with detailed clinical, histologic, somatic mutation, and transcriptomic data. Our results demonstrate an identifiable pattern of somatic and immune-associated transcriptomic differences between PD-L1–positive and PD-L1–negative stage III melanomas.

Patients and specimens

The study was undertaken with Human Ethics Review Committee approval and patient's informed consent. The Melanoma Institute Australia (MIA) database and archival files of the Department of Tissue Pathology and Diagnostic Oncology, Royal Prince Alfred Hospital, were utilized to identify melanoma patients who progressed to stage III disease, who had formalin-fixed paraffin-embedded (FFPE) tumor specimens, and had not received immunotherapy or BRAF/MEK inhibitors. All patients had a history of primary cutaneous melanoma. Patient demographics, primary tumor characteristics (age at primary diagnosis, Breslow thickness, Clark level, lymphatic invasion, mitotic rate, regression status, satellite status, T classification, sentinel node status, ulceration, vascular invasion, and time to first recurrence), and follow-up data were retrieved from the MIA Research Database.

Immunochemistry for PD-L1 (E1L3N) rabbit mAb

Whole FFPE block tissue sections were cut at 3 μm onto superfrost+ glass slides and stored at 4°C until IHC was performed (<2 weeks). IHC was performed on a Dako autostainer/PT-Link with high pH target retrieval buffer (Dako, K8005) as per the manufacturer's instructions. The primary antibody against PD-L1 (E1L3N-XP-Rb mAb; CST#13684) was incubated for 45 minutes at room temperature at a 1:500 dilution and visualized using the MACH3 Rabbit HRP polymer detection system (Biocare; M3R531) and DAB Chromogen Kit (Biocare; BDB2004) as per the manufacturer's instructions.

Pathologic assessment of PD-L1 IHC

PD-L1 IHC was evaluated as the percentage of positive tumor cells and the maximum intensity of immunohistochemical signal (0–3) were recorded. Positive staining of the surface membrane, cytoplasm, and nucleus of tumor cells was noted and scored independently. PD-L1 status was categorized as positive if ≥1% of tumor cells were found to express unequivocal membrane PD-L1. A cutoff for PD-L1 positivity of ≥1% was established in previous studies and verified in the current cohort via the ROC curve and Manhattan distance method (13).

RNAseq gene expression analysis

RNA sequencing (RNAseq) and clinical data were downloaded from the TCGA data portal (https://tcga-data.nci.nih.gov/) as described in The Cancer Genome Atlas (TCGA) research Network (14) and the results presented here are based upon data generated by the TCGA Research Network (http://cancergenome.nih.gov/). Gene expression data include normalized (level 3) data for the Skin Cutaneous Melanoma (SKCM) dataset. Samples were selected with a TCGA-EE ID label and sample type of regional lymph node. To identify relative transcriptomic differences between PD-L1–negative and PD-L1–positive stage III melanoma specimens, we calculated fold change for each gene, using edgeR version 3.8.6 (15) in R version 3.1.2. A procedure which is robust to outliers was used for generalized linear model parameter fitting (16). Fold-change values were exported for all genes and analyzed with version 2.2.0 of GSEA (17), using the GseaPreranked module. Default analysis parameters were used. Gene sets of the Canonical Pathways database of MSigDB version 5.0 were tested. A gene set with an FDR <0.05 was considered to be significantly enriched in genes which had large fold changes in a particular direction.

Statistical analysis

Correlations of PD-L1 IHC score with clinical features were performed using Spearman Rho method. Associations with PD-L1 status were performed using Kruskal–Wallis method. Meta-analysis of TCGA genomic classification was performed using either the Kruskal–Wallis method or Fisher exact test and Supplementary Table S1D (Patient-centric table, related to Figs. 15; Table 1) from the Genomic Classification of Cutaneous Melanoma (14). Univariate melanoma-specific survival analysis was carried out using ROC curves and the Manhattan method to dichotomize patients for each variable. Melanoma-specific survival was calculated from the date of surgical resection of stage III melanoma specimen to date of last follow-up (censored) or death from melanoma (event). Statistical analyses were performed using IBM SPSS Statistics 21 (18), Cutoff Finder version 2.0 (19), or Spotfire analytics software v6.5.2 (20).

Figure 1.

Immunohistochemical detection of PD-L1 in stage III melanoma. A, regional node metastasis with low-level lymphocyte infiltration demonstrating PD-L1 expression by a minority of tumor cells. B, regional node metastasis with high-level lymphocyte infiltration demonstrating PD-L1 expression in the majority of tumor cells (object lens 40×).

Figure 1.

Immunohistochemical detection of PD-L1 in stage III melanoma. A, regional node metastasis with low-level lymphocyte infiltration demonstrating PD-L1 expression by a minority of tumor cells. B, regional node metastasis with high-level lymphocyte infiltration demonstrating PD-L1 expression in the majority of tumor cells (object lens 40×).

Close modal
Figure 2.

PD-L1–negative status is associated with lower NSM burden in stage III melanoma (*, P = 0.017; Kruskal–Wallis).

Figure 2.

PD-L1–negative status is associated with lower NSM burden in stage III melanoma (*, P = 0.017; Kruskal–Wallis).

Close modal
Figure 3.

Gene Set Enrichment Analysis (GSEA) enrichment score (ES) curves for the top 6 of 31 significantly enriched pathways between PD-L1 positive and PD-L1 negative melanoma specimens. A, KEGG; Asthma. B, KEGG; autoimmune thyroid disease. C, KEGG; allograft rejection. D, KEGG; graft versus host disease. E, Reactome; PD-1 signaling. F, Reactome; phosphorylation of CD3 and TCR zeta ETA chains. The vertical black lines indicate the rank position (left to right; positively to negatively correlated) of each gene in the gene set.

Figure 3.

Gene Set Enrichment Analysis (GSEA) enrichment score (ES) curves for the top 6 of 31 significantly enriched pathways between PD-L1 positive and PD-L1 negative melanoma specimens. A, KEGG; Asthma. B, KEGG; autoimmune thyroid disease. C, KEGG; allograft rejection. D, KEGG; graft versus host disease. E, Reactome; PD-1 signaling. F, Reactome; phosphorylation of CD3 and TCR zeta ETA chains. The vertical black lines indicate the rank position (left to right; positively to negatively correlated) of each gene in the gene set.

Close modal
Figure 4.

Heat map shows z-scores for immune-associated genes differentially expressed by PD-L1 status in stage III melanoma specimens. Of 531 differentially expressed genes (FDR < 0.01), 141 were identified which cross-referenced with the ImmPort list of immune-related genes. The top quartile of differentially expressed immune-related genes is shown. False discovery rate values (FDR), P values, and log fold change (logFC) are also listed alongside of the Gene symbol. The majority of immune genes (64%) were up-regulated in the PD-L1positive groups (PD-L1 <0% of tumor cells).

Figure 4.

Heat map shows z-scores for immune-associated genes differentially expressed by PD-L1 status in stage III melanoma specimens. Of 531 differentially expressed genes (FDR < 0.01), 141 were identified which cross-referenced with the ImmPort list of immune-related genes. The top quartile of differentially expressed immune-related genes is shown. False discovery rate values (FDR), P values, and log fold change (logFC) are also listed alongside of the Gene symbol. The majority of immune genes (64%) were up-regulated in the PD-L1positive groups (PD-L1 <0% of tumor cells).

Close modal
Figure 5.

Associations of PD-L1 status with melanoma-specific survival (n = 52). A, negative PD-L1 status was associated with worse survival versus PD-L1positive/CD8Alow or PD-L1positive/CD8Ahigh (P = 0.007; log rank). B, PD-L1–negative stage III melanomas were associated with worse survival versus PD-L1 low or PD-L1 high tumors, [HR= 0.22 (0.07–0.68), P = 0.0042] and [HR = 0.35 (0.14–0.87), P = 0.019], respectively.

Figure 5.

Associations of PD-L1 status with melanoma-specific survival (n = 52). A, negative PD-L1 status was associated with worse survival versus PD-L1positive/CD8Alow or PD-L1positive/CD8Ahigh (P = 0.007; log rank). B, PD-L1–negative stage III melanomas were associated with worse survival versus PD-L1 low or PD-L1 high tumors, [HR= 0.22 (0.07–0.68), P = 0.0042] and [HR = 0.35 (0.14–0.87), P = 0.019], respectively.

Close modal
Table 1.

Predictors of outcome in stage III melanoma patients

ComparisonCut-offHR (CI)Log rank P
(A) PD-L1 positive vs. PD-L1 negative >0% 0.28 (0.12–0.66) 0.002 
(B) PD-L1 low vs. PD-L1 negative >0% 0.35 (0.14–0.87) 0.019 
(C) NSM burden high >203 0.35 (0.15–0.81) 0.010 
(D) CD8A high — 0.20 (0.05–0.87) 0.017 
ComparisonCut-offHR (CI)Log rank P
(A) PD-L1 positive vs. PD-L1 negative >0% 0.28 (0.12–0.66) 0.002 
(B) PD-L1 low vs. PD-L1 negative >0% 0.35 (0.14–0.87) 0.019 
(C) NSM burden high >203 0.35 (0.15–0.81) 0.010 
(D) CD8A high — 0.20 (0.05–0.87) 0.017 

NOTE: Optimal cutoffs calculated on the basis of ROC curve and Manhattan distance method. HRs and CIs calculated for each variable. A) PD-L1–positive status was a marker of better outcome. B) PD-L1 low patients (<10% PD-L1) had significantly better outcome than PD-L1–negative patients. C) NSM burden high (>203 NSM) was a marker of better outcome. D) High CD8A gene expression was a marker of better outcome.

Patients and tumor specimens

Cases were selected from TCGA data portal whose specimens had been obtained from resected stage III melanoma (lymph node metastasis) and for which matched archival pathology tumor blocks were available from the Department of Tissue Pathology and Diagnostic Oncology at the Royal Prince Alfred Hospital, Sydney, Australia. This study included a total of 52 surgically resected melanoma lymph node metastasis specimens from 52 patients. Median patient age at surgical resection for stage III melanoma was 57 years, and median patient follow-up from date of surgical resection for stage III melanoma was 2.5 years (0.08–10 years).

Immunohistochemical patterns of PD-L1 expression in stage III melanoma

PD-L1 showed marked intertumoral heterogeneity. When present, PD-L1 membrane staining was often strong (2 to 3+) and associated with weak cytoplasmic staining (Fig. 1). Cytoplasmic staining of sinus histiocytes and dendritic cells within residual lymphoid follicles was also frequently observed. Importantly, melanomas negative for PD-L1 did not show any significant degree of PD-L1 staining by intratumoral macrophages or infiltrating lymphocytes. In view of the observed staining pattern and predicted protein function, further analyses were performed utilizing the percentage of tumor cells with PD-L1 membrane staining in three representative regions of interest (5 mm2) within each lymph node metastasis. PD-L1 status was defined as positive when ≥1% of tumor cells exhibited PD-L1 membrane staining. There were 40 specimens (77%) with ≥1% PD-L1–expressing tumor cells and 12 specimens (23%) with 0% PD-L1–expressing tumor cells. The median nonzero PD-L1 positivity rate was 6.5% and the overall median PD-L1 positivity rate was 3.5%, indicating that even in positive cases the proportion of PD-L1–positive tumor cells is typically low. PD-L1 levels below 10% were observed in 42% of patients (PD-L1 low), and levels above 10% in 35% of patients (PD-L1 high).

We compared PD-L1 score by IHC to PD-L1 RNAseq expression data. There was a significant correlation between RNAseq and IHC (r2 = 0.711, P = 3.33E−09). However, while the overall correlation was strong there was disagreement between IHC and RNAseq expression values at the low end of the detection range. The disagreement between the two methods is pronounced such that assigning tumor PD-L1 status by gene expression alone is not reconcilable with immunohistochemical data and thus we contend that the later method remains the better standard for assigning tumor PD-L1 status (Supplementary Fig. S1).

Clinicopathologic analysis of the antecedent primary melanoma

We compared the PD-L1 status of stage III melanoma specimens with clinicopathologic data of the matched primary melanoma (21, 22). PD-L1 status of stage III melanoma specimens was not associated with either Breslow thickness or AJCC N-stage at primary diagnosis (P = 0.23 and 0.65; respectively; Kruskal–Wallis). There were no associations between PD-L1 status and the presence of ulceration or regression of the primary melanoma (P = 0.31 and P = 0.71; χ²) nor between PD-L1 status and sex (P = 0.55; χ²). Lower PD-L1 status was associated with a higher risk of death from melanoma (P = 0.033; χ²). PD-L1 status was not associated with the reported TCGA pathologic TIL score (P = 0.11; Kruskal–Wallis).

Somatic mutation profiles and PD-L1 status

We compared PD-L1 status with matched RNAseq somatic mutation profiles obtained for each tumor from TCGA. Nonsynonymous mutation (NSM) burden was assessed as the sum of nonsilent missense mutations derived from level 3 RNAseq mutation annotation files downloaded from the TCGA data portal for each patient specimen. The mean NSM burden was 518 [95% confidence interval (CI, 274.01–761.99)]. Low NSM burden was associated with negative PD-L1 status (P = 0.017; Kruskal–Wallis; Fig. 2) but did not associate with PD-L1 IHC score (P = 0.40; Kruskal–Wallis). There was a single lentigo maligna melanoma subtype (LMM) in the study cohort and this sample was noted as an outlier in the PD-L1–negative group as having high NSM burden and negative PD-L1 status; exclusion of this specimen increased the significance of the NSM burden association to PD-L1–negative status (P = 0.003; Kruskal–Wallis). A similar trend between PD-L1 status and NSM burden was observed in a smaller (n = 16) independent cohort of stage III melanomas (Supplementary Fig. S2).

Mutation status for BRAF, NRAS, and NF1 were assessed from the RNAseq data. NF1 mutations were restricted to PD-L1–positive tumors (P = 0.041; χ²), whereas neither BRAF nor NRAS mutations were associated with PD-L1 status. High NSM burden was also associated with NF1 mutation status (P = 0.0009; Kruskal–Wallis). NSM burden was not associated with CD8Ahigh gene expression as surrogate marker of immune response (rS = −0.05, P = 0.70). Other than NF1, we did not identify any other significantly mutated genes from the RNAseq mutation data that associated to PD-L1 status.

Transcriptome expression data

RNAseq gene expression values and the Bioconductor software package edgeR were used to create a ranked list of genes which showed differential expression (DE) between PD-L1–negative and PD-L1–positive tumors. Gene set enrichment analysis (GSEA) of the DE gene list identified 31 pathways with gene sets that were significantly enriched (FDR < 0.05) in the PD-L1–positive group but no pathways that were significantly enriched in the PD-L1–negative group (Supplementary Table S1). The 31 pathways all related to immune biology. The top 6 pathways included Reactome PD-1 signaling and Reactome phosphorylation of CD3 and TCR zeta chains (Fig. 3).

To further characterize PD-L1–associated immune-related genes, we used the fold-change data generated from edgeR and a cut-off FDR < 0.01 to select the top differentially expressed genes by PD-L1 status. A total of 531 genes passed this cutoff and were cross-referenced with the ImmPort comprehensive list of immune-related genes (23; http://immport.niaid.nih.gov). Among 531 differentially expressed genes, 141 (27%) were linked to immune-related biology by ImmPort. Of these, 90 immune-related genes were more highly expressed in PD-L1–positive tumors and 51 were more highly expressed in PD-L1–negative tumors (Fig. 4 and Supplementary Fig. S3; top quartile and all 141 immune-related genes, respectively). The observed immune gene signature was uniform across PD-L1–negative tumors. PD-L1 low-scoring tumors (PD-L1positive < 10%) had more heterogeneous immune gene signatures, whereas PD-L1 high-scoring tumors (PD-L1positive > 10%) had a uniform gene signature. Briefly, in addition to the T-cell–specific genes CD8A and CRTAM, PD-L1–positive tumors were significantly enriched for macrophage-specific genes (MSR1, CD58, CLEC7A, FCGR1A, FCGR1B, FCGR1C, FCGR3A) and showed upregulation of the chemokine-related genes CCL4, CCL8, CXCL9, CXCL10, and CXCL11, as well as the cytokine-specific genes IL13RA2 and IL7.

The DE gene list points to downregulation of MHC-I antigen presentation through lower expression of the PSME2 and B2M genes in PD-L1–negative tumors. The proteasome gene PSMB7 was more highly expressed in PD-L1–negative tumors, whereas the immunoproteasome alternate regulator subunit, PSME2, was more highly expressed in PD-L1–positive tumors (24). The B2M gene codes for the β2-microglobulin protein a component of MHC-I molecules and had lower expression in PD-L1–negative tumors.

Melanoma-specific survival

To characterize associations between PD-L1 status and immune response with clinical behavior, we examined melanoma-specific survival melanoma-specific survival from the date of surgical resection of stage III disease. Cut-off values were calculated for each variable based on ROC curve and Manhattan distance method (19). HRs and CIs were calculated for each cutoff (Table 1). Briefly, PD-L1–positive status was associated with better survival [HR = 0.28 (0.12–0.66), P = 0.002], as was CD8Ahigh gene expression as a surrogate for immune response [HR = 0.2 (0.05–0.87), P = 0.017], pathologic TIL score [HR = 0.42 (0.17–1.04), P = 0.052], and NSM burden [HR = 0.44 (0.22–0.88), P = 0.017].

We observed PD-L1–negative status was associated with worse survival than PD-L1/CD8low or PD-L1/CD8high tumors (Fig. 5A). CD8Ahigh gene expression was restricted to PD-L1–positive specimens. In addition, PD-L1low and PD-L1high were associated with better survival versus PD-L1–negative stage III melanoma [HR = 0.22 (0.07–0.68), P = 0.0042] and [HR = 0.35 (0.14–0.87), P = 0.019], respectively (Fig. 5B). A similar pattern for PD-L1 status and CD8A gene expression was observed in a survival analysis of an independent dataset (25) for which we had PD-L1 immunohistochemical data available (P = 0.014; log rank; Supplementary Fig. S4).

Meta-analysis of TCGA genomic classifications

We performed a meta-analysis to identify associations between PD-L1 status and seven melanoma genomic classification clusters from the TCGA's genomic classification of cancer (Oncosign clusters, Methylation clusters, Mutation clusters, RNAseq clusters, Protein array clusters, MiR clusters, UV signature clusters; ref. 14). Analysis of the methylation classification clusters indicated that the hypomethylation cluster was significantly associated with PD-L1–positive tumors (Supplementary Fig. S5A). As well, analysis of the mutation subtype classification cluster indicates that the NF1 subtype is restricted to PD-L1–positive tumors; however, the association was not statistically significant (Supplementary Fig. S5B). There were two melanomas in the current study that were identified as belonging to the non-UV signature subtype; both of these samples were PD-L1 negative. The remaining 50 samples were all UV signature subtype melanomas. The remaining melanoma genomic subtypes were not significantly associated with PD-L1 status.

It is unclear whether a difference in tumor PD-L1 status in stage III melanoma reflects phenotypic differences in immune responses, or whether PD-L1–negative status is an intrinsic quality of the tumor. As tumor PD-L1 is induced by IFNγ signaling through TILs, it is likely that PD-L1 tumor expression is best described as a reflecting innate adaptation by melanoma tumor cells to ongoing antitumor immunity (26) and yet PD-L1–negative tumors can frequently have associated TILs. Nevertheless, many PD-L1–negative tumors lack evidence of antitumor immunity, have a worse prognosis, and respond less frequently to immune checkpoint therapies (26), although lack of immunohistochemical evidence of PD-L1 expression does not preclude response.

Immunohistochemical studies of PD-L1 demonstrate that, when present, PD-L1 melanoma cell expression is a sporadic and heterogeneous event (8, 13). In this study, we compared stage III melanoma tumor specimens with low levels of PD-L1 tumor expression (<10%) to those with high levels of PD-L1 tumor expression (≥10%) and those with negative PD-L1 status. We observed that survival in patients with PD-L1–negative tumors was significantly worse when compared with patients who had either low-level PD-L1 [HR = 0.35 (0.14–0.87), P = 0.019] or high-level PD-L1 [HR = 0.22 (0.07–0.68), P = 0.0045]. Thus, even at immunohistochemical detection levels below 10% of tumor cells, PD-L1 can ascertain a significant aspect of melanoma tumor behavior and patient outcome, with implications for the selection of cutoffs in clinical trials and studies of PD-L1 as a biomarker (6). Furthermore, we observed that in stage III melanomas, high CD8A gene expression was restricted to PD-L1–positive tumors and these CD8Ahigh/PD-L1positive tumors were associated with better melanoma-specific survival. Furthermore, CD8Alow specimens that were also PD-L1 positive had better survival than patients whose tumors were both PD-L1 negative and CD8Alow.

Comprehensive studies of melanoma have identified transcriptomic subclasses enriched for immune gene expression that are significantly associated with improved patient survival (14, 25). In the current study, 14 (27%) stage III melanomas exhibited pronounced CD8A gene expression which was concurrent to tumor cell PD-L1 expression and was associated with better melanoma-specific survival. Patients with moderate to high PD-L1 levels displayed consistent immune gene signatures which were associated with increased representation of the PD-1 pathway and CD3 and TCR activation even in those lacking high CD8A gene expression. For patients with low PD-L1 expression (PD-L1 > 1% and < 10%), the presence of this immune signature was less pronounced. This might be due to the selection of high tumor content areas (low TILS) during specimen submission for TCGA genomic studies. Alternatively, this could be attributed to nascent and/or declining immune responses which were not detectable at the gene expression level at the time of tumor resection. Nevertheless, patients whose stage III specimens were PD-L1 low had a considerably better clinical prognosis and were phenotypically different in their immune-related gene expression biology to PD-L1–negative tumors. Given the low number of cases in this cohort, further studies powered to address the question of the clinical utility of PD-L1 and the establishment of appropriate cut-off points are needed, especially in PD-1–treated patients. Our data points to the utility of low-level PD-L1 detection in stage III melanoma specimens using IHC.

High PD-L1–expressing tumors demonstrated a strong macrophage-specific gene expression signal with enrichment for MSR1, CD16, and CD64 related genes (MSR1, CD58, CLEC7A, CXCL10, FCGR1A, FCGR1B, FCGR1C, and FCGR3A). While the role of macrophages in melanoma progression is unclear, gene expression as well as morphologic observations, indicate that there are notable differences in the degree of macrophage infiltration present in PD-L1–positive melanomas compared with PD-L1–negative tumors. Furthermore, macrophage PD-L1 expression is perhaps the most abundant source of intertumoral PD-L1 and understanding the role of PD-L1–positive macrophages in tumor progression could be important.

Although tumor-specific CD8+ T-cell responses often develop in melanoma patients, they frequently lead to the accumulation of functionally tolerant TILs within tumors, especially in lymph node metastases (27). While much of the recent focus in immunotherapy is geared toward reactivating exhausted T cells, immunotherapeutic approaches are needed for those tumors which have little or no apparent T-cell response. Interestingly, while the checkpoint genes PD-L2 and LAG3 were both more highly expressed in PD-L1–positive tumors, another checkpoint gene, CD276, was more highly expressed in PD-L1–negative tumors. CD276 is the target of the anticheckpoint drug MGA271 (28) currently undergoing a phase I clinical trial in combination with ipilimumab.

Low-level or downregulated MHC-I antigen presentation may be another feature of PD-L1–negative specimens. Higher expression of the nonimmunoproteasome subunit PSMB7 (24), and lower expression of B2M MHC-I subunit, could indicate that these tumors have downregulated antigen presentation as a means of immune escape. However, this might also be explained by the lack of TILs and the incumbent IFNγ signaling. Nevertheless, the lack of an immunoproteasome or MHC-I antigen presentation would suggest these tumors might be predisposed to resistance to T-cell–mediated immunotherapies (29). Thus, while GSEA indicated that immune pathways were the defining gene expression phenotype in PD-L1–positive tumors, there was also a novel immune gene signature associated with PD-L1–negative tumors and this signature may be an interesting basis for immunotherapy targets. Indeed, augmentation of NK cell activity through an anti-KIR antibody-based therapy is undergoing a phase-I clinical trial (ClinicalTrials.gov Identifier: NCT01714739) in combination with anti-PD-1 and anti-CTLA4 with goal of targeting MHC-I–deficient melanoma tumor cells (30, 31). Nevertheless, many immune-related genes are driven by IFNγ signaling and thus determining if any of these differentially expressed genes contribute to intrinsic immune evasion by melanoma cancer cells requires further study.

Melanoma has the highest average mutation burden of any cancer type; however, mutation burden still varies widely between melanomas, with many tumors at the lower end of the range (14, 32). Higher mutation levels in melanoma are associated with the degree of clinical benefit from ipilimumab but are not sufficiently predictive to be of clinical use (33). Across several cancer types, including melanoma, higher mutation levels are positively associated with the likelihood of immune reactive missense mutations capable of induction of a CD8+ antitumor immune response (34–38). Antitumor immunity has also been linked to higher mutation levels in cancer implicating neoantigens in affecting T-cell responses (39) and responses to both CTLA4 and PD-L1 therapies have been demonstrated to have a genetic basis largely dependent on the presence of specific types of patient neoantigens (33, 40, 41). In the current study, we found that NSM burden was highly varied among PD-L1–positive tumors, but in PD-L1–negative tumors remained significantly at the low end of the NSM burden range. This suggests that somatic mutation profiles might be a factor in determining not only clinical behavior but also PD-L1 status itself and could influence response to anti-PD-1 therapies.

A single outlier patient was noted in the PD-L1–negative group as having high NSM burden. This tumor was of the histologic subtype LMM, and the only LMM sample in the study. LMM is a subtype of melanoma that typically arises in chronically and severely sun damaged skin of elderly patients, and higher NSM burden levels are expected in LMM. This outlier does raise an interesting question as it was the only PD-L1–negative sample with high NSM burden. Sampling bias due to high levels of PD-L1 heterogeneity may also explain this outlier. Whether or not LMM presents a special case remains to be seen.

As high NSM burden increases the probability of an antigen-induced immune response, we predicted there would be an association between an immune gene signature and NSM burden. Interestingly, we did not observe any association between NSM burden and CD8A gene expression as a surrogate marker of immune response. Thus, while PD-L1–negative status was associated with low NSM burden, the effect of NSM burden on immune response likely diminishes beyond a threshold at which point other factors become dominant.

In summary, using IHC to assay PD-L1 status may be a robust method for determining the presence of low level, clinically relevant and/or therapeutically actionable, immune responses in stage III melanoma patients. PD-L1–negative status was a negative prognostic indicator, and was associated with a poor prognosis immune gene signature. Importantly, PD-L1–negative patients had significantly lower levels of somatic mutations when compared with tumors from PD-L1–positive patients. Overall, this suggests that somatic mutation levels may be an important determinant of PD-L1–associated antitumor immunity.

A.M. Menzies reports receiving speakers bureau honoraria from Bristol-Myers Squibb and Novartis and is a consultant/advisory board member for Chugai and MSD. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Madore, A.M. Menzies, G.V. Long, G.J. Mann, R.A. Scolyer, J.S. Wilmott

Development of methodology: J. Madore, A.M. Menzies, R.A. Scolyer, J.S. Wilmott

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Madore, R. Vilain, G.J. Mann, J.S. Wilmott

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Madore, D. Strbenac, R. Vilain, A.M. Menzies, J.Y.H. Yang, G.V. Long, G.J. Mann, R.A. Scolyer, J.S. Wilmott

Writing, review, and/or revision of the manuscript: J. Madore, D. Strbenac, A.M. Menzies, J.F. Thompson, J.Y.H. Yang, G.V. Long, G.J. Mann, R.A. Scolyer, J.S. Wilmott

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Strbenac, G.V. Long, G.J. Mann, J.S. Wilmott

Study supervision: J.S. Wilmott

The authors gratefully acknowledge the assistance of their colleagues at Melanoma Institute Australia and the Department of Tissue Pathology and Diagnostic Oncology at the Royal Prince Alfred Hospital.

R. Vilain is supported by a Cameron Fellowship through Melanoma Institute Australia. R.A. Scolyer is supported by the National Health and Medical Research Council Fellowship program. Funding support from the National Health and Medical Research Council, Cancer Institute New South Wales, the Melanoma Foundation of the University of Sydney, and Melanoma Institute Australia is also gratefully acknowledged.

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.

1.
Dong
H
,
Strome
SE
,
Salomao
DR
,
Tamura
H
,
Hirano
F
,
Flies
DB
, et al
Tumor-associated B7-H1 promotes T-cell apoptosis: a potential mechanism of immune evasion
.
Nat Med
2002
;
8
:
793
800
.
2.
Brahmer
JR
,
Tykodi
SS
,
Chow
LQ
,
Hwu
WJ
,
Topalian
SL
,
Hwu
P
, et al
Safety and activity of anti-PD-L1 antibody in patients with advanced cancer
.
N Engl J Med
2012
;
366
:
2455
65
.
3.
Garon
EB
,
Rizvi
NA
,
Hui
R
,
Leighl
N
,
Balmanoukian
AS
,
Eder
JP
, et al
Pembrolizumab for the treatment of non-small-cell lung cancer
.
N Engl J Med
2015
;
372
:
2018
28
.
4.
Le
DT
,
Uram
JN
,
Wang
H
,
Bartlett
BR
,
Kemberling
H
,
Eyring
AD
, et al
PD-1 blockade in tumors with mismatch-repair deficiency
.
N Engl J Med
2015
;
372
:
2509
20
.
5.
Robert
C
,
Schachter
J
,
Long
GV
,
Arance
A
,
Grob
JJ
,
Mortier
L
, et al
Pembrolizumab versus ipilimumab in advanced melanoma
.
N Engl J Med
2015
;
372
:
2521
32
.
6.
Larkin
J
,
Chiarion-Sileni
V
,
Gonzalez
R
,
Grob
JJ
,
Cowey
CL
,
Lao
CD
, et al
Combined nivolumab and ipilimumab or monotherapy in untreated melanoma
.
N Engl J Med
2015
;
373
:
23
34
.
7.
Thierauf
J
,
Veit
JA
,
Affolter
A
,
Bergmann
C
,
Grunow
J
,
Laban
S
, et al
Identification and clinical relevance of PD-L1 expression in primary mucosal malignant melanoma of the head and neck
.
Melanoma Res
2015
;
25
:
503
9
.
8.
Taube
JM
,
Anders
RA
,
Young
GD
,
Xu
H
,
Sharma
R
,
McMiller
TL
, et al
Colocalization of inflammatory response with B7-h1 expression in human melanocytic lesions supports an adaptive resistance mechanism of immune escape
.
Sci Transl Med
2012
;
4
:
127ra37
.
9.
Kluger
HM
,
Zito
CR
,
Barr
ML
,
Baine
MK
,
Chiang
VL
,
Sznol
M
, et al
Characterization of PD-L1 expression and associated t-cell infiltrates in metastatic melanoma samples from variable anatomic sites
.
Clin Cancer Res
2015
;
21
:
3052
60
.
10.
Massi
D
,
Brusa
D
,
Merelli
B
,
Falcone
C
,
Xue
G
,
Carobbio
A
, et al
The status of PD-L1 and tumor-infiltrating immune cells predict resistance and poor prognosis in BRAFi-treated melanoma patients harboring mutant BRAFV600
.
Ann Oncol
2015
;
26
:
1980
7
.
11.
Massi
D
,
Brusa
D
,
Merelli
B
,
Ciano
M
,
Audrito
V
,
Serra
S
, et al
PD-L1 marks a subset of melanomas with a shorter overall survival and distinct genetic and morphological characteristics
.
Ann Oncol
2014
;
25
:
2433
42
.
12.
Hino
R
,
Kabashima
K
,
Kato
Y
,
Yagi
H
,
Nakamura
M
,
Honjo
T
, et al
Tumor cell expression of programmed cell death-1 ligand 1 is a prognostic factor for malignant melanoma
.
Cancer
2010
;
116
:
1757
66
.
13.
Madore
J
,
Vilain
RE
,
Menzies
AM
,
Kakavand
H
,
Wilmott
JS
,
Hyman
J
, et al
PD-L1 expression in melanoma shows marked heterogeneity within and between patients: implications for anti-PD-1/PD-L1 clinical trials
.
Pigment Cell Melanoma Res
2015
;
28
:
245
53
.
14.
The Cancer Genome Atlas Network. 
Genomic classification of cutaneous melanoma
.
Cell
2015
;
161
:
1681
96
.
15.
Robinson
MD
,
McCarthy
DJ
,
Smyth
GK
. 
edgeR: a Bioconductor package for differential expression analysis of digital gene expression data
.
Bioinformatics
2010
;
26
:
139
40
.
16.
Zhou
X
,
Lindsay
H
,
Robinson
MD
. 
Robustly detecting differential expression in RNA sequencing data using observation weights
.
Nucleic Acids Res
2014
;
42
:
e91
.
17.
Subramanian
A
,
Tamayo
P
,
Mootha
VK
,
Mukherjee
S
,
Ebert
BL
,
Gillette
MA
, et al
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles
.
Proc Natl Acad Sci U S A
2005
;
102
:
15545
50
.
18.
IBM Corporation
.
SPSS for Windows; version
22; 
2013
.
19.
Budczies
J
,
Klauschen
F
,
Sinn
BV
,
Gyorffy
B
,
Schmitt
WD
,
Darb-Esfahani
S
, et al
Cutoff Finder: a comprehensive and straightforward Web application enabling rapid biomarker cutoff optimization
.
PLoS One
2012
;
7
:
e51862
.
20.
TIBCO Incorporated
. Spotfire; version 6.5.3.12; 
2015
.
21.
Murali
R
,
Brown
PT
,
Kefford
RF
,
Scolyer
RA
,
Thompson
JF
,
Atkins
MB
, et al
Number of primary melanomas is an independent predictor of survival in patients with metastatic melanoma
.
Cancer
2012
;
118
:
4519
29
.
22.
Menzies
AM
,
Lum
T
,
Wilmott
JS
,
Hyman
J
,
Kefford
RF
,
Thompson
JF
, et al
Intrapatient homogeneity of BRAFV600E expression in melanoma
.
Am J Surg Pathol
2014
;
38
:
377
82
.
23.
Bhattacharya
S
,
Andorf
S
,
Gomes
L
,
Dunn
P
,
Schaefer
H
,
Pontius
J
, et al
ImmPort: disseminating data to the public for the future of immunology
.
Immunol Res
2014
;
58
:
234
9
.
24.
Rivett
AJ
,
Bose
S
,
Brooks
P
,
Broadfoot
KI
. 
Regulation of proteasome complexes by gamma-interferon and phosphorylation
.
Biochimie
2001
;
83
:
363
6
.
25.
Mann
GJ
,
Pupo
GM
,
Campain
AE
,
Carter
CD
,
Schramm
SJ
,
Pianova
S
, et al
BRAF mutation, NRAS mutation, and the absence of an immune-related expressed gene profile predict poor outcome in patients with stage III melanoma
.
J Invest Dermatol
2013
;
133
:
509
17
.
26.
Taube
JM
,
Klein
A
,
Brahmer
JR
,
Xu
H
,
Pan
X
,
Kim
JH
, et al
Association of PD-1, PD-1 ligands, and other features of the tumor immune microenvironment with response to anti-PD-1 therapy
.
Clin Cancer Res
2014
;
20
:
5064
74
.
27.
Zippelius
A
,
Batard
P
,
Rubio-Godoy
V
,
Bioley
G
,
Lienard
D
,
Lejeune
F
, et al
Effector function of human tumor-specific CD8 T cells in melanoma lesions: a state of local functional tolerance
.
Cancer Res
2004
;
64
:
2865
73
.
28.
Loo
D
,
Alderson
RF
,
Chen
FZ
,
Huang
L
,
Zhang
W
,
Gorlatov
S
, et al
Development of an Fc-enhanced anti-B7-H3 monoclonal antibody with potent antitumor activity
.
Clin Cancer Res
2012
;
18
:
3834
45
.
29.
Chang
CC
,
Campoli
M
,
Restifo
NP
,
Wang
X
,
Ferrone
S
. 
Immune selection of hot-spot beta 2-microglobulin gene mutations, HLA-A2 allospecificity loss, and antigen-processing machinery component down-regulation in melanoma cells derived from recurrent metastases following immunotherapy
.
J Immunol
2005
;
174
:
1462
71
.
30.
Hölsken
O
,
Miller
M
,
Cerwenka
A
. 
Exploiting natural killer cells for therapy of melanoma
.
J Dtsch Dermatol Ges
2015
;
13
:
23
9
.
31.
Sottile
R
,
Pangigadde
PN
,
Tan
T
,
Anichini
A
,
Sabbatino
F
,
Trecroci
F
, et al
HLA class I downregulation is associated with enhanced NK-cell killing of melanoma cells with acquired drug resistance to BRAF inhibitors
.
Eur J Immunol
2016
;
46
:
409
19
.
32.
Alexandrov
LB
,
Nik-Zainal
S
,
Wedge
DC
,
Aparicio
SA
,
Behjati
S
,
Biankin
AV
, et al
Signatures of mutational processes in human cancer
.
Nature
2013
;
500
:
415
21
.
33.
Snyder
A
,
Makarov
V
,
Merghoub
T
,
Yuan
J
,
Zaretsky
JM
,
Desrichard
A
, et al
Genetic basis for clinical response to CTLA-4 blockade in melanoma
.
N Engl J Med
2014
;
371
:
2189
99
.
34.
Robbins
PF
,
Lu
YC
,
El-Gamil
M
,
Li
YF
,
Gross
C
,
Gartner
J
, et al
Mining exomic sequencing data to identify mutated antigens recognized by adoptively transferred tumor-reactive T cells
.
Nat Med
2013
;
19
:
747
52
.
35.
van Rooij
N
,
van Buuren
MM
,
Philips
D
,
Velds
A
,
Toebes
M
,
Heemskerk
B
, et al
Tumor exome analysis reveals neoantigen-specific T-cell reactivity in an ipilimumab-responsive melanoma
.
J Clin Oncol
2013
;
31
:
e439
42
.
36.
Linnemann
C
,
van Buuren
MM
,
Bies
L
,
Verdegaal
EM
,
Schotte
R
,
Calis
JJ
, et al
High-throughput epitope discovery reveals frequent recognition of neo-antigens by CD4+ T cells in human melanoma
.
Nat Med
2015
;
21
:
81
5
.
37.
Lu
YC
,
Yao
X
,
Crystal
JS
,
Li
YF
,
El-Gamil
M
,
Gross
C
, et al
Efficient identification of mutated cancer antigens recognized by T cells associated with durable tumor regressions
.
Clin Cancer Res
2014
;
20
:
3401
10
.
38.
Rooney
MS
,
Shukla
SA
,
Wu
CJ
,
Getz
G
,
Hacohen
N
. 
Molecular and genetic properties of tumors associated with local immune cytolytic activity
.
Cell
2015
;
160
:
48
61
.
39.
Brown
SD
,
Warren
RL
,
Gibb
EA
,
Martin
SD
,
Spinelli
JJ
,
Nelson
BH
, et al
Neo-antigens predicted by tumor genome meta-analysis correlate with increased patient survival
.
Genome Res
2014
;
24
:
743
50
.
40.
Rizvi
NA
,
Hellmann
MD
,
Snyder
A
,
Kvistborg
P
,
Makarov
V
,
Havel
JJ
, et al
Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer
.
Science
2015
;
348
:
124
8
.
41.
Van Allen
EM
,
Miao
D
,
Schilling
B
,
Shukla
SA
,
Blank
C
,
Zimmer
L
, et al
Genomic correlates of response to CTLA-4 blockade in metastatic melanoma
.
Science
2015
;
350
:
207
11
.

Supplementary data