The immune checkpoint LAG-3 is expressed by melanoma cells and correlates with clinical progression of the melanoma

ABSTRACT

Immune checkpoint blockers have substantially improved prognosis of melanoma patients, nevertheless, resistance remains a significant problem. Here, intrinsic and extrinsic factors in the tumor microenvironment are discussed, including the expression of alternative immune checkpoints such as lymphocyte activation gene 3 (LAG-3) and T-cell immunoglobulin and mucin domain-containing protein 3 (TIM-3). While most studies focus on immune cell expression of these proteins, we investigated their melanoma cell intrinsic expression by immunohistochemistry in melanoma metastases of 60 patients treated with anti-programmed cell death protein 1 (PD-1) and/or anti-cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) therapy, and correlated it with the expression of potential ligands, RNA sequencing data and clinical outcome. LAG-3 and TIM-3 were commonly expressed in melanoma cells. In the stage IV cohort, expression of LAG-3 was associated with M1 stage (p < 0.001) and previous exposure to immune checkpoint inhibitors (p = 0.029). Moreover, in the anti-PD-1 monotherapy treatment group patients with high LAG-3 expression by tumor cells tended to have a shorter progression-free survival (p = 0.088), whereas high expression of TIM-3 was associated with a significantly longer overall survival (p = 0.007). In conclusion, we provide a systematic analysis of melanoma cell intrinsic LAG-3 and TIM-3 expression, highlighting potential implications of their expression on patient survival.

Introduction

Immune checkpoint blockade targeting programmed cell death protein 1 (PD-1) and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) led to great improvements in survival of melanoma patients. However, only a subset of patients responds to these therapies, with about two-thirds of patients being primary or secondary resistant.^1–4 Studies investigating immune checkpoint blockade resistance focus on intrinsic and extrinsic mechanisms including alternative immune checkpoints such as lymphocyte activation gene 3 (LAG-3) and T-cell immunoglobulin and mucin domain-containing protein 3 (TIM-3).^5,6

LAG-3 is a type 1 membrane protein which is expressed on various immune cells.^7–10 Known ligands of LAG-3 include major histocompatibility complex class II (MHC class II),¹¹ liver and lymph node sinusoidal endothelial cell C-type lectin (LSECtin),¹² galectin-3,¹³ α-synuclein,¹⁴ and fibrinogen-like protein 1 (FGL-1).¹⁵ Inhibitory effects of LAG-3 on immune cells include inhibition of T-cell expansion or proliferation and cytokine secretion.^16,17 The immunoregulatory transmembrane receptor TIM-3 is also expressed on several immune cell types.¹⁰ Known TIM-3 ligands are carcinoembryonic antigen-related cell adhesion molecule 1 (CEACAM-1),¹⁸ galectin-9,¹⁹ phosphatidylserine,²⁰ and high mobility group protein B1 (HMGB1).²¹ LAG-3 and TIM-3 are often co-expressed with PD-1 on tumor infiltrating lymphocytes.^22–24 Gene expression of TIM-3 and LAG-3 was shown to be increased in T-cells of mice that acquired resistance to anti-PD-1 treatment in a lung adenocarcinoma model.²⁵ However, benefit from combination therapy with the anti-PD-1 antibody nivolumab and the anti-LAG-3 antibody relatlimab in melanoma patients was not dependent on immune cell expression of LAG-3.³

Even though immune checkpoints are classically expressed on immune cells as T-lymphocytes, first studies indicate that they can be expressed on cancer cells as well. PD-1 and CTLA-4 have been described to be expressed by malignant cells including melanoma cells^26–30 and first studies indicate that also LAG-3 and TIM-3 are expressed by cancer cells of several entities including melanoma.^31–39 However, little is known about melanoma cell intrinsic expression, especially for LAG-3, and a potential association with immunotherapy response.

We therefore investigated melanoma cell expression of LAG-3, TIM-3, and several of their ligands in metastases from a cohort of 60 melanoma patients treated with an anti-PD-1 antibody or/and the anti-CTLA-4 antibody ipilimumab and correlated the data with the clinical course of the disease.

Materials and methods

Patient material and data

Non-brain melanoma metastases were obtained from advanced melanoma patients treated at the Dermatooncology Department of the National Center for Tumor Diseases (NCT) Heidelberg, who gave written informed consent for the use of their samples for research purposes. The samples were collected between 2013 and 2020, before or shortly after the start of an immunotherapy targeting CTLA-4 (ipilimumab), PD-1 (pembrolizumab or nivolumab) or both (median time difference sample collection to start of therapy: IHC cohort: −20 days (range −135 - +6 days); FFPE RNA cohort: −22 days (range −79 - +6 days). Melanoma brain metastases of patients who gave written informed consent for the use of their samples for research purposes were obtained from the Biobank Department of Neurosurgery, Division Neurosurgical Research. Clinical data was obtained from routine clinical documentation. The M1 stage was defined according to the 8^th edition of the AJCC criteria.⁴⁰ Progression-free survival (PFS) and overall survival (OS) were defined as the time from the start of the respective immune checkpoint inhibitor (ICI) treatment to the time of progression or death (PFS) or death (OS), respectively. In case no event occurred, the date of the last contact was used for censoring. The best response was assessed in accordance with the Response Evaluation Criteria in Solid Tumors 1.1. (RECIST 1.1) criteria.⁴¹ The disease control rate (DCR, as CR + PR + SD) and objective response rate (ORR, as CR + PR) were calculated using the best response. Median follow-up time was 4.3 years for the IHC cohort and 3.1 years for the FFPE RNA cohort. The collection of patient material was approved by the Ethical Committee of the Medical Faculty of Heidelberg (S-091/2011, 005–2003). The retrospective analysis of clinical data without additional consent was approved by the Ethical Committee of the Medical Faculty of Heidelberg (S-454/2015).

Immunohistochemistry

If possible, the same FFPE tissue block was used for all experiments. If sample availability was limited, an FFPE tissue block from the same time point and region was used if available. FFPE metastasis samples were sectioned into 2–4 µm thin slides. Stainings were performed with the Leica BOND-MAX system (Leica Biosystems), using the BOND Polymer Refine Detection Kit (#DS9800, Leica Biosystems). Epitope retrieval was achieved by incubation with BOND ER1 solution (#AR9961, Leica Biosystems) or BOND ER2 solution (#AR9640, Leica Biosystems) for 20 min at 100°C (Table S1). Samples were incubated with the primary antibody (Table S1) for 30 min at room temperature. After antigen detection, slides were counterstained with hematoxylin and digitalized using a NanoZoomer2.0 HT (Hamamatsu Photonics) scanner or an Olympus SlideView VS200 scanner (OLYMPUS). Exemplary images were generated via exporting a section of the digitalized slide or via taking images using a KEYENCE BZ-X810 microscope.

Evaluation of immunohistochemical stainings

Stainings with a low protein expression frequency were evaluated using a percentage value or cutoff. Stainings with a common larger proportion of stained cells were scored via the H-scoring system. Therefore, we classified the staining intensity of tumor cells into the groups 0 (negative), 1⁺, 2^+, and 3⁺ and assessed the percentage of tumor cells in each of the groups. The H-score (range 0–300) was calculated according to the following formula: H-score = (%(1⁺) × 1) + (%(2⁺) × 2) + (%(3⁺) × 3).⁴² Based on the percentage cutoff or a cutoff at the median H-score, the expression levels were categorized into the groups “high/positive” or “low/negative”. DAB staining was used as routine staining in our laboratory. DAB stained cells can be discriminated from melanoma pigmentation by a different shade of brown coloring. The stainings were evaluated by two observers independently and in the case of a disagreement in grouping, a third experienced observer decided on the grouping.

RNA isolation from FFPE melanoma metastases

Two 10-µm-thick curls were cut from each FFPE block of a melanoma metastasis (n = 17). RNA was isolated using the ReliaPrep^TM FFPE Total RNA Miniprep System (#Z1002, Promega) according to the manufacturer’s instructions, except for DNase treatment. DNase treatment was instead performed after RNA isolation using the RapidOut DNA Removal Kit (#K2981, Thermo Fisher Scientific).

Bulk RNA sequencing

To quantify gene expression in FFPE tissue samples, a modified protocol of 3’ mRNA-seq was used. The approach was adapted from the mcSCRB-seq method^43–47 to adjust for a shorter RNA fragment length. 25 ng of total RNA were used as input for 3’ mRNA-seq. Samples were analyzed as technical duplicates in two different sequencing runs. To generate libraries, polyA⁺ RNAs were selected during cDNA synthesis (Maxima H Minus Reverse Transcriptase, #EP0753, Thermo Fisher Scientific) through annealing to a polydT oligo (5'/Bio/ACACTCTTTCCCTACACGACGCTCTTCCGATCT[BC₆]N₁₀T₃₀VN −3', IDT) containing an additional unique molecular identifier (UMI), a well-barcode and an Illumina adapter sequence. Second strand synthesis was achieved by RT-PCR using a template switch oligo (5' GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTrGrGrG − 3', Metabion) with an integrated Illumina adapter sequence. The well-barcode allowed early pooling of all samples directly after cDNA synthesis. Following purification (Sera-Mag Speed Beads #65152105050250, Thermo Fisher Scientific/pooling beads⁴⁴) and exonuclease I treatment (#M0293L, New England Biolabs), the pool was amplified by PCR (Terra Polymerase, #639270, TaKaRa) and again purified (AMPure Beads, #A63881, Beckman Coulter). The resulting library was sequenced on a NextSeq550Dx (Illumina) using a 75-cycle high output kit (#20024906, Illumina; read 1: 16 bp – UMI + barcode; read 2: 51 bp – Insert).

Bioinformatic analyses of RNA sequencing data

RNA-seq reads from bulk RNA sequencing were demultiplexed using the program bcl2fastq provided by Illumina. R version 4.1.2 was used for all further analyses. Quality control was performed using the command line tools fastQC⁴⁸ and multiQC.⁴⁹ Umitools⁵⁰ dedup was used for deduplication. The reads were aligned to the human reference genome hg38 using STAR⁵¹ aligner version 2.7.10b. The number of reads mapping to each gene was obtained from STAR using the parameter – quantMode GeneCounts. edgeR⁵² was used for differential expression analysis. McpCounter⁵³ was used for estimating the abundance of tumor immune and stromal cell populations. The deconvolute method from the R package Immunedeconv⁵⁴ was used to call the method QuantiSeq.⁵⁵

Statistical analyses and data visualisation

Statistics were performed using IBM SPSS Statistics Version 29 (IBM) or R version 4.1.2. The Person Chi-square test, or where applicable Fisher’s exact test/the exact Fisher-Freeman-Halton test, were used to analyze associations between groups. The exact p-value (two-sided) is reported. Correlations were analyzed using the Spearman correlation test. The inter-rater reliability of the classification of samples into the categories “high” and “low” was tested using Cohen’s kappa. PFS and OS were compared using Kaplan–Meier curves and the log-rank test. Hazard ratios were calculated using Cox-regression analysis. To calculate odds ratios for high LAG-3/TIM-3 expression, binary logistic regression was performed. For the multivariate logistic regression, the analyzed variables were included in one step. In all cases, a p-value of < 0.05 was considered as significant. Graphs were created using GraphPad Prism Version 9 (Dotmatics Limited), R version 4.1.2 (edgeR,⁵² ggplot2⁵⁶ and pheatmap⁵⁷), and SPSS IBM SPSS Statistics Version 29 (IBM).

Results

LAG-3 and TIM-3 are commonly expressed in tumor cells of melanoma metastases

We first analyzed melanoma cell intrinsic LAG-3 and TIM-3 expression in melanoma metastases collected before or shortly after the start of an immune checkpoint blocker therapy by immunohistochemistry (see Table S2 for patient characteristics). For LAG-3 we observed a wide spectrum of expression levels from none to strong expression (Figure 1a, Figure S1A). The staining was located mainly in the cytoplasm. Several metastases also showed a membrane staining. The samples evaluated for TIM-3 expression showed either none or up to a moderate expression (Figure 1b, Figure S2A). The staining occurred mainly in the cytoplasm and in single cases also a membrane staining was observed. Several samples showed some nuclear staining, but this was not considered for H-scoring. Tumoral expression of the immune checkpoint PD-1 was mostly negative. We only observed few metastases with a weak cytoplasmic staining and thus decided not to further analyze PD-1 expression by melanoma cells in our cohort.

Figure 1.

LAG-3, TIM-3, several of their ligands, and PD-L1 are expressed by tumor cells of melanoma metastases at differing levels. FFPE melanoma metastases were cut into 2–4 µm thin sections and were immunohistochemically stained with antibodies against the indicated proteins using the Leica BOND-MAX system. The sample number varies due to limited availability of the tumor tissue. (a) Sections were stained for LAG-3 (n = 60), α-synuclein (n = 29), galectin-3 (n = 26) and MHC class II (n = 28). Expression levels were categorized into the groups high/low according to a cutoff at the median H-score (LAG-3, α-synuclein, galectin-3) or a cutoff of < 5%/≥ 5% positive tumor cells (MHC class II). (b) Sections were stained for TIM-3 (n = 59), CEACAM-1 (n = 27), HMGB1 (n = 23) and galectin-9 (n = 24). The samples were classified into the groups high/positive or low/negative according to a cutoff at the median H-score (TIM-3, CEACAM-1, HMGB1) or at a percentage cutoff (galectin-9, < 5%/≥ 5% positive tumor cells). (c) PD-L1 stainings (n = 30) were scored as PD-L1⁻ or PD-L1⁺ (< 1%/≥ 1% positive tumor cells).

LAG-3 and TIM-3 ligands are expressed in tumor cells of melanoma metastases at varying levels

The LAG-3 ligands α-synuclein and galectin-3 showed a wide variability of melanoma cell intrinsic expression among the analyzed metastases (Figure 1a, Figure S1B, C). The staining for galectin-3 was mainly cytoplasmic, but some samples also showed a membrane staining. Focusing on the invasion front of the metastases, we evaluated membrane-located MHC class II expression of melanoma cells in the metastases by classifying samples into the groups negative/low and positive, according to a cutoff of < 5% or ≥ 5% positive cells (cutoff previously reported⁵⁸). About one-third of melanoma metastases was positive for tumor cell intrinsic MHC class II expression (Figure 1a, Figure S1D). Melanoma cell intrinsic expression of FGL-1 and LSECtin was very low or absent. Only single metastases showed a weak staining. We thus did not further analyze these markers.

Membrane staining of the TIM-3 ligand CEACAM-1 in melanoma cells of our samples was very variable (Figure 1b, Figure S2B). For HMGB1, we detected a range of cytoplasmic staining intensities of melanoma cells (Figure 1b, Figure S2C) and a nuclear staining. We considered the cytoplasmic staining for scoring, as described before.⁵⁹ The nuclear staining was largely uniform and thus not appropriate for differential expression analyses. Galectin-9 was expressed in only few of the analyzed melanoma metastases, mostly only by a low number of cells on a low to medium expression level. We thus grouped samples according to the percentage of positive tumor cells (<5% vs. ≥5%), resulting in a classification of 25% of samples as galectin-9^high (Figure 1b, Figure S2D).

As the PD-1 ligand PD-L1 was commonly discussed as a biomarker for response to anti-PD-1 therapy⁶⁰ we evaluated PD-L1 expression in melanoma cells based on a cutoff of < 1% vs. ≥1% (Figure 1c). Information about the PD-L1 status of three metastases was obtained from routine clinical histology reports. In total, 57.6% of the metastases were PD-L1⁺ (Figure S3).

Melanoma cell intrinsic TIM-3 and LAG-3 expression correlate with each other

We next analyzed the expression levels of LAG-3 and TIM-3 for correlations with other stained markers. For this analysis, we either tested associations between the classification groups (low vs. high) or where possible used the raw H-scores of scorer 1 and excluded any samples that were later classified into another expression group (high vs. low) by the third observer. We observed a significant positive correlation between combined cytoplasmic and membrane LAG-3 and TIM-3 expression (Spearman’s ρ = 0.334, p = 0.015, Figure 2a). This correlation was validated by analyzing the scores of scorer 2 according to the same principles (Spearman’s ρ = 0.293, p = 0.048). However, when considering the groupings of all samples into the categories low and high (and not using the raw H-scores), the association between LAG-3^low/high and TIM3^low/high samples showed only a trend, not reaching statistical significance (p = 0.116, Figure 2b). Correlation of LAG-3 and TIM-3 with their respective ligands did not show any significant correlations. Positivity of samples for PD-L1 was not associated with high melanoma cell intrinsic expression of LAG-3, TIM-3, or their analyzed ligands.

Figure 2.

LAG-3 and TIM-3 expression are associated with each other. (a) Scatter plot depicting the absolute H-scores of scorer 1 for LAG-3 and TIM-3, excluding those which were later classified into another group by the third observer. n(LAG-3/TIM-3) = 53. (b) Stacked bar chart depicting the association of samples classified as LAG-3^low/high in relation to the classification TIM-3^low/high. n(LAG-3/TIM-3) = 59. The association between the groups was tested using the Pearson Chi-square test. The exact p-value is reported.

Melanoma cell intrinsic LAG-3 but not TIM-3 expression is associated with melanoma stage and previous exposure to immune checkpoint inhibitors

Correlation of tumoral LAG-3 and TIM-3 expression (low vs. high) in metastases of patients with stage IV disease at the start of the respective therapy (56/60, 93.3%) with clinical data including previous exposure to immune checkpoint inhibitors, the metastasis stage, the history of brain metastases, the line of treatment, gender, age, the LDH level at the start of therapy and the BRAF V600 mutation status revealed a significant association between the expression of LAG-3 and the M1 stage of patients at the start of the respective immunotherapy (p < 0.001). Patients with a high tumoral LAG-3 expression were more frequently in higher M1 stages (Figure 3a). In line with this, there was a significant association between the brain metastasis history (indicating M1d stage) and LAG-3 expression (p = 0.002, Figure 3b). We thus also stained three melanoma brain metastases for LAG-3 and observed LAG-3 expression in all samples (Figure S4). In two cases, no immunotherapy treatment took place between the collection of the brain and the non-brain metastasis and both of these matched metastases showed pronounced LAG-3 expression (case #2, #3, Figure S4 B).

Figure 3.

Melanoma cell intrinsic LAG-3 expression is associated with clinical parameters. Stacked bar charts showing the association between LAG-3 expression of metastases from stage IV patients and M1 stage at the start of immunotherapy (a), the brain metastasis status at the start of immunotherapy (b) and a previous exposure of patients to adjuvant or non-adjuvant immune checkpoint inhibitors (c). Associations between the groups were tested using the exact Fisher-Freeman-Halton test (a) or the Pearson Chi-square test, for which the exact p-value is reported (b, c). (d, e) Forest Plots depicting the odds ratios for a high LAG-3 (d) or TIM-3 (e) expression in relation to different clinical parameters. The odds ratios were calculated using a univariate binary logistic regression analysis. Significant results are highlighted in light green. A p-value < 0.05 was considered as significant.

Interestingly, a previous adjuvant or non-adjuvant treatment with immune checkpoint inhibitors was associated with a higher tumoral LAG-3 expression (p = 0.029, Figure 3c). Based on these results we performed univariate binary logistic regression analyses for the clinical parameters and the LAG-3 group (high vs. low, Figure 3d) and again observed a significant effect for the M1d stage at the start of the respective immunotherapy (odds ratio (OR) = 17.47, 95% CI [2.07, 147.77], p = 0.009) and previous exposure to ICIs (OR = 5.39, 95% CI [1.31, 22.25], p = 0.02). We then analyzed both variables together in a binary logistic regression model and only the M1d stage remained significant (OR = 12.43, 95% CI [1.40, 110.20], p = 0.024), hinting toward the fact that tumor stage is correlated with LAG-3 expression independently of previous ICI exposure. For TIM-3 no significant associations with the tested clinical parameters were observed (Figure 3e, Figure S5).

Correlation of melanoma cell intrinsic expression of LAG-3 and TIM-3 with clinical benefit and survival

Next, we correlated the expression levels of LAG-3 and TIM-3 with clinical outcome of the patients with stage IV disease at the start of the respective therapy. In this stage IV cohort (n = 56) with a median age of 63 years (range 20–86) and 67.9% male patients, 23 patients (41.1%) received anti-PD-1 monotherapy, 18 (32.1%) anti-PD-1 plus anti-CTLA-4 therapy and 15 (26.8%) anti-CTLA-4 monotherapy. The ORR for the complete stage IV cohort was 35.7%, the DCR was 57.1% (range ORR 20.0–43.5%, range DCR 40.0–65.2%, depending on the treatment group).

Median PFS for patients with stage IV disease was 5.19 months (95% CI [2.38, 8.00]), median OS was 47.24 months (95% CI [31.78, 62.70], Figure S6A). Stage IV patients with previous exposure to ICIs did not show a significantly different PFS, but a shorter OS (median OS no exposure: 65.94 months, 95% CI [27.10, 104.79], median OS exposure: 10.68 months, 95% CI [0.00, 24.65], log-rank test: p = 0.036 (Figure S6B); hazard ratio (HR) = 2.13, 95% CI [1.03, 4.38], p = 0.041). As expected, the patients treated with anti-CTLA-4 monotherapy had the shortest PFS and OS compared to the other therapy arms (Figure S6C). However, these differences did not reach statistical significance in our small cohort.

The level of LAG-3 expression (high vs. low) did not significantly influence PFS in the complete stage IV melanoma patient cohort (p = 0.275); however, when only considering anti-PD-1 treated patients, we observed a separation of the cohort, in which patients with a high LAG-3 expression tended to have a shorter PFS (median PFS LAG-3^high: 3.32 months, 95% CI [2.42, 4.22], median PFS LAG-3^low: 17.31 months, 95% CI [0.71, 33.91], log-rank test: p = 0.088; HR = 2.20, 95% CI [0.87, 5.58], p = 0.096, Figure 4a). This trend was neither observed in the anti-CTLA-4-treated subgroup (p = 0.289) nor the anti-PD-1/anti-CTLA-4 combination therapy group (p = 0.550, Figure 4a) and did not translate into an OS benefit in our small cohort of patients (p(anti-CTLA-4 )= 0.697, p(all therapies) = 0.662, and Figure 4b). LAG-3 expression (high vs. low) of stage IV patients was not significantly associated with DCR (p = 0.787) or ORR (p = 0.781, Figure S7A).

Figure 4.

Associations of LAG-3 and TIM-3 expression with survival data. Kaplan–Meier plots depicting PFS or OS for the indicated treatment groups, stratifying the patients according to their LAG-3 (a, b) or TIM-3 (c, d) expression. The log-rank test was used to test for statistical differences between the groups. A p-value < 0.05 was considered as significant.

Correlations with clinical parameters for tumoral TIM-3 expression (high vs. low) of stage IV patients revealed no association for DCR (p = 1.0, Figure S7B) and ORR (p = 0.403, Figure S7B). Regarding PFS, we observed a trend toward shorter PFS of patients with high tumoral TIM-3 expression in the complete stage IV cohort (p = 0.101), but no statistical significance was reached in any of the treatment subgroups (p(anti-CTLA-4) = 0.185 and Figure 4c). However, stage IV patients with high tumoral TIM-3 expression demonstrated a longer OS under anti-PD-1 therapy (median OS TIM-3^high: 84.99 months, 95% CI [67.24, 102.74], median OS TIM-3^low: 18.86 months, 95% CI [1.89, 35.83], log-rank test: p = 0.007; HR = 0.20, 95% CI [0.06, 0.72], p = 0.014, Figure 4d). Again, this separation did not reach statistical significance in the complete cohort (p = 0.152), the combination therapy group (p = 0.562, Figure 4d) and the anti-CTLA-4 monotherapy group (p = 0.669).

RNA expression of LAG3 and HAVCR2 (TIM3) and their ligands in melanoma metastases

To get a more complete picture of the expression of immune checkpoints and their ligands in melanoma samples, RNA was isolated from 17 FFPE melanoma metastases blocks, which partly overlapped with the IHC cohort. The patient characteristics of this cohort are summarized in Table S2. The obtained RNA expression data covers tumor and immune cell expression, but cannot discriminate between both. With regard to the samples overlapping with the IHC cohort (10/17, 58.8%), one sample for LAG-3 and four samples for TIM-3 were classified as low/negative in terms of their tumor cell expression in IHC but showed RNA expression of LAG3 or HAVCR2 (encoding TIM-3) respectively, possibly hinting toward expression by immune cells. In general, the majority of samples showed HAVCR2 expression (12/17) and most samples also showed LAG3 expression (11/17), while 5 samples were negative for both mRNAs (Figures 5a, 6a). We also analyzed expression of the immune checkpoints PD-1 (PDCD1 gene), CTLA-4, V-domain immunoglobulin suppressor of T cell activation (VISTA, encoded by C10orf54), and T-cell immunoreceptor with immunoglobulin and ITIM domains (TIGIT). For PDCD1, CTLA4 and TIGIT, we observed a similar expression pattern as for LAG3 and HAVCR2, including positive and negative samples, while the gene encoding VISTA was expressed in all our samples (Figure 5a). As in our IHC analyses of tumor cell expression, LAG3 and HAVCR2 mRNA levels are correlated with each other (Spearman’s ρ = 0.932, p < 0.001, Figure 5b, Table S3). We additionally observed correlations between LAG3 and HAVCR2 expression levels with the expression levels of genes encoding other immune checkpoints such as PD-1, its ligand PD-L1 and VISTA. HAVCR2 was also correlated with TIGIT expression (Table S3).

Figure 5.

RNA expression analyses of immune checkpoints and LAG-3 and TIM-3 ligands in FFPE melanoma metastases. (a-d) The expression levels of selected genes in FFPE melanoma metastases were retrieved from the RNA sequencing data using edgeR.⁵² Gene expression values are presented as log2 (CPM +1) values. A pseudo count of 1 was added to the CPM values to eliminate negative values due to the log transformation. (a) Dot plot showing the expression levels of different immune checkpoint encoding genes. Lines indicate the median and the interquartile range. (b) Scatter plot of LAG3 and HAVCR2 expression values. Correlation was analyzed using the Spearman correlation test. A p-value < 0.05 was considered as significant. (c) Dot plot showing the expression values of LAG-3 ligand encoding genes. Lines indicate the median and the interquartile range. (d) Dot plot showing the expression values of TIM-3 ligand encoding genes. Lines indicate the median and the interquartile range. CPM: counts per million.

Figure 6.

LAG3 and HAVCR2 expression separates samples into two groups. (a) Dot plot showing gene expression values for LAG3 and HAVCR2, presented as log2 (CPM +1) values. A pseudo count of 1 was added to the CPM values to eliminate negative values due to the log transformation. (b) Multidimensional scaling (MDS) plot created using edgeR⁵² where distance between each pair of samples corresponds to the leading log2 fold change between the samples. Leading log2 fold change is the root-mean-square average of the log2 fold changes of the genes best distinguishing the samples. The MDS plot can be viewed as an unsupervised clustering of the samples based on the log fold changes.⁶¹ Percentages indicate the amount of variance explained by the respective dimension. (c) Heat map showing immune cell signatures for the two defined groups of samples (LAG3/HAVCR2 single/double positive vs. LAG3/HAVCR2 double negative). The abundances of the different immune and stromal cell types were inferred using MCP-counter.⁵³ CPM: counts per million.

In a next step, we analyzed the mRNA expression of LAG-3 and TIM-3 ligands in our samples (Figure 5c,d). In line with our results from analyzing tumor cells by immunohistochemistry, MHC class II gene expression was commonly detected. Among the non-MHC class II LAG-3 ligands, SNCA (encoding α-synuclein) and LGALS3 (encoding galectin-3) showed the highest median expression, whereas FGL1 and CLEC4G (encoding LSECtin) mRNA was rarely detected. We observed expression of all analyzed TIM-3 ligands. LAG3 expression significantly correlated with several MHC class II mRNA expression levels (Table S4, Figure S8).

LAG3 and HAVCR2 RNA expression correlates with immune cell profiles

As RNA from FFPE tissue samples includes RNA from tumor cells and all other cells from the tumor microenvironment including immune cells, we correlated the LAG3 and HAVCR2 expression with different immune cell profiles. LAG3 and HAVCR2 mRNA expression significantly correlated with several marker genes (retrieved from PanglaoDB⁶²) of B-cells and T-cells from the helper, cytotoxic and regulatory phenotype as well as macrophages and NK cells (Table S5). Moreover, we correlated LAG3 and HAVCR2 expression with the abundance of different immune and stromal cell populations using the microenvironment cell population (MCP) counter algorithm⁵³ and with the fraction of immune cell types determined using the quanTIseq algorithm.⁵⁵ With both methods, we observed a statistically significant positive correlation between HAVCR2 and CD8⁺ T-cells and B-cells, and between LAG3 and CD8⁺ T-cells (Table S6). LAG3 and HAVCR2 mRNA expression levels also positively correlated with expression of the majority of the 18 genes of the tumor inflammation signature⁶³ and several genes of the IFN-γ signature^64,65(Tables S7, S8).

For further analyses, we split the samples into two groups: one group, in which LAG3 and/or HAVCR2 were expressed, and another group in which neither LAG3 nor HAVCR2 were expressed (Figure 6a). A multidimensional scaling plot of all samples supported this grouping (Figure 6b). Only one sample was misplaced in another cluster. We could identify a series of differentially expressed genes between the two groups (Figure S9). Moreover, an analysis of different immune and stromal cell populations using MCPcounter⁵³ revealed a differential distribution of cell types between the two groups (Figure 6c). However, clinical outcome as DCR, ORR, PFS, and OS was not significantly different between both groups, though the limited number of samples has to be considered (Figure S10).

Discussion

In this study, we describe common LAG-3 expression in tumor cells of melanoma metastases. We are not aware of any study that systematically investigated and detected melanoma cell intrinsic LAG-3 expression in melanoma metastases at a large scale. Our results are in line with a study demonstrating expression of LAG3 mRNA in A375 melanoma cells.³⁴ Other studies did not observe melanoma cell intrinsic LAG-3 expression, but they only analyzed a small number of patients and/or used a different antibody than we did in our study.^66,67 Expression of LAG-3 by other malignant cell types has been described, e.g. for clear cell renal cell carcinoma,³³ non-small cell lung cancer,⁶⁸ diffuse large B-cell lymphoma³¹ and chronic lymphocytic leukemia.³² However, the malignant cells in the latter cancer entities are B-cells, which are reported to partly also express LAG-3.^31,69 One might initially expect to mainly observe a membrane staining for LAG-3 in melanoma cells, however intracellular staining has been described in malignant cells.^31,32,68 It might possibly arise from intracellular storage of LAG-3, as it was described for CD4⁺ T-cells.⁷⁰

We observed a melanoma cell intrinsic expression of the LAG-3 ligands FGL-1 and LSECtin only in single cases in our cohort. Differing results have been reported for the expression of LSECtin in melanoma cells; however, these studies differed in their used methodologies, cohort sizes and composition.^12,71 FGL-1 expression by tumor cells has been reported for other cancer entities, though also using different antibodies than in our study.^15,72 Our observation of melanoma cell intrinsic α-synuclein, galectin-3, and MHC class II expression is in line with the literature.^58,73,74 We did not observe a correlation of LAG-3 expression with one of its ligands when analyzing tumor cells in IHC. However, we observed a correlation of the mRNA expression of LAG3 with several MHC class II encoding genes in our cohort analyzing RNA from FFPE tissue samples, which is supported by a study in uveal melanoma, which in addition observed a correlation between LAG3 and the genes encoding LSECtin and galectin-3.⁷⁵

Little is known so far about the relevance of tumor cell intrinsic LAG-3 expression in melanoma. We here describe a trend toward an association of high LAG-3 expression in melanoma cells and a shorter PFS under anti-PD-1 treatment in our stage IV melanoma cohort, which we did not observe in the anti-PD-1/anti-CTLA-4 combination treatment group. This observation is generally in line with a study on soluble LAG-3 (sLAG-3) by our group. There we reported higher pre-treatment levels of sLAG-3 in patients with disease progression under anti-PD-1 treatment compared to patients with disease control, which we did not observe in the anti-PD-1/anti-CTLA-4 combination treatment group. The PFS was also significantly shorter in anti-PD-1 treated patients with pre-treatment sLAG-3 levels above the defined cutoff.⁷⁶ A connection between these two studies could be made if the sLAG-3 found in the serum of melanoma patients at least partly originates from tumor cells. Fröhlich et al. reported longer PFS under immune checkpoint blockade for melanoma patients whose analyzed samples showed a low methylation of the LAG3 gene at two tested loci in the promotor (associated with higher LAG3 mRNA expression).³⁴ However, this analysis was based on tissue samples, thus it did not only include tumor cells as our IHC study.

Our observations of the association of tumor cell intrinsic LAG-3 expression with PFS were not reflected in the OS probability in our cohort. Fröhlich et al. reported longer OS for a melanoma patient group with high LAG3 mRNA expression, however, as above, this analysis was not limited to tumor cells and the cohort contained primary and metastatic samples,³⁴ while our cohort for survival analyses only contained metastatic samples from patients with stage IV disease at the start of the respective treatment. Klümper et al. reported shorter OS in clear cell renal cell carcinoma patients with tumor cell intrinsic LAG-3 expression,³³ supporting an unfavorable effect of high tumor cell intrinsic LAG-3 expression. The effect of LAG-3 expression in tumor cells might also differ between tumor entities and tumor stages.

The RELATIVITY-020 trial investigated the effect of nivolumab and relatlimab in patients who progressed under anti-PD-1 or anti-PD-L1 containing therapies. The absolute values for ORR, DCR, median PFS and median OS showed a trend toward a more favorable performance in patients with high LAG-3 expression (cutoff ≥1% vs < 1%).⁷⁷ It would be interesting to perform similar analyses as in our study on a cohort which underwent anti-PD-1/anti-LAG-3 combination treatment as in the RELATIVITY-020 trial. This would allow to investigate whether subgroup analyses show a benefit for patients with high tumor-cell intrinsic LAG-3 expression, as anti-LAG-3 treatment in LAG-3^high patients might affect both, immune cells and tumor cells.

Our report of TIM-3 protein expression by melanoma cells is supported by the literature. Wiener et al. described the expression of TIM-3 in melanoma cell lines³⁷ and Schatton et al. reported the detection of HAVCR2 expression by melanoma cells via single-cell RNA sequencing in all analyzed patient samples (n = 14). Moreover, they immunofluorescently analyzed SOX-10⁺ (marking melanocytic lineage) cells and reported TIM-3 expression by a fraction of SOX-10⁺ cells in 7 out of 15 samples.³⁸ Cazzatto et al. also described preliminary immunohistochemistry results showing a cytoplasmic TIM-3 signal in melanoma cells.³⁹ These and our results are in disagreement with the study by Wang et al., which did not detect melanoma cell intrinsic TIM-3 expression via immunohistochemistry in extra- and intracranial melanoma samples of nine patients; however, the cohort size in this study was small and they used a different antibody than we did in our study.⁶⁶

All analyzed TIM-3 ligands were expressed by melanoma cells in our cohort, which is in line with the literature.^71,78–80 We did not detect a significant correlation between the expression of TIM-3 and any of its ligands in melanoma cells. Using TCGA data, Schatton et al. reported a positive correlation between HAVCR2 and LGALS9 mRNA levels in melanoma.³⁸ We observed a similar trend in our cohort of RNA from FFPE tissue samples; however, it did not reach statistical significance.

Schatton et al. describe a growth-suppressive effect of TIM-3 on melanoma cells³⁸ which is in principle in line with our observation of a shorter OS of patients with TIM-3^low samples. However, this difference only reached statistical significance in the anti-PD-1 treated subgroup. Other studies on tumor cell intrinsic TIM-3 expression in different solid cancer entities reported an association between higher TIM-3 expression in cancer cells and an unfavorable OS³⁵ or an unfavorable PFS but no significant effect on OS.³⁶ It might be that the consequences of TIM-3 expression in tumor cells differ between cancer entities.

Increased LAG-3 or TIM-3 expression by tumor cells might act via ligand binding and/or via independent additional roles in melanoma cells. It is important to elucidate the functional role of LAG-3 and TIM-3 expression in melanoma cells, as expression in these cells might have different effects than expression on immune cells. Studies addressing this question could also clarify whether the observed differences in survival are mediated by tumor cell intrinsic pathways or whether they are mainly influenced by other factors as a more common high LAG-3 expression in higher M1 stages. However, in studies analyzing RNA from melanoma metastases as a whole, the tumor cell intrinsic expression cannot be distinguished from expression by immune cells. We observed a correlation of LAG3 and HAVCR2 expression with immune cell marker genes and immune cell fractions/abundances in our transcriptomic analyses. This might indicate that the majority of the detected LAG3 and HAVCR2 transcripts in whole-metastasis analyses originates from immune cells or that immune cells are attracted to metastases with tumoral LAG3 or HAVCR2 expression. This question could be addressed via single cell RNA sequencing in future studies. Moreover, data from spatial imaging analyses might be combined with RNA sequencing data to further investigate associations between the detection of LAG3 and HAVCR2 expression and potentially increased immune infiltration in tumor samples. However, a predominant expression of LAG3 by immune cells is supported by the study of Fröhlich et al. reporting a significant correlation between LAG3 mRNA and the leukocyte fraction, but an inverse correlation with tumor purity of melanoma samples.³⁴ Moreover, this study reports a correlation of LAG3 expression with an IFN-γ signature consisting of five genes,³⁴ supporting our results.

Tumor cell intrinsic expression of PD-1 has been reported for several cancers, including melanoma,^26–28,81 though also negative results have been reported using the same antibody clone as we did.⁶⁶ We observed only few melanoma metastases with a weak PD-1 staining and therefore did not further correlate PD-1 expression with other markers. Differing results between studies might be based on different methods used and variability of the patient cohort composition.

Our study included two samples in the IHC cohort and one sample in the FFPE RNA cohort collected shortly after the start of the respective analyzed immunotherapy (1 and 6 days after or 6 days after the day of therapy start respectively). Some studies reported early changes in immune signatures or immune checkpoint expression after the start of an immunotherapy, however, the median time difference to the start of the therapy was longer in these studies.^82,83 As we aimed to achieve the maximum sample size possible and as the time difference was very small, we included these two samples in the cohort.

We are aware of the limitations of our study regarding sample size. A larger validation cohort would be desirable to verify the observed associations with OS and PFS. The estimations of hazard ratios for survival data provide an orientation, but have to be considered with care as survival curves partly cross and the sample size is limited. The results should be validated in another cohort of patients receiving anti-PD-1 monotherapy or even a combination with an anti-LAG-3 antibody. Moreover, tumor cell intrinsic LAG-3 expression has been reported in other cancer entities, such as clear cell renal cell carcinoma³³ and hence, to investigate theses effects in patients with other tumor entities receiving immunotherapies would be of interest. This could address the question whether tumor cell intrinsic immune checkpoint expression has likely generic or entity-specific effects. Given for example reports of pro- and anti-tumorigenic effects of tumor cell intrinsic PD-1 expression in different cancer entities,^{27,28,81,84,85} cancer entity-specific effects of immune checkpoint receptor expression seem likely. However, different standard treatments in different cancer entities might complicate the comparison of OS/PFS between cancer types in case immune checkpoint expression might interact with therapies. Moreover, melanoma is considered an immunogenic tumor (reviewed in⁸⁶), and associations of immune checkpoint receptor expression with clinical factors might also vary depending on the immunogenicity of the studied tumor models. Finally, it is highly desirable to obtain data on the function of LAG-3 expression by melanoma cells or other cancer cell types. Respective experiments in melanoma cells are ongoing.

Despite the discussed limitations, our study provides an important systematic analysis of the tumor cell intrinsic expression of LAG-3, TIM-3 and their ligands in melanoma metastases in combination with clinical factors and points toward a possible prognostic value of melanoma cell intrinsic LAG-3 expression for anti-PD-1 therapy resistance.

Funding Statement

This project was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project number 259332240 / RTG 2099 and a scientific grant by Bristol-Myers Squibb (CA209-7A6).

Disclosure statement

JCH has received honoraria from BMS, MSD, Novartis, Roche, Pierre Fabre, Sanofi, Almirall; consultant or advisory role: MSD, Pierre Fabre, Sunpharma; research funding: BMS; travel support: Pierre Fabre. NH has received a grant from Bristol-Myers Squibb and payment/honoraria from Merck Darmstadt KG and Roche. NP has personally received consulting fees from Glaxo Smith Kline and AstraZeneka UK, payment or honoraria from AstraZeneca, Illumina, PGDx, Bayer, Novartis and MSD and personally participated on a monitoring/advisory board for Janssen. This work was supported by a scientific grant by Bristol-Myers Squibb (see acknowledgements). MW, DM, SC, EMM, BL, RE and JR declare no additional conflicts of interest.

Data availability statement

The data supporting the article’s findings is available within the article and its supplementary materials. Further data can be provided upon reasonable request.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/2162402X.2024.2430066