Neoantigen Targetability in Progressive Advanced Melanoma

Immunotherapy with immune checkpoint inhibitors (ICI), aiming to reinvigorate T-cell immunity, has revolutionized the treatment and prognosis of patients with melanoma (1–3). The combination of nivolumab and ipilimumab resulted in a 5-year overall survival rate of approximately 50% for patients with advanced melanoma (2, 4). Another approach is adoptive transfer (ACT) of ex vivo–expanded tumor-infiltrating lymphocytes (TIL) or mixed lymphocyte tumor culture (MLTC) products. A phase II study showed a 36% objective response rate in heavily pre-treated patients (5), whereas a recent phase III randomized study revealed a response rate of 49% and a median progression-free survival of 7.2 months after TIL therapy (6). However, a deeper understanding of metastatic melanoma is required to optimize treatment strategies and achieve long-term outcomes.

In-depth analyses of the mechanisms underlying treatment responses in melanoma and other cancers have highlighted the crucial role of neoantigen-specific CD4⁺ and CD8⁺ T cells in driving antitumor immunity, along with the potential contribution of tumor-associated antigen (TAA)-reactive T cells (7–9). The presentation of (neo)antigens on tumor cells plays a significant role in determining the therapeutic efficacy of ICI, adoptive T-cell transfer, or (neo)antigen-based vaccination (10–13). However, primary and secondary immune escape mechanisms can hinder the success of these therapies, for example, by upregulation of immune checkpoint molecules, by immunosuppressive cell infiltration of the tumor, or by defects in IFNγ signalling (14–19). In particular, the loss of immunogenic antigens or defects in the antigen processing machinery can preclude the presentation of HLA class I/peptide complexes on the cell membrane, leading to failure of cytotoxic T cells to recognize and eliminate cancer cells (20). Consequently, the loss of targetable antigens poses a serious threat to the success of T-cell–based immunotherapies.

To address these concerns, we conducted a study using a unique cohort of patients with melanoma from whom sequential early and later metastases were isolated. Our aim was to investigate whether disease progression in advanced melanoma is associated with a reduced availability of potentially targetable neoantigens and TAAs and to determine whether continued immunotherapeutic treatment remains a viable option.

The study was approved by the Medical Ethical Committee of the Leiden University Medical Centre (protocol P04085 and L18039) and all patients provided written informed consent before participation. Patient samples and information were anonymized and this research was conducted according to the recommendations outlined in the Helsinki Declaration.

The studied patients with melanoma were included in clinical trials in the Leiden University Medical Center. Eligible patients were 18 years or older and had histologically proven stage IV or irresectable stage III cutaneous melanoma. Patients had progressive disease before treatment, with a WHO performance status 0 to 2 and a life expectancy of at least six months. On at least two different time points, lesions were obtained for the establishment of a tumor cell line and if enough material was available, also for TIL cultures. In addition, heparinized venous blood was collected before treatment for isolation of peripheral blood mononuclear cells (PBMC) by Ficoll gradient centrifugation. The isolated PBMCs were cryopreserved until further analysis. Patients were included in this study if enough patient material was available for detailed immunological characterization [including whole-exome and RNA sequencing (RNA-seq), mass spectrometry, flow cytometry, and spatial phenotyping].

Part of the obtained tumor tissue was used for standard pathological evaluation and, therefore, formalin-fixed, paraffin-embedded (FFPE). Another part of the tumor was used to establish melanoma cell lines in either the good manufacturing practices facility or the Laboratory of Medical Oncology (LUMC, Leiden, the Netherlands), as described previously (21). All melanoma cell lines were cultured in DMEM (Life Technologies), supplemented with 8% heat inactivated FCS, penicillin (100 IU/mL), streptomycin (100 μg/mL), and l-glutamine (4 mmol/L), all from Life Technologies. Short tandem repeat profiling was performed to authenticate the cell lines and they were bi-weekly checked for Mycoplasma contamination by PCR-based assays. The later metastatic (LM) melanoma culture of Mel7 was found to be Mycoplasma infected and could, therefore, not be taken along for downstream analyses. The tumor cell lines were used in assays within 2–4 weeks after thawing. If enough tumor material was available, TIL cultures were initiated in parallel as described previously (21). The procedures to obtain TIL and MLTC T-cell products are described in the Supplementary Methods.

Four-μm FFPE tissue sections were cut and placed on glass slides for immunohistochemical or immunofluorescence detection. Xylene deparaffinization and rehydration preceded the endogenous peroxidase block with 0.3% hydrogen peroxide in methanol for 20 minutes. Heat-induced antigen retrieval was performed for 10 minutes in preheated pH6.0 citrate buffer (10 mmol/L) for HCA2, HC10, HLA-DR, PD-L1, and TAP2 and pH9.0 Tris-EDTA buffer for β2m, PD-L2, and TAP1. The primary antibodies were diluted in PBS with 1% BSA and incubated overnight on the tissue slides with the following dilutions: mAb HCA2 (1:1,000, Nordic MUbio), mAb HC10 (1:3,200, Nordic MUbio), mAb β2m (1:4,000, clone EP2978Y, Abcam), mAb HLA-DR (1:200, clone TAL 1B5, Invitrogen), mAb PD-L1 (1:200, clone E1L3N, Cell Signaling Technology), mAb PD-L2 (1:200, clone D7U8C, Cell Signaling Technology), mAb TAP2 (1:50, clone TAP2.17, BD Biosciences), and TAP1-specific mAb NOB1 (1:10, kindly provided by Dr. Soldano Ferrone, Park Cancer Institute, Buffalo, NY). Detection with poly-horseradish peroxidase solution (Immunologic) was followed by signal development using DAB+chromogen (Dako) for 5 minutes. The slides were counterstained with hematoxylin for 30 seconds and dehydrated by increasing concentrations of alcohol solutions followed by xylene. Slides were mounted using Surgipath Micromount (Leica Microsystems Inc.). PD-L1 antibody expression was scored in every tumor section as the percentage of positive tumor cells and these scores were converted to negative (0%), intermediate (1%–5%), or positive (≥5%) according to common practice in the field. All other antibodies were scored against the staining intensity of the internal control—provided by stromal and immune cells—as positive, negative, weak, or heterogeneous (i.e., positive and negative scoring tumor areas; ref. 22; Supplementary Fig. S1). Classical HLA class I expression was evaluated using HCA2 (mainly anti–HLA-A) and HC10 (particularly anti–HLA-B/C) antibodies and loss of HLA class I was defined by the lack of staining with both antibodies (22, 23).

T-cell infiltrate was analyzed by a multi-color fluorescent staining of CD4, CD8, PD-1, and DAPI following the method published previously (24). A detailed description of the procedure is available in the Supplementary Methods.

To deal with limited tissue availability, cancer cell lines were established from the tumor material. The cell lines were cultured and harvested at approximately 70% confluency for DNA and RNA isolation with NucleoSpin Tissue kit (Macherey-Nagel) and NucleoSpin Tissue RNA kit (Macherey-Nagel), respectively, according to the manufacturer's instructions. Similarly, DNA was isolated from patients’ PBMCs as paired normal healthy tissue. Sequencing libraries were prepared from the isolated genomic DNA according to the manufacturer's instructions using the NEBNEXT Ultra II DBA Library Prep kit for Illumina (New England Biolabs) and the IDT xGEN Exome Target kit (Integrated DNA Technologies). RNA-seq libraries for 87.5% of the samples were generated using the NEBNext Ultra Directional RNA Library Prep kit for Illumina (New England Biolabs) following the manufacturer's instructions. The NEBNext rRNA Depletion kit (New England Biolabs) was used for rRNA depletion from the total RNA. The obtained paired-end, 150bp libraries were sequenced as listed in Supplementary Table S1.

Reads generated by RNA-seq were mapped against the same hg38 genome build using STAR (version 2.7.5a; RRID:SCR_004463). For exome sequencing, reads were mapped against the human reference genome (hg38) using BWA-MEM (RRID:SCR_010910). Picard Tools were used to remove duplicate reads (RRID:SCR_006525). Subsequently, variant calling was done using a combination of three software tools, muTect 2 (RRID:SCR_000559), varDict (RRID:SCR_023658), and Strelka (RRID:SCR_005109). The three variant sets generated were then combined into a single vcf file using GATK CombineVariants (RRID:SCR_001876). RNA-seq read counts for each variable allele were added to the identified variant chromosomal positions using the bam-readcount tool (RRID:SCR_023653). HLA typing from RNA-seq and exome sequencing data was done using Optitype (RRID:SCR_022279).

Variants were functionally annotated using the ensembl Variant Effect Predictor (VEP; RRID:SCR_007931). Variant calls annotated as protein-damaging were further investigated if at least one read with the alternative allele was present in the RNA-seq data. The selected variant calls were then visually inspected using Integrative Genome Viewer (IGV, Broad Institute) to exclude false positives (25–27). After inspection of each lesion independently, the early metastatic (EM) and LM melanomas were paired and re-inspected for variants that were only called in one of both. For all the variants considered to be true after visual inspection, 25 amino-acid (aa) sequences around the variant allele were extracted from the tumor RNA-seq data using the isovar tool (28). Peptides of variable sizes (8–12 mer peptides) were then extracted from the 51 aa peptide sequences, with the mutant amino acid placed at all possible positions. HLA-binding affinity predictions were performed for all the different peptides generated using NetMHC4.0 and NetMHCpan 4.0 (RRID:SCR_021651, RRID:SCR_018182). Short peptides that ranked as top 0.5% were considered strong binders and selected for further analysis. In addition, a restrictive list of TAA was composed, which only included genes that have been reported with a restrictive expression pattern in cancer-testis or melanocytic tissue (Supplementary Table S4).

The tumor-recognition by MLTC cultures or TIL (obtained as described in Supplementary Methods) was tested against autologous EM- and LM-derived cell lines that were either stimulated for 24–48 hours with IFNγ (100 IU/mL, Preprotech) to upregulate HLA expression, or left untreated. Tumor cells were harvested and washed to remove IFNγ used for stimulation, before the test. Briefly, a total of 1.5×10⁴ effector T cells (MLTC or TIL) were co-cultured with 3×10⁴ tumor cells in a total volume of 150 μL IMDM supplemented with 8% heat inactivated FCS, penicillin (100 IU/mL), streptomycin (100 μg/mL), and l-glutamine (4 mmol/L), all from Life Technologies. The test was performed in triplicate wells of a U-bottom 96-well plate. Medium alone was used as negative control and Dynabead CD3/CD28 human T activator beads (Thermo Fisher Scientific) or Staphylococcal Enterotoxin B (0.5 μg/mL) were used as positive control. After overnight incubation at 37°C the supernatant was harvested to determine the IFNγ or CCL4 secretion as a measure of recognition by ELISA (Mabtec and R&D Systems, respectively) according to the manufacturer's recommendations.

Tumor cell lines were stimulated for 24–48 hours with IFNγ (100 IU/mL) or control medium before harvesting. Single-cell suspensions were obtained using trypsin/EDTA (Gibco) and analyzed by multicolor flow cytometry. Cells were washed twice with FACS buffer [PBS (Braun) supplemented with 0.5% human serum albumin (Albuman)] and stained with cell surface antibodies specific for pan-HLA class I (HLA-ABC, MCA81A–AlexaFluor-647, Biolegend) and HLA class II (HLA-DP/DQ/DR, Bu26-FITC, Bioconnect) for 30 minutes on ice in the dark. Then the samples were washed twice with FACS buffer, followed by fixation in 1% paraformaldehyde (provided by the LUMC pharmacy). The fixed cells were measured with a flow cytometer (LSR Fortessa, BD Biosciences) and analysed with FlowJo software (LLC, version 10.7.1; RRID:SCR_008520).

Tumor cell lines with HLA class I expression on the cell surface and a decent in vitro growth rate were cultured as described above and during the last passage grown on 14-cm petri dishes to enable easy collection of the lysate. Approximately a total of 1×10⁹ cells were used for HLA class I peptide elution. Details on the mass spectrometry procedure and analysis are available in the Supplementary Methods. Table 2 lists the resulting, eluted neoantigen-derived peptides and Supplementary Table S5 lists the eluted TAA-derived peptides.

Ten 4-μm tissue slides were cut from FFPE sections and macrodissected to obtain tumor tissue based on sequential hematoxylin and eosin–stained sections, if required. In addition, to cover the spatial heterogeneity of the tumor, three punches were taken from the FFPE tissue and added to the sections. DNA isolation of this FFPE material was performed using the VERSANT Tissue Preparation Reagents and corresponding robot system (Siemens Healthcare GmbH, Erlangen, Germany). At least 100,000 cells were used as input for the DNA isolation with the Nucleospin Tissue kit (Macherey-Nagel) of cultured T-cell products. Library preparation was performed in house using the Ion AmpliSeq Library Kit Plus and the Oncomine TCRb short read assay for DNA according to the manufacturer's instructions (Thermo Fisher Scientific). The prepared libraries were sent to GenomeScan (Leiden, the Netherlands) for sequencing using Ion 530 chips on the Ion Genestudio S5 sequencer (Thermo Fisher Scientific). Analysis of the T-cell receptor (TCR) clones was performed with the IonReporter 5.14 software using the Oncomine TCRb-SR – w1.2 – DNA – Single Sample workflow (version 5.12) with default settings.

The human sequence data generated in this study are not publicly available due to patient privacy requirements but are available upon reasonable request from the corresponding author. Other data generated in this study are available within the article and its Supplementary Data Files.

We studied 8 patients with melanoma from whom we obtained resectable sequential metastatic lesions, referred to as EM and LM lesions, along with corresponding cell lines. In addition, we obtained MLTC T cells or TIL cultures from each lesion (Fig. 1). The EM and LM lesions were surgically resected between 2 and 103 months apart and were located in various organs (Supplementary Table S2). Following the surgical resection, the patients underwent various treatments, often involving multiple lines of therapy such as chemotherapy, radiotherapy, targeted kinase inhibitors, and immunotherapy, including checkpoint inhibition and ACT (Fig. 1). We have collected follow-up data for these patients until death or up to the time of writing this article.

We investigated the expression of HLA class I across subsequentially resected melanoma lesions by performing immunohistochemical detection of HLA class I on tumor tissue sections. In most melanomas (from 6 out of 8 patients) HLA class I was expressed (Table 1). Complete loss of HLA class I expression, as observed through negative immunodetection of both HLA-A allele (HCA2) and HLA-B/C-alleles (HC10) specific antibodies, was only observed in the LM of patient Mel5. Exome sequencing revealed a mutation (c.43_44delCT, p.L15fs43) in the B2M gene, likely responsible for impairing HLA class I complex formation and surface expression (29). Patient Mel1’s LM displayed a heterogeneous pattern of HLA class I expression, as detected through IHC, and specifically, the loss of HLA-B*15.01 was observed in the LM of Mel4 (identified through transcriptome sequencing; Table 1; Supplementary Table S3). Patient Mel6’s LM showed partial loss of HLA class I expression (negative immunodetection using the HCA2 antibody clone), along with loss of heterozygosity of the HLA class I genes (detected through both exome and transcriptome sequencing; Supplementary Table S3). Notably, the expression of transporter-associated with antigen-processing proteins 1 and 2 (TAP1 and TAP2), which play a critical role in antigen processing, was not detected in LM of Mel6, partly retained in LM of Mel1, and unaffected in the other 6 out of 8 LM lesions, indicating that the transport of antigens to the endoplasmatic reticulum for loading onto HLA class I molecules is still functional in the majority of lesions.

The HLA class I status of the melanoma samples was compared with the surface expression of these molecules on their corresponding cell lines using flow cytometry (Supplementary Fig. S2). With the exception of the LM of Mel5, all cell lines expressed HLA class I. Stimulation with IFNγ resulted in upregulated HLA class I or II expression in most of the cell lines, indicating that IFNγ unresponsiveness was not used as an evasion mechanism in these cells. The absence of HLA class I expression in the cell line derived from the LM of Mel5 was consistent with the lack of immunohistochemical detection of HLA class I and the presence of the B2M mutation indentified in this tumor sample. The cell line derived from the LM of Mel6 expressed HLA class I almost at the similar level as the cell line from the EM, despite partial loss of HLA and loss of TAP. Although the use of a pan-HLA class I antibody in flow cytometry may not provide enough distinction with respect to partial loss, our data may also suggest that either this cell line processes and presents TAP-independent epitopes (30) or that TAP-positive cells were still present in the LM of this patient and expanded during the establishment of the cell line.

The expression of HLA class II by melanoma cells plays a role in the recognition of TAAs or neoantigens by CD4⁺ T cells, which can contribute to clinical responses (31, 32). HLA-DR expression was detected in 5 out of 8 patients, and the levels of expression varied between EM and LM lesions. Specifically, HLA-DR expression as measured by IHC was lower in the LM of Mel1, 4, 5 and 8, whereas it was specifically increased in the LM of Mel3 (Table 1). Flow cytometric analysis of HLA-DR, -DP and -DQ expression, which is generally more sensitive than IHC, revealed the expression of HLA class II in EM and LM of Mel5, 6, and 8 as well as demonstrated that these alleles could be induced by IFNγ stimulation (Supplementary Fig. S2), suggesting the potential for CD4 T-cell–mediated recognition of tumors.

Cancer cells can acquire the expression of immune checkpoint molecules like programmed death ligand-1 (PD-L1) or PD-L2, which can limit T-cell–mediated immune responses against cancer cells (33, 34). Among the lesions of the eight patients studied, PD-L1 expression was observed in six patients, whereas neither lesion of patients Mel1 and Mel7 showed detectable PD-L1 expression. Interestingly, PD-L1 expression in melanoma cells was specifically increased in the LM lesions of three patients (Mel2, Mel3, and Mel5), whereas their EM lesions exhibited intermediate levels of PD-L1 expression (i.e., a maximum of 5% PD-L1⁺ tumor cells; Table 1). No changes in PD-L1 expression status were observed between the EM and LM in the remaining patients. PD-L2 expression was detected in three out of eight EM lesions and in two corresponding LM lesions, but was no longer detectable in the LM lesion of Mel2. PD-L2 expression was present in the later lesions of another three patients whereas their earlier lesions did not express PD-L2. In two patients PD-L2 was not detected (Table 1).

In summary, our findings revealed variations in the expression of HLA molecules and their associated antigen-processing components, and in the expression of inhibitory molecules between EM and LM in five out of eight patients. Although these changes may impact specific T-cell–based immunotherapeutic approaches and require careful consideration, our data indicate that, in theory, susceptibility to T-cell–mediated recognition of melanoma cells was preserved in seven out of eight patients' LMs.

T-cell infiltration was assessed using a multiplex immunofluorescence panel detecting CD4⁺ T cells, CD8⁺ T cells and programmed cell death protein-1 (PD-1) expression (Fig. 2A). Quantitative analysis showed comparable densities of infiltrating CD4⁺ and CD8⁺ T cells between EM and LM from the same patients, indicating overall similar immune infiltration levels (Fig. 2B and C). In patients Mel2, Mel5, and Mel8, higher numbers of CD4⁺ T cells were observed in LM lesions, whereas the LM lesions from Mel1, Mel6, and Mel7 showed lower numbers of CD4⁺ T cells as compared with EM lesions. However, these changes did not correlate with variations in HLA class I, HLA-DR, or PD-L1/2 expression. PD-1 expression was detected on a median of 11.4% (1%–22%) of CD4⁺ T cells and 36.9% (0%–78%) of CD8⁺ T cells. Fluctuations in the proportion of PD-1–positive CD4⁺ and CD8⁺ T-cell populations were observed but were not specifically associated with EM or LM lesions (Fig. 2D and E). In the LM of patients Mel1 and Mel6, the number of PD-1–positive CD8⁺ T cells was drastically lower, likely reflecting less CD8⁺ T-cell activation in the tumor microenvironment, which could be linked to the corresponding decreased expression of HLA class I, particularly HLA-A (Table 1). In general, the proportion of infiltrated CD4⁺ as well as CD8⁺ T cells was higher in lesions with higher HLA class I and PD-L1 expression, possibly reflecting the inflammation signatures reported previously (35, 36), but there were no overt differences between EM and LM lesions on a per patient basis (Fig. 2F–I).

In one patient (Mel2), the density of CD8⁺ T cells in the LM lesion was nearly 10 times higher compared with the EM (Fig. 2C). This patient showed disease stabilization after receiving ACT with MLTC-derived T-cell products obtained through repetitive stimulation of autologous PBMCs with the EM lesion-derived cell line. The ACT product, enriched for CD8⁺ T cells (70%), demonstrated tumor recognition exclusively in the CD8⁺ fraction (21). TC profiling was performed on the ACT product obtained from the EM from Mel2 and of DNA isolated from FFPE tumor tissue sections of the EM and LM. Strikingly, the frequency of shared clones between the ACT product and the tumor lesions increased from 9% in the EM to 45% in the LM; that is, 45% of the infiltrating T-cell clones in the LM consists of clones that are shared with the ACT product (Fig. 2J). This finding suggests that the increase in CD8⁺ T cells in the LM of Mel2 was a result of the ACT treatment and correlates with the observed disease stabilization in this patient. Similar observations of persisting clonotypes recognizing a limited number of neoantigens has been reported in patients with melanoma responding to ICIs (37).

Recognition of tumor cells by (expanded) TIL or MLTC cultures was assessed (Supplementary Table S2) and detected to cell lines from nearly all evaluable lesions by T cells through the specific production of IFNγ (or CC-chemokine CCL4) upon recognition of their cognate epitope presented by HLA at the tumor cell surface. The exception was the T-cell batch expanded from the LM of Mel8, which did not show significant recognition of the autologous EM and/or LM cell lines (Fig. 3). Low T-cell recognition was observed against the HLA class I–defective cell line derived from the LM of Mel5 (Supplementary Fig. S3). In line with the HLA class I defect, recognition was mediated by CD4⁺ T cells/HLA class II interaction, instead of CD8⁺ T cells (Supplementary Figs. S2 and S4). Despite the presence of TAP deficiency in the lesion, the expanded TIL product of Mel6 was able to recognize its corresponding cell lines, suggesting potential recognition of TAP-independent antigens (30), or the outgrowth of cells from the tumor that retained TAP expression. Remarkably, the T cells expanded from the TIL of the LM in Mel2 exhibited better recognition of both autologous EM and LM cell lines compared with the EM-derived MLTC culture. This finding supports the hypothesis that ACT treatment had a positive impact on the influx of tumor-specific CD8⁺ T cells in this patient's LM lesion (Fig. 2C).

In conclusion, both EM and LM lesions are recognized by tumor-reactive T cells and thus remain targets for the immune system, albeit that the type of T-cell response may vary and changes in the epitope repertoire can occur.

Whole-exome sequencing was performed on DNA extracted from EM and LM cell lines as well as patients’ PBMCs, to determine the somatic mutation landscape. The expression status of the mutations was determined by RNA-seq of the cell lines. On average, we detected 216 (107 to 360) non-synonymous, transcribed mutations, and the number of mutations was stable between EM and LM derived from the same patient (Fig. 4A). The majority of mutations, 79% (55%–93%), were shared at the DNA level between EM and LM. Among these shared mutations, an average of 73% (65%–81%) were also found to be transcribed in both lesions (Fig. 4B; Supplementary Fig. S5A).

The percentage of lesion-specific mutations at either DNA or RNA level accounted for approximately 20% of the total number of mutations for a given lesion (Fig. 4B; Supplementary Fig. S5A). We also investigated whether there was evidence of immunoediting between EM and LM, specifically focusing on mutations that could generate epitopes with high affinity for patient-specific HLA alleles. We determined the number of such mutated epitopes, referred to as “strong binders,” which are most likely to be presented at the cell surface and recognized by T cells (38). The number of strong binders was constant between lesions (Fig. 4C; Supplementary Fig. S5B), suggesting that if they were subjected to immune pressure, this does not eliminate all potential T-cell epitopes. In addition, differences in mutations between lesions may also be the result from clonal heterogeneity across metastatic lesions (39).

Furthermore, we evaluated the expression of 45 melanoma TAAs (Supplementary Table S4), which can also contribute to the antitumor response. On average, the cell lines expressed 19 (8–39) different TAAs, and there was no specific indication of TAA loss in LM compared with EM (Supplementary Fig. S6).

Overall, our findings suggest that, at the genomic level, there is a high number of potential neoantigens and TAAs available for T-cell targeting across EM and LM in progressive melanoma.

We conducted mass spectrometry analysis of peptides eluted from HLA class I molecules to examine cell surface presentation of T-cell antigens. This analysis was performed on five patient-derived cell lines, including two EM/LM pairs (Mel2 and Mel3) and one single EM lesion in which the corresponding LM lost HLA class I expression (Mel5). We detected eight neoantigen-derived peptides in the eluates of these cell lines. In Mel2, two peptides were identified in the EM sample and one in the LM (Fig. 4D; Table 2). In Mel3, four peptides were found in the EM sample, but none in the LM. Finally, one peptide was eluted from the EM of Mel5. These findings suggest a different cell surface presentation of neo-epitopes in the LM lesions compared with EM. Two of the mutations corresponding to the eluted peptides (RPL28 and SH2B3) were not present at the DNA level in LM, whereas the RNA expression of two other mutations (EML1 and VAT1L) was significantly lower in LM: 7.5- and 340-fold, respectively (Fig. 4D). Remarkably, strong T-cell–mediated immune responses were observed against the EM-derived EML1- and RPL28-derived peptides in the ACT products of these patients (Table 2), suggestive for ongoing immune pressure on these mutations. No T-cell recognition was observed to the other eluted peptides.

Furthermore, we identified 73 peptides that corresponded to 12 TAAs (Supplementary Table S5). Some of these TAA-derived peptides have been previously reported to be immunogenic (40, 41). The gene expression of most TAAs was similar between EM and LM, and even higher for MLANA in the LM of Mel2 (Fig. 4E). However, in the case of patient Mel5, where two peptides were eluted from EM but not LM due to loss of HLA class I expression in the late lesion, the gene expression of the two eluted peptides (MAGEA12 and MAGEC2) was lower in the LM. Compared with the changes observed in neo-epitope availability, these findings suggest that TAA expression remains stable across metastatic lesions over time.

Disease progression is common in late-stage melanoma with approximately half of the patients exibiting progression within one year of targeted therapies or whithin five years following treatment with ICI (2, 42, 43). Treatment resistance observed in these cases may be attributed to cancer cells exploiting immune escape mechanisms. Previous studies have demonstrated that the expression and presentation of tumor antigens is highly dynamic under the pressure of ongoing cancer immunity (44, 45). However, our findings indicate that T-cell–based therapies can still be applicable in heavily treated patients with late-stage melanoma. Despite heterogeneity and dynamics in the antigen-presenting capacity, T-cell infiltration, and (neo)antigen landscape in their tumors, there are still sufficient targetable antigens available for T-cell recognition. These findings align with recent observations of neoantigen-specific T-cell detection in patients with melanoma, where consistent detection of these T cells may correspond to sustained neoantigen expression and presentation, whereas sporadic detection may be attributed to the loss of these epitopes (37).

An in silico immunoediting prediction model applied to The Cancer Genome Atlas data indicated that most of the tumor types with evidence of T-cell immunity exhibited fewer neoantigens on average than predicted, except for melanomas (n = 99), which had the expected number of neoantigens, indicating limited immune pressure on the overall melanoma neoantigen landscape during evolution (46). These in silico predictions align with our observations that melanoma metastases developed at different points in time are targets for T cells. It appears that the moderate fluctuation in neoantigen availability reflects a balanced loss and gain of potentially immunogenic epitopes (“strong binders”), with a result that the number of potential targets for CD8⁺ T cells remains stable across metastatic tumor lesions for the majority of patients. We detected HLA class II expression in lesions from 5 out of 8 tested patients, indicating the potential for CD4⁺ T-cell–mediated reactivity and control of tumor outgrowth (31, 47). This is particularly important when HLA class I expression is completely lost, as observed in one of the lesions. Interestingly, neoantigen-specific CD4⁺ T cells are indeed frequently observed in patients with melanoma (32, 48).

Our study revealed that, except for one case, both the EM and LM tumor cell lines were recognized in vitro by MLTC or TIL products generated from either lesion. This highlights a window of opportunity for designing T-cell–based immunotherapies even in heavily pre-treated patients with progressive disease. These findings are further supported by a study demonstrating objective responses in half of the patients, following the adoptive transfer of autologous TIL in patients with advanced melanoma (6).

However, careful consideration is required when designing subsequent T-cell–based immunotherapies especially in late-stage progressive melanoma. Our results illustrate that reintroduction of ICI therapy could theoretically still be effective in a significant proportion of patients, as long as targetable (neo)antigens are still present. In case that TAA- or neoantigen-based vaccination (10, 49) or (neo)antigen-directed TCR-transgenic T-cell therapy is applied to enhance cancer-directed T-cell responses, one should validate the continued presentation of targeted (neo)antigens by tumor cells in later lesions and change the therapeutic targets when required, as well as redirect to CD4 epitopes in cases of complete loss of HLA class I. Similarly, ACT is still an option but may require the generation of new TIL batches from later lesions to maintain focus to the most appropriately presented targets in later-stage metastases. Our data also emphasize the importance to target multiple (neo)antigens or using polyclonal TIL or T-cell products obtained from multiple and later lesions to overcome potential escape due to tumor heterogeneity.

This study has a number of limitations. The tumor samples analyzed were obtained from a limited number of patients who received different treatments. The location of the melanomas differed as well as the time of sampling and intervals between sequentially obtained tumor samples. This makes it challenging to directly translate our results to clinical practice. We acknowledge that this variability introduces more than just temporal changes between two subsequent metastases, and the data may simply reflect variability of the genetic and immunological parameters assessed in this study between the two lesions. Even if this is the case, our data still indicate that such heterogeneity developed in time across these metastases, does not prevent them from being recognized by T cells.

In summary, the recognition of of early and later metastases by T cells demonstrates that late-stage melanomas, regardless of prior treatment, including progression on ICI, retain sufficient targetable (neo)antigens to warrant the continued pursuit of T-cell–based strategies for their treatment.

J. van den Bulk reports grants from KWF (Dutch Cancer Society) during the conduct of the study. S.H. van der Burg reports grants from KWF kankerbestrijding during the conduct of the study. S.H. van der Burg also reports personal fees from ISA pharmaceuticals; personal fees and other support from PCI Biotech and Mendus AB; and other support from Frame Pharmaceuticals outside the submitted work. No disclosures were reported by the other authors.

J. van den Bulk: Formal analysis, investigation, writing–original draft. E.M.E. Verdegaal: Conceptualization, funding acquisition, writing–review and editing. M. van der Ploeg: Investigation. M. Visser: Investigation. J.B. Nunes: Investigation. A.H. de Ru: Investigation. R.T.N. Tjokrodirijo: Investigation. M.E. Ijsselsteijn: Formal analysis, investigation. N.I. Janssen: Investigation. R. van der Breggen: Investigation. L. de Bruin: Investigation. P. de Kok: Investigation. G.M.C. Janssen: Investigation. D. Ruano: Formal analysis. E.H.W. Kapiteijn: Resources. P.A. van veelen: Conceptualization, formal analysis, funding acquisition. N.F.C.C. de Miranda: Conceptualization, funding acquisition, writing–review and editing. S.H. van der Burg: Conceptualization, funding acquisition, writing–review and editing.

We thank the patients for their willingness to participate in our ongoing studies on the immunotherapy of melanoma. This work was supported by the Dutch Cancer Society grant UL 2016–10815 (to E.M.E. Verdegaal, S.H. van der Burg, N.F.C.C. de Miranda, and P.A. van Veelen) and part of the research program Investment grant NWO Medium with project number 91116004 (to P.A. van Veelen), which is (partly) financed by ZonMw. J. van den Bulk was supported by an LUMC PhD fellowship. N.F.C.C. de Miranda is funded by the European Research Council (ERC) under the European Union's Horizon 2020 Research and Innovation Program (grant agreement no. 852832).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.