Melanoma cell lines Ma-Mel-63a (HLA-A*02:01, HLA-A*03:01, HLA-B*07:02, HLA-B*18:01, HLA-C06:02, BRAF-V600E, provided by Dirk Schadendorf, University Hospital Essen, Germany), SK-Mel-28 (RRID: CVCL_0526) (BRAF-V600E, HLA-A*11:01, HLA-B*40:01, HLA-C*03:03), UKE-Mel-105b (NRAS-Q61R, HLA-A*03:01, HLA-B*25:01, HLA-C*04:01, University Hospital Essen, Germany) were cultured at 37°C with 5% CO₂ under humidified atmosphere. Ma-Mel-63a and SK-Mel-28 were grown in RPMI1640 medium (Thermo Fisher) supplemented with 10% FCS (Capricorn) and 1% penicillin/streptomycin (Sigma-Aldrich). UKE-Mel-105b cells were cultured in DMEM high glucose HEPES medium (Gibco) supplemented with 10% FCS and 1% penicillin/streptomycin. A SK-Mel-28 vemurafenib and cobimetinib double resistant cell line was established. As the dual treatment with both inhibitors is challenging for most melanoma cell lines, first the resistance to vemurafenib (v) was introduced. Subsequently, the v-resistant cell lines underwent treatment with cobimetinib (c), to establish double resistant cell lines (SK-Mel-28 dr). For this purpose, cells were initially cultured in a low concentration of 15 nM vemurafenib (PLX4032, Cayman Chemical) in RPMI-CM (RPMI-1640 Medium with L-glutamine and sodium bicarbonate, Sigma Aldrich, + 10% FBS superior, Biochrom + 1% penicillin/streptomycin). The medium was refreshed every 2-3 days, and regular detachment of cells was performed to prevent excessive adherence. Upon reaching 80-90% confluence, the cells were split at a ratio of 1:4 and allowed to grow again to 80-90% confluence under the same inhibitor concentration. The doubling of the inhibitor concentration occurred only during the second 1:4 split. This iterative process continued until the cells adapted to grow in 2 µM Vemurafenib. Subsequently, a portion of the cells was cryopreserved, while the remaining cells were subjected to the same regimen with 2 µM vemurafenib and an additional 1.5 nM cobimetinib (GDC-0973, Cayman Chemical) until a final concentration of 200 nM cobimetinib was attained. The resistance profiles of the 2 µM v-resistant lines (r) and the 2 µM v- and 0.2 µM c-resistant lines (rr) were then evaluated through Western blot analysis.

Metabolic labeling was conducted using heavy (H) and medium-heavy (M) isotopically labeled amino acids arginine-13C6-15N4 (Arg10), lysine-13C6-15N2 (Lys8), leucine-13C6 (Leu6) or leucine-13C6-15N1 (Leu7) and arginine-13C6 (Arg6), lysine-D4 (Lys4), leucine-D3 (Leu3), respectively. All isotopically labeled amino acids were purchased from Eurisotop. Isotopic purity of isotopically labeled amino acids was checked by mass spectrometry. RPMI-1640 for SILAC was supplemented with 90 mg/L isotopically labeled arginine, 40 mg/L isotopically labeled lysine and optionally 50 mg/L isotopically labeled leucine. L-Proline was added in a concentration of 180 mg/L to suppress arginine to proline conversion. DMEM for SILAC was supplemented with 1 M HEPES (cell culture grade, Sigma Aldrich) to a final concentration of 24 mM, followed by 69 mg/L isotopically labeled arginine, 117 mg/L isotopically labeled lysine and 105 mg/L isotopically labeled leucine. Additionally, 10% dialyzed FCS (Capricorn) and 1% penicillin/streptomycin (Sigma Aldrich) was added to both RPMI-1640 and DMEM SILAC media. Cells were either expanded in SILAC medium for more than five cell division cycles, resulting in complete labeling, or labeled over a defined time interval with SILAC medium (pulsed SILAC, pSILAC) (25) while a treatment is applied. SK-Mel-28 wildtype (wt) cells were grown in RPMI SILAC medium containing medium-heavy labeled amino acids (RPMI M). SK-Mel-28 vemurafenib/cobimetinib double resistant cells (SK-Mel-28 rr) were grown in RPMI SILAC medium supplemented with heavy labeled amino acids (RPMI H). Ma-Mel-63a cells were grown in parallel in RPMI M and RPMI H, respectively. The following pSILAC experiments were conducted: Ma-Mel-63a cells were treated for 24h and 72h with 100 IU IFN-gamma (IFNγ, Peprotech) using RPMI M. In parallel, a vehicle control experiment with H₂O was conducted for 24h and 72h using RPMI SILAC H. Ma-Mel-63a cells were further treated with (i) a combination of 1 µM vemurafenib plus 100 nM combimetinib for 4h, 8h, 21h, 48h and (ii) 100 nM trametinib (MedChem Express) for 24h and 72h. The treatment was performed using RPMI H, while the vehicle control experiment was conducted in RPMI M containing 0.1% DMSO. Pulsed-SILAC experiments, including a vemurafenib + cobimetinib treatment and a control experiment were performed in the same manner for SK-Mel-28 wt cells and SK-Mel-28 dr for 24h and 72h. UKE-Mel-105b cells were treated with 100nM trametinib for 24h and 72h using DMEM for SILAC supplemented with heavy-isotope labeled amino acids (DMEM H) and DMEM with medium-heavy isotope labeled amino acids (DMEM M) for the corresponding vehicle control experiment (0.1% DMSO). For each cell line, an additional control experiment with mock treatment for both labeling conditions was performed. Cells were harvested, aliquoted at equal cell numbers and stored at -80°C until immunopeptidome analysis. For each SILAC condition 2-5x 10⁷ cells were used. Absence of mycoplasma was checked prior to experiments.

Normal human epidermal melanocytes, M2 (NHEM) from a 5-year-old (male, Caucasian) healthy donor (Helsinki permit was obtained) (HLA-A*03:01, HLA-A*68:01, HLA-B*07:02, HLA-B*44:02. HLA-C*07:02, HLA-C*07:04) were purchased from PromoCell and cultivated in Melanocyte Growth Medium M2 (PromoCell) until 90% confluency was reached. Cells were harvested for qualitative immunopeptidome analysis.

Viable cells were counted for each labeling state. Cell pellets were lysed in 2 ml lysis buffer per 1x 10⁸ cells for 1h at 4°C. Cell lysates were centrifuged at 16,000 x g for 20min at 4°C for preclearing. Precleared lysate of heavy-labeled cells was combined with lysate of the same number of medium-heavy-labeled cells. Immunoaffinity purification, HLA-I peptide isolation were conducted as described previously using W6/32 antibody (kindly provided by Hans-Georg Rammensee, Department of Immunology, University of Tuebingen, Germany) non-covalently bound to protein A Sepharose beads 4B (ThermoFisher Scientific), followed by solid phase extraction (SPE) with restricted access material (RAM) (16). Eluted HLA-I peptides were lyophilized and stored at -20°C until LC-MS/MS analysis.

5 µl of cell lysate was precipitated with the fourfold volume of acetone overnight at -20°C. Pellets were washed with acetone at -20°C. Precipitated proteins were dissolved in NuPAGE^® LDS sample buffer (Life Technologies), reduced with 50 mM DTT at 70°C for 10 min and alkylated with 120 mM iodoacetamide at room temperature for 20 min. Separation was performed on NuPAGE^® Novex^® 4-12% Bis-Tris gels (Life Technologies) with MOPS buffer according to manufacturer’s instructions. The gel was washed three times for 5 min with water and stained for 1 h with Simply Blue™ Safe Stain (Life Technologies). After washing with water for 1 h, each gel lane was cut into 15 slices. The excised gel bands were destained with 30% acetonitrile in 0.1 M NH₄HCO₃ (pH 8), shrunk with 100% acetonitrile, and dried in a vacuum concentrator (Concentrator 5301, Eppendorf, Germany). Digests were performed with 0.1 µg trypsin (Trypsin Gold, Mass Spectrometry Grade, Promega) per gel band overnight at 37°C in 0.1 M NH₄HCO₃ (pH 8). After removing the supernatant, peptides were extracted from the gel slices with 5% formic acid. Extracted peptides were pooled with the supernatant and analyzed by LC-MS/MS.

LC-MS/MS analysis of HLA-I peptides was performed as described previously (16). In brief, HLA-I peptides were dissolved in 30 μL of 2% ACN, 0.1% formic acid, and two subsequent LC-MS/MS runs were performed on an Orbitrap Fusion (Thermo Scientific) using a method for doubly charged peptides and a method for triple and single charged peptides. MS1 spectra were acquired with the following parameters: AGC target: 50,000, dynamic exclusion: 1, exclusion duration: 8s. MS2 spectra were acquired with the following paramters: intensity threshold: 10,000, collision energy: 30;35;40%, resolution: 60,000, AGC target: 50,000. ETD was applied for triply charged peptides only. MS data was analyzed by de novo sequencing with PEAKS XPro (Bioinformatics Solutions Inc.). Raw data refinement was performed with the following settings: (i) Merge Options: no merge; (ii) Precursor Options: corrected; (iii) Charge Options: 1−6; (iv) Filter Options: no filter; (v) Process: true; (vi) Default: true; (vii) Associate Chimera: yes. Parent Mass Error Tolerance was set to 10 ppm and Fragment Mass Error Tolerance was set to 0.02 Da. Enzyme was set to none.

To consider the medium-heavy and heavy labeled amino acids the following variable posttranslational modifications (PTMs) were allowed: Arg10 (Δm= 10.0083), Arg6(Δm= 6.0201), Lys8 (Δm= 8.0142), Lys4 (Δm= 4.0251), Leu6 (Δm= 6.0201) or Leu7(Δm= 7.0201), Leu3 (Δm= 3.0188). Additionally, Oxidation (M), pyro-Glu from Q (N-term Q), and carbamidomethylation (C) were included as variable PTMs. A maximum of 6 variable PTMs were allowed per peptide. Up to 10 de novo sequencing candidates were reported for each identified fragment ion mass spectrum, with their corresponding average local confidence (ALC) score. All de novo sequence candidates were matched against the 6-frame translated genome (GRCh38) and the 3-frame translated transcriptome (ENSEMBL release 90) using Peptide-PRISM (26). Peptides were then assigned to categories based on their genomic locations where they are encoded: (i) CDS, known protein; (ii) 5′-UTR; (iii) Off-frame, encoded within a protein coding region, but shifted by ±1; 3′-UTR; (iv) ncRNA; (v) Intronic; (vi) Intergenic; (vii) CDSintoIntron; (viii) OthersIntoIntron. Results were filtered to category-specific false discovery rate (FDR). NetMHCpan 4.0 was used to predict binding affinities for all identified MHC-I peptides. As per default, a cut-off of 0.5% rank for strong binders and 2% rank for weak binders was used. From all identified HLA-I peptides identified with Peptide-PRISM, a non-redundant custom fasta database was generated (category- and length-specific FDR < 10%). For each experiment, typically comprising one treatment and one mock condition for two different timepoints, an individual database was generated. After obtaining fasta databases, MaxQuant 2.4.2 (27) was used for quantitative analysis. Digestion mode was set to “no digestion”. Multiplicity for the SILAC-labeling was set to 3 (light, medium & heavy) with a maximum of 6 labels per peptide. Arg6, Lys4, Leu3 were selected as medium labels and Arg10, Lys8, Leu6 or Leu7 as heavy labels. FDR filtering was turned off by setting PSM FDR, protein FDR and site decoy fraction to 1. Minimum scores for modified and for unmodified peptides were set to 25. The second peptide option was enabled. Finally, the re-quantify option was used for improving quantification of large ratios. Apart from these adapted settings, MaxQuant default parameters were used. The MaxQuant peptides table was merged with the Peptide-PRISM results table by peptide sequence, and MaxQuant H/M ratios were used to generate scatter plots showing the modulation of the immunopeptidome. For improved ratio calculation, peptide H/M ratios were re-calculated from MaxQuant evidence table as follows ( Supplementary Table 1 ): for each experimental condition (“experiment type”, “timepoint”) the “median lg2 H/M ratio” for each peptide sequence was calculated from MaxQuant H/M ratios in the evidence table. When a peptide sequence feature was detected in both, medium and heavy labeling state (“Labeling State over Raw File” = 1,2 or 0,1,2), but was not recognized as a labeling cluster (due to large retention time shift of the differentially labeled species) the corresponding H/M ratios were re-calculated from the “summed Intensity H over Experiment” and the “summed Intensity M over Experiment” and used instead the original MaxQuant H/M ratios for calculating “median lg2 H/M ratio corrected”. To correct for potential mixing errors, “median lg2 H/M ratio corrected” was normalized on the median of the distribution. The resulting values in the columns “norm median lg2 H/M ratio corrected experiment type, timepoint” were used to generate the corresponding scatter plots. Intensity values (“log10 (intensity)”) used in the corresponding scatter plots were calculated from the summed intensities of all labeling states over one experimental condition.

For quantification quality assessment H/M ratios of type “multi” were classified as “reliable” in the column “Ratio reliability”, H/M ratios of type “iso” were classified as “unreliable”, except those that were detected in both, medium and heavy labeling state (“Labeling State over Raw File” = 1,2 or 0,1,2), but not recognized as a labeling cluster. These were corrected as described above and classified as “corrected”. The values from the column “Number of reliable and corrected ratios” can be used to color-code highly reliable and less reliable H/M ratios.

Publicly available immunopeptidome datasets of tumor and healthy cells ( Supplementary Table 2 ) (13, 18, 24, 28–45) were re-analyzed by de novo sequencing with PEAKS XPro and Peptide-PRISM as described above. We analyzed ~2500 MS raw data files from ~350 different patients/cell lines and identified more than 300,000 unique HLA-I peptides (<5% false discovery rate (FDR), ALC>50, only NetMHCpan-predicted binder), presented on ~140 different HLA allomorphs. Tumor data comprises samples from melanoma, meningioma, acute myeloid leukemia, colorectal cancer, ovarian cancer, glioblastoma and several other entities. Healthy control data includes the complete data set of the HLA ligand atlas project (13) (hla-ligand-atlas.org) comprising autopsy samples from 29 tissues from 13 donors, a large variety of samples from healthy blood donors, and some matched healthy tissue samples. We mapped all identified peptides with a unique genomic location to the human genome (hg38) and to their corresponding open reading frames (ORFs) and classified those ORFs as tumor-exclusive from which HLA peptides were detected exclusively in tumor samples.

Proteome analyses were performed by nanoLC-MS/MS on an Orbitrap Fusion (Thermo Scientific) equipped with a PicoView Ion Source (New Objective) and coupled to an EASY-nLC 1000 (Thermo Scientific). Peptides were loaded on a trapping column (2 cm x 150 µm ID, PepSep) and separated on a capillary column (30 cm x 150 µm ID, PepSep) both packed with 1.9 µm C18 ReproSil and separated with a 45-minute linear gradient from 3% to 30% acetonitrile and 0.1% formic acid and a flow rate of 500 nl/min. Both MS and MS/MS scans were acquired in the Orbitrap analyzer with a resolution of 60,000 for MS scans and 30,000 for MS/MS scans. HCD fragmentation with 35% normalized collision energy was applied. A Top Speed data-dependent MS/MS method with a fixed cycle time of 3 s was used. Dynamic exclusion was applied with a repeat count of 1 and an exclusion duration of 30s; singly charged precursors were excluded from selection. Minimum signal threshold for precursor selection was set to 50,000. Automatic gain control (AGC) was used with manufacturer’s standard settings for MS scans and MS/MS scans. EASY-IC was used for internal calibration. Raw MS data files were analyzed with MaxQuant version 2.0 (27). The search was performed against the UniProt reference proteome database (April 2024,104581 entries) with tryptic cleavage specificity, allowing 3 missed cleavages. Protein identification was under control of the false discovery rate (FDR; 1% FDR on protein and peptide spectrum match (PSM)). In addition to MaxQuant default settings, the search was performed against following variable modifications: Protein N-terminal acetylation, peptide N-terminal pyroglutamate formation at glutamine and oxidation (Met). Carbamidomethyl (Cys) was set as fixed modification. Arg6, Lys4 and Leu3 were set for medium-heavy SILAC labels and Arg10, Lys8 and Leu6 or Leu7 for heavy SILAC labels. Further data analysis was performed using R scripts developed in-house. For quantification of pSILAC-labeled proteins, the median was calculated from log2-transformed normalized peptide heavy-to-medium ratios (H/M) for each protein. Two ratio counts were required for protein quantification.

Protein extraction utilized RIPA lysis buffer from frozen dry pellets. Total protein concentration was determined via the BCA protein assay (Thermo Fisher). Equal protein quantities (12 µg per lane for 24h/48h/72h samples and 20 µg per lane for resistance validation) were combined with Pierce Lane Marker Reducing Sample Buffer (Thermo) and applied onto SDS gels (Mini-PROTEAN TGX Gels 8-16%, BioRad for 24h/48 h/72h samples and a 10% SDS gel for resistance validation). Two identical gels were loaded per time point, labeled Membrane 1 (M1) and Membrane 2 (M2), The Page Ruler Plus Prestained Protein Ladder, 10-250 kDa (Thermo), served as a marker. SDS-PAGE for protein separation was conducted in 1x Electrophoresis Buffer (25 mM Tris, 192 mM Glycine, 0.1% SDS) at a constant 120 V. Transfers onto 0.45 µm Nitrocellulose Membranes (Cytiva) occurred in 1x Blotting Buffer (48 mM Tris, 39 mM Glycine, 20% Methanol, 1.24 mM SDS) at a constant 100 V for 45 min using a CriterionTM Blotter (Bio Rad). Following Ponceau S staining to verify protein transfer integrity, membranes were treated with primary antibodies ( Supplementary Table 3 ), diluted to 1:1000 in TBS-T, 5% milk containing 0.01% NaN₃. For 24h/48h/72h samples, the application sequence was ERK followed by Vinculin or pERK, HLA-A, B, C followed by Vinculin. Resistant samples used primary antibodies pERK, Akt followed by Vinculin or pAkt, pERK followed by Vinculin. HRP-conjugated secondary antibodies ( Supplementary Table 3 ) were applied at a concentration of 1:2,000 in TBS-T/5% milk, depending on the species of the primary antibody. Development occurred with an ECL solution (Enhanced Chemiluminescence, 0.1 M Tris pH 8.5, 1.25 mM Luminol, 0.198 mM p-coumaric acid, 0.009% H₂O₂) using an Amersham Imager 600 (GE Healthcare Life Sciences).

For immunogenicity testing of selected peptide candidates, the in vitro priming strategy described by Bozkus et al. (46) was used with minor modifications. Peripheral blood mononuclear cells (PBMCs) of healthy donors were isolated from leukoreduction system chambers obtained from the Institute for Transfusion Medicine and Hemotherapy (University Hospital Wuerzburg, Germany). Therefore, Histopaque^®-1077(Sigma Aldrich) density gradient separation was performed. Isolated PBMCs were frozen as previously described (46) to maintain cell viability. HLA-typing was performed via HLA-surface staining with anti-HLA*02-APC-antibody (RRID: AB_2784362, Miltenyi Biotec) and anti-HLA-A*03-FITC-antibody (RRID: AB_2727190, Miltenyi Biotec), followed by flow cytometry analysis. HLA-A*02 and/or HLA-A*03 positive donors were used for in vitro priming assays. Frozen PBMCs were thawed and seeded in 96-well plates (Corning). PBMCs were cultured at 37°C with 5% CO₂ under humidified atmosphere and differentiated into monocyte derived dendritic cells (moDCs) using X-Vivo 15 Serum-free Hematopoietic Cell Medium (Lonza) medium supplemented with human GM-CSF (Gentaur), human IL-4 (Bio-Techne) and Flt3-Ligand (Bio-Techne) A fresh aliquot of donor cells was thawed and the MACS CD8+ T-cell isolation kit (Miltenyi Biotec) was used according to the manufacturers instruction to enrich for CD8 positive cells. Supernatants of each well, including non-adherent PBMCs were removed and substituted by the CD8 positive enriched T-cell fraction. Subseqeuntly, cells were matured with LPS (Sigma Aldrich), R848 (MedChem Express), IL-1β (Biotechne) and stimulated with peptides in a final concentration of 2µg/ml. The expansion of T-cells was supported by adding IL-2, IL-7 (Bio-Techne) and IL-15 (PeproTech) as described. On day 9, re-stimulation of peptide was induced according to the protocol of Boskus et al. (46). Cells were stained with surface antibodies anti-CD3-PE-Vio (RRID: AB_2725968, Miltenyi Biotec), anti-CD4-APC-Vio770 (RRID: AB_2726034, Miltenyi Biotec) and anti-CD8-FITC (RRID: AB_2659233, Miltenyi Biotec). Fixation and permeabilization were performed using the BD Cyofix kit and BD Perm/Wash kit (Becton Dickinson). Intracellular staining was conducted using anti-TNFα-APC (RRID: AB_2751984, Miltenyi Biotec) and anti-IFN-γ-PE (RRID: AB_2733717, Miltenyi Biotec) antibodies. A VioBlue-live-dead marker (Thermo Fisher) was included. Flow Cytometry was performed using the MACSQuant 16 Analyzer ( Supplementary Figure 1 ).

To get a first impression of the applicability, accuracy and reproducibility of pulsed SILAC (pSILAC) (25) for quantitative immunopeptidomics, we performed a proof-of-concept experiment to monitor the modulation of the immunopeptidome of a melanoma cell line by IFNγ. The effects of IFNγ on the immunopeptidome are well studied and could be used to assess the outcome of this quantitative approach. To this end we treated Ma-Mel-63a cells with 100 IU/mL IFNγ or vehicle control in medium-heavy (Lys4 & Arg6) or heavy (Lys8 & Arg10) SILAC medium, respectively. The differentially labeled IFNγ-treated and mock-treated cells were pooled after 24h and 72h, respectively. HLA-I peptides were isolated and analyzed by nanoLC-MS/MS. We used PEAKS for de novo peptide sequencing and Peptide-PRISM to identify both conventional peptides being part of the known proteome as well as cryptic peptides encoded in mRNA sequences that are not part of the protein-coding open reading frames, from non-coding RNAs, from introns or intergenic regions with a class-specific FDR of 10% as previously described (26). Next, we used MaxQuant for the quantification of all filtered peptides (27), and finally merged the results from Peptide-PRISM and MaxQuant (peptides table) ( Figure 1A ; Supplementary Table 4 ).

We observed a 2.1-fold median global upregulation of HLA peptides by IFNγ after 24h ( Supplementary Figure 2A ) and a 2.7-fold upregulation after 72h ( Figure 1B ), together with a strong up- and downregulation of many individual peptides, as well as a good correlation of HLA peptide-specific up- and downregulation at 24 and 72h ( Figure 1C ). In addition, we recognized the most profound upregulation of peptides presented on HLA-B (median ~4-fold) compared to peptides presented on HLA-A alleles (median ~2-fold) ( Figure 1D ), as described previously (15, 24). A comparison of quantitative immunopeptidomics and proteomics data showed no significant correlation, except for some proteins of the IFNγ-signaling pathway (e.g STAT, WARS) or the HLA peptide presentation pathway, such as HLA class I and HLA class II ( Figure 1E ; Supplementary Figure 2B ). The increased abundance of HLA peptides derived from these gene can be explained by the global IFNγ-induced expression of these genes on transcriptome and proteome level. These results are consistent with prior findings, including transcriptome studies and recent analyses by immunopeptidomics, as well as proteomics and genomics examining the effects of IFNγ (15, 23, 24, 44, 47–49). In summary, the proof-of-concept experiment indicated that the pSILAC approach is well suited for quantitative immunopeptidomics, gives reproducible results, and enables to monitor the modulation of the immunopeptidome in cell culture.

Next, we applied the optimized pSILAC workflow to study how the immunopeptidomes of two melanoma cell lines with overactive MAPK pathway, Ma-Mel-63a (BRAF V600E) and UKE-Mel-105b (homozygous NRAS Q61R), are modulated by pharmacological inhibition of the MAPK pathway. For this purpose, we employed the (i) MEK inhibitor trametinib and (ii) vemurafenib plus cobimetinib, an FDA- and EMA-approved BRAF/MEK inhibitor combination for BRAF^V600-mutated melanoma.

Concentrations of 100 nM trametinib and 1 µM vemurafenib plus 100 nM cobimetinib led to profound inhibition of the MAPK pathway ( Supplementary Figure 9 ) and thus were applied for all quantitative immunopeptidomics experiments described below. On average we identified ~5000 conventional HLA-I peptides derived from coding sequence (CDS) and ~500 cryptic HLA-I peptides for Ma-Mel-63a and for UKE-Mel-105b with Peptide-PRISM (category- and length-specific FDR < 10%). The high percentage (~10%) of cryptic peptides is due to the expression of HLA-A*03:01 in both cell lines. The immunopeptidomes of both melanoma cell lines exhibited substantial quantitative changes upon treatment with trametinib. ~10% of all HLA-I peptides showed significantly increased and another ~10% significantly decreased presentation, based on the 3-fold interquartile range (IQR) of the corresponding mock experiment ( Supplementary Figure 10 ). We found that peptides from melanoma differentiation antigens (MDAs, Supplementary Table 5 ), such as GTYEGLLRR and SLDDYNHLV (dopachrome tautomerase, DCT), ALDGGNKHFL, ALLAVGATK (premelanosome protein, PMEL) and ALLAGLVSL (tyrosinase, TYR) were consistently upregulated by trametinib ( Figures 3E–H ), which is in line with the previously described upregulation of MDAs early under MAPKi treatment in Ma-Mel-63a (52) and which is further in accordance with results published using a TMT-labeling approach (6).

Next, we aimed for identifying further tumor-associated or -specific T cell epitopes that are linked to an overactivated MAPK pathway. To classify HLA peptides as tumor-exclusive, we re-analyzed a large set of publicly available immunopeptidome data from tumor and healthy samples (for details see Methods) and identified conventional and cryptic HLA peptides with Peptide-PRISM. In addition, we analyzed the immunopeptidome of commercially available, primary melanocytes ( Supplementary Table 6 ). Since melanoma originates exclusively from melanocytes or their progenitor cells, melanocytes should be the best reference to identify truly melanoma-specific HLA peptides. To maximize the overlap sof HLA peptides with the two melanoma cell lines, we chose HLA-A*03:01-positive melanocytes. From the combined data we generated a database (tumor vs. healthy database, TvHdb) by mapping all identified HLA peptides to their corresponding open reading frames (ORFs) and classified those ORFs as tumor-exclusive, from which HLA peptides were detected only in tumor samples. With the information from TvHdb, we classified all HLA peptides from pSILAC experiments that derived from tumor-exclusive ORFs as “tumor” and all peptides from non-tumor-exclusive ORFs as “tumor & benign” ( Figure 3 ). In addition, we extracted from the TvHdb in how many different patients or cell lines the corresponding peptides have been detected and from which sample types (tumor entity or origin of healthy sample; Supplementary Tables 7–13 ).

As expected, the great majority of HLA peptides were not tumor-exclusive and probably represents self-peptides. HLA peptides from MDAs, such as DCT, PMEL and tyrosinase were recurrently detected exclusively in melanoma and melanocytes. One putative tumor-specific peptide with the sequence TTNARILAR, derived from urocortin-2 (UCN2) was downregulated upon MAPK-signaling inhibition in Ma-Mel-63a cells. This protein is known to activate ERK1/2 signaling and plays a role in the context of cardio protective effects (53). It was identified in 6 out of 12 HLA-A*03:01 positive melanoma patients. Transcriptomic analyses did not reveal any evidence for tumor-specific expression pattern of this gene (54). We therefore decided to not assess immunogenicity of this conventional HLA peptide. Essentially, no other tumor-exclusive HLA peptides were repeatedly detected in the pSILAC data from MAPK inhibition experiments. However, we reproducibly identified two tumor-exclusive cryptic peptides that changed upon MAPK pathway inhibition. The peptide RVYLDIVTPK (off-frame, HLA-A*03:01, IFT81) was reproducibly upregulated under all tested conditions and the peptide RLSSSLPSR (5’UTR, HLA-A*03:01, ST3GAL6) was reproducibly downregulated upon MAPK pathway inhibition in Ma-Mel-63a-cells ( Figure 3 ; Supplementary Figures 11, 12 ). In TvHdb RVYLDIVTPK (IFT81) was identified exclusively in melanoma samples (in 17 of 31 HLA-A*03:01/A*11:01/A*31:01-positive tumor metastases or patient-derived cell lines) and RLSSSLPSR (ST3GAL6) was identified in melanoma (in 9 of 31 HLA-A*03:01/A*11:01/A*31:01-positive melanoma samples) ( Supplementary Table 14 ). This could indicate a relevance of these peptides particularly in melanoma disease and progression. RVYLDIVTPK is an off-frame peptide, deriving from the coding region of IFT81 in a non-canonical reading frame ( Supplementary Figure 13 ). Ribosome profiling data [nuORF database (55)] indicated that this peptide derives from a short (34 amino acids) internal ORF with an alternative CTG start codon. The protein IFT81 (protein intraflagellar transport 81) is involved in cilia assembly. Interestingly, Jenks et al. described changes in cilia length and formation as a hallmark of (MEK) kinase inhibitor resistant cells (56). This supports our hypothesis, that the generation of the cryptic peptide RVYLDIVTPK (IFT81) could be linked to inhibited MAPK signaling and could derive from activated alternative signaling pathways. Although the exact mechanism still needs to be unraveled, our studies strongly indicate that the cryptic peptide RVYLDIVTPK is generated upon MEK-inhibitor exposure. According to the nuORF database, the cryptic peptide RLSSSLPSR (ST3GAL6) derives from an upstream open reading frame (uORF) also with an alternative CTG start codon ( Supplementary Figure 14 ). uORFs can regulate expression of downstream gene products. The translation of the corresponding uORF, associated with the generation of RLSSSLPSR could therefore suppress translation of ST3GAL6. A tumor suppressive role of ST3GAL6 (Type 2 lactosamine alpha-2,3-sialyltransferase) was indeed described by Li et al. (57), which could indicate that suppressed ST3GAL6 levels could be advantageous for tumor progression. Targeting this peptide under non-treatment conditions could be a promising immunotherapeutic strategy, provided that the tumor relies on ST3GAL6 suppression and the generation of the RLSSSLPSR peptide as byproduct of this tumor promoting mechanism.

To collect further evidence for the tumor specificity of the two cryptic peptides RVYLDIVTPK (IFT81) and RLSSSLPSR (ST3GAL6), we evaluated their immunogenicity applying an in vitro priming assay ( Figure 4 ). In addition to the two HLA-A*03:01-restircted cryptic peptides, we included one HLA-A*02:01- and one A*03:01-restricted peptide from DCT that showed increased presentation upon MAPK pathway inhibition. For in vitro priming we used PBMCs of two independent HLA-A*02- and HLA-A*03-positive healthy donors. The two self-peptides TLYEAVREV (HLA-A*02) and RILGPGLNK (HLA-A*03) derived from RPL10A with expected central tolerance were used as negative control. The peptides ELAGIGILTV (MART-1, HLA-A*02), YLQPRTFLL (SARS-CoV2, HLA-A*02) and ILRGSVAHK (Flu, HLA-A*03) served as positive controls.

Effector responses were evaluated by gating TNFα/IFNγ double positive cell populations of CD8⁺ T cells. The frequency of TNFα/IFNγ-positive T cells in all negative controls was below 0.06% and ranged between 5.33-21.3% for the positive controls. The HLA-A*02:01-restricted peptide SLDDYNHLV (derived from DCT) induced immunological responses in both healthy donors with fractions of 0.22% and 1.37% double positive cells, respectively, confirming the immunogenicity of this peptide as described earlier (58). In contrast, the HLA-A*03-restricted peptide GTYEGLLRR that also derived from DCT, induced fractions of only 0.021% and 0.072% double positive cells that were almost equal to negative control level. Cryptic peptides RLSSSLPSR (ST3GAL6) and RVYLDIVTPK (IFT81) showed only marginal responses in donor 1 (0.053% and 0.057% versus 0.031% in the negative control) and no responses in donor 2 (0.055% and 0% versus 0.056% in the negative control).

Long-term therapeutic efficacy of MAPKi is often limited due to acquired drug resistance and a major cause for clinical tumor progression or relapse. There are controversial results from randomized phase III studies concerning the anti-tumor activity of the combined treatment of BRAF-mutated metastatic melanoma with immune checkpoint and MEK inhibitors (59). In this context, we aimed to study how drug resistance contributes to immunopeptidomic alterations that can impact tumor immunogenicity. To this end we generated a MEKi-resistant melanoma cell line. Since Ma-Mel-63a cells that were long-term exposed to BRAFi and MEKi (Ma-Mel-63a dr) turned out to be still sensitive to BRAF and MEK inhibition ( Supplementary Figure 15 ), we induced BRAFi- and MEKi resistance in the BRAF^V600-mutated melanoma cell line SK-Mel-28 by long-term exposition to vemurafenib and cobimetinib. In contrast to SK-Mel-28 wt, the MAPK pathway was no longer inhibited by treatment with vemurafenib/cobimetinib for SK-Mel-28 dr ( Figure 5A ).

First, we conducted pSILAC experiments with SK-Mel-28 wt and SK-Mel-28 dr cells to unravel short-term immunopeptidomic changes upon BRAF- and MEK inhibition with a combination of vemurafenib and cobimetinib over 24h and 72h. In general, background alteration of the immunopeptidome in untreated cells were marginal for both, pSILAC- and SILAC-conducted experiments ( Supplementary Figure 16 ; Supplementary Table 12 ). As expected, BRAF and MEK inhibition induced perturbations of the SK-Mel-28 immunopeptidome ( Supplementary Figure 17 ) were similar to the mock condition. In analogy to melanoma cell lines Ma-Mel-63a and UKE-Mel-105b, SK-Mel-28 wt cells showed enhanced presentation of MDA-derived peptides upon MAPK pathway inhibition ( Figure 5B ). In contrast, SK-Mel-28 dr cells did not show this effect ( Figure 5C ). In addition, we recognized that two HLA peptides from NUF2 were downregulated upon MAPK pathway inhibition in SK-Mel-28 wt cells ( Figure 5B ). NUF2 exhibits a Cancer germline-like expression profile in GTEx (60) with predominant expression in testis and marginal expression in most other tissues. Expression of Cancer germline antigens (CGAs) get lost when stem cells differentiate and can be reactivated during oncogenesis when tumor cells dedifferentiate.

Next, we focused on cryptic peptides. We observed that many of these showed an altered presentation in vemurafinib/cobimetinib-treated SK-Mel-28 wt cells ( Figure 5B ). In contrast, in SK-Mel-28 dr cells, only a few cryptic peptides showed an altered presentation ( Figure 5C ) and most of these were not reproducibly detected for both time points ( Supplementary Figure 18 ). We identified two (according to TvHdb) putative tumor-specific cryptic peptides (RVYLDIVTPK, IFT81 and SSIRSWNNK, ESRP1) with increased presentation after short-term MAPKi treatment in wildtype as well as in resistant cells ( Figures 5B, C ), indicating that their presentation is not directly linked to MAPK signaling. Moreover, we identified four putative tumor-specific cryptic peptides from ZYG11A, RPL36A-HNRNPH2, CLIC1 and RPL31 that were decreasingly presented upon MAPKi in SK-Mel-28 wt but were unaffected in SK-Mel-28 dr cells ( Figures 5B–D ), indicating a direct link to MAPK signaling. A linkage to MAPK signaling has already been reported for one of these genes, chloride intracellular channel 1 (CLIC1), in the context of gastric cancer progression (61). All four HLA peptides have been recurrently identified in different tumor entities but never on healthy cells in TvHdb ( Supplementary Table 14 ). For instance, the peptide SVASTNPIK (RPL36A-HNRNPH2) has been identified in 10 tumor patients (melanoma, CRC, GBM), representing a recurrent putative tumor-specific T-cell antigen.

Subsequently, to study long-term alterations in the immunopeptidome that are linked to MAPK inhibitor resistance, we conducted a SILAC-based experiment to directly compare the immunopeptidomes of SK-Mel-28 wt and SK-Mel-28 dr cells. We found that formation of resistance was accompanied by substantial immunopeptidomic alterations, reflecting changes of the cellular state. We observed that the overall presentation of HLA peptides derived from MDAs (SOX10, PMEL, DCT) was diminished in SK-Mel-28 dr cells compared to SK-Mel-28 wt cells ( Figure 5E ), indicating an epithelial-to-mesenchymal-like transition during long term MAPKi exposure. We found additional evidence for this phenotype switch of drug-resistant cells towards a dedifferentiated cell state by RT-qPCR ( Supplementary Figure 19A ) and by proteome analysis ( Supplementary Figure 19B ). On transcriptome level, SK-Mel-28 dr cells showed elevated AXL and reduced MITF expression, representing a trend towards an AXL^high/MITF^low phenotype, which defines a dedifferentiated melanoma phenotype. While differentiated melanomas highly express MDAs, expression of MDAs is lost in dedifferentiated melanomas (62). Accordingly, we observed decreased protein levels of MDAs (DCT, PMEL) in the proteome of SK-Mel-28 dr compared to SK-Mel-28 wt cells. The observed diminished presentation of HLA peptides from MDAs suggests that patients with acquired MAPK inhibitor resistance might show decreased responsiveness to immunotherapeutic treatments targeting MDAs.

After that, we found that a number of putative tumor-specific cryptic peptides showed significantly increased presentation in MAPKi-resistant cells, while others exhibited substantially decreased presentation ( Figure 5E ). The peptides with increased presentation in SK-Mel-28 dr cells derived from EIF4B, FAM20B, LDHB, CYCS, C5orf22, RCAN1 and DIEXF and were recurrently identified in TvHdb in a variety of tumor patients ( Supplementary Table 14 ). For example, the peptides RVQILQILK (DIEXF) and STDLPILLK (C5orf22) were identified in 9 and 7 tumor patients, respectively. In general, upregulation of these cryptic peptides might mechanistically be explained by adaptive response of the tumor cells to stress conditions, such as pro-apoptotic or hypoxic conditions, induced by kinase inhibitor exposition. As a result, the expression of genes that adapt the cell state to overcome these unfavorable cellular conditions, is altered and the transcriptional landscape is changed. For instance, translation of off-frame cryptic peptides from transcripts of lactate dehydrogenase (LDHB) and cytochrome c (CYCS) could be a result of ongoing metabolic alterations to overcome kinase inhibitor-induced cell stress. Contretras Mostazo et al. reported that CYCS was upregulated in an imatinib-resistant CML cell line, confirming that mitochondrial respiration can be influenced by kinase inhibitor exposition (63). RCAN1 contributes to deregulation of calcineurin-involved signaling pathways, however, its role to cancer progression is contradictory. In melanoma, transcript levels of RCAN1 are elevated in comparison to normal samples, indicating that there might be an association to melanoma progression (64). Overall, putative tumor-specific cryptic peptides with enhanced presentation on kinase inhibitor-resistant melanoma cells, might represent attractive targets for immunotherapy, especially in combination with MAPKi treatment.

In contrast, putative tumor-specific cryptic peptides with diminished presentation on MAPKi-resistant melanoma cells are probably less attractive targets, but still reflect alterations of the cellular state. We identified cryptic peptides with reduced presentation on MAPKi-resistant cells from TRIAP1, SCARB1, AC058822.1, AL162457.2 and TNFRSF14 ( Figure 5E ; Supplementary Table 14 ). Two of these peptides derived from proteins that are related to necroptosis. Necroptosis is an alternative type of programmed cell death, in which processes of necrosis are involved. Necroptosis-related cryptic peptides derived from TNFRSF14, a TNF receptor gene and a long-non-coding RNA (lncRNA) AL162457.2. Reports of Liu et al. (65) indicate decreased overall survival in melanoma when this necroptosis-related lncRNA is expressed. Interestingly, findings of Sugaya et al. (66) show that the BRAF inhibitor dabrafenib attenuates necroptosis. These results combined with our data suggest, that necroptosis and expression of necroptosis-related genes is inhibited during long-term BRAFi exposure, finally leading to a decreased presentation of the corresponding cryptic HLA peptides.