Primary brain (PB) control samples (non-diseased cerebellum, n = 3) were obtained by an anatomical pathologist at the Royal Children’s Hospital (Melbourne, Victoria) as part of the rapid autopsy protocol of a patient with DIPG (HREC 34049), performed within 24 h of time of death. Samples were flash-frozen and cryopreserved until further use.

Six patient-derived DIPG cell cultures (SU-DIPG27, SU-DIPG35, SU-DIPG38, SU-DIPG43, SU-DIPG58, and SF7761) were maintained in working tumor stem medium (TSM) as described previously (5). First, tumor base medium (TBM) was prepared by mixing 1:1 ratio of Neurobasal-A medium (Gibco, Thermo Fisher Scientific, USA) and Dulbecco's Modified Eagle Medium F12 (D-MEM/F-12) (Gibco), supplemented with 10 mM HEPES (N-(2-Hydroxyethyl)piperazine-N'-(2-ethanesulfonic acid)) (Gibco), 1 mM MEM sodium pyruvate (Gibco), 0.1 mM Minimal essential media (MEM) nonessential amino acids (aa) (Gibco), 1× GlutaMAX-I supplement (Gibco), and 1× Antibiotic-Antimycotic (Gibco). Second, TBM was supplemented with B27 supplement (minus Vitamin A, Thermo Fisher Scientific, USA), human epidermal growth factor (H-EGF; 20 ng/ml; Shenandoah Biotech, USA), human basic fibroblast growth factor 154 aa (H-FGF-basic 154; 20 ng/ml; Shenandoah Biotech), human platelet-derived growth factor (H-PDGF-AA; 10 ng/ml; Shenandoah Biotech), H-PDGF-BB (10 ng/ml; Shenandoah Biotech), and 0.2% Heparin solution (2 mg/mL; STEMCELL Technologies, USA).

Next-generation sequencing (NGS) for HLA class I genotyping of SU-DIPG58, SU-DIPG38, and SF7761 was performed by the Victorian Transplantation and Immunogenetics Service (West Melbourne, Victoria, Australia), whereas genotyping of SU-DIPG27, SU-DIPG35, and SU-DIPG43 was performed by PathWest Laboratory Medicine WA (Perth, Western Australia, Australia).

Both the BB7.2 and W6/32 eluates from the small-scale elution of 4 HLA-A*02:01 expressing DIPG cell lines were spiked with a mixture of 11 indexed Retention Time (iRT) peptides to aid retention time alignment (39) and acquired on a Q-Exactive plus LC-MS (Thermo Fisher Scientific, USA). Samples were acquired in data-dependent acquisition (DDA) mode using online rapid separation liquid chromatography (RSLC) nano-HPLC (Ultimate 3000 UHPLC, Thermo Fisher Scientific, USA). The samples were injected onto a 100μm, 2cm nanoviper Pepmap100 trap column prior to separation on a 75 μm × 50 cm, Pepmap100 C₁₈ analytical column (Thermo Fisher Scientific, USA). The eluate was nebulized and ionized using a nano-electrospray source (Thermo Fisher Scientific, USA) with a distal coated fused silica emitter (New Objective, USA). The capillary voltage was set at 1.7 kV. The QE plus mass spectrometer was operated in the DDA mode to automatically switch between full MS scans and subsequent MS/MS acquisitions. Survey full scan MS spectra (mass to charge ratio (m/z) 375–1,600) were acquired in the Orbitrap with 70,000 resolution after accumulation of ions to a 5e5 target value with a maximum injection time of 120 ms. For MS/MS, dynamic exclusion was set to 15 s to avoid resequencing the same analyte. The 12 most intense charged ions (z ≥ +2) were sequentially isolated and fragmented in the collision cell by higher-energy collisional dissociation (HCD) with a fixed injection time of 120 ms, 35,000 resolution, and automatic gain control (AGC) target of 2e5 with isolation window of 1.8 m/z, intensity threshold 1.7e4, and scan range of 200–2,000 m/z.

The eight peptide fractions derived from each of the large cell pellets of SF7761 and SU-DIPG58 cell lines were analyzed on a Tribrid Fusion LC-MS (Thermo Fisher Scientific, USA) with addition of the iRT peptide mix. Peptides were loaded onto a PepMap Acclaim 100 C₁₈ trap column (5 µm particle size, 100 µm × 2 cm and 100Å (Thermo Fisher Scientific, USA) at 15 µL/min using an Ultimate 3000 RSLC nano-HPLC (Thermo Fisher Scientific, USA). Peptides were eluted and separated at a flow rate of 250 nL/min on an in-line analytical column PepMap RSLC C₁₈, 2 μm particle size, 75 μm × 50 cm, and 100 Å, (Thermo Fisher Scientific, USA) using 125-min linear gradient as described previously (37). Peptides were introduced using nano–electrospray ionization method into the Orbitrap Fusion MS at a source temperature of 275°C. All MS spectra (MS1) profiles were recorded from full ion scan mode 375–1,800 m/z, in the Orbitrap at 120,000 resolution with AGC target of 400,000 and dynamic exclusion of 15 s. The top 12 precursor ions were selected using top speed mode at a cycle time of 2 s. For MS/MS, a decision tree was made to aid selecting peptides of charge state +1 and +2–6 separately. For singly charged analytes, only ions falling within the range of m/z 800–1,800 were selected, whereas, for +2 to +6 m/z, no such parameter was set. The C-trap was loaded with a target of 200,000 ions with an accumulation time of 120 ms and isolation width of 1.2 amu. Normalized collision energy was set to 32 HCD, and fragments were analyzed in the Orbitrap at 30,000 resolution.

Immunopeptidomics MS/MS raw files generated on the Tribrid Fusion and QE Plus were exported and analyzed using Peaks 8.5 software (Bioinformatics Solutions Inc.) searching against the human proteome (Uniprot, accessed 15 June 2017; 20,182 entries). The following search parameters were used: error tolerance of 10 ppm using monoisotopic mass for precursor ions and 0.02Da tolerance for fragment ions; enzyme used was set to none with following variable modifications: oxidation at Met; deamidation at Asp and Gln; and phosphorylation at Ser, Thr, and Tyr. False discovery rate (FDR) was estimated using a decoy fusion method (40). For data processing, first, a 5% FDR cutoff was applied. Peptides matched to decoy database, iRT peptides, duplicates, and those detected in blanks were removed. Second, only peptides with 8–12 aa were retained with modified versions of the peptides also removed.

The binding motif of HLA peptides was interpreted using IceLogo (41) and binding affinity to different HLA alleles ascertained using NetMHC pan 4.0 RRID : SCR_021651 (42). Peptides were identified as either strong binders (SBs; based on rank threshold of 0 to 0.5) or weak binders (WBs; based on rank threshold of 0.5 to 2.0). Peptides identified were cross-referenced with Immune Epitope Database and Analysis Resource (IEDB) that lists all HLA-restricted peptides identified in humans (43). Whereas, to identify brain-specific proteins present in the data, Human Brain Protein Atlas (HBPA) (44) was used. The data were plotted in GraphPad Prism version 9.1 (GraphPad Software, USA, www.graphpad.com, RRID : SCR_002798).

Proteomics was performed on pHLA depleted cell lysates (Sections 2.1 and 2.3; n = 5) as well as part of primary non-diseased brain sample controls (cerebellum, n = 3) obtained from rapid autopsies of patients with DIPG (as described in Section 2.2), using sodium dodecyl sulfate (SDS)–based suspension trapping (S-TRAP) method. Briefly, lysates were thawed, vortexed, and centrifuged at 3,978g for 10 min at RT. Protein concentration was measured using Direct Detect Spectrometer (Millipore, USA). A volume equivalent to 400 µg of protein was mixed with an equal volume of 2× lysis buffer [10% SDS, 100 mM triethylammonium bicarbonate (TEAB; Sigma), pH 8.5]. After lysis, 20 mM (final concentration) of Tris(2-carboxyethyl) phosphine hydrochloride (TCEP; Sigma) was added to reduce the samples by incubation at 60°C for 15 min. For alkylation, 20 mM (final concentration) of iodoacetamide (IAA; Sigma) was added, and samples were incubated in the dark for 15 min at RT. Samples were then acidified using 1.2% phosphoric acid. Colloids were generated by adding S-TRAP binding buffer (90% methanol in 100 mM TEAB) in a 1:7 ratio of buffer to starting material. The resulting colloid samples were loaded onto the S-TRAP filter (Profiti, USA) and centrifuged at 3,978g for 30 s at RT. Columns were washed thrice using the binding buffer. Protein digestion in the S-TRAP filter was performed by using Trypsin (Sigma, USA) made in digestion buffer (50 mM TEAB) at a 1:60 ratio of enzyme to sample. Tubes were sealed and incubated at 37°C overnight. For peptide recovery, S-TRAP filters were washed sequentially with 80 µL of 50 mM TEAB followed by 80 µL of 0.1% FA and, finally, 80 µL of 50% ACN with 0.2% FA. Samples were concentrated using a speed vacuum concentration system (LABCONCO, USA) and reconstituted in 20 µL of 0.1% FA.

Proteomics samples were diluted 1:10 in 0.1% aqueous FA and spiked with iRT peptide mix. Samples were acquired using a hybrid trapped ion mobility (TIMS)-quadrupole time-of-flight (TOF) MS (Bruker TimsTOF pro, Bruker Daltonics, Germany) coupled to a nanoElute liquid chromatography system. The sample (1-2 µL) was loaded directly onto the C₁₈ analytical column (Aurora, IonOpticks) of 1.6 μm particle size and 75 μm × 25 cm and 120Å using a 20-min linear gradient of buffer A1 (2% ACN and 0.1% FA). Peptides were eluted from the column with flow rate set to 400 nL/min, by a linear step-wise gradient of buffer B1 (100% ACN and 0.1% FA) against buffer A1 initially to 17% over 60 min, then to 25% over 30 min, and 37% over 10 min followed by rapid rise to 95% over 10 min. DDA was performed with following settings: m/z range, 100–1,700 m/z; capillary voltage, 1,600 V; target intensity, 30,000; and TIMS ramp, 0.60 to 1.60 Vs/cm² for 166 ms. All SU-DIPG cell lines and SF7761 were acquired as technical replicates (n = 3), whereas PB samples were acquired as biological replicates from the same patient (n = 3).

For proteomics data analysis, the. TDF files generated on the Bruker TimsTOF for the samples were searched using MaxQuant computational proteomics platform version 2.0 (45). The following search parameters were used: N-terminal acetylation along with oxidation at Met and deamidation at Asp and Gln was set as variable modifications. Cys carbamidomethylation was set as a fixed modification. Bruker TIMS was selected under instruments with 20 ppm as first search peptide tolerance and 10 ppm as main search peptide tolerance. The enzyme specificity was set to Trypsin with a maximum of two missed cleavages being chosen. Quantification in MaxQuant was performed using the built-in extracted ion chromatogram-based label-free quantification (LFQ) (46). The spectra were searched by the Andromeda search engine against the human proteome (Uniprot, accessed 15 June 2017; 20,182 entries) including contaminants. Possible sequence matches were restricted to 8 to 25 aa, a maximum peptide mass of 4,600 Da. A FDR of 0.01% was set for proteomic analysis. Protein identification required at least one unique or razor peptide per protein group.

Further downstream analysis of the data was performed on the basis of LFQ values using Perseus software (version 1.4.0.6) (45). LFQ values were log2-transformed and filtered on the basis of valid values found in at least three samples in any group. Missing values were replaced from normal distribution. To identify proteins being significantly regulated in DIPG cell lines compared to PB, a two-sided t-test was performed at 0.01% FDR. For visualization in the form of heatmap, log2-transformed scores were normalized by Z-score. The −logP value was converted into −log10 P-value, and volcano plots was made using ggVolcanoR app (47) by plotting −log10 P-value vs. the log2 fold change difference.

Specific databases were used to annotate the proteomics data. To identify proteins expressed on the surface of cells, Cell Surface Protein Atlas (CSPA) (48) was used. To identify proteins that are targets of Food and Drug Administration (FDA)–approved drugs, we used the Human Protein Atlas (HPA) druggable proteome https://www.proteinatlas.org/humanproteome/tissue/druggable) (44). All databases were added under annotation in Perseus and used to filter rows.

Protein network analyses were performed using Cytoscape (version 3.7.2) RRID : SCR_003032 (49) plug-in ClueGO version 2.5.7 (50). The enrichment of proteins was determined by right-sided hypergeometric test with Bonferroni step down correction and kappa score threshold of 0.4.

Patient-derived SU-DIPG cell lines SU-DIPG 4, SU-DIPG 13, SU-DIPG 17, SU-DIPG 19, SU-DIPG 21, SU-DIPG 25, and SU-DIPG 33 were analyzed using cell surface proteomics in quadruplicate, using an amino-oxy-biotin labeling technique (51). Briefly, cells were labeled for surface sialic acid residue by biotinylating with 1 mM sodium meta-periodate (Thermo, cat. no. 20504), 200 μM amino-oxy-biotin (Biotium, cat. no. 90113), and 10 mM aniline (Sigma, stock no. 10.97 M) in PBS (pH 6.7) for 1 h at 4°C. Following this, the reaction was quenched with glycerol, cells were washed twice, and a small aliquot taken to test the efficiency of biotinylation, using a streptavidin-labeled fluorescent marker. The cells were then resuspended in lysis buffer [1% v/v Igepal CA-630, 150 mM NaCl, 1× protease inhibitor (complete, without Ethylenediamine tetraacetic acid (EDTA), Roche), 5 mM IAA, and 10 mM Tris-HCl (pH 7.6)]. Nuclei were removed by centrifugation at 2,800g for 10 min, at 4°C. The remaining biotinylated proteins in the lysate were enriched by incubation with high-capacity streptavidin-agarose (Pierce, Country) for 60 min at 4°C. After three washes with lysis buffer and PBS/0.5% (w/v) SDS, samples were incubated at RT for 20 min with 100 mM Dithiothreitol (DTT). Further washes with urea buffer [6 M urea and 100 mM Tris-HCl (pH 8.5)] were followed by a final incubation with urea buffer containing 50 mM IAA in the dark for 20 min. Subsequently washes were performed with urea buffer, PBS, and then water. The proteins were digested on-bead overnight in 100 µL of 50 mM ammonium bicarbonate containing 2 µg of Trypsin Gold. The following morning, peptides were collected by centrifugation, and tryptic fractions were analyzed for mass spectrometry analysis.

After sample processing, peptides were resuspended in 2% ACN/1% FA and separated by reversed-phase liquid chromatography on a M-class UHPLC system (Waters, USA) using a 250 mm × 75 μm column (1.6 µm C18, packed emitter tip; Ion Opticks, Australia) with a linear 90-min gradient at a flow rate of 400 nL/min from 98% solvent A (0.1% FA in Milli-Q water) to 35% solvent B (0.1% FA and 99.9% acetonitrile). The nano-UPLC was coupled on-line to an Impact II mass spectrometer equipped with a CaptiveSpray ionization source (Bruker Daltonics, Germany) and column-oven at 40°C (Sonation, Germany). The Impact II was operated in a data-dependent mode using a 1.5-s cycle time, switching automatically between one full-scan at 4 Hz and subsequent MS/MS scans for the remaining time with spectra rate determined using peptide intensity. The instrument was controlled using otofControl version 4 (Bruker).

Raw files were analyzed using MaxQuant (version 1.5.8.3). The database search was performed using the Uniprot Homo sapiens database (downloaded October 2020) plus common contaminants with strict trypsin specificity allowing up to two missed cleavages. The minimum peptide length was 7 aa. Carbamidomethylation of cysteine was a fixed modification, whereas N-acetylation of proteins N-termini and oxidation of methionine were set as variable modifications. During the MaxQuant main search, precursor ion mass error tolerance was set to 0.006 Da. PSM and protein identifications were filtered using a target-decoy approach at an FDR of 1% with the match between runs option enabled.

Downstream analysis of the data was performed using Perseus as mentioned in Section 2.8. Identification of proteins with transmembrane helices was performed using prediction server TMHMM (52, 53) (https://services.healthtech.dtu.dk/service.php?TMHMM-2.0) based on the PredHel (number of predicted helices present in the protein) score of 3 or more.

Cell surface HLA class I expression for the cell lines was measured by flow cytometry following surface staining using hybridoma supernatants for anti–HLA-A2 {BB7.2 [HB-82; American Type Culture Collection (ATCC)] (35); produced in-house}, anti–HLA-A3 [GAPA3 (HB-122; ATCC) (38); produced in-house], and pan HLA class I [W6/32 (HB-95; ATCC) (54); produced in-house] as primary antibodies followed by a goat anti-mouse Immunoglobulin G Phycoerythrin (IgG PE) secondary antibody [1:250 dilution (cat. no. 1030-09); Southern Biotech, cat. no. 1030-09, RRID : AB_2794297, USA] using a LSRII flow cytometer.

Cell surface protein targets were validated using antibody labeling of DIPG cell lines using flow cytometry. Briefly, single-cell suspensions of DIPG cells were antibody labeled with anti-CD276-PE (clone MIH42, BioLegend), anti-CD47-BV421 (clone CC2C6, BioLegend), or anti-CD63-APC (clone H5C6, BioLegend) (200 µL), and data were acquired on a Fortessa XX-20 with Diva software. Analysis of the flow cytometry data was performed using FlowJo software RRID : SCR_008520 (version 10.2, BD Biosciences, USA).

We characterized the HLA class I peptide repertoire of six DIPG patient-derived cell cultures (SU-DIPG27, SU-DIPG35, SU-DIPG38, SU-DIPG4, SU-DIPG58, and SF7761) using immunopeptidomics (55) ( Figure 1A ). To our knowledge, this represents the first detailed description of the immunopeptidome of DIPG. Prior to undertaking the antigen discovery approach, HLA genotyping was performed for all six cell lines revealing a total of 22 different HLA class I allotypes ( Supplementary Table 1 ). There were several shared allotypes among the cell lines, including HLA-A*02:01 (n = 4), A*01:01 (n = 2), A*68:01 (n = 2), B*44:02 (n = 2), and C*04:01 (n = 2). We assessed the level of HLA class I cell surface expression in the cell lines using flow cytometry. Allotype-specific antibodies such as BB7.2 (anti–HLA-A2) was used for SU-DIPG27, SU-DIPG35, SU-DIPG38, and SU-DIPG43 and GAPA3 (anti–HLA-A3) for SU-DIPG58. SU-DIPG35 demonstrated the highest detectable HLA-A2, with all cell lines showing different levels of HLA-A2 expression ( Supplementary Figure 1A ). W6/32 (anti-pan class I antibody) was used to stain for remaining HLA class I alleles across all six cell lines. As expected, HLA class I expression varied among different cell lines with the highest expression observed for SF7761 ( Supplementary Figure 1B ).

A total of 26,116 peptides were identified across all cell lines using LC-MS/MS ( Supplementary Table 1 ), of which 20,536 ( Supplementary File 1A ) were found to be non-redundant peptides of 8–12 aa in length. For the four HLA-A2⁺ cell lines, using an HLA-A2–specific antibody (BB7.2) for immunoprecipitation of HLA class I complexes resulted in identification of 7,459 HLA-A*02:01–restricted peptides ( Supplementary Table 1 ), non-redundant peptides are listed in Supplementary File 1B with the highest number being identified in SU-DIPG35 (2,671 peptides) and the lowest in SU-DIPG38 (1,418 peptides) ( Figure 1B ). All the HLA-A2 bound peptides followed canonical length distribution for HLA class I ligands, with the majority of peptides (~63%–70%) being nonamers (9 mers) ( Supplementary Figure 2A ). As expected, we observed adherence to the HLA-A*02:01 consensus binding motif for the 9-mer peptides across all cell lines, with Leu/Met at position 2 (P2) and Leu/Val at P9 ( Supplementary Figures 2B–E ). Similarly, for SU-DIPG58, the only HLA-A3⁺ cell line, 2,115 restricted peptides were identified using the allotype-specific GAPA3 antibody to immunopreciptate the HLA-A3 peptide complexes ( Figure 1B , Supplementary Table 1 ). The HLA-A3 peptides followed canonical HLA class I length distribution ( Supplementary Figure 2A ) and exhibited the canonical HLA-A3 binding motif ( Supplementary Figure 2F ) with Lys/Arg at P1 and P9 anchor positions.

In addition to HLA-A2 and HLA-A3 peptides, 16,542 peptides were identified in the pan HLA class I (W6/32) immunoprecipitations from the six DIPG cell lines, with the highest number (11,582) of peptides identified from SF7761 cells ( Figure 1B , Supplementary Table 1 ). Among the peptides identified using W6/32 immunoprecipitation, 9 mers represented 46%–57% of the entire peptide repertoire, followed by 10 mers with 14%–23% and 11 mers with 7%–15% ( Supplementary Figure 3 ). To assign HLA restriction to the peptides identified using the pan class I antibody, their binding affinities for the relevant HLA allotypes were ranked using NetMHC 4.0 (42). All HLA allotypes were assigned peptide ligands predicted to be both SBs and WBs, which were collectively used to plot the binding motif for each allele using Icelogo ( Supplementary Figures 4A–F ).

The source protein landscape of DIPG cell lines was explored to gain insight into the class of proteins contributing toward the immunopeptidome. HLA class I presented peptides that were derived from a total of 6,312 non-redundant source proteins; most of the proteins contributing to HLA class I peptides came from cytoplasmic proteins followed by nucleus and membrane proteins ( Figure 1C ). Further analysis of the immunopeptidome revealed that, in all cell lines (regardless of antibody used for immunopeptidomics), 69% to 89% of the source proteins contributed to only one peptide being presented at the cell surface ( Supplementary Figures 5A, B ).

Comparative analysis between the four HLA-A*02:01–positive SU-DIPG lines showed that there were 408 peptides shared ( Figure 1D , Supplementary File 1C ) across them. On the basis of their log2-transformed intensity observed in immunopeptidomics data, the top 40 peptides were selected, and a heatmap was plotted. As expected, some of the peptides were derived from the most abundant proteins involved in transcription (ribosomal proteins RS8, RS16, and RL19), translation (elongation and initiation factors), and filament protein such as Vimentin ( Figure 1E ).

Next, the immunopeptidomics data were cross-referenced with the HBPA database to identify peptides presented from brain-enriched proteins. A total of 1,783 brain-enriched peptides, restricted to different HLA class I allotypes, were identified ( Supplementary File 2A ). This included proteins such as neurofilament medium (NEFM) neurochondrin (NCDN), neurocan core protein (NCAN), and neuroligin-3 (NLGN3) and brevican core protein (BCAN). Of the total 1,783 peptides identified, we observed a high contribution derived from PTPRZ1 protein (77 peptides), as well as SOX proteins, ASTN1, and CD99 ( Figure 1F ). Notably, four peptides SILDIVTKV (RFTN2), AIIDGVESV (PTPRZ1), KVFAGIPTV (PTPRZ1), and NLDTLMTYV (NLGN4X) ( Supplementary Table 2 ) were also reported as immunogenic in patients with GBM providing strong evidence that these peptides should also be explored in patients with HLA-A2⁺ DIPG (29, 56). Furthermore, we identified eight peptides from PEG10 (with two peptides restricted to HLA-A*02:01) and two peptides from PEG3 proteins ( Table 1 ). Both proteins are potentially oncogenic HERV (57). Of the 1,783 peptides identified, 1,082 peptides were found previously reported in the publicly available IEDB database that may be of therapeutic significance ( Supplementary File 2B ). Additional interrogation of publicly available IEDB T-cell data revealed that, among the peptides derived from brain-specific proteins, there were 25 known CD8⁺ T-cell epitopes ( Supplementary File 2C ).

The immunopeptidome data from all six DIPG cell lines were further interrogated for the presence of known TAAs and CTAs using two different databases: Tumor T-cell Antigen Database (TANTIGEN; http://cvc.dfci.harvard.edu/tantigen) (58) and CTdatabase (http://www.cta.lncc.br/) (59). Within the combined DIPG patient cell line dataset, a total of 977 non-redundant cancer associated peptides were identified, with 891 contained within the TANTIGEN database ( Supplementary File 3 ), 80 from the CTdatabase ( Supplementary File 4A ), and six overlapping peptides ( Supplementary File 4B ) that were present in both databases ( Figure 2A ). Varying numbers of TAAs were identified across all DIPG cell lines with the most peptides (516) identified in SF7761 ( Figure 2B ). Across the entire dataset, 19 peptides have been previously reported as T-cell epitopes in the TANTIGEN and CTdatabase ( Supplementary Table 2 ) in ovarian (60) and renal (61, 62) cancers, of which 13 peptides are restricted to HLA-A2. Two peptides KIQEILTQV (IGF2BP3) and TMLARLASA (CSPG4) have been reported immunogenic in patients with GBM (56) ( Supplementary Table 2 ). Some other source proteins contributing to the highest number of TAA peptides included CSPG4 (31 peptides), NPM1 (14 peptides), and SOX proteins (11 peptides) ( Figure 2C ).

Furthermore, the investigation of source proteins contributing to CTA peptides revealed several interesting candidates. The source protein for peptide IVDPGYLGY was interleukin-13 receptor subunit alpha-2 (IL13Rα2), which has been reported to be overexpressed in 85% of DIPG cases (63) and an effective CAR T-cell target for glioblastoma tumor regression (64). Other peptides of interest included peptides previously reported as immunogenic and derived from PReferentially expressed Antigen in MElanoma (PRAME), Melanoma associated antigen A1 (MAGEA1), and Ephrin receptors proteins ( Table 1 , Figure 2D ).

Of the 977 total CTA- and TAA-derived potential immunogenic peptides identified, 12 peptides from proteins including IL13Rα2, PRAME, and MAGE family were selected ( Table 2 ) and validated by comparing their LC-MS/MS characteristics to their synthetic counterparts. Eleven comparative synthetic and endogenous peptide MS/MS fragmentation patterns matched with high Pearson correlation values ranging from 0.92 to 0.99 ( Supplementary Figure 6 ) with a difference in retention time of ±3 min. Representative mirror plots for two peptides of interest from proteins IVDPGYLGY (IL13Rα2) and SLLQHLIGL (PRAME) solidify the data outlined above ( Figure 3 ). This dataset provides an ideal starting point for pHLA targeting therapeutics for patients with DIPG.

To understand the systemic dysregulation occurring and driving DIPG, protein expression and patient-derived DIPG cell cultures (SU-DIPG27, SU-DIPG35, SU-DIPG43, SU-DIPG58, and SF7761) were investigated and compared to non-diseased PB. Whole proteomics was performed on cell lysate depleted of pHLA using S-TRAP technique followed by enzymatic digestion with trypsin and LC-MS/MS analysis. A total of 6,947 proteins were identified across the entire dataset at 1% FDR (cell lines and PB samples). The number of proteins identified per sample ranged between 1,247 and 4,143 ( Figure 4A , Supplementary Files 5A–F ). Comparative analysis between DIPG cell lines and brain samples showed that 820 proteins are shared between all samples, whereas 626 proteins were only identified in non-diseased PB and not detected in DIPG cells ( Figure 4A ). Pathway analysis of the 820 shared proteins was performed using the Cytoscape plug-in ClueGO (50). Importantly, pathways related to axon and synapse development were found to be significantly enriched against the background GO term datasets ( Supplementary Figure 7 ), reflecting the neuronal origin of the DIPG cell lines and reiterating their relevance for studying pediatric brain tumors.

Furthermore, the peak areas expressed as LFQ values (maxLFQ) for proteins were transformed and filtered resulting in robust quantification of 4,996 proteins across the dataset ( Supplementary File 6A ). To identify proteins enriched in DIPG cell lines compared to PB, the transformed LFQ values were used, and Student’s t-test was performed using Perseus software. On the basis the test, the log fold change and P-value were used to make volcano plot depicting proteins enriched in DIPG cell lines. Of the 4,996 proteins, a total of 2,025 proteins were found to have higher expression in DIPG cell lines ( Supplementary File 6B ). Notably, analysis of the DIPG enriched proteins performed using GO terms revealed that majority of them were involved in pathways related to nucleobase metabolism, mRNA synthesis and processing, cell cycle, and histone modification ( Figure 4B ). Proteins that had the highest log fold change in DIPG cell lines were Chromodomain Helicase DNA Binding Protein 4 (CHD4), Flap endonuclease 1 (FEN1), DEAD-box helicase 46 (DDX46), X-Ray Repair Cross Complementing 6 (XRCC6), Non-POU domain containing octamer binding (NONO), and Serine Hydroxymethyltransferase 2 (SHMT2) ( Supplementary Figure 8 ). These proteins are known chromatin remodelers and play a role in DNA repair after damage.

In addition, histone proteins (H2A, H2B, and H3) were found to be more abundantly expressed in the DIPG cell lines, along with proteins involved in the antigen processing and presentation pathway including HLA-A, HLA-B, and HLA-C along with TAP1, TAP2, TAP binding protein (TABBP) and proteasome activator and regulatory subunits such as Proteasome Activator Subunit 1 (PSME3), 26S proteasome non-ATPase regulatory subunit 7 (PSMD7), and 26S proteasome non-ATPase regulatory subunit 11 (PSMD11) ( Supplementary Figure 8 ).

To identify cell surface proteins, we cross-referenced our data with the CSPA database and found a total of 273 (18%) nominal cell surface proteins enriched in our DIPG dataset ( Supplementary File 6C ). This included cluster of differentiation (CD) markers along with extracellular matrix (ECM) proteins such as Integrins and Fibronectin (FN1). Next, we sought to identify proteins or pathways in DIPG that could be a potential drug target that we used as HPA druggable proteome (a list that comprises of 812 proteins that are targets of current FDA-approved drugs). Of the 2,025 proteins enriched in DIPG cell lines, we identified 44 proteins that could be potential targets of various drugs ( Supplementary File 6D ). As expected, some of the proteins identified included known drug targets such as histone deacetylases (HDAC1 and HDAC2), cyclin dependent kinases (CDK4 and CDK6), and other proteins like Poly(ADP-Ribose) Polymerase 1 (PARP1) and DNA (cytosine-5)-methyltransferase 1 (DNMT1). Another interesting target identified was the TSPO -(Translocator protein) that is known to be expressed exclusively in brain cancers, including GBM (65), and can be used for delivering drugs across the blood brain barrier (66) along with DNA topoisomerase I (TOP1), Lipoprotein Lipase (LPL), and DNMT1 ( Figure 4C ). Some other proteins of interest include the surface protein FN1 and the enzyme butyrylcholinesterase (BCHE), both being known as unfavorable prognostic markers for different cancers (67–69) along with EphA2.

Immunotherapies such as bispecific antibodies and CAR T cells require the identification of cancer-specific cell surface proteins. Therefore, mapping the DIPG cell surface proteome allowed the identification of potential and novel CAR targets. To our knowledge, this represents the first detailed description of the surfaceome of DIPG. To capture DIPG cell surface proteins, we biotinylated surface glycoproteins with amino-oxy-biotin to facilitate streptavidin pulldown and identification by LC-MS/MS. Seven DIPG cell lines, namely, SU-DIPG4, SU-DIPG13, SU-DIPG17, SU-DIPG18, SU-DIPG21, SU-DIPG25, and SU-DIPG33, were processed for cell surface protein identification ( Figure 1A ). A total of 3,033 proteins were found to be expressed on the surface of 7 SU-DIPG cell lines ( Supplementary File 7A ). Of the 3,033-proteins, a total of 799 shared proteins were detected across all seven cell lines ( Supplementary File 7B ). The top 50 cell surface proteins based on spectral intensity identified using this approach included proteins belonging to family of solute carrier (SLC) such as SLC3A2, SLC1A5, SLC16A1, SLC1A4, SLC6A8, SLC12A2, and SLC29A1 proteins along with several CD markers ( Figure 5A ). When cross-referenced with CSPA, 421 proteins were identified ( Supplementary File 7C ), which included nine CD markers including CD276 (B7-H3), CD47, and CD63. Some other targets of interest include EGFR protein, Human epidermal growth factor receptor 3 (HER3), and Ephrin proteins (from both EphA and EphB) along with IL13αR.

CD276 is an immune checkpoint protein that is expressed at a high level across all seven DIPG cell lines. There is increasing evidence that suggest that it is a good immunotherapy target for pediatric brain tumors (70). Indeed, there is a CD276 targeting CAR T-cell immunotherapy phase I clinical trial investigation (BrainChild-03) in pediatric CNS tumors open and currently recruiting (NCT04185038) (21).

The expression of CD47 is known to facilitate immune evasion by tumors by escaping phagocytosis by macrophages (71), and antibody-mediated inhibition of CD47 is a clinically active of area of immunotherapy (72), with Gilead’s Magrolimab being the most clinically advanced (NCT04313881). CD63 is a lysosomal and exosomal-associated membrane protein of the tetraspanin family that is the only known membrane receptor to interact with TIMP-1 an inhibitor of MMPs such as ADAM-10 (73).

Next, we validated the cell surface protein expression of some selected cell surface proteins including CD276, CD63, and CD47. A panel of five DIPG cell lines (SU-DIPG4, SU-DIPG13, SU-DIPG17, SU-DIPG21, and SU-DIPG33) was screened using surface antibody labeling and flow cytometry ( Figures 5B–D ). When compared to the control samples, all SU-DIPG cell lines labeled strongly with antibodies particularly SU-DIPG4 and SU-DIPG13 which had high CD47 expression, whereas SU-DIPG33 had the highest CD276 expression, validating our proteomics findings.

Furthermore, to identify membrane proteins in the dataset, we parsed the 800 shared proteins through the TMHMM server, and, based on a PredHel score of 3 or more, we identified the top 75 proteins ( Supplementary File 7D ). This also included two CD markers (CD47 and CD63) along with several SLC proteins associated with neuronal pathophysiology (74).