Immunogenic epitopes containing missense mutations can still be generated through MHC-I pathwayWe looked for missense mutations on the TCGA-GDC cancer database, and we found that the EGFR, IDH1, PTEN, and TP53 proteins were the most frequently mutated proteins in GBM. A total of 49, 5, 57, and 64 missense mutations were uncovered for EGFR, IDH1, PTEN, and TP53, respectively (Supplementary Table S1). From this list of mutations, 4 occurred in immunogenic epitopes of EGFR, 1 in IDH1, 2 in PTEN, and 22 in TP53 (Supplementary Table S2). Immunogenic epitopes were recovered from IEDB.Next, we used the combined predictors of proteasomal processing, TAP transport, and MHC binding to score and identify the probability of presentation of the mutated epitopes in selected proteins. The total score for wild-type and mutated epitopes and their respective upstream/downstream epitopes were compared (Supplementary Figs. S1 and S2). Our analysis revealed no significant differences in the mean total scores of mutated epitopes compared to their respective wild-type counterparts. This suggests that the missense mutations do not impact the presentation of these epitopes. However, when directly comparing the pair of wild-type and mutated epitopes we found that in 39 pairs, the wild-type and mutated epitope total scores were below the mean total score, and the difference in the total score between the wild-type and the mutated epitope was higher than 0.5 (arbitrary value). Because this result suggests that the mutation might influence the peptide generation in the processing pathway, these 39 pairs were not considered for downstream analysis. We selected the 83 remaining pairs for further structural analysis.Structural analysis of MHCs in the context of wild-type and mutated epitopesAs explained in the Materials and Methods section, we kept the same MHCflurry cutoff value for both wild-type and mutant epitopes. Because HLA-Arena models only pMHC-I complexes that pass through MHCflurry cutoff, there were cases where the pair (wild-type/mutant) was not modeled (Supplementary Table S3). In our case, 24 pairs out of 83 were modeled in the context of different HLA molecules (Table 1) and used for downstream analyses.
Table 1 Epitopes with their respective alleles modeled through structural analysis using the HLA-Arena platform. The mutated amino acids are highlighted (missense) according to the GDC-Cancer.To evaluate the similarity of the TCR-interacting surface of the remaining 24 pMHC-I complex pairs, we performed a hierarchical clustering analysis (HCA) of the electrostatic potential collected from 46 regions of interest (Supplementary Fig. S3). The use of HCA for selection of similar pMHC-I pairs based on electrostatic features was validated elsewhere19,20,21. Figure 1A shows the HCA for TP53-derived peptides, while Fig. 1B illustrates the image similarity between two distinct peptide-MHC class I (pMHC-I) pairs (wild-type/mutant). Although qualitative differences and similarities in the pMHC-I complexes can be observed, the HCA offers a quantitative approach to emphasize these characteristics more distinctly, at the same time allowing the clustering of similar pairs. For TP53 epitopes, 8 pairs of pMHC complexes were clustered together.Figure 1HCA of the electrostatic potential and exemplification of their electrostatic and structural similarity. (A) Dendrogram of the TP53 protein demonstrating the pairs of pMHC complexes, the red rectangles indicate the pairs preserving structural characteristics. (B) Top view of pMHC complexes highlighting the electrostatic similarity between wild-type and mutated epitopes from TP53 protein (red, white, and blue represent negative, neutral, and positive charges, respectively). AU, Approximately Unbiased; BP, Bootstrap Probability.The HCA results for IDH1 proteins showed a different pattern, positioning the wild-type epitope on a distant branch compared to the mutated epitopes. Since we are interested in the pair of epitopes where the wild-type is structurally similar to the mutated counterpart, we did not pursue additional analysis on IDH1 epitopes (Fig. 2).Figure 2Dendrogram illustrating the HCA of the IDH1 protein, focusing on four pMHC complexes: WVKPIIIGRHAY-WT, WVKPIIIGHHAY-R132H, WVKPIIIGGHAY-R132G, and WVKPIIIGCHAY-R132C. The analysis reveals distinct clustering, with the mutated pMHC complexes R132G and R132C forming one cluster, followed by the appearance of R132H. The wild-type (WT) complex is positioned as the most distant group, highlighting the significant variance from the mutated forms.Filtering targeted epitopes for GBM therapyTo pursue new targets for GBM therapy, we selected only the epitope pairs that presented better-predicted results in all analyses (i.e., sequence-based prediction and HCA–based structural analysis). The rationale behind this approach is to select epitopes where the described mutation has a lower chance of affecting the CD8 T cell-specific immune response. In total, six TP53 epitopes stand out (Table 2). Since immunogenicity is crucial, we submitted each pMHC-I to an immunogenicity prediction tool (TLImm)17. We observed that the scores were similar to or even better (3 out of 6 cases) than the well-known immunogenic wild-type peptide.
Table 2 Pairs of epitopes were selected due to their better-predicted results in all analyses (i.e., processing pathway and presentation of MHC-I antigens, modeling of pMHC-I complexes, and hierarchical clustering).To further explore structural features in a dynamic environment, we decided to run 100 ns molecular dynamics simulation for all 12 pMHC-I complexes selected from Table 2. We extracted the main quantitative features from each simulation, such as Root Mean Square Deviation (RMSD), Root Mean Square Fluctuation (RMSF), Radius of Gyration (RoG), and mean contacts between the epitope and the MHC-I (Table 3).
Table 3 Epitopes resulting from the prediction showed a strong binding affinity for MHC-II.One of the measures reflecting pMHC-I stability is the RMSD. In most cases, the average RMSD for both the protein and epitope was lower for the mutated peptide (Table 3). The effect size of this difference, quantified using Cohen’s d descriptor, ranged from medium to huge (see details in the Methods section). In all simulations, the RMSD reached a plateau before the 100 ns mark, indicating that the simulation time was appropriate (Fig. 3). Interestingly, the density plot analysis of the protein RMSD revealed primarily single peaks, except for the HLA-A*02:01-LLGWNSFEV complex, which displayed two distinct peaks. When we analyzed the epitope RMSD we could observe that more pMHC-I complexes showed multiple peaks, likely because the calculation involved only 9 to 10 residues (Supplementary Fig. S4). Nonetheless, the epitope RMSD remained stable throughout the simulation in all cases.Figure 3Protein RMSD (in Å) for 6 selected pMHC-I pairs. The graphs on the left show the RMSD along time (in nanoseconds), while the graphs on the right shows the RMSD density plot for the whole simulation. Wild-type peptides are colored in blue, while the mutated counterpart is colored in orange. The mutated residue is shown in green.We were also interested in evaluating the free energy surface (FES) landscape for each of the pairs analyzed (Figs. 4 and 5). The FES can be used to understand the stability and conformational changes of molecular systems because it is derived from RMSD and RoG values. Complexes HLA-A*02:01-ALNKMFCQL/ALNNMFCQL and HLA-A*23:01-EYLDDRNTF/EYLDDRNIF show a slightly different distribution of energy minima, but FES plots support the hypothesis that the mutation in these cases does not drastically destabilize the peptide-MHC complex. Other complexes show a more complex scenario, indicating that the mutation can introduce alternative conformations (e.g., HLA-A*02:01-LLGRNSFEV/LLGWNSFEV, which aligns with the protein RMSD plots) or additional flexibility (e.g., HLA-B*07:02-ALNKMFCQL/ALNNMFCQL, consistent with the epitope RMSF values in Table 3). Still, and supported by RMSD analysis, the mutation was not sufficient to destabilize any of the pMHC complexes analyzed.Figure 4Free energy surface (FES) landscape (left) and lowest energy structures retrieved from simulation (right) for complexes HLA-A*02:01-ALNKMFCQL/ALNNMFCQL, HLA-A*02:01-LLGRNSFEV/LLGWNSFEV, and HLA-A*23:01-EYLDDRNTF/EYLDDRNIF. The FES plots depict the stability and conformational changes of the peptide-MHC complexes. In the middle, the MHC structures are shown with the peptide atoms highlighted, where the mutated residues are marked in green. On the right, the electrostatic potential surface of the peptide-MHC (pMHC-I) complex is displayed, illustrating the potential impact of mutations on the complex’s surface properties.Figure 5Free energy surface (FES) landscape (left) and lowest energy structures retrieved from simulation (right) for complexes HLA-A*24:02-TYSPALNKMF/TYYPALNKMF, HLA-B*07:02-RPILTIITL/RPILTISTL, and HLA-B*57:01-LAKTCPVQLW/LTKTCPVQLW. The FES plots depict the stability and conformational changes of the peptide-MHC complexes. In the middle, the MHC structures are shown with the peptide atoms highlighted, where the mutated residues are marked in green. On the right, the electrostatic potential surface of the peptide-MHC (pMHC-I) complex is displayed, illustrating the potential impact of mutations on the complex’s surface properties.We sought to determine if the electrostatic potential would change when comparing the modeled pMHC-I models against the lowest energy structures generated during the molecular dynamics (MD) simulation (Supplementary Fig. S5), and we also compared the lowest energy structures to the wild-type/mutated pairs. We observed primarily topographical modifications, along with differences in the distribution of charges around the TCR-interacting surface for some of the pMHC-I complexes. For instance, in the HLA-A*02:01-LLGRNSFEV/LLGWNSFEV complex, the substitution of Arginine with Tryptophan resulted in a loss of positive charges (Fig. 4). In the HLA-B*07:02-RPILTIITL/RPILTISTL complex, despite the modification occurring at the C-terminus of the epitope, a new positive charge emerged at the N-terminal region near residue 2 (Fig. 5). Additionally, the HLA-B*57:01-LAKTCPVQLW/LTKTCPVQLW complex exhibited significant topographical changes, likely due to the need to accommodate the more hydrophilic Threonine residue in the 10-mer peptide (Fig. 5). This adaptation resulted in a pronounced negative charge at the N-terminal part of the epitope and a gradual loss of the negative charge at the C-terminus.Finally, we aimed to gain a better understanding of the contact map during the simulation between the peptide (wild-type/mutant) and the respective MHC-I (Fig. 6). We focused on contacts within the range of 0.4 nm (4 Å), a distance that captures key interactions such as hydrogen bonds, van der Waals interactions, hydrophobic contacts, and potential salt bridges. These interactions are crucial for the stability and specificity of the peptide-MHC-I complex and can provide insights into the effects of mutations on peptide binding and presentation.Figure 6Contacts performed in the range of 0.4 nm between peptide and MHC-I. The line graphs represent the number of contacts performed along the simulation (wild-type peptide in blue and mutated peptide in orange). The circular alluvial plots show the residues interacting with the wt/mutated residue. The links are in blue (wt residue) or orange (mut residue). In each sector, the color blue, orange, or purple represent the MHC residues that are being contacted by the wt residue, by the mut residue or by both, respectively.We observed that the contacts tended to increase or remain stable in most of the pMHC-I complexes, with the exceptions illustrated in Fig. 6A,F. It is clear that most of the time, the same MHC-I residues were contacted by both wild-type and mutated residues throughout the simulation. However, in some specific cases, like the HLA-A*02:01-LLGRNSFEV/LLGWNSFEV complex (Fig. 6B), MHC-I residues were mainly contacted by the mut residue. One particular noteworthy case is the HLA-A*24:02-TYSPALNKMF/TYYPALNKMF complex (Fig. 6D). In this case, the number of contacts between the mutated epitope and the MHC-I was five times greater than those of the wild-type peptide. This significant increase suggests that the mutation greatly enhances the interaction between the epitope and the MHC-I. Additionally, some key contacts were unique to the mutated epitope and MHC-I, involving residues such as HIS114 and MET97, which were not observed in the wild-type interactions. These unique contacts may positively contribute to binding affinity and specificity observed in the mutated complex.MHC-II binding analysis of the selected epitopesBecause of the importance of CD4 T cells in helping CD8 T cells and exerting antitumor immune response22, we used a sequence-based MHC-II peptide predictor (https://services.healthtech.dtu.dk/services/NetMHCII-2.3/) to rank the binding capacity of the epitopes selected on Table 3. As a result, we obtained 9 epitopes (wild-type and mutated) that showed strong binding affinity to the alleles described for MHC-II (Table 3). Alleles were selected according to epitope prediction based on HLA class II binding in the human population23. These data indicate that the selected epitopes can also bind MHC-II inducing a CD4 T cell response.
Mutations in glioblastoma proteins do not disrupt epitope presentation and recognition, maintaining a specific CD8 T cell immune response potential
