Background
Gliomas are the most prevalent brain tumor with dismal outcomes, with a median survival time of 12–15 months after diagnosis and a 5-year survival rate of no more than 3% [
1,
2]. One factor contributing to gliomas’ refractoriness is relatively low immune responses, namely responses that are “immunologically cold” or immunosuppressive. Although some of immunotherapy agents like depatuxizumab [
3] manifested curing effects in recurrent glioblastoma patients, other agents including bevacizumab [
4,
5] and PD-1/PD-L1 antibodies such as nivolumab [
6,
7] and the like, which have been proven effective for the treatment of other tumors, are of limited value for primary glioma patients. Besides, although chimeric antigen receptor T cell (CAR-T) therapy was approved for the treatment of B-cell malignancies by the American Food and Drug Administration, it does not have a satisfying effect on solid tumors such as gliomas, which is partly due to the immunosuppressive microenvironment inside solid tumors [
8]. However, the personalized tumor vaccine that is based on neoantigens produced by gliomas provides a new avenue for glioma treatment [
9]. However, identification of the neoantigens appropriate for use in CAR-T therapy for gliomas is challenging.
In most situations, cancers arise from an accumulation of genetic mutations or damage to DNA [
10]. Neoantigens derive from these non-synonymous genetic mutations, including single nucleotide variations (SNV), chromosomal deletions and insertions, gene fusion, and alternative splicing [
11]. As the presented neoantigens might elicit T-cell-mediated anti-tumor immunity specifically against tumor cells, they are considered promising immunotherapy targets [
12]. Recently, neoantigens derived from non-coding regions of RNAs have received attention [
13,
14]. Of these, long-non-coding RNAs (lncRNAs) are especially notable for their crucial role in modulating immune responses [
15,
16] and their ability to translate short peptides [
17]. LncRNAs have also been reported to translate neoantigens that can be presented by major histocompatibility complex class 1 molecules (MHC I), which contributes to essential cellular immunosurveillance [
17] and expands the range of immunopeptidomes that can be targeted for immunotherapy [
18].
The translation of lncRNAs can be modulated by many factors, including small or short open-reading frames (smORF or sORF) [
19], eIF4E [
20], and N6-methyladenosine (m6A) modification. lncRNAs lack canonical ORF codes for over 100 amino acids [
19]. Therefore, they are considered untranslatable until sORF or smORF, which encode small peptides, are identified inside them [
21]. eIF4E is a translation initiation factor that weakly binds to 5′ caps of RNA after phosphorylation, which induces inhibition of mRNA translation and facilitates interaction between lncRNA and ribosomes [
22]. Additionally, m6A modification has been proven to affect mRNA translation [
23]. Studies have shown that m6A-modified sites serve as translation initiation sites for circular RNA [
21]. Recent studies have provided further evidence that lncRNA translation that produces micropeptides is also affected by m6A modification [
24]. Moreover, TransLnc [
25] and LncPep [
26], two recently created databases, have recorded the potentially translatable lncRNAs that have been identified through experimental evidence or algorithmic deduction. Both databases use m6A modification as an index in the evaluation of lncRNA translation potential. By calculating micropeptides' binding affinities to MHC I and MHC II, TransLnc also evaluated their potential to be presented by MHC complex so that they might act as neoantigens. Results indicate that m6A plays an important role in both lncRNA translation and neoantigen production from lncRNA. Moreover, select RNA modification processes, such as pseudouridine (Ψ), N1-methyladenosine (m1A), and 5-methylcytosine (m5C), also participate in the modulation of mRNA translation, the launching of immune responses [
27,
28], and the modification of lncRNA [
29‐
32]. However, whether these non-m6A modifications are related to lncRNA-mediated immune processes and neoantigen production has yet to be explored.
Here, we investigate the signatures of neoantigen-coding lncRNAs that could be related to m6A or non-m6A modifications in the TransLnc database. We compared two sets of lncRNAs by analyzing gene expression data of gliomas from The Cancer Genome Atlas (TCGA) and discovered that the signature of non-m6A-related neoantigen-coding lncRNAs has better efficacy in the prognostic model. We then established a scoring model using a clustering model of the non-m6A-related neoantigen-coding lncRNAs’ signature. This score is highly correlated with immune infiltrations, glioma cell development, glioma patient prognoses, and tumor mutation burden (TMB) of T-cell positive regulators. After investigating the correlation between this score and gliomas’ T cell receptor (TCR) repertoires, we found that the score model is positively correlated with the expression levels of widely expressed neoantigen-active TCR clonotypes. In summary, these results indicate that the model of non-m6A-related neoantigen-coding lncRNAs is a promising tool for determining glioma patient prognoses, and it also provides widely targetable T cell clonotypes for potential CAR-T therapy for the treatment of gliomas.
Methods
Data preparation
mRNA sequencing data were downloaded from TCGA database (
https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga), and validation data were downloaded from the Chinese Glioma Genome Atlas (CGGA) database (
http://www.cgga.org.cn/). Two datasets from the CGGA, “mRNAseq_325” and “mRNAseq_693”, were employed; these are labelled herein as CCGA325 and CCGA693, respectively. There were 169 glioblastoma multiforme (GBM) samples and 529 low-grade glioma (LGG) samples in the training cohort of TCGA dataset, and there were 139 GBM samples and 186 LGG samples in the CGGA325 dataset. Also, 249 GBM samples and 444 LGG samples were included in the CGGA693 dataset. Verhaak classification was performed, as described above [
33].
Single-cell sequencing (scSeq) data were obtained from the Gene Expression Omnibus (GEO) database (
https://www.ncbi.nlm.nih.gov/geo/), from which GSE84465 was selected for analysis. This dataset contained 3589 cells. Furthermore, another dataset from the CGGA database containing 6148 cells was also included. GSE129671 was used for SCENIC analysis. The “Seurat” package from R was applied to normalize the count data from these datasets, and the “FindMarkers” function was used to identify unique gene markers in every cluster. Cell distribution was displayed using the “tSNE" or “umap” functions. The cells were annotated as described in a study that also used the GSE84465 dataset [
34] after undergoing slight modifications.
TCR sequencing data were acquired from the GEO database, and two datasets, GSE79338 and GSE188620, were used for analysis. The GSE79338 dataset contained TCR data from normal brain tissue as LGG and GBM were used to explore unique TCR patterns or clonotypes in gliomas as compared to normal brain tissue in this study. Two samples, GBM09 and GBM13, were excluded as they had relatively fewer TCR clonotypes than other GBM samples. The GSE188620 dataset contained TCR sequencing data before and after GBM-cell lysate vaccination, and it was used in this study for the validation of selected TCR patterns.
Neoantigen activation score (NAS) model
In accordance with the literature, the high- to intermediate-level modifications (Ψ and m5C) and ultra-low-level modifications (m1A) were used as the main non-m6A modifications in this study [
35]. Their regulators were identified in previous studies [
36‐
38]. The cluster model was constructed using the “ConsensusClusterPlus” package in R. We used “km” (k-means) as the cluster algorithm and “euclidean” as the distance function. Two clusters, scored with continuous numbers between 2 and 9, were considered to have the best clustering results that showed the highest clustering reliability (Additional file
2: Fig. S2A).
Next, the differential expression genes (DEGs) were identified by comparing cluster 1 to cluster 2 using the “limma”, “edgeR” and “DESeq2” packages in R with a threshold of false discovery rate (FDR) < 0.05 and |log2(fold change)|> 1 (Additional file
12: Table S1 and Additional file
13: Table S2). For DEGs of CGGA693 in non-m6A-related NAS model, no gene could meet this threshold in limma and DESeq2 analyses so we used FDR < 0.05 and |log2(fold change)|> 0.4 in edgeR to keep the number of DEGs similar to CGGA325 dataset. Then the DEGs were confirmed as the intersection of these three parts. A univariate cox regression was then conducted in order to identify the genes associated with significant survival outcomes. Following this, the principal components of these significant genes were calculated, and TCGA samples’ NASs were calculated using the following formula according to a previous study [
34].
$$\mathbf{N}\mathbf{A}\mathbf{S}=\mathbf{k}\mathbf{*}\sum \left({\mathbf{G}\mathbf{e}\mathbf{n}\mathbf{e}}_{\mathbf{H}\mathbf{R}>1}\mathbf{*}\left(\mathbf{P}\mathbf{C}1+\mathbf{P}\mathbf{C}2\right)\right)-\mathbf{k}\mathbf{*}\sum \left({\mathbf{G}\mathbf{e}\mathbf{n}\mathbf{e}}_{\mathbf{H}\mathbf{R}<1}\mathbf{*}(\mathbf{P}\mathbf{C}1+\mathbf{P}\mathbf{C}2)\right)$$
GeneHR>1 and GeneHR<1 represent the expression levels of genes with a hazard ratio (HR) that is higher or lower than 1 in survival analyses, respectively. Gene expression levels are in fragments per kilobase of transcript per million mapped reads (FPKM). To minimize the NAS value without altering its prognostic efficacy, k is set as 0.0001 to keep most of the NAS value ranging from ten to thousands.
After this, the characteristics of cluster 1 and 2 were identified through the use of the support vector machine (SVM) that is embedded in R package “e1071”, indicated by the function “svm”. For parameters of this function, the kernel “radial” was applied, and k-fold cross was set to 10. The randomly selected 520 samples (75% of all) in TCGA data were used as training phase data while the rest 174 samples (25% of all) were used as testing phase data. There were two clusters resulting from SVM-based prediction: cluster 1 and cluster 2. The cluster model was then reproduced in the CGGA325 and CGGA693 datasets using the “predict” function. The NASs were calculated through a similar process.
Overall survival outcome prediction
The TCGA, CGGA325, and CGGA693 samples were divided into cluster 1 and cluster 2 and into high and low NAS groups by the cluster model or by the NAS. Kaplan-Meier analysis was used to determine the different groups' overall survival time. The receiver operation characteristic (ROC) curve was generated, and the area under the curve (AUC) was calculated for each model.
Gene function enrichment
Gene set variation analysis (GSVA) enrichment of gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) was performed using the “GSVA” package in R. The GO and KEGG enrichment analysis was conducted using the “clusterProfiler” package in R. This package was also employed for gene set enrichment analysis (GSEA) of the DEGs between high and low NAS groups.
Immune microenvironment analysis
The “ESTIMATE” package in R was applied to analyze the infiltration ratio of immunocytes and stromal cells and to estimate the immune score and tumor purity within every TCGA, CGGA325, and CGGA693 sample. The immune cell components in the microenvironment were analyzed using the CIBERSORT algorithm.
RNA velocity and inter-cell communication analysis
RNA velocity was analyzed using the “scVelo” and “velocyto” packages in Python 3.8.8. This analysis revealed the evolutionary pathways of tumor cells based on RNA velocity. Intercellular communication was analyzed using the “celltalker” package in R, and differential ligand-receptor pairs were extracted.
Transcription factors analysis
For the RNA sequencing (RNA-seq) data, Expression2Kinases (X2K,
https://maayanlab.cloud/X2K/) was employed to compare upstream transcription factors of DEGs between low and high NAS groups in TCGA, CGGA325, and CGGA693 datasets. For the scSeq data, we applied pySCENIC algorithm to construct the transcription factors’ regulatory network. The human transcription factors data were downloaded from cisTarget databases (
https://resources.aertslab.org/cistarget/) for the network construction. The activation of transcription factors was estimated using the “AUCell” package in R.
Least absolute shrinkage and selection operator (Lasso) analysis
For the simplify of NAS calculation, lasso analysis was performed with R package “glmnet”. The datasets of TCGA, CGGA325 and CGGA693 were merged into one by their common DEGs. For the parameters of “glmnet” function, we used family = "cox", alpha = 1, nlambda = 100.
TCR-neoantigen peptide pairs identification
The TCR clonotypes were clustered into patterns using the GLIPH2 algorithm (
http://50.255.35.37:8080/), and the binding probability between selected TCR clonotypes and potential neoantigen peptides was analyzed using the DLpTCR algorithm (
http://jianglab.org.cn/DLpTCR/). Differences in the clonotypes before and after vaccination were analyzed using the “scRepertoire” package in R.
Patients and tissue specimens
For immunohistochemistry staining, the paraffin-embedded glioma tissues of WHO grade II (n = 3), III (n = 3) and IV (n = 3) were acquired from glioma patients who underwent surgery in Department of Neurosurgery, Qilu Hospital of Shandong University. And the normal brain tissues (n = 3) were from craniocerebral trauma patients whose normal brain must be partially resected for decompression. And for the quantitative real-time PCR, the frozen glioma tissues for RNA extraction of WHO grade II (n = 4) and IV (n = 4) were also acquired from glioma patients in Department of Neurosurgery, Qilu Hospital of Shandong University.
Immunohistochemistry (IHC) staining
The formalin-fixed and paraffin-embedded tissues were sectioned into 4 μm slices. Antigen retrieval was conducted in boiled sodium citrate buffer (pH 6.0). Then endogenous horseradish peroxidase was blocked with 3% H2O2 and the tissues were blocked with 10% normal goat serum. Then the slides were incubated with primary antibodies targeting TMSB10 (Elabscience #E-ab-15878, 1:100 dilution), vimentin (VIM) (Cell Signaling #5741S, 1:200 dilution) and PD-L1 (Invitrogen #14–5982-82, 1:100 dilution). Then target protein was visualized with DAB with standard protocols. The cell nuclei were stained with hematoxylin. Then images were obtained with an Olympus inverted microscope.
Cell lines and cell culture
LN229, U118MG, A172, U251MG and Jurkat cells were obtained from Culture Collection of the Chinese Academy of Sciences. The GBM#P3, GBM#BG5 and GBM#BG7 were patient-derived glioblastoma stem-like cells isolated from glioblastoma specimens and were functionally characterized [
39,
40]. The LN229, U118MG, A172, U251MG glioma cell lines were cultured in Dulbecco’s modified Eagle medium (Macgene, #CM10017) with 10% fetal bovine serum (FBS). The Jurkat cell was cultured in RPMI-1640 (Macgene, #CM10041) with 10% FBS. The GBM#P3, GBM#BG5 and GBM#BG7 were cultured in Neurobasal™ medium (Gibco/Thermo Fisher Scientific, # 21,103,049) supplemented with 10 ng/ml basic fibroblast growth factor (bFGF; PeproTech, #100-18B), 20 ng/ml epidermal growth factor (EGF; Thermo Fisher Scientific, # PHG0311L) and 2% B-27™ Neuro Mix (Thermo Fisher Scientific, # A1895601).
Quantitative real-time PCR
Total RNA from cells or tissues was extracted with the RNA-Quick Purification Kit (#RN001, ESscience Biotech). 1000 ng of total RNA was synthesized into cDNA with Hifair
® III 1st Strand cDNA Synthesis SuperMix (Yeasen Biotechnology, #11141ES10). Then Hieff
® qPCR SYBR Green Master Mix (Yeasen Biotechnology, #11201ES03) was applied for amplification in the quantitative real-time PCR. The volume of reaction mixture was 10 μl and the reaction procedure was set according to the manufacturer’s two-step protocol. The primer sequences were listed in Additional file
14: Table S3.
Cell counting kit-8 (CCK-8) assay
The Jurkat cells were seeded into 96-well plates at a density of 1 × 105 per well with BAPTA-AM at 0, 10, 20 and 40 μM. Then CCK-8 assay was conducted following manufacturer’s protocol (Yeasen Biotechnology, #40203ES76) at the time of 0, 24, 48 h after seeding. The OD values were measured at the wave length of 450 nm.
Calcium colorimetric assay
Jurkat cells were seeded into 6-well plate at a density of 5 × 105 per well and were treated with BAPTA-AM at 0, 10, 20 and 40 μM for 48 h. Then calcium colorimetric assay was performed with Calcium Colorimetric Assay Kit (Beyotime Biotechnology, #S1063S) following manufacturer’s protocol. Briefly, after counting under microscope, cells were collected by 600 g centrifugation of 5 min. Then cells were treated with lysis buffer and the lysate were centrifugated again at 12000 g for 5 min. Then the supernatant was extracted and used for calcium colorimetric assay to detect calcium concentration. Then calcium mass per million cells was calculated.
Co-culture and ELISA assay
Jurkat cells were seeded into 6-well plate at a density of 5 × 10
5 per well. And they were activated by adding 2 μg/ml soluble anti-CD3 (Proteintech, #60,181–1-Ig) and 1 μg/ml soluble anti-CD28 (Proteintech, #65,099–1-Ig) for 24 h, as previously described [
41]. Then activated Jurkat cells were seeded into 24-well plate at a density of 1 × 10
5 per well and BAPTA-AM was added to activated Jurkat cells at 0, 10, 20 and 40 μM for 24 h. At the same day of BAPTA-AM treatment, the LN229 cells were seeded into 96-well plates at a density of 3 × 10
3 per well. After BAPTA-AM treatment, Jurkat cells were added to LN229 cells at 2 × 10
3 per well. After 48 h co-culture, the supernatant was extracted for ELISA assay detecting IFN-γ (Proteintech, #KE00146). The remaining LN229 cells were used for CCK-8 assay.
Statistical analysis
Shapiro-Wilk test was applied to normality test. The Mann-Whitney test was used to compare two groups of data that did not subject to normal distribution and Student’s t-test was employed to compare two groups of data that did. And the ANOVA test was employed to conduct multiple comparisons. The log-rank test was used to determine the overall survival outcome, and the Spearman test was used to analyze the correlations between two sets of data. Analyses were conducted using R (version 4.1.3), Python (version 3.8.8), and GraphPad Prism (version 8.3.0).
Discussion
In this study, we investigated neoantigen-coding lncRNAs using the TransLnc database, and we found that neoantigen-coding lncRNAs related to non-m6A modifications, including pseudouridine, m5C, and m1A, have the potential to play a role in the prediction of glioma patients’ prognoses. The cluster model that was based on the 13 selected non-m6A-related neoantigen-coding lncRNAs could predict the prognosis in all gliomas and LGG. Furthermore, the NAS that was based on the cluster model predicted prognosis better in both GBM and LGG. The NASs based on the non-m6A-related neoantigen-coding lncRNAs together with cluster models were more accurate than those based on the m6A-related models. Moreover, a higher NAS indicates more aggressive gliomas, both at the tumor level and at the cellular level. Enrichment analysis suggested that immune pathways, including endogenous antigen processing and presentation as well as T-cell-mediated immunity, were enriched in high NAS samples. Moreover, NAS was positively correlated with immune infiltration, including CD8 T cell infiltration, but high NAS gliomas also exhibited more PD-L1 expression suggesting an immunosuppression environment, which is consistent with the fact that a higher NAS predicts worse survival outcomes. Further analysis of the immunosuppression of microenvironment T cells revealed that several positive regulators of T cells downregulate functions, including Ca
2+ flux and DNA repair. The Ca
2+ flux-related gene AHNAK also manifested most of the mutations present in TCGA dataset. Transcription factor analysis indicated that the expression of stemness-related transcription factors was elevated in the high NAS group compared to the low NAS group. Additionally, cell communication analysis confirmed that the high NAS group showed more inter-cell communication between immune cells and inflammation-related glioma cells and that high neoplastic glioma cells might escape T cell binding by downregulating IFNGR1. Finally, we identified several unique TCR CDR3 patterns that widely exist in glioma tissues, two of which (%GSTDTQYF and MNEQ) were significantly elevated in glioma tissues after tumor-cell lysate vaccination. And increased levels of these two neoantigen-reactive TCR patterns were found in high NAS gliomas, suggesting NAS model was also correlated with neoantigen response. Therefore, non-m6A-related neoantigen-coding lncRNAs play an essential role in neoantigen-related immune microenvironment of gliomas, which is a conclusion that provides a potential avenue for future CAR-T therapy, which could have wide targetability among glioma patients. Our non-m6A-related NAS model exhibited higher prognostic efficacy in TCGA dataset than m6A-related NAS model, m6A/non-m6A clustering model, age, gender and grade (Fig.
1G). In all three RNA-seq databsets, non-m6A-related NAS model also showed obviously better prognostic effect than m6A-related NAS model and was close to WHO grade model (Additional file
1: Fig. S1E). Moreover, the non-m6A-related NAS model was also highly associated with aggressive subtypes in both RNA-seq and scSeq data. It could also predict immune infiltration and T cell-glioma cell interaction via IFNGR1 pathway. These advantages make it a tool better than classic pathological grading in aspect of studying antitumor immunity.
LncRNAs are commonly considered as transcripts that cannot code peptides, but some small peptides encoded by lncRNAs were recently discovered, making it a new and interesting field in the study of non-coding RNAs (ncRNAs). Some of the m6A modification sites were found to participate in regulating ncRNA translation as elements similar to internal ribosome entry sites [
21,
54]. Moreover, these small peptides are also involved in the progression of different kinds of cancers. For example, the micropeptide 53aa encoded by the lncRNA HOXB-AS3 plays a role in metabolic reprogramming in colon cancer, and it suppresses the proliferation of colon cancer cells [
55]. Another study showed that a 130aa-peptide known as SRSP encoded by the lncRNA LOC90024 plays a role in modulating mRNA splicing by binding to SRSF3 and thereby promoting the proliferation, migration, and invasion of tumor cells [
56]. The current study has also proven that Yin Yang 1-binding micropeptide (YY1BM) encoded by LINC00278 can be more efficiently translated when m6A modification is demethylated by ALKBH5, and it protects esophageal squamous carcinoma cells from apoptosis induced by nutrient deprivation [
24]. Thus, peptides translated from lncRNA have a strong effect on cancer cell biology and can be regulated by RNA modification such as m6A. In our study, we found non-m6A modification model exhibited better prognostic efficacy compared with m6A modification model according to TransLnc database. In addition, lncRNA-encoded micropeptides have been proven to affect immunity functions. The lncRNA Aw112010 encodes a small peptide in murine macrophages, which promotes inflammation in mucosa immunity [
57]. Another peptide, miPEP155 encoded by MIR155HG, has been reported to affect antigen presentation by binding to a chaperone: HSC70 [
58,
59]. Furthermore, the antigen encoded by lncRNAs can also be present and involved in cellular immunosurveillance [
17]. Moreover, the m6A reader YTHDF1 is involved in promoting neoantigen degradation by regulating levels of lysosomal proteases through translation [
60]. This is consistent with our results in that NAS is positively associated with immune infiltration and T-cell-induced immunity as well as with antigen presentation pathways. The TransLnc database [
25] collected all predicted encoded peptides from lncRNAs together with the possible m6A modification status of lncRNAs. The binding affinities of these peptides to MHC complexes were also calculated using NetMHCpan [
61], which suggests that RNA modification plays a significant role in antigen production from lncRNA. Our work confirms that the signature of several neoantigen-coding lncRNAs from TransLnc correlates with glioma patients’ survival outcomes. Additionally, the neoantigen peptides provided by TransLnc were predicted to have a high probability of binding to potential neoantigen-reactive TCR clonotypes, which might be potential therapeutic targets for immunotherapy.
For tumors with immunosuppressive microenvironments such as gliomas [
62], canonical immunotherapies have not exhibited satisfying outcomes in primary gliomas, and some such therapies have only benefited patients with recurrent gliomas [
3‐
7]. Our study shows that tumor-related inflammation was inhibited in high-grade and high NAS glioma groups even though immune infiltration, including the level of CD8 + T cells, increased. IHC results also revealed higher PD-L1 levels in higher NAS gliomas. Therefore, innovations in gliomas immunotherapies are urgently required. The neoantigens belong to immunopeptides that specifically present on the surface of tumor cells. Due to its specificity in tumor immunity, these are considered promising targets for immunotherapy [
63]. Recently, the neoantigen vaccine’s ability to induce T-cell-mediated immunity has been widely discussed [
9,
63,
64]. Some circumstantial evidence advocates for the reinvigoration of exhausted T cells in tumor patients. An in vivo experiment showed that the vaccination can achieve beneficial outcomes when neoantigen-reactive T cells express exhausted markers before vaccination [
65]. Despite the fact that it was difficult to reinvigorate exhausted T cells after vaccination, it was suggested that, even after a tumor grows and possible immunosuppression is established, T-cell-mediated anti-tumor immunity can be facilitated by neoantigen vaccination. Our results suggest that non-m6A-related neoantigen-coding lncRNAs play a crucial role in determining glioma prognosis, and the study screens widely existing unique TCR clonotypes that could recognize potential neoantigens encoded by selected lncRNAs. High NAS gliomas are also found to contain more neoantigen-reactive TCR patterns, indicating NAS model together with lncRNA-derived micropeptides is associated with neoantigen-related immune response. The identified TCR-neoantigen pairs could provide universal targets for CAR-T therapy. Moreover, the protein-coding abilities of selected lncRNAs and the binding ability of potential micropeptides to TCR might be validated in further researches.
Our study established a NAS model based on non-m6A-related lncRNAs that were predicted to encode neoantigen peptides. This model exhibited its abilities to predict prognosis of glioma patients, immune infiltration in gliomas and neoantigen expression in tumor vaccine therapy. The correlations between NAS-related genes and PD-L1 was also verified by IHC. Besides, the screening of TCR-neoantigen binding pairs also provided several neoantigen-reactive TCR patterns that might be utilized for CAR-T therapy. But the effect of these TCR patterns needs more biological experiment verification in the future.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.