In tumor cells, mutated genes are transcribed into mRNA and translated into corresponding mutated proteins. These mutated proteins are degraded by the proteasome into many short peptides (69), which are recognized by antigen-presenting cells (APCs). After entering the endoplasmic reticulum, short peptides are further processed by N-terminal peptidases and bind HLA in tumor cells to form peptide-HLA complexes (pHLA). These pHLA complexes leave the endoplasmic reticulum and are transported to the cell membrane, where they are recognized by TCR and trigger an antitumor immune response (70) (Fig. 2). The mutated short peptides that induce an immune response in tumor cells are referred to as neoantigens.

The core of cancer immunotherapy is the presence of tumor antigens in tumor cells, which are recognized by the immune system. Based on their specificity, tumor antigens are divided into tumor-associated antigens (TAAs) and tumor-specific antigens (TSAs) (Fig. 3).

Initially, research on tumor vaccines focused on TAAs such as mucin 1 (71,72), human epidermal growth factor receptor 2 (73,74), telomerase reverse transcriptase (75) and tyrosinase (76). However, TAAs are not only selectively expressed or overexpressed in tumor cells but also in some non-tumor cells (77). As the body typically develops immune tolerance to these TAAs, it is difficult to stimulate a strong immune response; thus, immunotherapy targeting TAAs may be ineffective. In addition, targeting TAA may cause the immune system to attack normal tissue, which in turn may trigger autoimmune responses and diseases (78), making TAA not an ideal therapeutic target (79,80).

By contrast, TSA originates from mutations in the cancer genome, is present only in tumor cells and is virtually absent in normal tissues, making it highly tumor-specific. TSA also has greater immunogenicity and better MHC affinity and is not affected by central immune tolerance (81). Because normal tissue cells do not express TSA, immunotherapy targeting TSA does not cause off-target damage to non-tumor tissue and has a better safety profile (82). Therefore, TSA, as a tumor neoantigen, shows potential for clinical applications.

The generation of neoantigens is typically associated with genomic mutation and the number and types of mutation vary according to the cancer type (83,84), including single nucleotide variants (SNVs), insertions/deletions (INDELs) (85) and gene fusion (86,87). Point mutations account for 95% of tumor mutations, whereas INDELs and frameshift mutations constitute the remaining portion (88). Changes in amino acid sequences and spatial structures caused by INDELs, or frameshift mutations are more pronounced, and mutated peptides have a stronger affinity for MHC, making them more likely to be recognized by T cells as neoantigens (89). Therefore, research has mainly focused on SNVs and INDELs (90), which frequently occur in tumor cells and trigger abnormal protein translation under certain conditions (such as altered RNA splicing (91-93) or imbalances in translational regulation (94)), leading to the formation of neoantigens. Trials of neoantigen vaccines for melanoma indicate that neoantigens derived from SNVs expand T cell populations (95) and induce disease regression (58,96). Although neoantigens generated from SNVs and INDELs have been widely studied (90,97,98), their clinical applications remain limited. These neoantigens often have high patient specificity (99-101), meaning that the mutation may not be the same across patients, resulting in poor universality of neoantigen vaccines and limited clinical efficacy. Fusion gene-derived neoantigens have higher immunogenicity (102). Wei et al (103) collected 6,552 tumor samples from The Cancer Genome Atlas (TCGA), including colon adenocarcinoma, stomach adenocarcinoma, uterine corpus endometrial carcinoma, melanoma and predicted a total of 67,502 fusion neoantigens. Compared with SNVs and INDELs, fusion genes can generate up to 6-fold more candidate neoantigens and up to 11-fold more specific candidate neoantigens. Moreover, 5.8% of the fusion neoantigens in TCGA are shared between patients. However, such fusion genes are relatively rare events (104). Splice variants, endogenous retroelements (105) and other tumor-specific processes may also generate neoantigens. In addition to neoantigens derived from mutations in protein-coding genes, peptides produced from non-coding regions have immunogenic potential. Non-coding transcripts can be created from non-coding exons, introns and untranslated regions, as well as non-canonical reading frames within coding regions (106). Laumont et al (107) discovered numerous antigens from non-coding regions, including both mutated and unmutated antigens, by studying traditional non-coding sequences in patients with leukemia and lung cancer. Huang et al (108) found that peptides encoded by circular RNA (circRNA) not only exhibit immunogenicity but can also effectively induce specific T cell responses, thereby triggering a strong antitumor immune response. These findings provide novel ideas for tumor immunotherapy, suggesting the development of vaccines or immunotherapy strategies using circRNAs and peptides encoded within tumor cells.

Tumor mutation sites are filtered by performing exome sequencing of tumors and adjacent normal tissue (110,113), aligning the sequence reads with the reference genome and using mutation identification tools (Table I) (114-132) to detect somatic mutations (such as SNVs and INDELs). RNA-seq can identify neoantigens generated from gene fusions, along with other biological information, such as gene expression levels and copy number variations (133), which can refine the selection of tumor mutation sites. Combining WES data with RNA-seq can confirm whether mutated genes are expressed in tumor cells, thereby filtering more reliable candidate neoantigens.

HLA is a key molecule in the immune system that is involved in antigen recognition. It is encoded by the HLA gene complex located on chromosome 6p21.3 and is primarily involved in functions such as antigen recognition and presentation and immune response regulation, exhibiting a high degree of genetic polymorphism (134). Based on structural and functional differences, HLA is divided into three classes: HLA I, HLA II and HLA III (135). HLA I and II encode molecules that bind and present antigens, allowing cytotoxic T lymphocytes to interact with mature HLA cell surface proteins through antigen-binding grooves and enabling the immune system to recognize and respond to antigens. HLA I genes present endogenous antigens (such as tumor antigens) to CD8⁺ T cells (136), activating cytotoxic immune responses, and thereby helping the immune system identify and eliminate tumor cells. By contrast, HLA II molecules are primarily responsible for presenting exogenous antigens, serving a key role in initiating immune responses in CD4⁺ T cells (136), and participating in immune regulation. HLA III genes are located between HLA I and II (135), and functions of these genes remain unclear. In the application of tumor vaccines, HLA typing affects the prediction of neoantigens (137), thereby affecting vaccine preparation. Therefore, accurate HLA typing is key for personalized immunotherapy.

In clinical practice, HLA typing commonly employs three methods: PCR sequence-specific primers and oligonucleotides and sequence-based typing (SBT) (138). Among these, the SBT method has become the gold standard for HLA typing, as recommended by the World Health Organization, because of its high efficiency and accuracy. However, these typing methods are labor-intensive and expensive. Therefore, with the widespread adoption of next-generation sequencing (NGS) technology, second-generation sequencing is gradually being used for HLA gene typing (139). Compared with other methods, NGS technology offers advantages such as high throughput and speed, and the ability to simultaneously detect classical HLA I and II genes, as well as non-classical HLA I genes (140,141). Compared with sequence-specific oligonucleotide probe (SSOP) methods, NGS-based genotyping reduces reagent consumption, minimizes technical variability between experiments, and significantly improves the resolution of HLA allele typing (142). Furthermore, current HLA typing prediction software can achieve up to 99% accuracy in predicting HLA class I alleles (143). Several tools are available for predicting HLA molecular types (Table II) (144-160), among which, OptiType, Polysolver and PHLAT are considered the best performing HLA I typing prediction tools (161,162). HLA typing obtained from WES shows better predictive results than RNA-seq data (149).

HLA molecules exhibit high numbers of polymorphisms, with each HLA molecule recognizing a range of peptides and presenting antigenic peptides to T lymphocytes, thereby triggering an immune response. Although the number of short peptides generated by mutations is high, tumor antigens are often masked by polysaccharides or other molecules, making it difficult to process and effectively present these antigenic peptides. Ultimately, the number of peptide segments that can bind HLA and be recognized by T cells to trigger an immune response is limited (163). Of 100-200 peptides generated by mutations, only one may bind to a specific HLA molecule (164). Therefore, a large number of candidate false-positive neoantigen peptides exist between the identification of tumor somatic mutations through sequencing and the recognition of neoantigens by TCRs that lead to an immune response.

Traditional screening of candidate positive peptides relies on cytological experiments such as competitive binding (165) and enzyme-linked immunospot (ELISPOT) assays (166). However, depending solely on these experimental methods is often time-consuming and labor-intensive. Therefore, techniques such as genomics and bioinformatics to predict the affinity between HLA and antigen peptides and select suitable tumor antigens are key in immunotherapy (167).

Currently, a number of peptide-HLA binding affinity datasets are available for the training and validation of prediction software, including the Immune Epitope Database (IEDB) (168), SYFPEITHI (169), and MHCBN (170). The development of online prediction algorithms for peptide-MHC affinity is progressing, with the accuracy of predictions improving (171,172). The discovery of allele-specific motifs (173) has enabled algorithms to provide more precise predictions (174). Bioinformatics software tools have emerged to predict the affinity between HLA and antigen peptides; commonly used tools include NetMHC (175), NetMHCpan (176), SMM (177) and SMMPMBEC (178). These prediction tools primarily rely on three types of algorithms: Machine learning (ML), structure-based approaches, and linear regression (LR). ML-based methods predict peptide-binding affinity by learning a function that maps a peptide to regions of binding affinity based on available known binding peptides (179). Representative software includes the artificial neural network-based method NetMHC (175) and the pan-specific method NetMHCpan (176). Structure-based methods predict peptide-binding affinity by calculating the minimum free energy of the pHLA (180) using residue-based statistical energy functions, quantitative structure-activity relationship analysis and quantitative sequence-activity models (181,182). However, owing to the limited number of available crystal structures, the prediction speed and accuracy of these methods are relatively low (183). The LR-based method, a matrix-based approach, predicts peptide-binding affinity by constructing a matrix from the alignment of peptides that represent the motif information (184). As the linear computational complexity is lower than the non-linear computational complexity of other methods, these methods can quickly predict the binding affinity. In addition to these mainstream prediction algorithms, certain software programs integrate multiple methods or use their own methods. For example, TripHLApan (185) combines triple-encoding matrices and recurrent neural networks, ForestMHC (186) uses a random forest classifier and Anthem (187) develops an aggregating one-dependence estimators model based on Bayesian ensemble methods to predict the binding affinity between peptides and HLA (Table III) (169,175-178,184-226).

Prediction algorithms for HLA I molecules are mature, but there are relatively few algorithms for HLA II. Unlike HLA I molecules, the peptide-binding groove of HLA II molecules is open, leading to variations in both the length of peptides that bind to HLA-II and the position of the binding core (227). The most common peptide lengths that bind to HLA II range from 13 to 25 amino acids (228), while class I peptides typically range from 8 to 15 amino acids (195). To the best of our knowledge, there is no database regarding the interactions between HLA II and neoantigens. Therefore, compared with the prediction methods for HLA I, the binding predictions for HLA II are in a developmental stage and face more challenges in practical applications (229).

Despite the progress in predictive algorithms, currently available tools cannot reliably predict the immunogenicity of peptides. Several comparative studies have been conducted to evaluate the performance of software packages in predicting affinity (230-234). The results show that artificial neural network algorithms exhibit better predictive performance, with no notable differences between the methods. Concurrently, there are considerable differences in the results predicted by these bioinformatics software programs, depending on the algorithm, MHC type and peptide length (233). Regarding decision thresholds, certain commonly recommended thresholds [half maximal inhibitory concentration (IC50)≤500, percentile rank ≤2] can lead to low sensitivity, potentially overlooking many MHC ligands (235,236). Additionally, applying a lower binding affinity threshold (IC50≤5,000) generally does not improve prediction performance. The strong binding affinity threshold resulted in the most stable accuracy across different predictors, but also resulted in the lowest accuracy values. Using strong (IC50≤50) and intermediate (IC50≤500) binding affinity thresholds, provides high predictive specificity (233).

To predict pHLA binding affinity more accurately, integrated platforms for prediction methods have emerged (237), such as MHCcombine (233) and IEDB (168). Compared with individual methods, using multiple approaches is more helpful in screening for neoantigens, as algorithms vary in modeling strategies, training datasets, and predictive performance. By combining multiple approaches, the limitations of individual methods can be compensated for, improving both the reliability and accuracy of predictions. This enables a more comprehensive and effective identification of potential neoantigen candidates. However, their predictive accuracy remains unsatisfactory, especially in clinical application (238). Other factors affect the final immunogenicity of the predicted epitopes. These include gene expression, RNA splicing, protein processing and overall patterns of peptide loading and presentation by MHC (239,240). Therefore, the prediction of HLA-peptide binding needs to be improved based on these factors.

Full activation of T cells requires two signals. The first signal originates from the specific binding of the TCR to pHLA. The second signal originates from costimulatory molecules expressed by APCs that interact with and activate T cells (245). Numerous tools are available for predicting the recognition of neoantigen-specific T cells, which assists in the preliminary screening of the immunogenicity of neoantigens. Commonly used tools such as NetCTL (246) and NetCTLpan (247) generate a comprehensive score by analyzing MHC presentation, proteasomal C terminal cleavage, and transporter associated with antigen processing (TAP) transport efficiency (rather than directly predicting T cell binding) to identify potential cytotoxic T lymphocyte (CTL) epitopes within protein sequences. There are also machine and deep learning techniques to predict TCR and pHLA binding (248). These prediction tools screen neoantigens further, reducing the experimental workload and the consumption of scarce experimental materials.

Traditional antigen screening methods rely primarily on immunochemical luminescence to detect the strength of T cell effector functions in response to antigen stimulation. When TCRs recognize their target antigens, T cells are activated and subsequently release cytokines such as IL-2 and IFN-γ, as well as granzyme and perforin, leading to the lysis and death of target cells. Numerous detection methods assess the immunogenicity of antigens based on the levels of cytokine release, such as ELISA (249) and ELISPOT (250,251). The most common method is the ELISPOT technique, which evaluates TCR antigen reactivity by detecting changes in the cytokines secreted by T cells in response to antigen stimulation through antibodies (252). This technique is widely used to assess the responses of CD4⁺ and CD8⁺ T cells to antigens or mitogenic stimulation (253,254). Owing to its simplicity and high sensitivity, the ELISPOT assay is currently the primary method used for immunological validation.

Chromium release assay is used to measure the cytotoxic activity of effector cells in vitro (255). This method quantifies the specific killing of target cells by T cells by co-culturing T cells with target cells labeled with the radioactive isotope ⁵¹Cr, then detecting the number of radioactive pulses of ⁵¹Cr released into the supernatant following lysis of the target cells (256,257). The Cr release assay has the advantages of accurate results and high reproducibility; however, owing to the short half-life of ⁵¹Cr, it is difficult to perform multiple measurements. In addition, ⁵¹Cr is radioactive, posing health risks to researchers and requiring training in the handling of radioactive materials. These factors limit the applicability of this method.

Organoids refer to a technology involving the culture of pluripotent stem cells derived from human tissue in a semi-solid matrix in the presence of mitogens and pathway regulators (258). Compared with traditional cancer cell lines, in vitro tumor organoids have greater clinical relevance in tumor immunology: Their heterogeneous three-dimensional structure and spatial arrangement reflect the genomic, morphological and physiological characteristics of the original tumor, making them more likely to accurately predict patient response to drugs (259,260). This provides more precise efficacy and effectiveness data for drug screening, aiding identification of resistance mechanisms or understanding the reasons for treatment failure. Co-culturing tumor organoids with autologous immune cells can demonstrate the cytotoxic ability of T cells against tumor organoids and test the immunogenicity of tumor vaccines (261). Organoids effectively retain the characteristics of the tumors within the patient (262), contributing to the development of personalized treatment plans.

Humanized mouse models have been created by transplanting human tissue (such as the fetal thymus and liver tissue), peripheral blood mononuclear cells (PBMCs), or CD34⁺ human hematopoietic stem cells and their progenitors (263,264) into immunodeficient mice to simulate the human immune system. Subsequently, human cancer cell line- and patient-derived xenografts are implanted for immune validation experiments (Fig. 5). Among humanized mouse models, PBMC-based models are the most widely used due to their relative ease of implementation and lower cost (265,266). These humanized mouse models effectively simulate human immune responses to determine tumor immune evasion mechanisms, test novel immunotherapeutic strategies, and validate the safety and efficacy of drugs (267). Humanized mouse models have achieved notable results in the testing and validation of immunotherapy and are widely used in preclinical research (268,269). Camacho et al (270) assessed a carcinoembryonic antigen (CEA) vaccine using transgenic non-obese Diabetic/severe combined immunodeficiency (NOD/scid)-DR1 mice and detected CEA-specific cellular immune responses. In another study, Spranger et al (271) developed PBMC-NOD/scid IL2Rg^null (NSG) mouse models to evaluate melanoma-associated antigen MART-1 vaccine, and the results showed that the vaccine could induce antigen-specific CD8⁺ T cell responses. These findings suggest that humanized mouse models effectively recapitulate human immune cell responses to tumor antigens and serve as valuable tools for studying anti-tumor immune responses.

Melanoma is a highly aggressive and metastatic malignant tumor that is often associated with ultraviolet radiation exposure. Owing to its high tumor mutational burden (TMB) (272,273), melanoma is an ideal target for neoantigen-based immunotherapy (274,275).

In 2015, Carreno et al (95) conducted a phase I clinical study of neoantigen DC vaccines for patients with advanced cutaneous melanoma, which demonstrated that the neoantigen DC vaccine enhances spontaneous HLA I-restricted T cell responses, providing experimental evidence for the induction of T cell responses by neoantigen vaccines.

Sahin et al (96) reported the first application of an RNA-based polyepitope vaccine for melanoma in humans. All patients exhibited T cell responses and two patients showed notable T cell infiltration into metastatic lesions following vaccination, leading to neoepitope-specific tumor cell killing. Notably, one patient achieved complete remission after receiving the vaccine in combination with the PD-1 blockade, demonstrating the feasibility, safety and antitumor activity of the vaccine in clinical application. In addition, Ott et al (58) showed that personalized neoantigen vaccines induce CD8⁺ T cell responses in 16% of cases and CD4⁺ T cell responses in 60% of cases; among six patients, four remained recurrence-free 25 months post-vaccination, while the two patients who experienced recurrence achieved complete tumor regression following PD-1 inhibitor treatment. Ott et al (276) further evaluated the effects of the personalized neoantigen vaccine NEO-PV-01 in combination with PD-1 blockade in patients with advanced melanoma, NSCLC and bladder cancer: Vaccination effectively induced neoantigen-specific CD4⁺ and CD8⁺ T cell responses, with no severe adverse events, further supporting the safety and efficacy of this combination therapy.

Hu et al (59) assessed the clinical effects of the NeoVax vaccine in eight patients with surgically resected stage IIIB/C or IVM1a/b melanoma and observed persistent neoantigen-specific T cell responses as well as T cell clonal expansion and diversification. Notably, when combined with ICIs, this approach may be more effective in controlling metastatic tumor progression. Mørk et al (277) developed a personalized neoantigen peptide vaccine, EVX-01, which includes a novel CD8⁺ T cell-inducing adjuvant, CAF^®09b, for treating metastatic melanoma. When combined with ICIs, all five patients exhibited EVX-01-specific CD4⁺ T cell responses, and the vaccine induced durable neoantigen-specific cellular immune responses. Following combination of EVX-01 immunotherapy and anti-PD-1 therapy, two patients achieved partial response (PR) and one patient achieved complete response (CR). All immune-related adverse effects were grade 1, further demonstrating the potential of personalized neoantigen vaccines in combination with ICIs.

In the KEYNOTE-942 trial (278), the efficacy of a personalized neoantigen vaccine, mRNA-4157, in combination with pembrolizumab for advanced melanoma was evaluated. The results showed that this combination therapy significantly prolonged recurrence-free survival (RFS) and decreased recurrence and mortality rates, providing new research directions for adjuvant selection and combination of multiple immunotherapeutic approaches.

Lung cancer is the most common cancer worldwide, accounting for 12.4% of newly diagnosed cases (1). The application of neoantigen vaccines for the treatment of lung cancer has also progressed. Phase I/II clinical trials are currently evaluating the safety and immunogenicity of neoantigen vaccines (279-284) and their role as adjuvant interventions for other immunotherapies.

Ding et al (279) treated 12 patients with metastatic lung cancer using a DC vaccine based on neoantigens. The disease control rate was 75%, with a progression-free survival (PFS) of 5.5 months and an overall survival (OS) of 7.9 months. All treatment-related adverse events were grade 1-2, indicating that this therapeutic approach has a favorable safety profile.

Recent studies on tumor neoantigen vaccines have demonstrated promising efficacy in the treatment of NSCLC (280,281). Li et al (282) reported the results of a phase I clinical trial involving 24 patients with stage III/IV NSCLC who received personalized neoantigen peptide vaccines. Following vaccination, the median PFS was 6.0 and the median OS was 8.9 months. Additionally, all patients exhibited a significant increase in the frequency of neoantigen-specific CD8⁺ T cells in peripheral blood following the vaccination process. In the KEYNOTE-603 study (280), 16 patients received treatment with the mRNA-4157 vaccine, including 11 patients with NSCLC; of these patients, 14 remained disease-free throughout the study. This vaccine has progressed to a phase III clinical trial (281).

The efficacy of neoantigen vaccines in combination with other treatment modalities for lung cancer is widely recognized (283,284). As demonstrated by Ott et al (276), the combination of neoantigen vaccines with ICIs can enhance the immune system antitumor response. In a case report (283), a patient with advanced squamous cell carcinoma who was resistant to PD-1 blockade was treated with a neoantigen DC vaccine combined with ICIs. The results showed significant tumor regression, with OS of 48 months and no severe adverse reactions. Additionally, in a phase Ib clinical trial, Awad et al (284) used the personalized neoantigen vaccine NEO-PV-01 in combination with pemetrexed, carboplatin and pembrolizumab as first-line treatment for advanced NSCLC. Following vaccination, neoantigen-specific CD4⁺ and CD8⁺ T cell responses were observed along with increased infiltration of CD4⁺ T cells, which exhibited a pronounced cytotoxic phenotype. These findings validate the potential efficacy of combining neoantigen vaccines with ICIs.

McCann et al (285) explored the feasibility of personalized neoantigen vaccines in patients with lung cancer and low TMB using a mouse model. In patients with an unfavorable tumor microenvironment, it was possible to identify antigens capable of effectively activating neoantigen-specific T cells (285). This provides new insights and evidence for personalized immunotherapy in lung cancer with low TMB.

Gliomas are the most common primary intracranial tumors of the brain, with a poor prognosis and limited curative treatment options (286). Gliomas are considered immune-privileged tumors (287), leading to limited effectiveness of immunotherapy. However, an increasing number of immunotherapeutic strategies have been developed (288).

Isocitrate dehydrogenase 1 (IDH1) (289,290) mutationassociated vaccines are currently the most common neoantigen vaccines for gliomas (99,291-294), particularly the IDH1 arginine-to-histidine substitution at codon 132 (IDH1 R132H) mutation, which is widely present in grade II and III gliomas as well as 80% of secondary glioblastomas (295). Schumacher et al (99) confirmed that IDH1 R132H contains immunogenic epitopes suitable for mutation-specific vaccination. This mutated peptide is presented by MHC class II molecules and induces a mutation-specific CD4⁺ T-helper-1 immune response (99). Additionally, the study employed an MHC-syngeneic humanized mouse model to administer the IDH1 R132H-specific peptide vaccine. The results demonstrated that the vaccine effectively induced a specific anti-tumor T cell immune response and enhanced anti-tumor efficacy (99). Pellegatta et al (292) established a glioma xenograft mouse model carrying the IDH1 R132H mutation and immunized mice with five peptide vaccines containing the mutation. The results showed a significant increase in peripheral CD8⁺ T cells following immunization, accompanied by elevated IFN-γ levels and the generation of anti-IDH1 antibodies, indicating that the vaccine effectively activated an antitumor immune response.

Platten et al (293,294) developed a tumor vaccine targeting IDH1 mutations and conducted a phase I clinical trial in humans. Following vaccination with IDH1-specific peptides, vaccine-induced immune responses involving multiple MHC alleles were observed in 93.3% of participants. Furthermore, the vaccine demonstrated promising clinical efficacy, with 3-year PFS and OS rates of 0.63 and 0.84, respectively. Among patients with a positive immune response, the 2-year PFS rate was 0.82. Moreover, related adverse events were classified as grade 1. The IDH1 R132H mutation-specific vaccine has demonstrated good immunogenicity and safety in preclinical and clinical studies (99,292,293,294).

Due to the notable intratumoral heterogeneity of gliomas, research on glioma neoantigen vaccines has primarily focused on developing personalized vaccines: For example, Hilf et al (296) conducted a phase I clinical trial of the Glioma Actively Personalized Vaccine Consortium, in which researchers designed personalized vaccines based on transcriptomic and immunopeptidomic mutation analyses of tumors. The results showed that most patients responded to at least one immunopeptide, with a significant increase in CD8⁺ T cell numbers and a transition to a memory phenotype. Additionally, patients exhibited new epitope-specific responses, primarily TH1-type CD4⁺ T cell responses, while maintaining a good safety profile. In a phase I/Ib study, Keskin et al (60) adopted a similar strategy using multi-epitope personalized neoantigen vaccines to treat patients with glioblastoma: Following vaccination, patients developed multifunctional CD4⁺ and CD8⁺ T-cell responses to neoantigens.

A phase III clinical trial (297) evaluated the efficacy of a personalized peptide vaccine in HLA-A24-positive patients with recurrent glioblastoma. Although the trial showed OS of vaccinated patients was 8.4 months, which was slightly higher than the 8.0 months in the placebo group, the difference in OS and PFS between the two groups was not significant. Thus, although personalized peptide vaccines exhibit immunostimulatory potential, their clinical benefits require further validation. These studies indicate that the design and application of personalized neoantigen vaccines holds promise for the treatment of gliomas.

HCC has a high recurrence rate (298-301). Therefore, studies on neoantigens in HCC have primarily focused on anti-recurrence treatments (302-304). Liu et al (302) used neoantigen-reactive T cell-based immunotherapy combined with tomotherapy to treat a patient with advanced HCC. The patient achieved long-term PFS following treatment, demonstrating the potential of the combined therapeutic approach in the treatment of liver cancer. Cai et al (303) designed a personalized neoantigen vaccine for 10 patients with resectable HCC and portal vein branch invasion; RFS was prolonged after vaccination, and five patients who exhibited a strong neoantigen response after full vaccination had longer RFS than the others.

Peng et al (304) conducted a phase II clinical trial in which a personalized neoantigen DC vaccine was combined with neoantigen-activated T cell therapy to treat 10 patients with HCC who underwent curative resection or radiofrequency ablation. The results showed that 70% of the patients developed a neoantigen-specific T cell response, and 71.4% remained recurrence-free within 2 years of treatment. All adverse events were grade 1 or 2. Shen et al (305) treated patients with unresectable HCC patients with radiotherapy combined with an immune therapy composed of personalized peptide vaccination (PPV) peptides, DC, and CTL (PPV-DC-CTL). After radiotherapy and 1-3 cycles of PPV-DC-CTL treatment, the Alpha-fetoprotein (AFP) levels in six patients significantly decreased. Imaging assessments showed PR in three patients, while the other three patients had stable disease. The response rate was 33%, and the disease control rate was 66%. The study found that the regimen was safe and well-tolerated. The aforementioned studies indicate that combining neoantigen-based immunotherapy with other treatments is safe and feasible, and it may help reduce recurrence of HCC after curative treatment, or effectively treat HCC.

Guo et al (306) designed a neoantigen-based DC vaccine, Neo-MoDC, for treating metastatic gastric cancer. Clinical trial results showed that patients who received the neo-MoDC vaccine alone generate a neoantigen-specific T cell response. When combined with ICIs, a stronger immune response is triggered, leading to complete regression of all tumors within 25 months (306).

Yu et al (307) conducted a personalized neoantigen vaccine study on six patients with microsatellite-stable colorectal cancer who experienced recurrence or metastasis following surgery and chemotherapy. Neoantigen-specific immune responses were observed in 66.67% of the vaccinated patients, and four patients remained progression-free until the completion of the clinical trial.

Compared with cancers such as melanoma and lung cancer, the development of neoantigen-based therapies for other tumor types is in its early stages. The personalized neoantigen vaccine NEO-PV-01, developed by Ott et al (276), has also made breakthrough progress in the treatment of bladder cancer. After synthesizing the antigen peptides, they were mixed with the adjuvant polyinosinic-polycytidylic acid stabilized with poly-L-lysine and carboxymethylcellulose (poly-ICLC) (308) for bladder cancer therapy, demonstrating the vaccine's safety and efficacy (276). Zeng et al (309) treated a patient with advanced collecting duct carcinoma (CDC) using a personalized neoantigen vaccine combined with NRTs. Following treatment, both the primary CDC tumor and metastatic lesions remained in a stable disease state for 9 months, suggesting that neoantigen-based immunotherapy may improve the PFS of patients with advanced CDC.

Chen et al (310) developed a personalized neoantigen vaccine containing 20 neoantigen peptides for seven patients with advanced pancreatic cancer. The median and vaccine-associated OS and PFS rates were 24.1, 8.3 and 3.1 months, respectively. Notably, one patient had a vaccine-related OS of 21 months, and the abundance of antigen-specific TCR clones increased from 0 to nearly 100%, demonstrating the potential of neoantigen vaccines to activate specific T cell subsets and eliminate cancer cells.

Rojas et al (54) treated 16 patients with surgically resected pancreatic ductal adenocarcinoma using a personalized mRNA neoantigen vaccine in combination with multiple therapies. The results showed that at least one neoantigen-specific CD8⁺ T cell response was induced in half of the patients, and responders had significantly prolonged median RFS. Based on these promising results, a global randomized clinical trial will be conducted to validate the efficacy of this therapy.

Recent studies have identified neoantigens derived from RNA splicing aberrations in numerous types of cancer (93,311-313). These previously uncharacterized shared neoantigens are widely expressed in tumors such as gliomas, mesothelioma and prostate and liver cancer (314). It is hypothesized that these neoantigens are endogenously generated and presented by tumor cells under physiological conditions, effectively triggering neoantigen-specific CD8⁺ T cell responses to eradicate cancer cells (313,315). This provides a molecular foundation for addressing tumor heterogeneity in immunotherapy.