Although effective responses have been observed in adoptive cell therapy, adverse effects have become a Gordian knot in many trials. Patients treated with lymphocytes modified with high-affinity TCR-T cells that target MART-1 or gp100 exhibited severe destruction in normal tissues where melanocytic cells were present, including the skin, eyes, and inner ears, which was due to the expression of MART-1 in normal cells (10). Patients with metastatic colorectal carcinoma treated with TCR-T cells targeting CEA exhibited severe inflammatory colitis, possibly because CEA is expressed in the normal mucosa of the colon (11). Three of nine patients treated with MAGE-A3-specific TCR-T cells experienced mental disturbances, and two of them died of leukoencephalopathy (12). The adverse effects of CAR-T therapy are similar to those of TCR-T therapy and include CAR-T cell-related encephalopathy syndrome (CRES), off-tumor effects and acute respiratory distress syndrome, which are due to on-target off-tumor recognition and killing; in addition, cytokine release syndrome (CRS) is the most frequent side effect of CAR-T therapy (13–15).

Different cell therapies have different solutions to eliminate side effects (Figure 1). For TCR-T therapy, a switch is inserted into TCR-T cells, such as the inducible caspase 9 safety switch, herpes simplex virus thymidine kinase, or a truncated human epidermal growth factor receptor, which makes the TCR-T cells lethally sensitive to the related ligands; when side effects occur, the reaction can be terminated through the administration of the related ligand (16–18). In addition, reducing the affinity is also an effective method, because, as with many other cell surface receptors, αβ TCRs bind to peptide-MHC (pMHC) complexes with a very low affinity (~1–50 μM) (19); thus, if TCRs have higher affinity with their cognate, they might enhance the ability to recognize the relatively low-affinity antigen in normal cells, which will cause severe side effects. To address the side effects of CAR-T therapy, numerous approaches have been developed to manipulate the activity and depletion of adoptive T cells. These approaches include all methods applied in TCR-T cell side effect control; for example, to rapidly ablate CAR-T cells, inducible caspase (iCasp9) is expressed on CAR-T cells, and, similar to the treatment of TCR-T cell side effects, this is done with the addition of a dimerizing drug that activates iCasp9 signaling and leads to apoptosis (20, 21). Other suicide genes, such as epitope tags, are also used to control CAR-T cell reactions (22). In addition, the affinity altering method is applied in CAR-T cell side effect control (23). Furthermore, switchable CARs have been designed to increase the safety and manipulate the activity at human will, reducing the cytotoxicity without the deletion of programed cells. The strategy is to separate the antigen-binding domain from the signal transduction domain through a peptide neoepitope (PNE). The PNE is designed to contact an antibody that recognizes and binds a specific antigen on cancer cells, while the PNE can be recognized by antibodies on the adoptive T cells and thus acts as a bridge between the antigen-binding domain and the signal transduction domain (24, 25). In addition, logic gate CAR, using AND, NOT, and OR logic gates, can also increase specificity and reduce side effects. The most prevalent AND gate CAR is synNotch CAR (26), which imitates the Notch signaling pathway activation mechanism; this strategy contains a synthetic Notch receptor (synNotch) that releases a transcription factor once the receptor recognizes its cognate, in turn driving the expression of a CAR specific for a specific antigen. The NOT gate CARs can distinguish normal cell antigen from cancer cell antigen through the coexpression of an inhibitory CAR (iCAR) that dampens the T cell response when normal cell antigen is present (27). The OR gate CAR is similar to the bispecific CAR, which integrates CD3ζ and the costimulatory domain with two CARs recognizing different antigens (28).

These safety mechanisms may have limitations, both in TCR-T and CAR-T immunotherapies, such as a relatively slow induction efficiency, and inherent leakiness can lead to residual TCR/CAR expression in the absence of the inducer (29), which limits the wide application of adoptive cell therapy and threatens the safety of patients. The most crucial event in the development of ACT is to find high-specificity neoantigens, which will reduce the tedious safety control methods and help to make ACT a widely applied therapy. The potential advantages and disadvantages to apply TCR-T or CAR-T based on neoantigens for cancer therapy were discussed in Table 1. Although there are tremendous obstacles and undiscovered mechanisms in the immune system, efforts in related studies make a difference, and some methods are available for the discovery of neoantigens.

Recently, a Chinese research group devised another kind of method to screen neoantigens in patients who suffer from cancers (47), with the expectation to identify neoantigens in a timely and convenient manner. They created an off-shelf neoepitope peptide library. They first mined high-frequency mutant genes in nine types of human malignant solid tumors with the TCGA and COSMIC databases and found genes with frequencies >10% in the COSMIC database. There were 21 mutant genes with frequencies >10%; among these mutant genes, 29 hotspot mutations were selected as candidate targets to build the shared neoantigen peptide library, which covered 9.49~89.56% of cancer patients in the TCGA database, with a median coverage of 23.04%. They selected the high-frequency HLA-A class I gene product subtypes, HLA-A*11 (A*1101), HLA-A*02 (A*0201, A*0203, and A*0206), and HLA-A*24 (A*2402), and determined the possibility of these candidate peptides binding to HLA-A with five different algorithms: BIMAS, IEDB, NetMHC3.4/NetMHC4.0, NetCTL1.2, and SYFPEITHI. After the analysis of these algorithms, 44 shared neoepitope peptides were selected for peptide synthesis. In the clinic, the patients who underwent targeted sequencing were retrieved against the neoantigen library, and the researchers found that the mutations in 13 patients corresponded with the shared neoantigen peptide library. In addition, HLA-A alleles were also matched, and then immunogenic neoantigen identification was conducted by detecting the secretion of IFN-γ using Cytometric Bead Array and ELISPOT. Eventually, immunogenic neoantigens were identified in six patients through the neoantigen peptide library (Figure 3).

This method is more convenient and time-saving than WES combined with MS, but there are also some problems that need to be pointed out. First, the capacity of the shared neoantigen database must cover as many candidate peptides as possible to incre ase the accuracy of the identification of neoantigens. Second, the accuracy of the prediction algorithm also limits the outcomes, as with the previous method, WES combined with MS. However, with an increasing number of neoantigens being identified, this kind of method is a promising approach to neoantigen screening.

Li et al. developed an entirely fantastic method to discover T cell antigens through trogocytosis (48). Trogocytosis is a biological phenomenon that happens during cell conjugation by which cells share membranes and membrane-associated proteins (49). It has been reported that trogocytosis is a bidirectional physiological activity, but Li et al. found that target cells with supraphysiological levels of epitopes can extract membranes and membrane-associated proteins from interacting T cells; this phenomenon can be tracked by the acquisition of T cell membrane proteins. Using this characteristic of T cell-target cell reactions, they developed a TCR ligand discovery platform that could distinguish target cells that present genetically encoded epitopes cognate to orphan TCR-transduced T cells with fluorescence-activated cell sorting from a target cell library. They established a Jurkat cell line expressing F5-TCR or 1G4-TCR and K562 cells expressing their cognate single-chain trimer of HLA-A2/MART1 or A2/NYESO1. They coincubated these two kinds of cell lines and found that trogocytosis occurred from T cells to target cells and was scaled with pMHC (peptide-MHC) density; therefore, they labeled the orphan TCR of T cells and coincubated with cognate target cells, and they sorted the target via FACS with high specificity. Next, they determined whether this platform can be applied in a real tumor model; that is, whether or not this method can identify the cognate neoepitope for a tumor-infiltrating lymphocyte-derived orphan TCR from a custom library of privately mutated, subject-specific neoepitopes. They chose metastatic melanoma cells as the target cell. First, they identified private mutations through exome and RNA sequencing and then predicted which of these mutations generated neoepitopes that would be presented by HLA-A*02:01. A neoepitope-reactive TCR was isolated through a pMHC multimer panel. A neoepitope SCT cDNA library comprising 3,251 unique neoepitopes ranging from 8 to 12 amino acids in length was constructed, and K562 cells were transduced with this library. They then coincubated K562 cells with neo-TCR-transduced Jurkat cells and performed two rounds of sorting and neoepitope identification through next-generation sequencing. Eventually, they found the highest rank peptide was derived from ubiquitin-specific peptidase 7, and they verified the neoepitope by measuring the ability of the neo-TCR to induce cytotoxicity in the recognition of mutUSP7-K562 target cells (Figure 4).

This method is a new breakthrough in T cell antigen discovery. In comparison with pMHC yeast display (50), the method has some advantages. First, it does not need to produce orphan TCR protein reagents, and the orphan TCRs are expressed in their natural context. Second, the extent of trogocytosis can be controlled through the coexpression of CD8 molecules on the donor cells because it can increase the avidity between T cells and cognate target cells, which could be used to adjust the screening specificity. However, the neoepitope discovery is based on the exome sequence and RNA sequence, and prediction with bioinformatics tools, which has the drawbacks mentioned before, in addition to the construction of neoepitope libraries, is laborious and time-consuming, which limits the wide application of this method.

These different screening methods have their own characteristics, which are applicable to specific situations. The advantages and disadvantages of different approaches for neoantigen identification are listed in Table 2. In early research, serological analysis of recombinant expressed cDNA clones (SEREX) was demonstrated to be a useful method to detect tumor- and tumor-associated antigens in a variety of malignancies (51, 52). This kind of method utilizes antigen and antibody reactions to identify tumor antigens. Similar to ELISA, a tumor cDNA library first needs to be constructed and transfected into Escherichia coli; the expressing antigens are transferred to the membrane and then coincubated with patient serum antibody; and enzyme-conjugated anti-human IgG antibodies are used to generate a color reaction to identify the tumor-associated antigens. Compared to SEREX, these methods mentioned before not only can screen neoantigens but also, in a high throughput way, can improve the accuracy. Recently, Kula et al. developed another method for T cell epitope discovery, and the mechanism of this method is almost the same as that of Li, but they used lentiviral delivery of antigen libraries into target cells rather than an SCT library, and they coincubated these transduced target cells with T cells if the target cells triggered T cells to secrete granzyme B; the antigens transduced would be identified by next-generation sequencing (53). Joglekar et al. described the use of chimeric receptor called signaling and antigen-presenting bifunctional receptors (SABRs) in a cell-based platform for TCR antigen discovery. By connecting a MHC/peptide complex with CD3ζ-CD28 intracellular signaling domain, they devised a novel reporter system for antigen discovery to screen thousands of antigenic epitopes, and identified the targets recognized by public TCRs of known specificities (54).

Epidermal growth factor receptor variant III (EGFRvIII) is the variant of EGFR in human tumors that is widely distributed (56). This variant results from a deletion and a mutation in the exon of EGFR, which create a tumor-specific and immunogenic neoantigen (57). Previous studies have demonstrated that in both human and mouse, EGFRvIII can induce the autoimmune response, and the EGFRvIII-specific antibodies could even be isolated from breast cancer patients (58, 59). Marcela Maus conducted a first-in-human study of the treatment of glioblastoma with anti-EGFRvIII CAR-T therapy (60); it is impressive that after the infusion of anti-EGFRvIII CAR T cells, the cancer was suppressed, without evidence of off-tumor toxicity or cytokine release syndrome. Shuangyin Han constructed chimeric EGFRvIII scFv-ICOS-CD3ζ (EGFRvIII CAR) using lentivirus and found that there was a robust increase in IFN-γ secretion after coincubating anti-EGFRvIII CAR T cells with EGFRvIII-expressing U87 cells; moreover, EGFRvIII CAR T cells inhibited the in vivo growth of EGFRvIII-expressing glioma cells in a xenograft mouse model regardless of intravenous or intratumor injection (61). Yu et al. transduced human NK cell lines NK-92 and NKL and primary NK cells with a lentivirus containing anti-EGFR CAR to evaluate the anti-GB efficacy; they found that intracranial administration of anti-EGFR CAR NK-92 cells resulted in efficient suppression of tumor growth and significantly prolonged the survival of tumor-bearing mice (62).

KRAS is a pivotal oncogene in numerous human cancers; is the upstream activator in many signaling pathways, especially the MAP kinase pathway; and determines the division and metabolism of cells (63). The predominant variants of KRAS are site mutations at codon 12, in particular G12D and G12V, which account for 60–70% of pancreatic cancers and 20–30% of colorectal cancers (64). Interestingly, there is a hypothesis that EGFR and KRAS mutations have functionally equivalent roles in lung tumorigenesis because mutations of both EGFR and KRAS are rarely found in the same tumors (65). Qiong J Wang isolated one T cell receptor with high affinity for the mutated KRAS variants G12V and G12D, transduced peripheral blood lymphocytes (PBLs) with these TCRs and found that these genetically engineered PBLs could recognize multiple HLA-A*11:01⁺ tumor lines that destroy these target cells; in addition, the adoptive transfer of these transduced PBLs could significantly reduce the growth of tumors in a xenograft model (66). Tran et al. identified a polyclonal CD8⁺ T cell from a patient with metastatic colorectal cancer which responded to KRAS G12D mutation tumor cells; objective regression was observed after adoptive transfer TILs specifically targeted KRAS G12D (67). As a driver mutation, the KRAS mutant is conceptually attractive since it is tumor-specific and biologically important to tumor progression and is likely to be expressed in all kinds of tumors (68), which makes the KRAS mutant a hot spot for adoptive cell therapy.

MYD88 is a Toll-like receptor (TLR) adaptor protein, and ~90% of certain non-Hodgkin lymphomas (NHLs) have the Leu265Pro (L265P) mutation, which is a driver mutation (69). Signaling studies have shown that the mutation of MYD88 at site 265 triggers tumor growth by activating the NF-κB signaling pathway. As MYD88^L265P is a widely occurring and tumor-specific mutation in NHL, Nelde et al. predicted potential MYD88^L265P-containing HLA ligands for several HLA class I restrictions in silico. Three HLA-B*07-restricted peptides and one HLA-B*15-restricted peptide were identified, and they found that MYD88^L265P-derived peptides can induce mutation-specific and functional immune responses in vitro (41).

Isocitrate dehydrogenase type 1 (IDH1) usually mutates in the development of a subgroup of gliomas (70). The arginine residue (Arg132) in the catalytic pocket tends to mutate during tumorigenesis, resulting in different enzymatic functions that catalyze the production of the oncometabolite 2-hydroxyglutarate (2-HG). In 2-HG metabolism-deficient patients, the excess accumulation of 2-HG will foster brain tumor growth. Dang et al. found that human malignant gliomas harboring IDH1 mutations resulted in an increase in 2-HG, which contributes to the formation and malignant progression of gliomas (71). There is a statistic showing that more than 70% of diffuse grade II and grade III gliomas carry the most frequent mutation, IDH1^R132H (72, 73). Theresa Schumacher found that mice whose MHC molecules were deficient and that were vaccinated with the human MHC class I/II with IDH1^R132H p123-142 experienced tumor regression (74). Michael Platten has shown that IDH1^R132H is an immunogenic tumor antigen that induces mutation-specific CD4⁺ T cell and antibody responses that are capable of controlling IDH1^R132H-expressing tumor growth in vivo in MHC-humanized A2.DR1 mice after vaccination; because of its CD4⁺ T cell-dependent manner, IDH1^R132H is suitable for adoptive cell therapy (75).

The identification and characterization of mutant p53 pR175H was conducted by Steven Rosenberg's group (76). Mutant p53 pR175H is a kind of neoantigen found in a subset of patients with cancer. p53 has been described as “the guardian of the genome” because of its role in conserving stability by preventing genome mutation; thus, the mutation of p53 has a serious adverse effect on normal cells and accelerates the process of carcinogenesis as a driver mutation. Lo et al. screened tumor-infiltrating lymphocytes for the recognition of mutated neoantigens in a patient with metastatic colorectal cancer in a HLA-A-dependent manner; they found that the minimal peptide epitope of pR175H was HMTEVVRHC, and the screened TIL also mediated the recognition of p53 pR175H⁺ colon, breast, and leukemia cell lines after transduction with a retrovirus encoding HLA-A*0201 (76).

The MUC1 membrane mucin is a high-molecular-mass glycoprotein encoded by the muc1 gene, also called CD227, which was first identified through a monoclonal antibody binding experiment (77). It is well-known that MUC1 promotes cell growth and survival (78). Evidence shows that the overexpression of MUC1 is related to cell adhesion inhibition and the increased metastatic potential of tumor cells, especially in breast cancer (79). Although MUC1 in cancer cells is not directly produced by gene mutation, but its aberrant glycosylation and conformation changes are partly due to the mutation of other genes such as loss-of-function mutations in Cosmc gene in cancer cells (80). By confocal microscopy of immunostaining assay, Avery Posey found the Tn (GalNAcα1-O-Ser/Thr) glycoform of MUC1 (Tn-MUC1) expressed intracellularly in normal tissue such as human kidney, but at the cell membrane in several cancers. Based on that, his team developed an anti-Tn-MUC1 CAR-T cell that recognized the cancer-specific cell surface expressing Tn-MUC1, and demonstrated the target-specific cytotoxicity and suppression of tumor growth in xenograft models of T cell leukemia and pancreatic cancer (81). This finding demonstrates that Tn-MUC1 might be a potential target for cancer therapy. In early research, Wilkie et al. developed dual-targeted CAR-T cells by coexpressing ErbB2- and Tn-MUC1-specific CARs with respective costimulatory domain. They found that “dual-targeted” T cells kill ErbB2⁺ tumor cells efficiently, but their proliferation requires coincubation with the target cells with both Tn-MUC1 and ErbB2 expression (82).

In addition to the neoantigens mentioned above, there are many other antigens that have been discovered either by WES or by chance, such as the GAS7 mutant, CSNK1A1, and HAUS3 (83); these antigens deepen our understanding of tumorigenesis and development and can be developed as potential targets for cancer treatment. From a clinical perspective, preferred neoantigens would be formed by epitopes encoded by mutations that are shared across patients and, to reduce the risk of immune escape, locate to driver genes that are essential for tumor survival. On the other hand, identification of neoantigen-specific lymphocytes is also a promising and helpful way to enrich the treatment of cancer, Steven Rosenberg has developed an intact method to screen neoantigen-specific lymphocytes, they aim at tumor infiltration lymphocytes and use PD-1 as a biomarker to detect T cells that target neoantigens (84, 85). When the technology of neoantigen mining is well-established, the finding and application of neoantigens will save time and labor.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.