Neoantigens in precision cancer immunotherapy: from identification to clinical applications

Introduction

Antitumor treatment has entered a new era since the successful application of immunotherapies, of which a growing number have emerged. Immune checkpoint inhibitors (ICIs), adoptive transfer of T/natural killer (NK) cells, tumor-associated antigen (TAA) vaccines, oncolytic viruses, and immunomodulators have been proven to be effective in clinical practice.[1–4] However, even ICIs, the most popular immunotherapeutic agents, are only permitted to treat a few specific types of cancer. Results from a large number of clinical trials show that the objective response rates (ORRs) of most cancer immunotherapies are far from satisfying, let alone their association with frequent immune-related adverse events (irAEs).[5,6] Therefore, cancer immunotherapy must be more precise to achieve higher ORRs and fewer irAEs. Facing this challenge, pioneers have made considerable efforts to discover appropriate targets. Among the targets discovered, neoantigens are ideal for precision cancer immunotherapy.[7]

Neoantigens are abnormal peptides specifically expressed by malignant cells that are present on their surfaces. Most neoantigens are products of accumulating mutations in somatic cells. In virus-associated cancers, such as cervical cancer, neoantigens may be the products of the open reading frames of viral genomes.[8] In contrast to TAAs, neoantigens can activate CD4+ and CD8+ T cells exempt from central tolerance because they are totally non-self for the immune system.[9] Identification of neoantigens by researchers can be traced back to the 1980s.[10] Initially, complementary DNA library screening was used to identify neoantigens.[11] With the application of next-generation sequencing (NGS) technology and bioinformatic algorithms, the prediction of neoantigens became cheaper, easier, and faster.[12] However, prediction results are sometimes unsatisfactory because of the false-positive rate. By combining NGS with mass spectrometry and bioinformatics tools, the identification of neoantigens becomes more accurate. Thus, precision cancer immunotherapy targeting neoantigens has become more feasible [Figures 1–4].

Figure 1:

T cell response induced by neoantigens. DC: Dendritic cell; NeoAg: Neoantigen; T: T cell.

Figure 2:

The workflow of neoantigen identification. The whole workflow can be divided into five steps. The first step is sequencing of the cancer cells by WGS, WES or RNA-seq, etc. The following steps are variants calling, HLA typing and interactions prediction (interactions among HLA-neoantigen-TCR) based on the sequencing data. The last step is screening the interactions in vitro. ELISA: Enzyme-linked immunosorbent assay; GLIPH: Grouping of lymphocyte interactions by paratope hotspots; HLA: Human leukocyte antigen; INDELs: Short insertions and deletions; RNA-seq: RNA sequencing; SNPs: Single nucleotide polymorphisms; TCR: T cell receptor; WES: Whole exome sequencing; WGS: Whole genome sequencing.

Figure 3:

Dynamic neoantigen tracking is used to evaluate the prognosis of cancer patients. This evaluation strategy can be more sensitive and reliable than imaging examination. Individualized gene panels are the key point to make this strategy sensitive and economical. ctDNA: Circulating tumor DNA.

Figure 4:

Therapies based on neoantigens. CAR: Chimeric antigen receptor; DC: Dendritic cell.

Immunotherapies targeting neoantigens are safe, precise, and have demonstrated their potential in numerous clinical trials. Personalized neoantigen vaccines have proven to be feasible, safe, and effective.[9] Adoptive cell therapy targeting neoantigens has also achieved encouraging results.[13] Neoantigens are also important biomarkers for immunotherapy. Taking all these points into consideration, we think it is quite significant to make a comprehensive review of neoantigens and to discuss their future roles in immunotherapy.

Immune Responses Induced by Neoantigens Immunogenicity of neoantigens

In 1988, researchers discovered that neoantigens could promote the proliferation of cytotoxic T cells in a mouse tumor model.[11] Eight years later, another group of researchers discovered that neoantigens could be recognized by tumor-infiltrating lymphocytes (TILs) in a patient with melanoma.[14] An increasing number of studies have verified the immunogenicity of neoantigens. Briefly, immunogenicity is the ability of neoantigens to activate adaptive immune responses. Neoantigens are treated as non-self by the immune system because they are abnormal peptides expressed by mutated genes in cancer cells. Without the limitation of central tolerance, neoantigens can induce immune responses, as has been proven (Table 1).

EGFR-TKI: Epidermal growth factor receptor-tyrosine kinase inhibitor; 6MHP: 6 Melanoma helper peptide vaccine; mFOLFIRINOX: Leucovorin+5-fluorouracil+irinotecan+oxaliplatin; PD-1: Programmed death-1; PD-L1: Programmed death ligand 1; RFA: Radiofrequency ablation; RT: Radiotherapy.

However, because of the genomic instability of cancer cells, the heterogeneity of neoantigens is so significant that their immunogenicity varies. The structure of the neoantigen is not the only factor that affects its immunogenicity. The way it is processed and presented is also an important factor. After release from cancer cells, neoantigens are captured by antigen-presenting cells (APCs), such as dendritic cells (DCs) and macrophages. Neoantigens are degraded into small peptides by the proteasome and then loaded into major histocompatibility complex (MHC) molecules and presented on the surface of APCs.[15] Then, neoantigen-loaded APCs migrate into tumor-draining lymph nodes. When the relative location of APCs and T cells is convenient for interaction, specific T cells recognize neoantigen-MHC complexes by T cell receptors (TCRs). Hence, the spatial location of APCs is another critical factor.

T cell responses

The successful presentation of highly immunogenic neoantigens does not necessarily induce an effective T cell response. Multiple negative co-stimulatory molecules and mechanisms impair T cell activation. For example, immune checkpoints are negative mechanisms that inhibit T cell activity. Many clinical trials focusing on ICIs have proved that immune checkpoint blockade (ICB) is an effective strategy to restore the activity of T cells. ICB has been shown to enhance the durable neoantigen-specific T-cell responses.[16] Perumal et al[17] found that the activation and expansion of neoantigen-specific T-cells are improved after treatment with ICIs. Their work showed that therapy based on neoantigens can induce stronger T cell responses in combination with ICB.

In addition, there are many positive co-stimulators of T cell activity. Both interleukin (IL) 2 and IL-15 stimulate the activation and proliferation of CD8+ T cells and NK cells. Although numerous clinical studies have proven that neoantigen vaccines or adoptive neoantigen-specific T cells can achieve satisfactory anti-tumor immunity alone, combination with positive co-stimulators can improve anti-tumor immunity.[18,19] In addition, the IL-2/CD25 fusion protein has been shown to amplify neoantigen-specific CD4+ and CD8+ T cell responses.[20] In contrast to IL-2 alone, the IL-2/CD25 fusion protein is more selective, effective, and less toxic. The implication of this study is that the traditional positive co-stimulators, which may have severe side effects, can be engineered into safer ones. This will accelerate their application in immunotherapy, especially therapies based on neoantigens.

Memory T cells

Neoantigens can induce effector T cells and memory T cells. After exposure to neoantigens, specific T cells mature from naïve cells to effector cells and then eventually to exhausted cells or memory cells. This is the foundation of long-term remission in patients with cancer after immunotherapy based on neoantigens. The number of memory cells among TILs correlates with the outcomes of patients receiving immunotherapy.[21] Although CD8+ T cells are considered the most common sub-type of memory T cells,[22,23] Hu et al[24] showed that neoantigen vaccines can induce CD4+ memory T cells that persist for years in patients with melanoma. Further analysis of long-term TCR sequencing data and flow cytometry data showed that the clonal memory T cells were CD4+ αβT cells. A recent study showed that memory T cells can recognize specific antigens as well as shared ones.[25] This finding suggests that personalized neoantigen vaccines can be applied to multiple patients with the same somatic mutation.

Neoantigen Identification

During the past several years, a number of analytical pipelines for neoantigen discovery have been proposed to predict peptides that have the potential to induce tumor immune responses related to T-cell activation.[26–28] Binding affinity between the peptides and MHC is successfully used to select neoantigens against colorectal tumor[29], melanoma[9] and glioblastoma.[30,31] Fitness model integrating binding affinity between MHC and peptide with sequence similarity of peptide and virus sequence is used to identify neoantigens in melanoma, lung and pancreatic cancer.[32,33] Differential agretopicity index model evaluating the binding affinity difference with MHC between wild-type peptide and mutant peptide is applied upon melanoma and lung cancer.[34] High quality neoantigen model is successfully developed to determine neoantigens for primary isocitrate dehydrogenase wildtype glioblastoma.[35] NeoScreen, a recently developed in vitro TIL expansion and screening methodology, is reported to enable the selective expansion of neoantigen-targeting TILs against melanoma, colon, lung and ovarian cancers.[36] NeoScreen first applies computational methods to identify neoantigen candidates, which are then pulsed into the engineered B cells for antigen presentation. The B cells are cultured with tumor cells and TILs, which are isolated for identifying the neoantigens and TCRs.[36] DLpTCR, a multimodal ensemble deep learning framework, is able to predict the likelihood of interaction between single/paired chains of TCR and peptide presented by MHC.[37] DeepImmuno is a deep learning based method able to predict immunogenic peptides for T-cell immunity.[38] However, no consensus approaches have been established by mathematical and statistical models, which have great variations among neoantigen identification methods. The general workflows are quite similar and mainly include pre-processing of raw data, read alignment, somatic mutation calling, human leukocyte antigen (HLA) allele typing, peptide inference, peptide-MHC binding prediction, TCR-peptide-MHC complexes (pMHC) interaction estimation, and in vitro immune screening. The general approaches for neoantigen identification and applications have been extensively reviewed elsewhere.[39–41] In this review, we summarize the state-of-the-art computational tools for neoantigen analysis and provide an extensive discussion of critical concepts and practical guidance for each analysis step.

Somatic variant calling

The pre-requisite for inferring neoantigens is the selection of appropriate tools for the identification of somatic variants. A large variety of mutation types can produce neoantigens in cancer, including single nucleotide variants (SNVs), short insertions and deletions (INDELs), gene fusions, exon-exon junctions, intron retentions, and alternative splicing events. Numerous somatic variant callers are applied to DNA sequencing data (whole genome sequencing [WGS], whole exome sequencing [WES], or targeted amplicon data) of tumor and matched non-tumor samples to identify somatic mutations including SNVs and INDELs. However, the sensitivity and false positive rates of these methods are highly variable, and this leads to substantial differences among called variants.[42] Popular SNV calling methods include SAMtools, VarScan2, MuTect, VarDict, SomaticSniper, and Strelka.[43] MuTect2, Strelka, GATK, SAMtools, VarScan, and VarDict are also able to call INDELs while Pindel is specifically designed for large INDEL calling.[44] For high-allelic-fraction somatic SNV calling, the impact of caller selection is generally weak. MuTect and Strelka have high sensitivity for detecting low-allelic-fraction SNVs. VarScan2 and SomaticSniper provide less sensitivity for calling somatic SNVs of the low-allelic-fraction.[45] VarScan2 sensitivity can be improved by tuning the minimum allele fraction threshold from the default value of 0.2 to a lower value but at the cost of significantly compromised specificity.[45] SAMtools with default settings reports any possible changes in nucleotides with low specificity and with a high false-positive rate. VarDict calls a few more sites than expected in the benchmark data, which suggests its low sensitivity in SNV calling. Multiple callers and repeats significantly decreased false-positive calls for SNVs and INDELs. A manual review of somatic mutations from callers in the Integrative Genomics Viewer can further reduce false positives. Gene fusions can be detected from RNA sequencing (RNA-seq) data using various tools including FusionHunter, FusionMap, MapSplice, TopHat-Fusion, BreakFusion, SOAPfuse, EricScript, and FusionCatcher.[46] The performance of gene fusion detection tools largely depends on the quality, read length, and number of reads of RNA-seq data.[47] Overall, neoantigen identification requires a sensitive, accurate, and comprehensive somatic variant calling pipeline that can robustly detect all variant classes that are relevant to a tumor type.

HLA allele typing

T cells recognize neoantigens presented by APCs on MHC I or II molecules, which are encoded by the HLA gene complex located in highly polymorphic regions of chromosome 6p21.3 and have over 12,000 alleles.[48] As HLA genes are unique to each individual, it is essential to have accurate HLA haplotyping for neoantigen prediction. Sequence-specific PCR amplification, which is laborious and expensive, is the gold standard for clinical HLA typing. Computational HLA typing is becoming a popular approach and uses patients’ WGS, WES, or RNA-seq data from a peripheral blood or skin sample. Multiple tools including seq2HLA, PHLAT, HLAMatchmaker, HLAreporter, HLAforest, HLAminer, and xHLA were developed to precisely type class I HLA.[41]

However, class II HLA typing is still challenging and unreliable, with a few benchmarking studies reporting that PHLAT, HLA-VBSeq, seq2HLA, xHLA, and HLA-HD show comparable accuracies with WES and RNA-seq data.[49] Due to the varying capturing efficiency of DNA from HLA genes, it is critical to carefully examine the read coverage from WES/RNA-seq for HLA genes or to construct ensemble methods to produce optimal prediction.[50,51]

Neoantigen HLA-class I allele interactions

About 8-11 N- and C-terminal peptide residues are involved in binding with MHC class I molecules, which then present the peptides to cytotoxic CD8+ T lymphocytes to elicit T cell immunity.[52] It is necessary to predict the binding affinity of mutant peptides to HLA-class I alleles to prioritize mutant peptides. A few computational methods have been developed to evaluate the binding affinity between HLA-class I alleles and mutant peptides and include scoring function-based methods, machine learning-based tools, and consensus methods. The performances of HLA-class I-peptide binding prediction methods have reached a high level (auROC > 0.95). Scoring function-based methods prioritize mutant peptides by calculating sequence features including sequence similarity, amino acid frequencies, position-specific scoring matrices (PSSMs), and BLOSUM matrices. SYFPEITHI, RANKPEP, PickPocket1.1, SMMPMBEC, PSSMHCpan 1.0, and MixMHCpred 2.0.1 are scoring function-based approaches.[53] Machine learning tools such as NetMHC 4.0,[54] NetMHCstabpan 1.0,[55] NetMHCpan 4.0,[56] ConvMHC,[57] HLA-CNN[58] and MHCflurry 1.2.0[59] assign a mutant peptide as a binder or non-binder by constructing a training model using extracted representative features. Consensus-based tools such as NetMHCcons 1.1[60] integrate several peptide-MHC binding prediction methods in a weighted manner. The most widely used is NetMHC-pan 4.0, which uses artificial neural networks trained on a combination of more than 85,0000 quantitative binding affinities and mass-spectrometry-eluted ligand peptides.

Neoantigen HLA-class II allele interactions

Neoantigens are presented by MHC class II molecules to CD4+ T cells or “helper T cells”, which may then stimulate humoral or cell-mediated immune response pathways.[61] Precise prediction of peptide-MHC interactions is important for identifying neoantigens. Unlike MHC class I molecules, class II molecules, which include an α-chain and a β-chain, have higher variability with an open binding pocket on both ends that allows a larger range of peptides to bind. The peptides that bind to MHC class II molecules have a length range of 13 to 25 amino acids. Computational models for MHC class II peptide binding prediction are usually built on matrix-based approaches and ensembles of artificial networks. Popular methods include ARB, MHCPRED, PROPRED, RANKPEP, SMM-align, SVMHC, SVRMHC, SYFPEITHI, MultiRTA, and NetMHCIIpan.[51] A very recent transformer neural network model, BERTMHC, was reported to outperform other models[62,63] Consensus methods that combine multiple models may help improve performance, but rigorous ongoing efforts are needed to further improve the effectiveness of existing models.

TCR-pMHC interaction

Not all peptides presented by MHC molecules are immunogenic. TCRs must contact both the peptide and MHC molecules to elicit an immune response. There are remarkable similarities in the topology of TCR binding to pMHC irrespective of MHC class I or class II restriction. TCR-pMHC interactions can help further narrow down true neoantigens out of MHC-presented peptides. Very recently, a number of machine learning computational methods constructed based on complementarity determining region 3 (CDR3) sequences or additional cellular information such as SETE,[64] ERGO,[65] NetTCR-2.0,[66] and TCRMatch[67] have been reported to predict TCR-pMHC interactions. However, the prediction models of TCR-pMHC interactions have poor performance owing to the limited availability of training data. There is an urgent need for the development of cost-effective and accurate computational methods to assess TCR-pMHC binding.

In vitro immune screening

In vitro screening of T cell responses triggered by candidate neoantigens may provide direct evidence that a given neoantigen is immunogenic. T cell responses elicited by neoantigens may cause clonal expansion of T cells, upregulation of activation markers on the cell surface, effector cell differentiation, cytotoxicity induction, or cytokine secretion. A popular approach is to assess the activation of T cells in peripheral blood mononuclear cells by neoantigens in vitro (generally 2–4 days of incubation). An alternative method for in vitro T cell stimulation is to use Escherichia coli that expresses predicted neoantigens in autologous APCs, which are then incubated with T cells derived from the patient's PBMCs. T cells are then harvested and processed for various assays. Production of cytokines such as tumor necrosis factor α and interferon γ can be measured through enzyme-linked immunospot assay[68] or intra-cellular cytokine staining. Additionally, the clonality and diversity of the TCR repertoire suggest whether neoantigen-triggered T cell responses have occurred,[69] and TCR clonotyping (identification and characterization of CDR3 sequences of T cell clones) has been used to identify clonal T cell responses to neoantigens. CDR3 sequences can be identified using output data from focused assays such as Adaptive, ClonTech, or CapTCR bulk tissue RNA-seq[70] and single-cell RNA-seq.[71] It is essential to characterize the diversity of the repertoire, determine the pairing of TCR α (TRA) and TCR β (TRB) clonotypes, and pair T cell clones with their target neoantigens. MiXCR was developed to extract TCR sequences from raw data of both bulk and single-cell sequencing and then group them into identical clonotypes.[72] MIGEC[73] was designed for methods using unique molecular identifiers, and TraCeR,[71] which is specific for single-cell methods, is able to identify paired α–β sequences derived from the same clonally expanded cells. Oligoclonal T cell populations with consistent CDR3 motif sequences are reported to be able to recognize the same neoantigen,[74] therefore, it is possible that there is one-to-one mapping between T cell clones and neoantigens. Grouping of lymphocyte interactions by paratope hotspots can identify CDR3 motifs across T cells in bulk TCR sequencing.[75] Collectively, in vitro immune screening can further filter out immunogenic neoantigens, which have the potential to improve the survival of patients with cancer.

Clinical Applications of Neoantigens Biomarkers for immunotherapy

To optimize the efficiency of immunotherapy, researchers have investigated numerous prognostic biomarkers such as tumor mutation burden, microsatellite instability, mismatch repair deficiency, TILs, and PD-L1 expression. Recently, more attention has been paid to a new biomarker, the neoantigen. Lauss et al[76] showed that a higher neoantigen load is associated with better progression-free survival and overall survival (OS) of patients with melanoma treated with adoptive T-cell therapy. Another study revealed that specific neoantigens contribute to the long-term survival of patients with pancreatic cancer.[33] Furthermore, the investigators observed the selective loss of neoantigens with metastatic progression, which implied that neoantigens could be used as biomarkers for disease progression and patient survival. Similarly, Ren et al[77] reported that neoantigens can be used to predict OS independent of TILs and other biomarkers in breast cancer. They showed that higher HLA-I- or HLA-II-restricted neoantigen load correlated with better OS. For patients with adenoid cystic carcinoma (ACC), researchers found that a low neoantigen load was correlated with a poor response to anti-PD1 therapy and a high cancer recurrence rate. The investigators suggested that successful immunotherapy for ACC requires a “hot” immune microenvironment, namely, more TILs and neoantigens. Therefore, neoantigens can be utilized as positive prognostic biomarkers similar to TILs. However, Quintana et al[78] showed that tumors with more TILs have lower neoantigen loads. Contrary to previous studies, their results hint at a negative correlation between TILs and neoantigens. Their hypothesis is that failure of lymphocytes to migrate into the tumor leads to the survival of tumor cells and neoantigens. Thus, the negative correlation between TILs and neoantigens suggests that neoantigens can be used as negative prognostic biomarkers, contrary to TILs. The reason why they reach a conclusion different from others may be that their data are not comprehensive because they exclude patients with TILs percentage between 5% and 50%, who may be the major group who need to be considered. Furthermore, Balachandran et al[33] revealed that it is the quality rather than the quantity of neoantigens that correlate significantly with prognosis. Similarly, Rosenthal et al[79] proved that high clonal neoantigen burden was associated with better prognosis while the sub-clonal or total neoantigen burdens were not. Therefore, we can evaluate the prognosis before and during immunotherapy by tracking high-quality neoantigens. Our previous work showed that high-quality neoantigens, tracked by circulating tumor DNA (ctDNA) sequencing, can be used to predict the prognosis of patients with non-small cell lung cancer (NSCLC).[80] Specifically speaking, because of the high heterogeneity of somatic mutations among patients, it is infeasible to track neoantigens by ctDNA sequencing using common gene panels, which would be far too complicated and costly. Thus, we proposed the method of using the individually customized panels to track neoantigen evolution during ICB treatment. Our method is proved to be more sensitive, economical and feasible in clinical practice, making sequencing-based personalized management in immunotherapy possible.

Therapies based on neoantigens

Therapies based on neoantigens can be divided into two types: neoantigen vaccines and neoantigen-specific T-cells. Neoantigen vaccines include four main categories: peptide vaccines, mRNA vaccines, DNA vaccines, and DC vaccines. Each category has its own technical route and clinical characteristics.

Peptide vaccines based on neoantigens are widely used in clinical practice. Ott et al[9] synthesized 20 peptides that target personal tumor neoantigens. Vaccinated with these peptides, 80% (4/6) of melanoma patients survived without recurrence for 25 months. The other 2 patients who experienced recurrence achieved complete recession after subsequent anti-PD-1 therapy. The investigators discovered that peptide vaccines can activate pre-existing neoantigen-specific T cells and induce a wide range of new specific T cells. In a glioblastoma trial, researchers have shown that neoantigens can ignite specific circulating CD4+ and CD8+ T cell responses and increase TILs. This trial also proved that circulating neoantigen-specific T-cells can be recruited to intra-cranial tumors.[30] Mueller et al[81] revealed that a H3.3K27M-specific vaccine could induce neoantigen-specific T cell responses in 39% (7/29) of participants. Patients with a positive response had a significantly longer OS than the others. Recently, another study showed that it is feasible and safe to combine peptide vaccines with anti-PD-1 therapy in melanoma, NSCLC, and bladder cancer. The 1-year OS rates were 96%, 83%, and 67% for patients with melanoma, NSCLC, and bladder cancer, respectively, which were better than the historical data from anti-PD-1 monotherapy studies.[82]

Kreiter et al[83] showed that neoantigen mRNA vaccines can induce cytotoxic T lymphocyte responses and reshape the tumor microenvironment associated with tumor control in mice. Furthermore, Cafri et al[84] designed an mRNA vaccine encoding up to 20 neoantigens for patients with gastrointestinal cancer. The mRNA vaccine has been shown to safely elicit neoantigen-specific T cell responses. Unfortunately, the investigators observed no objective responses among the four patients in the trial. As mentioned by other researchers, instability may be a critical obstacle for mRNA vaccines.

DNA has also used to produce neoantigen vaccines. Researchers have shown that synthetic DNA vaccines – recombinant plasmids that can produce neoantigens – can generate robust T cell responses in mice.

View original article

CHINESE MEDICAL JOURNAL

Like

分享书签

0 0 0 0 0 0 0

More from this channel

Neoantigens in precision cancer immunotherapy: from identification to clinical applications

留言 (0)