Emerging potential of immunopeptidomics by mass spectrometry in cancer immunotherapy

Abstract

With significant advances in analytical technologies, research in the field of cancer immunotherapy, such as adoptive T cell therapy, cancer vaccine, and immune checkpoint blockade (ICB), is currently gaining tremendous momentum. Since the efficacy of cancer immunotherapy is recognized only by a minority of patients, more potent tumor‐specific antigens (TSAs, also known as neoantigens) and predictive markers for treatment response are of great interest. In cancer immunity, immunopeptides, presented by human leukocyte antigen (HLA) class I, play a role as initiating mediators of immunogenicity. The latest advancement in the interdisciplinary multiomics approach has rapidly enlightened us about the identity of the “dark matter” of cancer and the associated immunopeptides. In this field, mass spectrometry (MS) is a viable option to select because of the naturally processed and actually presented TSA candidates in order to grasp the whole picture of the immunopeptidome. In the past few years the search space has been enlarged by the multiomics approach, the sensitivity of mass spectrometers has been improved, and deep/machine‐learning‐supported peptide search algorithms have taken immunopeptidomics to the next level. In this review, along with the introduction of key technical advancements in immunopeptidomics, the potential and further directions of immunopeptidomics will be reviewed from the perspective of cancer immunotherapy.

In this review, we discuss the rapidly advancing field of immunopeptidomics research for “cancer's dark matter” and its utility in the field of cancer immunotherapy along with the latest developments in mass spectrometry technology.

Abbreviations

DDA

data‐dependent acquisition

DIA

data‐independent acquisition

DIM

differential ion mobility

HLA

human leukocyte antigen

ICB

immune checkpoint blockade

IM

ion mobility

MRM

multiple‐reaction monitoring

MS

mass spectrometry

NCP

noncanonical ORF‐derived protein

ORF

open reading frame

PDC

patient‐derived cell

PRM

parallel reaction monitoring

PTM

post‐translational modification

Ribo‐seq

ribosomal sequencing

RNA‐seq

RNA sequencing

TIM

trapped ion mobility

tMS2

targeted MS2

TSA

tumor‐specific antigen

WES

whole‐exome sequencing

1. INTRODUCTION

Mass spectrometry (MS) is an analytical instrument that can obtain an over view of proteins (proteome), peptides (peptidome), and binding partner analysis (interactome) against the target of interest. ¹ In recent years, MS analysis has been significantly improved through synergies with advances in deep sequencing and deep learning technologies (Figure 1). Together with the development of the latest mass spectrometers and peripheral devices, the file size of raw data obtained by MS analysis is now greater than ever before. One research subject that has largely benefited from this progress is immunopeptidomics, the analysis of immunopeptides by MS. ²

FIGURE 1.

Current trends in interdisciplinary research in cancer immune therapy including immunopeptidomics by liquid chromatography–tandem mass spectrometry (LC‐MS/MS). To explore the “dark matter” of the cancer immunopeptidome, the collaboration of multiomics, data science, and engineering is required. Skills in molecular biology and biochemistry are also implicitly and necessarily required to prepare appropriate samples for immunopeptidomics in this interdisciplinary field. ORF, open reading frame; SNVs, single‐nucleotide variants; Indels, insertions and deletions; TEs, transposable elements.

Immunopeptides are antigen peptides extracellularly presented by major histocompatibility complex (MHC) class I and class II; the human MHC is called human leukocyte antigen (HLA). The class I complex is comprised of immunopeptides, α‐chain, and β2M as a trimer (Figure 1, center). In the α‐chain, the region of the immunopeptide‐binding groove, which forms a pocket to immobilize an immunopeptide, is known to be highly polymorphic (https://hla.alleles.org/nomenclature/index.html), ³ , ⁴ and this polymorphism results in immune diversity over individuals. The polymorphisms in the binding groove of the α‐chain create different consensus motifs that restrict the binding affinity of immunopeptides. Hence, the genetic combination of HLA alleles defines the immunopeptidome associating with immune diversity in individuals. Class I immunopeptides have long been studied for cancer immunotherapy due to its potential of immunogenicity against cytolytic T lymphocytes (CTLs, i.e., CD8⁺ TCs) against tumor cells. ⁵ , ⁶ , ⁷ , ⁸ Furthermore, the development of cancer vaccines has also been considered to be a major challenge in cancer immunology to prevent and fight against cancer. ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ In order to obtain a potential therapeutic cancer antigen from thousands of immunopeptide sequences, previous research has relied on a prediction system which usually provides dozens to hundreds of candidate sequences for further screening steps. The presentation machinery of class I immunopeptides is highly complex, involving genetic alteration, source translation, and the cancer microenvironment (Figure 2). Thus, prediction algorithms using in silico translated protein databases from genomic templates, lacking the node to calculate the multiple rate‐limiting factors in the intracellular antigen presentation machinery, often includes false‐positive candidates. Immunopeptides are, simply speaking, degradation products of intracellular proteins, but the first and basic premise here is that the source of immunopeptides should be translated into protein/peptide levels in the first place. The translated products include normal proteins, mutated proteins, and noncanonical ORF‐derived proteins (NCPs). In addition, among the nascent proteins, misfolded proteins and the population called short‐lived proteins (SLiPs) and defective ribosomal products (DRiPs) are also known to become antigen sources for HLA class I immunopeptides. ¹⁶ To trim these source proteins/peptides into a more desirable property as immunopeptides to make the HLA complex, proteasomal degradation and endopeptidase activity are required. More recently, the ubiquitin proteasome system (UPS)‐independent source supply has been also reported. ¹⁷ This foretells the possibility that there still remains the unidentified source and the presentation machinery for class I immunopeptides. Therefore, it is still difficult to obtain reliable TSA candidates without taking these biological factors into consideration. On the other hand, the existence of the T cell (TC) repertoire in cancer patients will be the concrete proof of the bona fide TSA, while gaining the single clone of TCs without plausible antigen information also requires tremendous efforts.

FIGURE 2.

Presentation machinery for class I immunopeptides at a glance: from genomic blueprint to T cell recognition. Successful transcription without nonsense‐mediated decay (NMD) is the first barrier for somatic mutation‐carrying immunopeptides. Next, both canonical open reading frame (ORF)‐ and noncanonical ORF‐derived mRNAs have to be fully translated into peptide/protein levels. Aberrant translation can happen by microenvironmental factors, such as IFN‐γ‐induced tryptophan deficiency or DNA damage. Short‐lived proteins (SLiPs) and defective ribosomal products (DRiPs) are also known as a source for class I immunopeptides. The translation step is an error‐prone process that produces misfolded nascent proteins that will be degraded by intracellular degradation. ¹⁴ Enzymatic activity of proteasome can be varied by the combination of subunits and two distinct types, called constitutive proteasome and immunoproteasome, which are known to associate with the immunopeptide presentation machinery in tumor cells. A relatively rare splicing event within the proteasome cavity can produce spliced peptide as class I immunopeptide source. Of note, in the antigen‐presenting cells (APCs), the cross‐presentation of class I immunopeptides can happen from the phagocytosed sources of extracellular debris and pathogens. ¹⁵ TAP proteins transport the source polypeptides into the endoplasmic reticulum, and further enzymatic cleavage by ERAP proteins trims the source polypeptides into appropriate length for human leukocyte antigen (HLA) complex formation. Stable HLA complex with immunopeptides/epitopes is extracellularly presented for immunosurveillance by T cells. APC, antigen‐presenting cell; NCP, noncanonical ORF‐derived protein.

These difficulties in cancer immunotherapy research have been gradually resolved by latest immunopeptidomics with appropriate MS analysis options and the contribution of these refined immunopeptidomics. Indeed, the discovery of tumor‐specific antigens and predictive markers for ICB treatment associated with personal immunopeptidome has been increasingly reported.

In this review, the application and the future potential of immunopeptidomics in cancer immunotherapy research will be discussed by introducing and outlining the key technologies and the progress of this interdisciplinary research field.

2. TRANSFORMATIVE TECHNOLOGIES THAT SHED LIGHT UPON THE DARK MATTER OF THE IMMUNOPEPTIDOME

2.1. Inextricable link between deep nucleic acid‐sequencing technologies and immunopeptidomics for new landscapes

To find a cancer‐specific antigen, it is inevitable to interrogate the individual genomic/transcriptomic information and incorporate the cancer specific sequences into personalized/customized database for peptide search. By whole‐genome sequencing (WGS) and whole‐exome sequencing (WES), sample‐specific mutations, including single‐nucleotide variants (SNVs), ¹⁸ insertion–deletion (in‐del) frameshifts, ¹⁹ , ²⁰ and onco‐viral sequences can be obtained. ²¹ , ²² RNA sequencing (RNA‐seq) and long‐read sequencing can provide the transcription information of NCPs for immunopeptides. These sources for immunopeptides and supporting sequencing technologies were well summarized in the previous reviews. ²³ , ²⁴ Technically, all the translated products derived from successfully translated mRNAs without nonsense‐mediated decay (NMD) can be a source for immunopeptides (Figure 2). Currently, sources for immunopeptides can be roughly classified into two groups: canonical open reading frame (ORF)‐derived proteins and NCPs. Immunopeptides derived from NCPs are also called “cryptic peptides.” These NCPs have been overlooked as a “dark matter” of proteomics for a long time. It becomes more general and feasible to adopt sample‐specific genome/transcriptome information for MS search as “proteogenomics” to uncover NCPs, a new landscape for immunopeptides. ²⁵ Although TSAs derived from somatic mutations uniquely found in tumor cells should be the best and ideal therapeutic targets from the viewpoint of specificity, the limited number of shared somatic mutations across cancer types and patients become a disadvantage to develop a practical target. As the translation from noncanonical ORF becomes a general knowledge in biology, whether the tumor‐specific and immunogenic antigen can be identified from these noncanonical ORF‐derived sequences, irrespective of somatic mutation, is of great interest. From this perspective, ribosome sequencing (ribo‐seq) and polysome sequencing (poly‐seq) are also actively introduced into immunopeptidomics to reflect the status even closer to the actual translation step than RNA‐seq. Transcriptomics data can include disease‐specific translations by alternative splicing such as aberrant translation under tryptophan deficiency ²⁶ , ²⁷ or the pioneer round of translation after DNA damage. ²⁸ Although it has often been overlooked, and thus should be noted, the mutation in the exon–intron junction region can also evoke aberrant splicing ²⁹ , ³⁰ that still translated within the canonical ORF. Therefore, strictly speaking, to discern the true “noncanonical” ORF‐derived immunopeptides, it may be necessary to confirm that the desired sequence never arises from the sample‐unique somatic mutations by aberrant splicing. As is known, the tumor‐specificity of an antigen is the critical factor when developing a safe immunotherapy.

The current task of NCP immunopeptidomics is to expand the unified sequence of long non‐coding RNAs (lncRNAs) and untranslated regions (UTRs). The current task of NCP Immunopeptidomics is to expand the unified sequence annotation of non‐canonical ORFs, such as those of long non‐coding RNAs (lncRNAs) and untranslated regions (UTRs). Recently, to make use of NCPs for better MS analyses, including immunopeptidomics, the standardized annotation of 7264 translated ORFs has been established. ³¹ In this project, the authors focused on constructing a comprehensive reference annotation resource to reciprocally improve the usage of Ribo‐seq data in the scientific community. For that purpose, the seven datasets from the key project for genome‐wide Ribo‐seq identification projects in the last 5 years ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ were curated. Among the annotated noncanonical ORFs, the consortium identified nearly 3085 as redundant (shared) that were identified from more than two samples, while the remaining 4179 sequences were found to be solely unique to the individual samples. In cancer immunopeptidomics, Ribo‐seq is still a relatively new technology, and the amount of information in tumor specificity and the recurrence is still limited; thus, these points should be addressed. It is expected that more Ribo‐seq data will be integrated into the standardized noncanonical database to clarify the homology of noncanonical translation across samples as well as to explore the “dark matter” of the immunopeptidome deeply.

The existence of “cryptic” immunopeptides derived from NCPs has long been a proposed concept. Owing to a combination of innovations in the fields of omics technologies, in the last few years there have been finally urgent attempts to identify tumor‐specific and immunogenic neoantigens from cancer dark matter. Examples of previously identified tumor‐specific/‐associated NCP‐derived neoantigens by immunopeptidomics are listed in Table 1. In a report that attempted to ensure tumor specificity based on sample‐specific genome and transcriptome information using multiple deep‐sequencing technologies, one sequence of an NCP‐derived neoantigen from patient‐derived cells (PDCs) of melanoma was reported as tumor specific and immunogenic. ³⁹ The authors also reported the four tumor‐specific NCP‐derived immunopeptides from lung cancer tissues, but the immunogenicity was not validated. ³⁹ Another example is a publication which employed Peptide‐PRISM, ⁴¹ a method to identify NCP‐derived immunopeptides using virtual translation against all possible reading frames of human reference genome and reference transcriptome without requiring the sample‐specific RNA data source. By utilizing Peptide‐PRISM, three NCP‐derived neoepitopes were identified from melanoma PDCs and reported as immunogenic. ⁴⁰ However, the authors mentioned that the transcriptomic source of these three neoepitopes may not be completely tumor specific, based on the screening of GEPIA using transcript datasets of TGCA and GTEx healthy controls. Still, the induced TCs against these three candidate sequences exhibited preferred immunogenicity for tumor cells rather than normal cells; thus, it can be stated that there remains a possibility that tumor‐specific/‐associated neoepitopes identified by Peptide‐PRISM could be a desirable target for cancer immunotherapy. Although further validation such as tumoricidal effect in an in vivo model is awaited because the immunogenicity of the NCP‐derived neoantigen was confirmed at the in vitro level, these reports are an important milestone that enlightens us about the potential of NCP‐derived neoantigens as a therapeutic target. On the other hand, we have occasionally noticed that in some sequences identified as NCP‐derived immunopeptides, there were sequences that may also be derived from canonical proteins. Considering that the transcriptional and translational complexity of the human genome and transcriptome results in a large number of entries, filtering methods for NCP validation should be carefully designed. In addition, the degree of homology of genomic sequences outside the CDS region among individuals has still been ambiguous. Furthermore, the in‐del variations outside the canonical‐ORFs have been reported to be associated with disease risk. ⁴² This indicates great diversity in noncanonical transcripts and the proteins produced from them among individuals. This can be kept in mind for researchers who employ reference genome/transcriptome datasets to identify NCP‐derived immunopeptides. As a further perspective for clinical application, what do we know about the recurrence of tumor‐specific/‐associated immunogenic NCP‐derived neoantigens?

TABLE 1.

Examples of tumor‐specific/‐associated NCP‐derived neoantigens identified by MS‐based immunopeptidomics.

Source gene description	Key method for immunopeptidomics	Clinical sample source	Noncanonical template source	Immunopeptide sequence	Length (aa)	Noncanonical type	Tumor specificity validation used	Immunogenicity validation	Ref.
RP11‐566H8.3	“Integradated proteogenomics”	Lung squamous carcinoma (snap‐frozen tissue)	WES Total RNA‐seq Single RNA‐seq Ribo‐seq	STYITKNFK	9	lncRNA	Immunopeptidome: not found in control (normal tissue) samples Immunopeptidome: not found in control (healthy samples) of ipMSDB Gene expression: TPM <1 in 1–3 tissues out of 30 healthy tissue controls by GTEx Gene expression: absent at the RNA level in the adjacent normal tissues	Not validated	39
HAGLROS				KVLAGTVLFK	10	lncRNA
HAGLROS				VLAGTVLFK	9	lncRNA
RP11‐398 K22.12				ILSSHATTRK	10	lncRNA
ABCB5		Melanoma (PDC)		KYKDRTNILF	10	dORF		Autologous & Induced TCs, ELISpot
HOXC13	“Peptide‐PRISM”	Melanoma (PDC)	3‐frame translated transcriptome (Ensembl90), six‐frame translated genome (hg38)	SSLPLASGVFKK	12	5’‐UTR	Immunopeptidome: not found in control (normal tissue) samples Immunopeptidome: not found in HLA Ligand Atlas (benign tissue samples) Minor to negative immunogenicity against normal cells	Induced TCs 4‐1BB Upregulation *HLA‐over expressed	40
ZKSCAN1				RVAEITGIVKK	11	5’‐UTR
C5orf22				STDLPILLK	9	Noncoding spliced variant

^*Immunogenicity validated nonC‐derived neoantigens are shown in bold style.

Abbreviations: HLA, human leukocyte antigen; MS, mass spectrometry; NCP, noncanonical open reading frame‐derived protein; PDC, patient‐derived cell; Ref., Reference; WES, whole‐exome sequencing.

One of the latest preprints reported the value of tumor‐specific recurrent NCP‐derived TSAs in glioblastoma. ⁴³ The authors defined tumor‐specific alternative splicing as “neojunctions” by curating publicly available RNA datasets from multiple cancers. The extracted tumor‐wide neojunction‐derived TSAs exhibited immunogenicity and tumoricidal effect induced by CD8⁺ TC clones. Further, the authors identified cognate neojunction‐derived TSAs from a patient‐derived glioblastoma cell line by immunopeptidomics. The identification of potential recurrent TSAs from NCPs for cancer immunotherapy is indeed making much progress.

2.2. Ion mobility, a new ion separation technology in MS

With the expansion of the landscape for the immunopeptidome, broader identification of immunopeptides is demanded. In proteomics, which generally use trypsin digests, the digestion of 10,000 molecules of a protein will result in 10,000 partial peptide fragments with the same sequence. In immunopeptidomics, the sample is generally prepared by immunoprecipitation of HLA complexes that include a variety of immunopeptide species. If the sample immunopeptidome consists of 5000 sequence species, each immunopeptide sequence has to be identified from only two peptides. This difference should be noted, as peptide identification in immunopeptidomics may become more delicate than proteomics. (The real immunopeptide identification should be affected by the abundance of HLA alleles and the source protein, as well as the ionization efficiency.) General sample preparation for immunopeptidomics is described in previous publications. ¹⁸ , ⁴⁴ , ⁴⁵ In order to increase the number of immunopeptide identifications, prefractionation by column chromatography has often been used, as in proteomics. However, this process has been reported to lose not a negligible amount of immunopeptides during sample preparation. ⁴⁶ , ⁴⁷ To avoid this kind of unfavorable sample losses, a new ion separation technology, called ion mobility (IM), is emerging. IM is basically the moving speed of ion in a given electric field. In MS, the analytes (here, immunopeptides) have to be ionized. IM is affected by the structural bulkiness of each ion and is therefore differentiated by the peptide sequence (Figure 3A). The IM system applies this ion property to separation by producing a quasi‐electric field. To date, two types of IM have been developed and introduced in MS. The first one is differential IM (DIM), which uses a so‐called compensation voltage (CV) to let the ions pass through at a given voltage (Figure 3B). For this IM, external devices, such as FAIMS‐Pro (Thermo Fisher Scientific), are available for certain mass spectrometers. The other type of IM is trapped IM (TIM) and is fully installed as an MS system such as timsTOF (Bruker). In this mass spectrometer, the ions are literally trapped in a separation module and accumulated (Figure 3C). A new dimension of ion separation by timsTOF yields a collisional cross section (CCS) to support peptide identification. Both DIM and TIM have advantages in creating a new dimension of ion separation without column purification by the physicochemical properties of peptides. DIM has an advantage of reducing background noise, and TIM has an advantage of accumulating precursor ions. We have previously reported that our DIM‐MS‐based immunopeptidomics exhibited improved identification efficiency in immunopeptidomics from cell line samples and further identified neoepitopes directly from 40 mg of colon cancer tissues. ¹⁸ This technology supported the identification of KRAS‐G12V carrying and another private mutation‐carrying neoepitope directly from clinical tissue as well. By comparative analysis of the in‐depth personal immunopeptidome, which includes thousands of immunopeptide sequences identified from patients individually, tumor‐exclusive peptide trimming was delineated. ¹⁸ These results indicate the advantage of IM in immunopeptidomics, which requires a sensitive analysis. Other IM‐MS‐based immunopeptidomics methodologies have been acknowledged by other groups as well to show their potential to expand the identification of immunopeptides. ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² Not long ago, one of the challenges in immunopeptidomics was reported to be the unfeasibility of analysis from a small amount of sample. ⁵³ However, the situation has been improving due to the advancements of IM, improved sensitivity, and the scan speed by the mass spectrometer.

FIGURE 3.

Ion mobility (IM), a new dimension for ion separation that supports immunopeptidomics. (A) A general notion of ion mobility. Molecules with the same molecular formula but distinct structure (bulkiness) can be distinguished by mobility. (B) Schematic diagram for a type of differential ion mobility (DIM). By generating sequential quasi‐electric fields that function as an ion filter, the passage of ion species is restricted by compensation voltage (CV). (C) A schematic diagram of trapped‐ion mobility (TIM). By generating a gradient electric field and gas flow, the ions are separated and accumulated at the equilibrium point of both factors within the separation device. Ramping down the voltage of the quasi‐electric gradient, trapped ions can selectively pass through the device.

2.3. Global, targeted, and hybrid identification of immunopeptides and the application in cancer immunology research

As we previously reviewed, proteomics by MS has analytical options (e.g., global and targeted proteomics) to identify analytes. ¹ The same can be applicable for immunopeptidomics. If the focus is on the identification of more variations of immunopeptide sequences, global MS operated by data‐dependent analysis (DDA) or data‐independent analysis (DIA) methods should be the first choice to try. On the other hand, if there exists already a specific immunopeptide of interest, targeted MS operated by multiple‐reaction monitoring (MRM), parallel‐reaction monitoring (PRM), or targeted MS2 (tMS2) approaches may also be opted for. The global MS approach can interrogate MS data against throughout the entire grounds like proteome. In the targeted‐MS approach, by acquiring MS data only for pre‐defined analytes, increased sensitivity against a lesser amount of immunopeptides can be expected. Indeed, our recent targeted‐immunopeptidomics analyses by DIM‐assisted targeted MS2 with Orbitrap detector (Thermo Fisher Scientific) revealed that oncogenic KRAS carrying neoepitopes, which have been overlooked in the past, can be identified from the general cancer cell lines. This result indicated the possibility that more neoepitopes with driver mutations are yet to be identified and also suggested that the analytical sensitivity of exploratory analytical methods, such as DDA, may still be insufficient to identify useful neoantigens. MS can provide different analysis methods depending on the research purpose, global immunopeptidomics (DDA or DIA methods), and further deeper searches by targeted MS (methods such as MRM, PRM, and tMS2).

To overcome the trade‐off relationship in identification comprehensiveness and analytical sensitivity, hybrid DIA, which combines DIA and PRM within a single analysis, has been described in recent reports. ⁵⁴ It seems that simultaneous analysis for global and targeted MS by this kind of a hybrid method is highly desirable in immunopeptidomics, as it can be performed with a limited amount of clinical samples.

Lately, comparative analyses of personal immunopeptidomics performed by DDA and DIA from lung cancer tissues have revealed the association of the class II immunopeptidome and tumor immune status, while the class I immunopeptidome has failed to stratify the tumor status. ⁵⁵ Although the publications of in‐depth immunopeptidomics for personal immunopeptidome analysis by nontumor and tumor samples from the same individual are still limited, in order to delineate the class I immunopeptidome signature for tumor heterogeneity, simply more sensitive or higher‐resolution immunopeptidomics, such as spatial immunopeptidomics or single‐cell proteomics, may be required. To date, microfluidics‐assisted single‐cell proteomics has been reported, and this was also applied to immunopeptidomics. ⁴⁷ , ⁵⁶ Although this is still a high‐end analytical method and requires a relatively special device and an ultrasensitive mass spectrometer that enables single‐cell proteomics, it is expected that further progress in single‐cell immunopeptidomics will allow us to understand tumor heterogeneity more in detail based on the immunopeptidome profile.

2.4. Implementation impacts of deep/machine learning and de novo sequencing in immunopeptidomics

The bigger MS raw data become in size, the more efficient search method will be required as technical key to identify more immunopeptides from MS raw data. The time performance to search peptides from the gigantic database becomes another challenge.

Even under the better separation by IM, the MS2 spectra used for peptide spectrum match (PSM) can still become chimeric. To discern the chimeric spectra separately into a singular PSM to expand the identification, the deep/machine learning‐assisted system, using additional metrics for peptide search below, is becoming an essential component for better MS analyses. From this aspect, the rescoring system for PSM using artificial intelligence (AI) prediction nodes has become a more standard technique. In addition to the previously used metrics, mass to charge ration (m/z), retention time (RT) and IM, the additional metrics, such as predicted spectral angle (SA) and the predicted RT, have been introduced to enhance the (immune)peptide identification. ⁵⁷ , ⁵⁸ , ⁵⁹ These synergic progresses both in MS hardware and the supportive search systems make robust identification possible from the growing large‐scale MS raw data. As handling of gigantic MS data has become more generalized, machine‐learned search‐assisting algorithms play a greater role in this field.

In addition to the existing commercially based software which identify peptides based on a PSM, development of script‐based high‐performance peptide search algorithms have also been reported. The search scripts of MSFragger and FragPipe, together with machine learning nodes of MSBooster and MSFragger‐Glyco, have been developed recently to enhance the identification of immunopeptides and glycosylated immunopeptides, respectively. ⁶⁰ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ For more variations of immunopeptides with post‐translational modifications (PTMs), PROMISE has been developed, which delineated the preference of the PTMs in a certain HLA allotype. ⁶⁵

Not only these kinds of boosting systems for peptide search but also de novo sequencing is gaining attention due to its potential in exploratory immunopeptidomics for NCPs. For example, Peptide‐PRISM utilized de novo candidates to annotate PSM onto noncanonical ORFs. ⁴¹ The most unique feature of de novo sequencing is that it arithmetically identifies peptides independently of the database, that is, candidate peptide sequences can be obtained without a database. One example of de novo sequencing software is PEAKs, which utilizes a machine/deep learning boost system for immunopeptide search (DeepNovo) ⁶⁶ and the developed sequencing algorithms Novor, ⁶⁷ pNovo 3, ⁶⁸ PointNovo/PGPointNovo, ⁶⁹ , ⁷⁰ SMSNet, ⁷¹ and Casanovo. ⁷² The current trend in processing the immunopeptidome by de novo sequencing is first to calculate peptide sequences arithmetically de novo, then to use an appropriate database for peptide search, and finally to investigate the unannotated sequences separately. It has been noted that the true identification rate using PEAKS software is lower than that using other existing database search software. ⁶⁵ This point should be kept in mind to comprehend the results obtained by de novo sequencing‐based peptide search.

3. EFFORTS TO DISCERN THERAPEUTICALLY POTENT NEOEPITOPES FROM THE VAST IMMUNOPEPTIDOME

In order to utilize the latest gigantic immunopeptidomics data in cancer immunotherapy research, it is required to discern the bona fide immunopeptides/neoantigens from irrelevant species based on criteria of intracellularly, naturally processed; actually presented; and immunogenically potent. Previously, the prediction algorithm based on a binding affinity and other metrics for actual presentation have been utilized to extract the possibly presented immunopeptides by binding affinity or proteasomal cleavage. ⁷³ , ⁷⁴ , ⁷⁵ However, considering the current size of expanded immunopeptidomics data, the validation of immunogenicity, such ELISPOT and tetramer assays by using cognate synthetic peptides, becomes a huge burden in research. In addition, due to intracellular degradation, which is easily affected by the steric structure of the source protein, using linearized partial peptides for validation assays is still fraught with the possibility of obtaining false‐positive results. Therefore, identification of bona fide therapeutic targets by MS is now the big agenda. Also here, the latest large‐scale immunopeptidomics offers the dataset to establish machine/deep learning‐boosted algorithms that effectively discern the immunogenic neoantigens.

Recently, auxiliary scores have been invented to extract which mutation‐carrying epitopes are more likely to be presented or immunogenic as an immunopeptide. The following are examples of these new supportive metrics to support actual presentation and immunogenicity of candidate neoantigens.

3.1. Antigen hotspots

It has been reported that immunopeptides do not equally emerge from the entire length of the source protein. ⁷⁶ Based on this feature of immunopeptide presentation machinery, antigen “hotspots” were revealed by annotating the immunopeptide sequences enrolled in the large‐scale database with each source protein and elicited which regions of the protein were likely to be presented as immunopeptides (Figure 4A). ⁷⁶ On this notion, it is expected that the mutation‐carrying epitope is plausible to be presented when the mutation of interest locates on a hotspot region (Figure 4B,C). The prioritization by this metrics has improved the efficacy of antigen presentation by up to 50%. ⁷⁶ However, the following possibilities can be overlooked for neoantigen presentation under this concept (Figure 4D): first, the case of substitution of originally deleterious amino acid residue in anchor position of a binding motif into a significantly favorable one; and second, the case of introduction of a new cleavage site by the substituted amino acid that produces new source peptides, and so on. Therefore, a system to rescue the false‐negatively excluded candidates should be anticipated.

FIGURE 4.

Toward a true immunogenic neoantigen identification. New metrics that support immunogenic neoantigens. (A–C) A metric by an antigen hotspot for the prediction of presentation. (A) Previously identified known immunopeptides were mapped onto each source protein to delineate the preferred region for class I immunopeptide presentation. (B) If somatic mutation A (indicated by red color) in cancer cells is located on the “hotspot,” this mutation‐carrying immunopeptide will be more likely to be presented. (C) Mutation B (indicated by blue color), not located on the “hotspot,” is regarded as unlikely to be presented under this metric. (D) Mutational impacts that may be overlooked by hotspot metrics (indicated by green color). Somatic mutation that affects enzymatic recognition for peptide trimming and the anchor position may be underestimated by hotspot metrics. This false prediction may be avoidable by combining the metrics for proteasome degradation. Predicting antigen presentation by hotspot metric alone may falsely predict the immunopeptide with deleterious mutation in anchor position (indicated in black X) as to be presented. This false prediction should be avoided by using this metric with other metrics, such as human leukocyte antigen (HLA) allotype‐specific binding affinity. (E–G) A metric by antigen quality for the prediction of immunogenicity. (E) If the substituted amino acid has similar chemical properties to the wild type, this epitope presentation is recognized as self‐antigen, and T cells are immune tolerable without exhibiting immunogenicity. (F) If the substituted amino acid has dissimilar chemical properties to the wild type, this epitope is more likely to be recognized as non‐self‐antigen that evokes immunogenicity. (G) Current limitation under the metric of antigen quality. The post‐translational modifications (PTMs), including already known immunogenic antigen‐producing glycosylation, ubiquitination, and the chemical modification, are yet to be included in this metric. Due to the lack of immunopeptides and corresponding immunogenicity assay datasets, at least to date, this metric is only available to the major HLA allotype, HLA‐A*02:01. TCR, T cell receptor.

3.2. Antigen quality

Another concept to predict therapeutically relevant epitopes is based on immunological tolerance (Figure 4E,F). The concept of antigen quality was invented by considering how a mutated epitope differs from its original state and how similar it is to the known immunogenic antigens, together with the actually validated immunogenic antigen datasets. ⁷⁷ , ⁷⁸ , ⁷⁹ According to this concept, the chemical properties of the mutated amino acid are more different from the wild‐type amino acid, and the epitope will be regarded as a non‐self, alienated antigen and therefore more likely to be immunogenic. While this prediction is currently applicable only to the major HLA allotypes with higher frequency, of which the TC immunogenicity validation dataset is available.

The latest publication reported that from datasets of WES and RNA‐seq (without MS data) together with a large‐scale screening dataset of neoantigen immunogenicity a machine learning‐refined prediction algorithm for immunogenic neoantigens was developed. During the developmental steps, immunogenicity was found to correlate with novel metrics such as antigen hotspots, binding promiscuity, and the physiological role of the mutated gene in cancer cells, enabling the prediction of immunogenicity beyond common metrics previously used. By this approach, improved prediction accuracy by up to 30% has been achieved. ⁸⁰ However, to date these metrics rely on the amino acid sequences of candidate epitopes and does not take PTMs into account (Figure 4G). It has been reported that the immunopeptidome with PTMs, such as phosphorylation, glycosylation, and methylation, on TC receptor recognition position exhibited a preferred affinity for certain HLA allotypes. ⁶⁵

Therefore, prediction algorithms based on immunopeptidomics and genomics/transcriptomic data are now being reorganized and integrated, and their combination heralds the completion of better prediction algorithms that can be used in clinical settings in the near future.

4. CONCLUSION

Currently, there are two major challenges in cancer immunotherapy research: the selection/combination of immunotherapy in light of the patient's medical history and cancer heterogeneity, and the identification of the optimal antigen for antigen‐centric immunotherapy. Until a few years ago, due to the complexity of human HLA types, the lack of pipelines for applying individual genomic/transcriptomic information including NCPs into a custom database for MS and the insufficient depth of MS analysis itself were the hindrances to tackle these issues. Although there is still a long way ahead, technologies and methods are currently being established to get the critical insights for cancer immunotherapy, and multidisciplinary research is being conducted at a rapid pace.

In this review, the key technologies contributing to the latest immunopeptidomics research and offering solutions for current challenges to improve cancer immunotherapy were summarized. At present, the identification phase of naturally presented immunopeptide species is being gradually saturated. It should be noted that the sequence species confirmed within the immunopeptidome is influenced by the zygosity of HLA alleles. Thus, that does not necessarily imply the “abundance” of immunopeptide presentation. Immunogenicity can be affected by the quality as well as the quantity of presented antigens, indicating that quantitative MS will become a next milestone in immunopeptidomics to validate the efficacy of immunotherapy.

In clinical settings, HLA allotypes complicate the prediction of presentation efficiency due to their diverse binding properties. This is because genetically distinct HLA by allotype is still structurally similar in binding pocket by supertype classification. ⁸¹ HLA allotypes with similar biding pocket structure can be expected to share/compete with the same immunopeptide in antigen presentation. The development of normalization according to sample‐specific HLA allotypes by motif similarity and the determination of the threshold counts of antigen presentation together with antigen quality should contribute to the better understanding of antigen immunogenicity that leads to the prediction of immunotherapy response. The interdisciplinary team science for immunopeptidomics will become increasingly important for the realization of precision cancer immunotherapy in the future.

Due to the rapid progress of MS and its related field, cancer immunotherapy, there remain countless significant publications and more details unmentioned in this review. Essential knowledge about immunopeptidomics has been described by an enormous number of publications, which were recently excellently summarized in the literature review of Admon et al. ⁸²

We hope non‐native MS researchers will be interested in this unique technology and that immunopeptidomics will continue to play a role in cancer immunotherapy research.

AUTHOR CONTRIBUTIONS

Yuriko Minegishi: Conceptualization; writing – original draft; writing – review and editing. Yoshimi Haga: Writing – review and editing. Koji Ueda: Writing – review and editing.

FUNDING INFORMATION

This work was partly supported by the Development of Technology for Patient Stratification Biomarker Discovery of the Japan Agency for Medical Research and Development (20ae0101074s0302). This work was supported by JSPS KAKENHI Grant Number 23K04971 to Y.M. Funding support from SHIMADZU Corporation.

CONFLICT OF INTEREST STATEMENT

K.U. is an editorial board member of Cancer Science. Y.M. and Y.H. have no conflicts of interest.

ETHICS STATEMENTS

Approval of the research protocol by an Institutional Reviewer Board: N/A.

Informed Consent: N/A.

Registry and the Registration No. of the study/trial: N/A.

Animal Studies: N/A.

Minegishi Y, Haga Y, Ueda K. Emerging potential of immunopeptidomics by mass spectrometry in cancer immunotherapy. Cancer Sci. 2024;115:1048‐1059. doi: 10.1111/cas.16118