Exploring the dynamic landscape of immunopeptidomics: Unravelling posttranslational modifications and navigating bioinformatics terrain

3.1. Sample preparation

The immunopeptidomics workflow consists of multiple steps and a standardized protocol for this workflow has been published by Purcell et al. (2019). In this review, we aim to provide a concise overview of the crucial steps involved in immunopeptidome analysis. This analysis differs significantly from classical proteomic approaches, presenting unique challenges throughout the workflow, as extensively reviewed by Faridi et al. (2018). Many hampering factors have been circumvented by improvements in the efficiency of isolating immunopeptides, the rapid development of instrument sensitivity, and improved bioinformatic pipelines, resulting in the ongoing project of mapping of the human immune‐peptidome project (HIPP) (Caron et al., 2017). The starting material in an immunopeptidomics experiment can consist of cell lines, primary cells, and fresh or fresh‐frozen tissue samples. Although the recent advances in the sensitivity of mass spectrometry (e.g., the increased focus on single‐cell proteomics) have reduced the amount of material needed for an immunopeptidome analysis, the amount of sample material available remains a crucial limiting factor. An interesting development in that aspect is the use of patient‐derived organoids (PDO). Hereby, a small patient biopsy can be multiplied to a larger sample, with the intent of mimicking the patient's initial biopsy condition (Wang et al., 2022).

Figure 2 provides an overview of the current immunopeptidomics workflow. Two well‐established techniques for isolating HLA‐bound peptides from tissue or cells are mild acid elution (MAE) (Sugawara et al., 1987) and Immunoprecipitation (Subramanian, 2002). The commencement of the immunopeptidomics workflow with immunoprecipitation involves the lysis of tissue or cells. This critical step employs a mild detergent to ensure the solubilization of MHC, a transmembrane protein, for subsequent processing. Monoclonal antibodies (mAbs) are then used for the immunoaffinity capture of MHC‐peptide complexes. The selection of mAbs is contingent upon the research objective and the HLA allomorph of the investigated cells. A comprehensive list of commonly used mAbs and their references can be found in the protocol published by Purcell et al. (2019). Depending on the research question at hand, using specific antibodies for a particular HLA allele may offer higher specificity. Most commonly used for HLA class I isolation is the mouse mAb W6/32, which is specific to the heavy chain of HLA‐A, ‐B, and ‐C (Brodsky & Parham, 1982). This allows for the monomorphic isolation of HLA class I peptides. Polymorphic antibodies, such as ME1 against HLA‐B07, BB7.2 against HLA‐A02, and GAP.A3 against HLA‐A*03, target and recognize determinants carried by specific alleles (Hilton & Parham, 2013). For HLA II on the other hand, the pan‐HLA class II antibody supposedly pull down the HLA molecules DR, DQ, and DP. However, immunopeptidomics studies have demonstrated poor specificity towards DQ and DP molecules, hence resulting in low yields of peptides bound to these loci and underrepresentation of DQ and DP in datasets(Nilsson et al., 2023; Racle et al., 2019). Additionally, even though a lot of studies focus on DRB1 alleles, it has been shown that DRB3, 4, and 5 play an important role in forming the HLA II immunopeptidome, underlying the importance of full HLA typing (Kaabinejadian et al., 2022). Following immunoprecipitation, the nonspecific lysate components are thoroughly washed away, leaving only the MHC‐peptide complexes. The MHC‐peptide complex is then eluted from the mAbs using a mild acid, dissociating the peptide from the MHC. Eluted immunopeptides undergo additional enrichment processes to reduce complexity and eliminate contaminants. Standard techniques for enrichment are solid‐phase extraction (SPE) (Bassani‐Sternberg et al., 2010) and ultrafiltration using a molecular cutoff (MWCO) filter (Sturm et al., 2021). These techniques enhance the sensitivity and selectivity of subsequent analyses. During the enrichment procedure, it is crucial to remove the intact proteins that are part of the MHC complex, such as beta 2 microglobulin and the MHC alpha chain, as their presence interferes with downstream analysis. Utilizing restricted access material (RAM) into solid‐phase extraction enhances the recovery of immunopeptides, particularly favoring hydrophobic ones. This improvement is attributed to the selective non‐retention of contaminating proteins by the RAM material, allowing for the application of a higher concentration of organic solvent during the elution process (Bernhardt et al., 2022). Kuznetsov et al. (2020) published a thorough review emphasizing various techniques for the isolation of the immunopeptidome.

MAE, in contrast to immunoprecipitation, utilizes viable cells to isolate peptides presented by the MHC on cell surfaces. In this method, cells are treated with a mild acid, causing peptides to separate from the MHC. Importantly, it's highlighted that peptides associated with MHCII are believed to remain unaffected by this process (Sugawara et al., 1987). When evaluating the two methods using B‐cell lymphoblasts, the Immunoprecipitation technique results in a significant 6.4‐fold increase in the detection immunopeptides (Lanoix et al., 2018). Furthermore, a different study comparing an enhanced MAE technique with immunoprecipitation revealed a nearly 50% overlap in the number of identifications between the two approaches (Sturm et al., 2021). There is a crucial need for a high‐throughput analysis methodology to enable the application of immunopeptidomics in mid to large‐sized cohorts and for future comparative studies. As Sian et al. suggested, the semi‐automated workflow revolves around the extraction of MHC‐bound peptides through for instance magnetic beads for immunoprecipitation, performed on an automated platform employing 96‐well plates. According to their protocol, it is possible to simultaneously process 12 samples, identifying roughly 400 to 13,000 distinct peptides, derived from cell quantities ranging from 0.5 to 50 million (Lim Kam Sian et al., 2023). However, it should be noted that in a high throughput set‐up the MS analysis and the subsequent data analysis are very likely to become the bottleneck as both are time‐consuming steps in the analysis. Considering the significance of both high throughput and sensitivity, Li et al.'s research team developed a workflow utilizing a microfluidics platform. This platform facilitates automated liquid handling and minimizes sample transfers. Through this streamlined process, the team successfully identified more than 4000 and 5000 MHC‐I‐restricted immunopeptides from 0.2 million RA957 cells (human B‐cell line) and 5 mg of melanoma tissue, respectively (Li et al., 2023).

3.2. Mass spectrometry analysis of isolated immunopeptides

The enriched and cleaned‐up peptides are then subjected to liquid chromatography (LC) tandem mass spectrometry (MS) analysis. Selecting the proper instrumentation is crucial for achieving comprehensive immunopeptidome coverage. To acquire purified immunopeptides effectively, it is advisable to employ a high‐end, state‐of‐the‐art LC‐MS/MS system. Given the biochemical similarities of immunopeptides, a high‐resolution LC system is essential. Nano LCs are favored over microflow LCs in immunopeptidomics approaches, primarily due to the usual low sample quantities and the need for the highest possible sensitivity. Some studies suggest the inclusion of charge‐enhancing additives such as dimethyl sulfoxide or m‐nitrobenzyl alcohol to enhance performance (Nielsen & Abaye, 2013; Van Wanseele et al., 2019). However, the effectiveness of improving electrospray ionization and sensitivity may vary, depending on factors such as the emitter source type and specific parameters like source voltage, temperature, and gas flow. For the mass spectrometer, opting for a high‐mass accuracy analyzer is a requirement as it enhances the reliability of peptide sequence identification. It is important to note that depending on the MHC allele of the sample, a substantial number of singly charged peptides may be present. Strategies for increasing the charge state of these peptides using chemical derivatization and isobaric labeling have been shown (Chen et al., 2018; Pfammatter et al., 2020).

The ongoing improvement of mass spectrometry instruments has advanced the field of immunopeptidomics immensely. For example, cutting‐edge mass spectrometers coupled with ion‐mobility separation and/or trapping devices (IMS) offer increased identifications and improved signal‐to‐noise ratios, further enhancing the immunopeptidome analysis process (Meier et al., 2018). Moreover, a recent study has revealed that MS/MS spectra from timsTOF instruments exhibit more reproducibility at low abundances compared to MS/MS spectra from Orbitrap instruments (Hoenisch Gravel et al., 2023).

In the realm of immunopeptidomics, three primary methods are employed for identifying and quantifying immunopeptides. Most studies in this field rely on data‐dependent acquisition (DDA) to maximize the amount of information acquired within an experiment (Caron, Kowalewski, et al., 2015). In DDA, the ionized immunopeptides are first detected in a survey scan (MS1 scan). The most abundant precursor ions detected are further isolated for fragmentation, generating individual peptide spectra (fragmentation spectra) and maximizing the information extracted from the experiment (Gatlin et al., 2000). However, in DDA, precursor ion selection follows a straightforward intensity‐dependent rule, resulting in the (semi‐)random isolation of precursor ions for fragmentation. This randomness can lead to inconsistent peptide identification when analyzing the same sample repeatedly, compromising reproducibility. Consequently, DDA is less suitable for accurately quantifying immunopeptides across different samples. On average, approximately 20% of the selected MHC I immunopeptides have been observed to vary between replicate analyses of the same sample (Caron, Espona, et al., 2015). On the contrary, data‐independent acquisition (DIA) adopts a different approach by acquiring fragmentation data for all ions within predefined precursor isolation windows, providing a comprehensive map of a given sample. Since its introduction in 2004, several DIA strategies have been described and reviewed in detail (Chapman et al., 2014; Distler et al., 2014; Sajic et al., 2015). DIA captures comprehensive data by simultaneously fragmenting and analyzing all precursor ions in a sample, creating a detailed digital map of the sample's molecular components. However, since chimeric spectra are produced from fragments of multiple peptide precursors rather than individual fragmentation spectra, successfully implementing a DIA strategy requires the development of a comprehensive spectral library. A specific constraint in DIA experiments is the need for specialized approaches to identify individual peptides, which require prior knowledge of the peptides' fragmentation and chromatographic behavior. This information is typically obtained through the construction of spectral libraries based on previous DDA analysis of similar samples, or through the use of retention time and fragmentation spectra modeling based on peptide sequences. These approaches are particularly challenging for immunopeptidomics, as immunopeptides can vary significantly between samples depending on the HLA genotypes present (Lou & Shui, 2024). This can be challenging, especially when investigating noncanonical peptides and neoantigens. In proteomics, a spectral library is typically generated from experimental data using multiple data‐dependent data sets. Since the immunopeptidome is by definition patient‐specific, one cannot easily resort to publicly available datasets to construct a library. In essence, ideally a specific library would have to be generated per individual to be comprehensive enough to be of use for potentially actionable immunopeptides. Limited availability of clinical biopsies makes it difficult to generate DDA data for a spectral library while still having enough material for another DIA analysis. Recent prediction models, however, allow in silico generation of spectral libraries, offering a solution to the challenge of relying on clinical biopsies for spectral library creation. A recent study implementing a DIA immunopeptidomics workflow using spectral libraries of increasing complexity reported compelling results. Utilizing a complex multi‐MHC spectral library of previously measured DDA results resulted in a twofold increase in peptide identification compared to a sample‐specific spectral library and a threefold increase compared to a DDA approach. Additionally, the study demonstrated the successful application of DIA for neoantigen discovery by analyzing DIA data with predicted MS/MS spectra (generated by a trained model) of clinically relevant immunopeptides (Gessulat et al., 2019; Pak et al., 2021). Nevertheless, the utility of these in silico prediction models is constrained when it comes to identifying PTMs. In the study mentioned earlier, peptides containing cysteine were deliberately omitted. This decision was driven by the fact the trained model for spectral library prediction assumes the carbamidomethylation of all cysteines, reflecting the characteristics of the training data upon which the models rely, derived from proteomics data.

An alternative method for the identification and quantification of immunopeptides is the targeted data acquisition mode. In contrast to DDA and DIA, a targeted approach offers high specificity, sensitivity, reproducibility, quantitative accuracy, and a wide dynamic range (Gallien et al., 2014; Peterson et al., 2012; Picotti & Aebersold, 2012). Typically conducted in Selected/Multiple Reaction Monitoring (S/MRM) and Parallel Reaction Monitoring (PRM) modes, this method is commonly employed on triple quadrupole and quadrupole Orbitrap instruments (Peterson et al., 2012; Picotti & Aebersold, 2012). It is important to note that a targeted approach necessitates prior knowledge of the specific target of interest (Croft et al., 2013; Gubin et al., 2014; Hassan et al., 2014; Hogan et al., 2005; Tan et al., 2011). In certain instances, studies employing Selected/Multiple Reaction Monitoring (S/MRM) for both relative and absolute quantification have been conducted. For instance, in a study by Hassan et al. (2014), heavy‐labeled peptides loaded on MHC class I were utilized for accurate quantitation of peptide yield isolated using immunoprecipitation (IP) (Hassan et al., 2014). Utilizing isobaric labeling for targeted measurements enables the simultaneous analysis of multiple samples. A recent technique known as TOMAHAQ employs synchronous precursor selection (SPS‐MS3) to enhance quantitative accuracy. This approach uses a dedicated MS3 fragmentation event for quantification, effectively eliminating interference from co‐isolated or co‐fragmented peptides (Pollock et al., 2021; Rose et al., 2019). Applying this methodology in immunopeptidomics extends quantitative sensitivity to the low amol/µL level (Pollock et al., 2021). Despite the advantages of targeted approaches, the drawback of requiring prior knowledge about the investigated targets limits their standalone applicability. Consequently, the implementation of targeted approaches appears to be more complementary to other acquisition methods in immunopeptidomics and its use is largely restricted to validation studies.

Another crucial factor influencing immunopeptidome coverage is the choice of fragmentation technique. In a study by Mommen et al. (2014), the combination of electron‐transfer/higher‐energy collision dissociation (EThcD) led to a threefold increase in the identification of MHC‐I immunopeptides compared to traditional fragmentation techniques like collision‐induced dissociation (CID), higher‐energy collisional dissociation (HCD), or a combination thereof. EThcD outperforms conventional methods in sequencing peptides bound to HLA‐B7 and ‐B27 molecules, which are challenging due to the preferred presence of proline or arginine as an anchor residue as the second amino acid in immunopeptide's sequence. ETD cannot cleave proline's N‐Cα bond, and CID/HCD faces difficulties with internal arginine residues. EThcD's dual‐ion series overcomes these limitations, leading to an almost twofold improvement in identifying peptides associated with HLA‐B7 and –B27 peptides.

4.1. PTMs—A key role in healthy and diseased cells

The regulation of cellular activities, including the biological activity of proteins, cellular localization, and the assembly of protein complexes, often relies on PTMs that occur naturally on proteins. Enzymes like kinases, phosphatases, and glycosyl‐ or acyltransferases, among others, catalyze these modifications, which are reversible in most cases. The common naturally occurring PTMs, such as phosphorylation, glycosylation, methylation, acetylation, ubiquitination, and citrullination, are essential for cellular processes, including for proper immune responses (Mann & Jensen, 2003; Seo & Lee, 2004; Zavala‐Cerna et al., 2014). An overview of the common PTMs in eukaryotic cells can be seen in Figure 3.

As previously noted, the modifications of proteins can result in modified immunopeptides, which are then presented by MHC I/II molecules, possibly triggering T‐cell responses. Despite their potential significance, PTMs on immunopeptides remain understudied. This primarily stems from the limitation of the methods employed. In general, detection of PTMs requires highly sensitive mass spectrometers, dedicated fragmentation techniques, specific enrichment strategies, and/or dedicated data analysis. These methods have been applied extensively in proteomics. However, it is only with the recent advancements in mass spectrometry instrumentation, particularly in terms of enhanced sensitivity, that the extension of these analyses to immunopeptides has become feasible.

4.2. Experimental challenges in mass spectrometry analysis of PTMs

MS/MS offers a range of essential analytical capabilities for determining the amino acid sequence of a peptide and, in most cases, not only identifying a PTM but also pinpointing the specific amino acid residue it affects. However, analyzing PTMs using mass spectrometry can be challenging due to several factors. Numerous reviews discuss the challenges and approaches for identifying PTMs at the protein and peptide levels (Doll & Burlingame, 2015; Kim et al., 2016; Smith & Rogowska‐Wrzesinska, 2020). While many challenges in detection and localization from a proteomics perspective also apply to immunopeptidomics, there are key challenges that significantly impact PTM detection in immunopeptidomics.

To analyze immunopeptides with PTMs using mass spectrometry, they must be ionized. However, certain PTMs can affect a peptide's ionization efficiency, making them more difficult to detect than their unmodified counterparts (Gao & Wang, 2007). Once a modified immunopeptide is ionized, the PTM can alter the peptide's fragmentation pattern, complicating the accurate interpretation of the mass spectra (Kim et al., 2016). Additionally, post‐translationally modified peptides are often less abundant than unmodified peptides, further complicating their detection (Larsen et al., 2006).

Moreover, identifying immunopeptides with PTMs does not necessarily attribute the modification to a biological origin. For example, methionine oxidation can occur during sample preparation, and asparagine deamidation is sensitive to pH, temperature, and buffer type, underscoring the importance of selecting appropriate sample preparation methods and materials. These sample preparation artifacts can ultimately complicate data interpretation (Liu et al., 2013; Morand et al., 1993; Pace et al., 2013).

Confidently identifying PTMs on a peptide requires the PTM to persist throughout the sample preparation process and downstream MS/MS analysis. Chemical stability is crucial for efficient PTM detection in MS/MS, with stable PTMs like acetyl‐lysine maintaining integrity and causing a 42 Da mass increment for the intact peptide. Less stable PTMs, such as phosphoserine and phosphothreonine, often undergo neutral loss (‐H₃PO₄) during MS/MS, resulting in a mass deficit of 98 Da in fragment ions (Annan et al., 2000). These Δm values prove valuable for annotating PTM peptide MS/MS spectra, aiding in the identification of modified peptides. However assignment of modified residues requires often a specific approach. In electron transfer dissociation (ETD), the protein or peptide is fragmented using low‐energy anions produced by the chemical ionization of fluoranthene, ETD typically causes fragmentation along the N‐Cα bonds, preserving the modified amino acids (Yu & Veenstra, 2021). Beyond sample preparation and MS/MS analysis, challenges persist in the data analysis phase of identifying PTMs. These challenges will be further explored in the forthcoming data analysis chapter.

4.3. Significance and applications of modified immunopeptides

The elusive question concerning modified immunopeptides is, whether they do exist in vivo, and what their relevance is. Can they trigger a specific T‐cell response? As mentioned earlier, PTMs play a crucial role in the function and structure of proteins. Since the immunopeptidome mirrors the proteome, various PTMs should be present on MHC I and MHC II (Admon & Bassani‐Sternberg, 2011). In a couple of studies, the presentation and recognition by cytotoxic T‐lymphocytes of modified peptides have been demonstrated for phosphorylated and glycosylated peptides in particular (Andersen et al., 1999; Haurum et al., 1999). More recent MS and bioinformatics analysis suggest that modified peptides may account for 12%–25% of the total immunopeptidome (Prus et al., 2019). Additionally, research has demonstrated that the physiochemical alterations caused by PTMs can influence the binding affinity of peptides to specific allotypes, further indicating that PTMs may have an impact on antigen presentation (Sidney et al., 2018).

Aberrant PTMs have been confirmed to result in antigenic peptides and the cognate T‐cell receptor (TCR) recognition in various instances (Apostolopoulos et al., 2003; Vlad et al., 2002). Notably, these immunogenic peptides, arising from dysregulated PTMs in cancer cells, represent an understudied class of potentially tumor‐specific (neo)antigens. These neoantigens hold great potential as lead candidates to be used in anticancer immunotherapy. Moreover, because they do not involve patient‐specific or tumor‐specific mutations, they have the potential of being shared across individuals independent of the specific tumor‐associated mutations, making them promising ‘off‐the‐shelf’ targets for immunotherapy.

Accumulating evidence suggests that alternative sources of cancer neoantigens, such as gene fusions, alternative splicing variants, translated noncoding sequences ‐and PTMs, hold promise as novel targets for immunotherapy. Particularly, neoantigens derived from gene fusions, recurrent mutations in cancer driver genes, noncoding regions, and aberrant PTMs are more likely to be shared among patients, providing a readily available resource for immunotherapy (Capietto et al., 2022; Smith et al., 2019; Zhou et al., 2019). In the upcoming chapters, we will explore the PTMs that exhibit the highest potential as neoantigens, spotlight the modified immunopeptides currently identified and briefly describe the challenges in PTM detection. Advancements and Applications in Glycosylation Analysis Glycosylation is essential for physiological and pathological cellular activities; advancements in analytical techniques have fueled progress in the study of glycobiology during the last decade. For a more detailed overview of the function of glycosylation in health and disease, we refer to the review written by Reily et al. (2019).

A significant function of glycans is to contribute to protein stability and solubility, including appropriate molecular orientation and reduction of nonspecific protein–protein interactions. However, glycosylation can also be present on peptides presented by the MHC and form tumor‐specific neoantigens (Malaker, Penny, et al., 2017; Marino et al., 2015). While glycopeptides have been isolated from MHC‐II molecules for some time, the evidence for the existence of MHC‐I glycopeptides was initially limited (Chicz et al., 1993). The first documentation of glycosylated MHCI immunopeptides was reported by Haurum et al. (1999). Their groundbreaking work demonstrated that naturally presented MHC‐I peptides include a small subset of glycopeptides, primarily featuring O‐β‐N‐acetylglucosamine (GlcNAc) substitutions on serine and threonine residues. Using synthesized glycopeptides, they further established that these peptides were immunogenic in mice and could be transported by the TAP transporter for antigen presentation (Haurum et al., 1999).

Characterizing and quantifying intact glycopeptides from complex datasets remains challenging due to their inherent glycan heterogeneity, unique ionization and separation characteristics, and relatively low abundance compared to non‐modified peptide counterparts. In a study conducted by Mukherjee, ion mobility‐coupled mass spectrometry devices were used to separate glycosylated peptides by their collisional cross‐section (CCS). Ion mobility emerged as a promising tool for characterizing glycopeptides due to their distinct physical properties. By isolating glycopeptides and optimizing glycan‐dedicated stepped collision energy, the extraction of information is improved, resulting in less ambiguous and more complete fragmentation spectra. Although this study was limited to trypsin‐digested peptides, the analysis should also be applicable to immunopeptidomics samples (Mukherjee et al., 2023).

Glycosylation plays a role in many autoimmune diseases and cancer, therefore providing an interesting target to study in the context of immunopeptidomics. Rheumatoid arthritis (RA) is linked to specific MHC class II alleles and is characterized by a recurrent autoimmune response in the joints. Collagen type II (CII), the primary component of hyaline cartilage, has been hypothesized as a potential autoantigen in rheumatoid arthritis because CII‐specific antibodies are commonly identified in RA patients. In a study using transgenic mice expressing human DR4 and human CD4, T cells predominantly recognized the immunodominant type II collagen antigen in its glycosylated form (Bäcklund et al., 2002; Corthay et al., 1998). In tuberculosis, CD8⁺ T cells recognize the glycosylated peptide presented by HLA‐E, a nonclassical MHC I molecule (Harriff et al., 2017). Additionally, the alteration in protein glycosylation within tumor cells has a decisive impact in all stages of the disease. This could impact the HLA‐epitopes and lead to the overexpression of standard glycoproteins and tumor‐specific glycoproteins (Peixoto et al., 2019). Several studies have shown T‐cell reactivation towards glycosylated epitopes presented by MHC I and II, underlining the importance of these epitopes (Kastrup et al., 2001; Unanue et al., 2016).

In a study profiling the prevalence of PTMs among a subset of HLA allotypes, about 17%–38% of the identified HLA I‐bound peptides were found to be posttranslationally modified. Deamidated peptides were the second most common PTM found in the immunopeptidome, accounting for 2.5%–7% of the detected modified peptides. A large proportion, nearly 59%, of the deamidated peptides, has the common N‐glycosylation motif NX (S/T), suggesting that most deamidations originate from glycosylated proteins that have undergone deglycosylation via the ER‐associated protein degradation pathway (ERAD) (Mei et al., 2020). In a study conducted by Malaker, Penny, et al. (2017), investigating melanoma cell lines, a total of 93 MHC II‐bound glyco‐peptides were identified (Malaker, Ferracane, et al., 2017). Out of these peptides, a vast majority carried the glycosylation on flanking residues, not interfering with the peptide binding domain of the MHC. Recent research has identified glycosylated HLA‐DR‐bound peptides in dendritic cells pulsed with the SARS‐CoV‐2 spike protein. The glycosylation profile of the HLA‐II‐bound S‐protein‐derived peptides shows substantial trimming of glycan residues compared to the S protein, likely due to antigen processing (Parker, Partridge, et al., 2021).

5.1. Bioinformatic challenges in the identification of PTMs

In the identification of modified peptides, researchers typically have two options. They can either specify a predetermined list of potentially occurring PTMs in a closed search or perform an open modification search. During a closed search, a narrow precursor mass window is used to select candidate peptides to compare the experimental spectrum against. Only the specified PTMs will be considered, leaving any unspecified PTMs undetected. As each PTM that is added will increase the search space drastically, usually only a handful of PTMs are considered during a closed search. However, an open modification search approach allows unbiased detection of any PTM. During open modification searching, a wide precursor mass window is used, allowing the matching of a spectrum from a modified peptide to its unmodified counterpart in the database. Consequently, a mass shift, wherein the identified peptide‐spectrum match (PSM) deviates in mass from the precursor ion, can be observed. These mass shifts are often attributable only to a single known PTM or, in some cases, a combination of PTMs, neutral losses, or chemical modifications. While an open search enables the handling of numerous unidentified PTMs, in immunopeptidomics, it is encumbered by two main challenges. First, the search space expands significantly, resulting in longer search times. Second, the search outcomes are prone to errors, often stemming from the misidentification of peptide sequences and the mis‐localization of PTM sites, necessitating additional postprocessing steps or manual validation of the results compared to a conventional closed search approach. Recently, new tools have been emerging to resolve some of the main challenges in open modification searching such as reducing the search time, and the identification of localization of PTMs (An et al., 2019; Avtonomov et al., 2019; Han et al., 2011).

Not only is the presence of many PTMs challenging, but the nature of these PTMs can also make their detection difficult. Modified peptides are usually less abundant, making identification more challenging than their unmodified counterpart, especially in complex samples, where acquiring high‐quality spectra is difficult. While easily overlooked, each PTM may also impose different challenges. Take some of the previously mentioned PTMs as an example. Deamidation and Citrullination cause a mass shift of 0.9840 Da, roughly corresponding to the mass of a single hydrogen atom (1.0078 Da). The minor mass shift of approximately 1 Da between the deamidated peptide and unmodified counterpart causes the deamidated peptide to overlap with the isotopic pattern of the unmodified peptide. Without employing proper separation techniques, distinguishing between the deamidated and unmodified peptides can be highly challenging (Badgett et al., 2017). Phosphorylation is a labile PTM, especially in the case of phosphothreonine and phosphoserine. These modified amino acids are prone to spontaneous dephosphorylation, causing a mass shift of −80 Da, making it troublesome to localize the modification. Glycosylation is also a challenging PTM to analyze because of the variability in the attached polysaccharides. Adding to this challenge is that glycosylation is often a set of modifications in a peptide with a high heterogeneity, resulting in not a single mass shift being associated with glycosylation, but rather a broad distribution. Also, the glycans are high molecular weight modifications, potentially shifting the molecular weight of the peptide out of the detection region of the MS. Lastly, conventional fragmentation of glycosylated peptides results into fragmentation of the glycan, yielding a spectrum containing information about the nature of the modification instead of the peptide sequence or the location of the modification. This can be circumvented through fragmentation by electron‐capture dissociation. A popular approach to detect glycosylation in peptides is by removing the polysaccharides from the peptide and analyzing both separately (Mann & Jensen, 2003). Recently, a deep learning framework, called DeepGlyco, was developed. DeepGlyco is capable of predicting fragment spectra of glycopeptides. Leveraging these predictions makes it possible to create spectral libraries for processing steps (Yang & Fang, 2024).

A final bioinformatic challenge regarding PTMs is the localization of the modification within the peptide. Localization is a crucial aspect of MS‐based proteomics in general to properly understand the protein function and regulation. In immunopeptidomics, this is of even greater importance given the relevance of immunopeptides and their localization of the modification in terms of T‐cell recognition and MHC binding affinity. While many search engines can identify modified peptides, only a few algorithms have incorporated score statistics for modification site localization. In the following sections, we will mention spectrum search engines suited to address these issues. Additionally, there are also various postprocessors available for PTM localization such as A‐score (Beausoleil et al., 2006), phosphoRS (Taus et al., 2011), and DeepFLR (Zong et al., 2023). Notably, phosphoRS and DeepFLR are specifically designed for localization of phosphorylated residues.

5.2. Overview of the data analysis workflow

Figure 4 presents a general workflow for analyzing mass spectrometry data. In this discussion, we will provide detailed insights into the data analysis aspects, specifically focusing on nonspecific enzymatic digestion and open modifications searching. We will begin by exploring various peptide identification methods and then examine the potential benefits of preprocessing steps, such as spectral clustering and proteogenomics. Finally, we will delve into rescoring as a post‐processing technique.