The physiological interactome of TCR-like antibody therapeutics in human tissues.

Selective binding of TCR-like antibodies that target a single tumour-specific peptide antigen presented by human leukocyte antigens (HLA) is the absolute prerequisite for their therapeutic suitability and patient safety. To date, selectivity assessment has been limited to peptide library screening and predictive modeling. We developed an experimental platform to de novo identify interactomes of TCR-like antibodies directly in human tissues using mass spectrometry. As proof of concept, we confirm the target epitope of a MAGE-A4-specific TCR-like antibody. We further determine cross-reactive peptide sequences for ESK1, a TCR-like antibody with known off-target activity, in human liver tissue. We confirm off-target-induced T cell activation and ESK1-mediated liver spheroid killing. Off-target sequences feature an amino acid motif that allows a structural groove-coordination mimicking that of the target peptide, therefore allowing the interaction with the engager molecule. We conclude that our strategy offers an accurate, scalable route for evaluating the non-clinical safety profile of TCR-like antibody therapeutics prior to first-in-human clinical application.

Cryptic MHC-E epitope from influenza elicits a potent cytolytic T cell response.

The extent to which unconventional forms of antigen presentation drive T cell immunity is unknown. By convention, CD8 T cells recognize viral peptides, or epitopes, in association with classical major histocompatibility complex (MHC) class I, or MHC-Ia, but immune surveillance can, in some cases, be directed against peptides presented by nonclassical MHC-Ib, in particular the MHC-E proteins (Qa-1 in mice and HLA-E in humans); however, the overall importance of nonclassical responses in antiviral immunity remains unclear. Similarly uncertain is the importance of 'cryptic' viral epitopes, defined as those undetectable by conventional mapping techniques. Here we used an immunopeptidomic approach to search for unconventional epitopes that drive T cell responses in mice infected with influenza virus A/Puerto Rico/8/1934. We identified a nine amino acid epitope, termed M-SL9, that drives a co-immunodominant, cytolytic CD8 T cell response that is unconventional in two major ways: first, it is presented by Qa-1, and second, it has a cryptic origin, mapping to an unannotated alternative reading frame product of the influenza matrix gene segment. Presentation and immunogenicity of M-SL9 are dependent on the second AUG codon of the positive sense matrix RNA segment, suggesting translation initiation by leaky ribosomal scanning. During influenza virus A/Puerto Rico/8/1934 infection, M-SL9-specific T cells exhibit a low level of egress from the lungs and strong differentiation into tissue-resident memory cells. Importantly, we show that M-SL9/Qa-1-specific T cells can be strongly induced by messenger RNA vaccination and that they can mediate antigen-specific cytolysis in vivo. Our results demonstrate that noncanonical translation products can account for an important fraction of the T cell repertoire and add to a growing body of evidence that MHC-E-restricted T cells could have substantial therapeutic value.

Interferon-α promotes neo-antigen formation and preferential HLA-B-restricted antigen presentation in pancreatic ß-cells.

Interferon (IFN)-α is the earliest cytokine signature observed in individuals at risk for type 1 diabetes (T1D), but its effect on the repertoire of HLA Class I (HLA-I)-bound peptides presented by pancreatic ß-cells is unknown. Using immunopeptidomics, we characterized the peptide/HLA-I presentation in in-vitro resting and IFN-α-exposed ß-cells. IFN-α increased HLA-I expression and peptide presentation, including neo-sequences derived from alternative mRNA splicing, post-translational modifications - notably glutathionylation - and protein cis-splicing. This antigenic landscape relied on processing by both the constitutive and immune proteasome. The resting ß-cell immunopeptidome was dominated by HLA-A-restricted ligands. However, IFN-α only marginally upregulated HLA-A and largely favored HLA-B, translating into a major increase in HLA-B-restricted peptides and into an increased activation of HLA-B-restricted vs. HLA-A-restricted CD8+ T-cells. A preferential HLA-B hyper-expression was also observed in the islets of T1D vs. non-diabetic donors, and we identified islet-infiltrating CD8+ T-cells from T1D donors reactive to HLA-B-restricted granule peptides. Thus, the inflammatory milieu of insulitis may skew the autoimmune response toward epitopes presented by HLA-B, hence recruiting a distinct T-cell repertoire that may be relevant to T1D pathogenesis.

Novel canonical and non-canonical viral antigens extend current targets for immunotherapy of HPV-driven cervical cancer.

Current immunotherapeutic approaches for human papillomavirus (HPV)-driven cervical cancer target the viral oncogenes E6 and E7. We report viral canonical and alternative reading frame (ARF)-derived sequences presented on cervical tumor cells, including antigens encoded by the conserved viral gene E1. We confirm immunogenicity of the identified viral peptides in HPV-positive women, and women with cervical intraepithelial neoplasia. We observe consistent transcription of the E1, E6, and E7 genes in 10 primary cervical tumor resections from the four most common high-risk HPV subtypes (HPV16, 18, 31, and 45), suggesting the suitability of E1 as therapeutic target. We finally confirm HLA presentation of canonical peptides derived from E6 and E7, and ARF-derived viral peptides from a reverse-strand transcript spanning the HPV E1 and E2 genes in primary human cervical tumor tissue. Our results extend currently known viral immunotherapeutic targets in cervical cancer and highlight E1 as an important cervical cancer antigen.

Long non-coding RNA-derived peptides are immunogenic and drive a potent anti-tumour response.

Protein arginine methyltransferase (PRMT) 5 is over-expressed in a variety of cancers and the master transcription regulator E2F1 is an important methylation target. We have explored the role of PRMT5 and E2F1 in regulating the non-coding genome and report here a striking effect on long non-coding (lnc) RNA gene expression. Moreover, many MHC class I protein-associated peptides were derived from small open reading frames in the lncRNA genes. Pharmacological inhibition of PRMT5 or adjusting E2F1 levels qualitatively altered the repertoire of lncRNA-derived peptide antigens displayed by tumour cells. When presented to the immune system as either ex vivo-loaded dendritic cells or expressed from a viral vector, lncRNA-derived peptides drove a potent antigen-specific CD8 T lymphocyte response, which translated into a significant delay in tumour growth. Thus, lncRNA genes encode immunogenic peptides that can be deployed as a cancer vaccine.

Immunopeptidomic Analysis of BoLA-I and BoLA-DR Presented Peptides from Theileria parva Infected Cells.

The apicomplexan parasite Theileria parva is the causative agent of East Coast fever, usually a fatal disease for cattle, which is prevalent in large areas of eastern, central, and southern Africa. Protective immunity against T. parva is mediated by CD8+ T cells, with CD4+ T-cells thought to be important in facilitating the full maturation and development of the CD8+ T-cell response. T. parva has a large proteome, with >4000 protein-coding genes, making T-cell antigen identification using conventional screening approaches laborious and expensive. To date, only a limited number of T-cell antigens have been described. Novel approaches for identifying candidate antigens for T. parva are required to replace and/or complement those currently employed. In this study, we report on the use of immunopeptidomics to study the repertoire of T. parva peptides presented by both BoLA-I and BoLA-DR molecules on infected cells. The study reports on peptides identified from the analysis of 13 BoLA-I and 6 BoLA-DR datasets covering a range of different BoLA genotypes. This represents the most comprehensive immunopeptidomic dataset available for any eukaryotic pathogen to date. Examination of the immunopeptidome data suggested the presence of a large number of coprecipitated and non-MHC-binding peptides. As part of the work, a pipeline to curate the datasets to remove these peptides was developed and used to generate a final list of 74 BoLA-I and 15 BoLA-DR-presented peptides. Together, the data demonstrated the utility of immunopeptidomics as a method to identify novel T-cell antigens for T. parva and the importance of careful curation and the application of high-quality immunoinformatics to parse the data generated.

Ionizing Radiation Drives Key Regulators of Antigen Presentation and a Global Expansion of the Immunopeptidome.

Little is known about the pathways regulating MHC antigen presentation and the identity of treatment-specific T cell antigens induced by ionizing radiation. For this reason, we investigated the radiation-specific changes in the colorectal tumor cell proteome. We found an increase in DDX58 and ZBP1 protein expression, two nucleic acid sensing molecules likely involved in induction of the dominant interferon response signature observed after genotoxic insult. We further observed treatment-induced changes in key regulators and effector proteins of the antigen processing and presentation machinery. Differential regulation of MHC allele expression was further driving the presentation of a significantly broader MHC-associated peptidome postirradiation, defining a radiation-specific peptide repertoire. Interestingly, treatment-induced peptides originated predominantly from proteins involved in catecholamine synthesis and metabolic pathways. A nuanced relationship between protein expression and antigen presentation was observed where radiation-induced changes in proteins do not correlate with increased presentation of associated peptides. Finally, we detected an increase in the presentation of a tumor-specific neoantigen derived from Mtch1. This study provides new insights into how radiation enhances antigen processing and presentation that could be suitable for the development of combinatorial therapies. Data are available via ProteomeXchange with identifier PXD032003.

HLA variants have different preferences to present proteins with specific molecular functions which are complemented in frequent haplotypes.

Human leukocyte antigen (HLA) genes are the most polymorphic loci in the human genome and code for proteins that play a key role in guiding adaptive immune responses by presenting foreign and self peptides (ligands) to T cells. Each person carries up to 6 HLA class I variants (maternal and paternal copies of HLA-A, HLA-B and HLA-C genes) and also multiple HLA class II variants, which cumulatively define the landscape of peptides presented to T cells. Each HLA variant has its own repertoire of presented peptides with a certain sequence motif which is mainly defined by peptide anchor residues (typically the second and the last positions for HLA class I ligands) forming key interactions with the peptide-binding groove of HLA. In this study, we aimed to characterize HLA binding preferences in terms of molecular functions of presented proteins. To focus on the ligand presentation bias introduced specifically by HLA-peptide interaction we performed large-scale in silico predictions of binding of all peptides from human proteome for a wide range of HLA variants and established which functions are characteristic for proteins that are more or less preferentially presented by different HLA variants using statistical calculations and gene ontology (GO) analysis. We demonstrated marked distinctions between HLA variants in molecular functions of preferentially presented proteins (e.g. some HLA variants preferentially present membrane and receptor proteins, while others - ribosomal and DNA-binding proteins) and reduced presentation of extracellular matrix and collagen proteins by the majority of HLA variants. To explain these observations we demonstrated that HLA preferentially presents proteins enriched in amino acids which are required as anchor residues for the particular HLA variant. Our observations can be extrapolated to explain the protective effect of certain HLA alleles in infectious diseases, and we hypothesize that they can also explain susceptibility to certain autoimmune diseases and cancers. We demonstrate that these differences lead to differential presentation of HIV, influenza virus, SARS-CoV-1 and SARS-CoV-2 proteins by various HLA alleles. Taking into consideration that HLA alleles are inherited in haplotypes, we hypothesized that haplotypes composed of a combination of HLA variants with different presentation preferences should be more advantageous as they allow presenting a larger repertoire of peptides and avoiding holes in immunopeptidome. Indeed, we demonstrated that HLA-A/HLA-B and HLA-A/HLA-C haplotypes which have a high frequency in the human population are comprised of HLA variants that are more distinct in terms of functions of preferentially presented proteins than the control pairs.

Corrigendum: Elucidation of the Signatures of Proteasome-Catalysed Peptide Splicing.

[This corrects the article DOI: 10.3389/fimmu.2020.563800.].

The Choice of Search Engine Affects Sequencing Depth and HLA Class I Allele-Specific Peptide Repertoires.

Standardization of immunopeptidomics experiments across laboratories is a pressing issue within the field, and currently a variety of different methods for sample preparation and data analysis tools are applied. Here, we compared different software packages to interrogate immunopeptidomics datasets and found that Peaks reproducibly reports substantially more peptide sequences (~30-70%) compared with Maxquant, Comet, and MS-GF+ at a global false discovery rate (FDR) of <1%. We noted that these differences are driven by search space and spectral ranking. Furthermore, we observed differences in the proportion of peptides binding the human leukocyte antigen (HLA) alleles present in the samples, indicating that sequence-related differences affected the performance of each tested engine. Utilizing data from single HLA allele expressing cell lines, we observed significant differences in amino acid frequency among the peptides reported, with a broadly higher representation of hydrophobic amino acids L, I, P, and V reported by Peaks. We validated these results using data generated with a synthetic library of 2000 HLA-associated peptides from four common HLA alleles with distinct anchor residues. Our investigation highlights that search engines create a bias in peptide sequence depth and peptide amino acid composition, and resulting data should be interpreted with caution.

Integral Use of Immunopeptidomics and Immunoinformatics for the Characterization of Antigen Presentation and Rational Identification of BoLA-DR-Presented Peptides and Epitopes.

MHC peptide binding and presentation is the most selective event defining the landscape of T cell epitopes. Consequently, understanding the diversity of MHC alleles in a given population and the parameters that define the set of ligands that can be bound and presented by each of these alleles (the immunopeptidome) has an enormous impact on our capacity to predict and manipulate the potential of protein Ags to elicit functional T cell responses. Liquid chromatography-mass spectrometry analysis of MHC-eluted ligand data has proven to be a powerful technique for identifying such peptidomes, and methods integrating such data for prediction of Ag presentation have reached a high level of accuracy for both MHC class I and class II. In this study, we demonstrate how these techniques and prediction methods can be readily extended to the bovine leukocyte Ag class II DR locus (BoLA-DR). BoLA-DR binding motifs were characterized by eluted ligand data derived from bovine cell lines expressing a range of DRB3 alleles prevalent in Holstein-Friesian populations. The model generated (NetBoLAIIpan, available as a Web server at www.cbs.dtu.dk/services/NetBoLAIIpan) was shown to have unprecedented predictive power to identify known BoLA-DR-restricted CD4 epitopes. In summary, the results demonstrate the power of an integrated approach combining advanced mass spectrometry peptidomics with immunoinformatics for characterization of the BoLA-DR Ag presentation system and provide a prediction tool that can be used to assist in rational evaluation and selection of bovine CD4 T cell epitopes.

HLA binding of self-peptides is biased towards proteins with specific molecular functions.

Human leukocyte antigen (HLA) is highly polymorphic and plays a key role in guiding adaptive immune responses by presenting foreign and self peptides to T cells. Each HLA variant selects a minor fraction of peptides that match a certain motif required for optimal interaction with the peptide-binding groove. These restriction rules define the landscape of peptides presented to T cells. Given these limitations, one might suggest that the choice of peptides presented by HLA is non-random and there is preferential presentation of an array of peptides that is optimal for distinguishing self and foreign proteins. In this study we explore these preferences with a comparative analysis of self peptides enriched and depleted in HLA ligands. We show that HLAs exhibit preferences towards presenting peptides from certain proteins while disfavoring others with specific functions, and highlight differences between various HLA genes and alleles in those preferences. We link those differences to HLA anchor residue propensities and amino acid composition of preferentially presented proteins. The set of proteins that peptides presented by a given HLA are most likely to be derived from can be used to distinguish between class I and class II HLAs and HLA alleles. Our observations can be extrapolated to explain the protective effect of certain HLA alleles in infectious diseases, and we hypothesize that they can also explain susceptibility to certain autoimmune diseases and cancers. We demonstrate that these differences lead to differential presentation of HIV, influenza virus, SARS-CoV-1 and SARS-CoV-2 proteins by various HLA alleles. Finally, we show that the reported self peptidome preferences of distinct HLA variants can be compensated by combinations of HLA-A/HLA-B and HLA-A/HLA-C alleles in frequent haplotypes.

Elucidation of the Signatures of Proteasome-Catalyzed Peptide Splicing.

Proteasomes catalyze the degradation of endogenous proteins into oligopeptides, but can concurrently create spliced oligopeptides through ligation of previously non-contiguous peptide fragments. Recent studies have uncovered a formerly unappreciated role for proteasome-catalyzed peptide splicing (PCPS) in the generation of non-genomically templated human leukocyte antigen class I (HLA-I)-bound cis-spliced peptides that can be targeted by CD8+ T cells in cancer and infection. However, the mechanisms defining PCPS reactions are poorly understood. Here, we experimentally define the biochemical constraints of proteasome-catalyzed cis-splicing reactions by examination of in vitro proteasomal digests of a panel of viral- and self-derived polypeptide substrates using a tailored mass-spectrometry-based de novo sequencing workflow. We show that forward and reverse PCPS reactions display unique splicing signatures, defined by preferential fusion of distinct amino acid residues with stringent peptide length distributions, suggesting sequence- and size-dependent accessibility of splice reactants for proteasomal substrate binding pockets. Our data provide the basis for a more informed mechanistic understanding of PCPS that will facilitate future prediction of spliced peptides from protein sequences.

The Choice of HLA-Associated Peptide Enrichment and Purification Strategy Affects Peptide Yields and Creates a Bias in Detected Sequence Repertoire.

Understanding the most appropriate workflow for biochemical human leukocyte antigen (HLA)-associated peptide enrichment prior to ligand sequencing is essential to achieve optimal sensitivity in immunopeptidomics experiments. The use of different detergents for HLA solubilization as well as complementary workflows to separate HLA-bound peptides from HLA protein complex components after their immunoprecipitation including HPLC, C18 cartridge, and 5 kDa filter are described. It is observed that all solubilization approaches tested led to similar peptide ligand identification rates; however, a higher number of peptides are identified in samples lysed with CHAPS compared with other methods. The HPLC method is superior in terms of HLA-I peptide recovery compared with 5 kDa filter and C18 cartridge peptide purification methods. Most importantly, it is observed that both the choice of detergent and peptide purification strategy creates a significant bias for the identified peptide sequences, and that allele-specific peptide repertoires are affected depending on the workflow of choice. The results highlight the importance of employing a suitable strategy for HLA peptide enrichment and that the obtained peptide repertoires do not necessarily reflect the true distributions of peptide sequences in the sample.

Identification of an Unconventional Subpeptidome Bound to the Behçet's Disease-associated HLA-B*51:01 that is Regulated by Endoplasmic Reticulum Aminopeptidase 1 (ERAP1).

Human leukocyte antigen (HLA) B*51:01 and endoplasmic reticulum aminopeptidase 1 (ERAP1) are strongly genetically associated with Behçet's disease (BD). Previous studies have defined two subgroups of HLA-B*51 peptidome containing proline (Pro) or alanine (Ala) at position 2 (P2). Little is known about the unconventional non-Pro/Ala2 HLA-B*51-bound peptides. We aimed to study the features of this novel subpeptidome, and investigate its regulation by ERAP1. CRISPR-Cas9 was used to generate an HLA-ABC-triple knockout HeLa cell line (HeLa.ABC-KO), which was subsequently transduced to express HLA-B*51:01 (HeLa.ABC-KO.B51). ERAP1 was silenced using lentiviral shRNA. Peptides bound to HLA-B*51:01 were eluted and analyzed by mass spectrometry. The characteristics of non-Pro/Ala2, Pro2, and Ala2 peptides and their alteration by ERAP1 silencing were investigated. Effects of ERAP1 silencing on cell surface expression of HLA-B*51:01 were studied using flow cytometry. More than 20% of peptides eluted from HLA-B*51:01 lacked Pro or Ala at P2. This unconventional group of HLA-B*51:01-bound peptides was relatively enriched for 8-mers (with relatively fewer 9-mers) compared with the Pro2 and Ala2 subpeptidomes and had similar N-terminal and C-terminal residue usages to Ala2 peptides (with the exception of the less abundant leucine at position Ω). Knockdown of ERAP1 increased the percentage of non-Pro/Ala2 from 20% to ∼40%, increased the percentage of longer (10-mer and 11-mer) peptides eluted from HLA-B*51:01 complexes, and abrogated the predominance of leucine at P1. Interestingly knockdown of ERAP1 altered the length and N-terminal residue usage of non-Ala2&Pro2 and Ala2 but not the Pro2 peptides. Finally, ERAP1 silencing regulated the expression levels of cell surface HLA-B*51 in a cell-type-dependent manner. In conclusion, we have used a novel methodology to identify an unconventional but surprisingly abundant non-Pro/Ala2 HLA-B*51:01 subpeptidome. It is increased by knockdown of ERAP1, a gene affecting the risk of developing BD. This has implications for theories of disease pathogenesis.

Identification of antigens presented by MHC for vaccines against tuberculosis.

Mycobacterium tuberculosis (M.tb) is responsible for more deaths globally than any other pathogen. The only available vaccine, bacillus Calmette-Guérin (BCG), has variable efficacy throughout the world. A more effective vaccine is urgently needed. The immune response against tuberculosis relies, at least in part, on CD4+ T cells. Protective vaccines require the induction of antigen-specific CD4+ T cells via mycobacterial peptides presented by MHC class-II in infected macrophages. In order to identify mycobacterial antigens bound to MHC, we have immunoprecipitated MHC class-I and class-II complexes from THP-1 macrophages infected with BCG, purified MHC class-I and MHC class-II peptides and analysed them by liquid chromatography tandem mass spectrometry. We have successfully identified 94 mycobacterial peptides presented by MHC-II and 43 presented by MHC-I, from 76 and 41 antigens, respectively. These antigens were found to be highly expressed in infected macrophages. Gene ontology analysis suggests most of these antigens are associated with membranes and involved in lipid biosynthesis and transport. The sequences of selected peptides were confirmed by spectral match validation and immunogenicity evaluated by IFN-gamma ELISpot against peripheral blood mononuclear cell from volunteers vaccinated with BCG, M.tb latently infected subjects or patients with tuberculosis disease. Three antigens were expressed in viral vectors, and evaluated as vaccine candidates alone or in combination in a murine aerosol M.tb challenge model. When delivered in combination, the three candidate vaccines conferred significant protection in the lungs and spleen compared with BCG alone, demonstrating proof-of-concept for this unbiased approach to identifying new candidate antigens.

Contribution of proteasome-catalyzed peptide cis-splicing to viral targeting by CD8+ T cells in HIV-1 infection.

Peptides generated by proteasome-catalyzed splicing of noncontiguous amino acid sequences have been shown to constitute a source of nontemplated human leukocyte antigen class I (HLA-I) epitopes, but their role in pathogen-specific immunity remains unknown. CD8+ T cells are key mediators of HIV type 1 (HIV-1) control, and identification of novel epitopes to enhance targeting of infected cells is a priority for prophylactic and therapeutic strategies. To explore the contribution of proteasome-catalyzed peptide splicing (PCPS) to HIV-1 epitope generation, we developed a broadly applicable mass spectrometry-based discovery workflow that we employed to identify spliced HLA-I-bound peptides on HIV-infected cells. We demonstrate that HIV-1-derived spliced peptides comprise a relatively minor component of the HLA-I-bound viral immunopeptidome. Although spliced HIV-1 peptides may elicit CD8+ T cell responses relatively infrequently during infection, CD8+ T cells primed by partially overlapping contiguous epitopes in HIV-infected individuals were able to cross-recognize spliced viral peptides, suggesting a potential role for PCPS in restricting HIV-1 escape pathways. Vaccine-mediated priming of responses to spliced HIV-1 epitopes could thus provide a novel means of exploiting epitope targets typically underutilized during natural infection.

High-throughput detection of low abundance sialylated glycoproteins in human serum by TiO2 enrichment and targeted LC-MS/MS analysis: application to a prostate cancer sample set.

Glycopeptide enrichment can be a strategy to allow the detection of peptides belonging to low abundance proteins in complex matrixes such as blood serum or plasma. Though several glycopeptide enrichment protocols have shown excellent sensitivities in this respect, few reports have demonstrated the applicability of these methods to relatively large sample cohorts. In this work, a fast protocol based on TiO2 enrichment and highly sensitive mass spectrometric analysis by Selected Reaction Monitoring (SRM) has been applied to a cohort of serum samples from prostate cancer and benign prostatic hyperplasia patients in order to detect low abundance proteins in a single LC-MS/MS analysis in nanoscale format, without immunodepletion or peptide fractionation. A peptide library of over 700 formerly N-glycosylated peptides was created by data dependent analysis. Then, 16 medium to low abundance proteins were selected for detection by single injection LC-MS/MS based on selected-reaction monitoring. Results demonstrated the consistent detection of the low-level proteins under investigation. Following label-free quantification, four proteins (Adipocyte plasma membrane-associated protein, Periostin, Cathepsin D and Lysosome-associated membrane glycoprotein 2) were found significantly increased in prostate cancer sera compared to the control group. Graphical abstract ᅟ.

MS-Rescue: A Computational Pipeline to Increase the Quality and Yield of Immunopeptidomics Experiments.

LC-MS/MS has become the standard platform for the characterization of immunopeptidomes, the collection of peptides naturally presented by major histocompatibility complex molecules to the cell surface. The protocols and algorithms used for immunopeptidomics data analysis are based on tools developed for traditional bottom-up proteomics that address the identification of peptides generated by tryptic digestion. Such algorithms are generally not tailored to the specific requirements of MHC ligand identification and, as a consequence, immunopeptidomics datasets suffer from dismissal of informative spectral information and high false discovery rates. Here, a new pipeline for the refinement of peptide-spectrum matches (PSM) is proposed, based on the assumption that immunopeptidomes contain a limited number of recurring peptide motifs, corresponding to MHC specificities. Sequence motifs are learned directly from the individual peptidome by training a prediction model on high-confidence PSMs. The model is then applied to PSM candidates with lower confidence, and sequences that score significantly higher than random peptides are rescued as likely true ligands. The pipeline is applied to MHC class I immunopeptidomes from three different species, and it is shown that it can increase the number of identified ligands by up to 20-30%, while effectively removing false positives and products of co-precipitation. Spectral validation using synthetic peptides confirms the identity of a large proportion of rescued ligands in the experimental peptidome.