Game of ‘Omes: RiboSeq Expands the MHC I Immunopeptidome

Abstract

Peptide ligands presented by cell-surface MHC class I molecules enable T cells to eradicate intracellular pathogens and cancers. The presented peptide repertoire, the class I immunopeptidome, is generated from each cell’s translatome in a highly biased manner to avoid overrepresenting highly abundant translation products. The immunopeptidome can only be defined by mass spectrometry. Here, we review recent advances in immunopeptidomics, focusing on using ribosome profiling (RiboSeq) as the optimal MS database to optimize the false and failed discovery rates and relate these findings to the contribution of defective ribosomal products (DRiPs) and cellular quality control mechanisms to MHC class I antigen processing and presentation.

Introduction

MHC class I molecules (MHC I) are expressed by nearly all cell types in dragons, Mothers of dragons, and other jawed vertebrates. MHC I present short peptides derived from proteins synthesized by antigen-presenting cells (APCs), enabling CD8⁺ T cell immunosurveillance of viruses and other intracellular pathogens. By presenting peptides associated with oncogenic transformation, they also provide the basis for CD8⁺ T cell-based cancer immunotherapy, one of the more active, lucrative, and promising areas of clinical immunology research [1].

A critical question in MHC class I biology, highly relevant to cancer immunotherapy due to difficulties in identifying suitable target peptides [2, 3], is how MHC class I peptide ligands (MAPs) are selected for MHC I presentation. Every protein in the translatome (the repertoire of proteins synthesized by ribosomes) is a potential source of MAPs, both as:

1. DRiPs (defective ribosomal products): translation products that result from errors in transcription, translation, or post-translational maturation.

2. SLiPs (short-lived proteins): correctly translated intrinsically disordered proteins that fail to reach a stable conformation, e.g., due to the absence of a suitable binding partner.

3. Retirees: mature proteins degraded at their natural half-life (typically a first-order process following synthesis).

If MAP selection were random, the immunopeptidome (the repertoire of peptides presented by MHC I) would be dominated by peptides from the 250 or so housekeeping proteins that account for ~75% of the proteome by mass [4]. From an ever-increasing number (over 2 million, climbing rapidly) of mass spectrometry (MS) identified MAPs [5], it is clear that peptide selection is highly skewed from molar representation (though this conclusion is undermined by the lack of peptide quantitation inherent to MS; as a result, peptides are nearly always scored as present/absent). The extreme bias in peptide generation was recognized in the 1990s from the presentation of viral peptides nearly immediately following infection in the face of overwhelming levels of competing cell proteins [6]. This extended remarkable findings from the Boon laboratory that T cells can recognize peptides generated from undetectable levels of translation of transfected gene fragments (or eventually introns), leading to the Pepton hypothesis [7].

Here, we review recent progress in understanding how peptides are selected for the MHC class I presentation.

Defining the MHC Class I Immunopeptidome

Examining immunopeptidome kinetics of influenza A virus and SARS-CoV-2 infected cells by MS, Wu et al. and Weingarten-Gabbay et al., respectively, showed that viral MAPs are detected before or simultaneously with the viral source protein [8, 9], extending evidence that DRiPs account for the vast majority of viral peptides.

In a tour de force immunopeptidomics study, Sarkizova et al. generated 95 HLA-A, B, C, and -G engineered mono-allelic cell lines, acquiring MS profiles of >185,000 MAPs that mapped to 10,649 human genes accounting for 91% of human genes detected in the proteome, and 89% of transcripts in the transcriptome. MAPs derived from 1,517 gene products were detected neither in the proteome nor the transcriptome, which authors attributed to their low expression levels. Large, highly abundant proteins represented the top 50 identified MAP source proteins with high peptide coverage [10]. Although ribosomal profiling data for the parental cell line in this study were available, the MS peptide spectra were searched against the database of annotated genes and transcripts, potentially missing many unannotated translation products [10•].

RiboSeq to the Rescue

Although advances in MS instrumentation have greatly expanded the MHC I immunopeptidome, the false discovery rate (FDR) remains an important limitation [11]. A key component of the FDR is the size of the potential peptide database used to identify MS-determined masses. While the entire genome potentially encodes peptides, less than 1% of the genetic information (including all six potential reading frames from a given DNA sequence) is translated. An RNAseq database is better, but only a fraction of mRNA-encoded information is translated. Further, RNAseq typically omits introns, a known source of peptides, and is frequently limited to annotated genes, likely comprising less than 20% of the transcriptome. Ribosome profiling (RiboSeq), which entails deep sequencing of mRNA fragments protected from nuclease digestion by translating ribosomes, provides the information (translation initiation and termination sites, reading frame, read counts) to generate a snapshot of protein translation [12]. Since endogenous peptides, by definition, originate from translated proteins, RiboSeq provides the ideal database for MS peptide identification.

Several pioneering groups have applied RiboSeq to the MS-based discovery of cancer cell peptides. Using a human melanoma cell line, Chong et al. [13••] were the first to report the existence of hundreds of MAPs derived from non-canonically translated products, many of which derived from upstream open reading frames (uORFs) present on annotated mRNAs. Ouspenskaia et al. [14••] used RiboSeq to build an MS search dataset from 29 human samples (cell lines, normal and cancer tissues) and applied this to a published raw MS immunopeptidome dataset from cells expressing one of 92 HLA-A, B, or C alleles. The resulting ORF dataset reduced the potential ORFs in the standard RNAseq-based human transcriptome by 25-fold, with a 10-fold reduction vs. the typical RNAseq dataset in a human cell line. Despite the dramatic trimming of the MS search dataset, they discovered an additional 6,501 high-confidence MAPs originating from 3,261 novel or unannotated open reading frames (nuORFs), contributing 3.3% to the immunopeptidome and 16% to all detected source proteins with more than one MAP detected [14••].

Comparing the translatome, proteome, and immunopeptidome of three human B lymphoma cell lines, Ruiz Cuevas et al. found a similar disparity of MAP source proteins present in proteome vs. immunopeptidome [14••, 15••]. Notably, while nuORFs were translated as efficiently (determined by RiboSeq reads/RNAseq reads) as canonical (annotated) MAP source proteins, they generated MAPs five times more efficiently per translation event [15••]. NuORFs derived from diverse translation events, including up-stream and down-stream ORFs (dORFs); frameshifting; translation of alternative and overlapping ORFs; translation initiation from cognate non-AUG codons; stop-codon read-through; translation of long non-coding RNAs, intronic sequences, transposable elements, and endogenous retroviral sequences. NuORFs MAP source proteins were only 10% the length of standard proteins (49 vs. 504 residues), with more predicted disorder and instability and more likely to initiate on non-AUG start codons [13••, 14••, 15••]. The small size and likely increased degradation of nuORF proteins would explain their absence from the MS-defined proteome. Their over-representation as MAP sources is consistent with DRiPs being a more efficient source of MAPs than retirees.

Remarkably, 26 nuORFs found by Ouspenskaia et al. are the precise length of the identified MAPs, i.e., ready-made MAPs [14••]. Erhard et al. reported that for 54% of cryptic protein-derived MAPs, the predicted start codon occurs within ten amino acids from the MAP N-terminus [16••]. This can possibly be explained by the short length of the cryptic proteins and start codon misassignment (i.e., the actual translation products are longer). Further, Malka et al.’s remarkable discovery of abundant translation from 5’-uncapped and polyadenylated transcripts generated via alternative cleavage and polyadenylation (APA) of mRNAs [17••] could contribute to this bias and the numerous novel isoform-derived MAPs observed by Ruiz Cuevas et al. [15••]. Malka et al. further showed that global APA induction, occurring in many tumor cells and triggered by activation of normal cells, increases MHC I immunopeptidome diversity more than 3-fold [17••].

In Erhard et al., ~20% of cryptic protein-derived MAPs represent the C-terminus of the translation product. Since 54% of these MAPs have N-terminal extensions of 10 or fewer residues, this implies that they are ready-made for TAP transport (18 or fewer residues), avoiding the need for cytosolic protease trimming and likely greatly enhancing their antigenicity [16••]. A similar observation was made by Chong et al., who found enrichment of C-terminal derived MAPs not only from cryptic proteins but also from similarly short canonical proteins [13••].

Jaeger et al. [18] developed a Cre-inducible MHC affinity-tagged mouse K^b molecule that enables immunopeptidome interrogation of specific cell types, in this case, oncogene-induced tumors. This enabled the discovery of 438 tumor-specific peptides, including 192 peptides not detected using standard methodology. Combined RNAseq and RiboSeq analysis of in vitro cultured cells revealed no correlation for tumor-specific peptides with source mRNA levels or translational efficiency. Rather, the authors concluded that specific features of the source proteins favored their presentation [18]. It is of obvious interest to investigate in this system the dependence of nascent protein synthesis (i.e., the DRiPiness of source proteins) on tumor-specific peptide generation.

Studying induced pluripotent human stem cells (iPSCs), Apavaloaei et al. found a set of pluripotency-associated MAPs (paMAPs), absent in coding transcripts from differentiated and stem cells but present and shared among some cancer types. Half of these paMAPs originated from transcripts annotated as non-coding. The immunogenicity of some of the paMAPs offers their promise as cancer immunotherapy targets [19].

Investigating cancer-associated protein arginine methyltransferase 5 (PRMT5) and its target, master transcription regulator E2F1, Barczak et al. [20] discovered an effect on long non-coding RNA (lncRNA) gene expression. Importantly, they showed that inhibition of PRMT5 or knockout of E2F1 altered the repertoire of lncRNA-derived MAPs on tumor cells. The lncRNA-derived MAPs were immunogenic, driving a potent anti-tumor response in prophylactic and therapeutic mouse cancer models. MAPs originated from short ORFs (<100 aa) with weak Kozak translation initiation sequences. Furthermore, the most immunogenic peptides used in the prophylactic vaccine model had a relatively low expression in thymocytes compared to other tissues, suggesting that lncRNA-derived antigens can avoid central tolerance. A potential limitation of this study is the use of standard oligo(dT) RNAseq to construct the lncRNA-derived peptide MS search database, preventing the detection of non-polyadenylated lncRNA-derived MAPs [20].

Bartok et al. linked IFN-γ signaling and tryptophan (Trp) metabolism to a new class of aberrant MAPs generated by ribosomal frameshifting [21••]. IFN-γ induces indoleamine 2,3-dioxygenase 1 (IDO1), which depletes Trp pools, causing ribosome stalling at Trp codons and, surprisingly, accumulation of ribosomes downstream of Trp codons revealed by RiboSeq (“W-bumps”). This leads to either ribosomal frameshifting [21••, 22] or Trp-to-Phe codon reassignment [23••], both processes generating aberrant translation products presented as cancer-specific MAPs, expanding the targetable cancer immunopeptidome [21••, 22, 23••].

Pioneer mRNA translation: a major source of MHC class I associated peptides?

Irradiation is known to enhance and alter the composition of the MHC I immunopeptidome [24, 25•]. Uchihara et al. [25] reported that DNA damage, regardless of mechanism, enhances MHC I cell surface expression in 8 human cancer cell lines, presumably due to increased MAP supply, as originally reported by Reits et al. [24]. Increased surface MHC I required ATR-AKT-mTORC1-p70-S6K kinase signaling. S6K appeared to stimulate the pioneer round of translation (PRT) associated with nonsense-mediated decay (NMD) based on siRNA CBP20 and SMG1 knockdown experiments. By contrast, siRNA knockdown of canonical translation initiation factor 4E (eIF4E), which replaces CBP20 after PRT, did not affect class I expression [25•]. Future studies should aim to confirm an overall effect on class I supply and examine how the MS-defined immunopeptidome is influenced by NMD-translated peptides.

Further, knowing just how many rounds of translation occur in the NMD process is crucial. A single pioneer translation round seems insufficient to generate sufficient peptides to have much effect on the immunopeptidome, given an efficiency of less than one MAP generated per 50 source proteins degraded [26]. Indeed, the association of NMD factors CBP20/CBP80 with polysomes [27, 28] supports NMD surveillance based on multiple translation rounds.

Continuing their ground-breaking studies on the contribution of PRT to immunosurveillance, Sroka et al. [29••] generated a transgenic mouse with the SIINFEKL model peptide inserted into the second β-globin gene intron. This mouse demonstrated CD8⁺ T cell tolerance to SIINFEKL sufficient to enable SIINFEKL-expressing cancer cells to escape immunosurveillance. Pre-spliced globin mRNA was primarily detected on monosomes and disomes, in contrast to spliced mRNA, which was nearly exclusively present on polysomes. Within introns, SIINFEKL-containing polypeptides are initiated by CUG and translated by nuclear ribosomes, as most directly shown by proximity ligation immunofluorescence [29••].

Evidence for such nuclear translation, a highly contentious issue for the translation field, continues to accumulate. Park et al. reported that in HeLa cells, pioneer mRNA translation occurs in close proximity to nuclear export pores, possibly in the inner aspect of the nuclear membrane [28]. Robust nuclear translation was detected in heat or acidosis-induced nuclear amyloid-like bodies in the nuclei of human cells by fluorescent puromycin labeling of ribosome-associated nascent chains in inert cells [30•]. This supports previous findings of puromycin-detected nuclear translation [31] and functional evidence for peptide presentation from influenza mRNA sequestered in the nucleus by lacking a 5’ cap [32].

Regardless of the translation site, evidence accrues for NMD-associated translation of antigenic peptides and the more general issue of intron-encoded peptides. Weinstein-Marom et al. [33] showed that the addition of an NMD-sensitive 3’ non-coding region to a protein-coding sequence increases degradation of the encoded protein and enhances peptide generation, presumably reflecting NMD surveillance. Chu et al. [34] reported that adding a 3’ NMD domain to mRNA enhances proteasome degradation of the upstream encoded protein and transcript degradation. This process is abrogated by the knockdown of NMD components UPF1 and SMG1 but not SMG1 phosphorylation of UPF1, which is required for mRNA degradation.

As 3–10% of all mRNAs in human cells are natural substrates of NMD [35] (and possibly more based on recent long-read sequencing studies [36]), NMD may constitute an important source of antigenic peptides.

Mechanistically Defining DRiPs

ER-targeted proteins have long been known (e.g., viral glycoproteins, ovalbumin) to be a major source of viral and host peptides. Approximately ~30% of newly synthesized proteins are targeted to the ER, typically by N-terminal signal peptides (SP) [37, 38]. Since proteins that achieve their native conformation in the secretory pathway are secreted or shed into extracellular fluid or degraded in lysosomes, peptides from these proteins that load in the ER must derive from DRiPs.

Cosma et al. [39•] tracked peptides from ER-targeted vs. mislocalized virus-encoded proteins based on the conversion of Asn to Asp upon cytosolic PNG’ase removal of an Asn-linked glycan in the target peptide [40, 41] and T cell discrimination of SIINFSKL vs. SIIDFSKL. This revealed that DRiPs, in the form of a minute fraction of translation products that fail to enter the ER, account for nearly all presented peptides.

Trentini et al. [42•] examined the contribution of ribosome-associated quality control (RQC) to MAP supply by generating degron controllable, SIINFEKL reporter proteins with or without stop codons. Absent stop codons, ribosomes stall on the 3’ poly-A tail, which induces RQC-mediated degradation of the stalled polypeptides, mediated by LTN and likely other E3 ligases. Such stalling results in efficient nascent protein degradation and K^b-SIINFEKL generation, even when degron-mediated protein misfolding is inhibited by a small molecule degron stabilizer, consistent with DRiP synthesis.

Expanding the study to the immunopeptidome by comparing wt to LTN knockout cells revealed that RQC contributes to ~3% of MAPs by mass and ~5% by diversity; these numbers are minimal values since LTN knockout does not wholly abrogate RQC. LTN-dependent MAPs were longer and enriched for source proteins containing more than ten transmembrane domains (TMDs) [42•]. Trentini et al. proposed that longer proteins contain more introns, increasing the probability of ribosome collisions and RQC-mediated protein degradation. Further, co-translational insertion of multi-pass transmembrane proteins into the target membrane is also likely to increase ribosome stalling and RQC. Based on the comparison between LTN-dependent and independent MHC I presented peptide source proteins in regard to their degradation profiles, i.e., following exponential degradation (efficient biosynthesis/low DRiP rate) or non-exponential degradation (error-prone biosynthesis/high DRiP rate), the authors propose that RQC may facilitate early sampling of stable viral proteins, consistent with the DRiP hypothesis [42•, 43, 44]. Indeed, most viral genomes likely contain secondary RNA code (e.g., packaging signals, secondary structures, etc.) [45], which in case of overlap with the protein-coding region, possibly leads to translation stalling and RQC activation.

Conclusion: Defining the Dark Immunopeptidome

Advances in ‘omics instrumentation, techniques, and analysis have greatly expanded the known immunopeptidome. Still, the number of unaccounted masses in immunopeptidome MS datasets remains vast. Since all endogenous peptides are translation products, RiboSeq provides the key to their identification, which may ultimately depend on advances in identifying post-translational amino side chain modifications that alter encode peptide mass. MS is unable, at present, to accurately quantitate peptides, which is obviously essential to defining the immunopeptidome and understanding how peptides are generated. Ultimately, single cell immunopeptidomics is required to characterize the uniformity of MAP expression in a given cell population, a critical issue in T cell killing of tumor cells.

The remarkable success of checkpoint inhibitors in treating cancer, while limited to partial success in a handful of cancers, offers the promise of a broad role of immunotherapy in treating cancer. Recent discoveries described in this review indicate that changes in the translatome associated with oncogenesis and natural and therapy-induced inflammation generate cancer-associated peptides suitable for vaccine targeting. This has led to a surge of interest in the antigen processing field that will lead to translatable discoveries and insight into fundamental cellular processes involved in synthesizing and degrading proteins.