2-3 pages summary

The advent of immunotherapy has revolutionized the treatment of many forms of cancer. It is now well estab- lished that T cells have the ability to reject tumours upon binding to antigenic peptides, derived from endogenous cellular proteins or exogenous viral proteins, presented by the major histocompatibility complex (MHC) on the surface of tumour cells. Several promising immuno- therapeutic anticancer approaches, such as therapeutic vaccines and T cell receptor engineered T cells (TCR- T cells) for adoptive cell therapy, rely on the identification of suitable target antigens1. Historically, the focus has been on three classes of tumour antigens: tumour- specific somatic non- synonymous mutation- derived neoantigens; cancer germline antigens; and antigens derived from viral genes that are expressed by virally infected tumour cells (for example, E6/E7 from human papilloma virus)2. Clinical studies have revealed remarkable outcomes both for TCR- T cell therapy targeting cancer germline anti- gens and for neoantigen- based vaccines1. For instance, TCR- T cells targeting NY- ESO-1, a tumour- specific shared germline antigen, have been shown to mediate sustained antigen- specific antitumour effects in patients with multiple myeloma, as well as several other can- cer types3–5. Further, personalized vaccines targeting mutation- derived neoantigens have been shown to elicit strong neoepitope- specific T cell responses in patients with melanoma (an immunologically ‘hot’ tumour with a high tumour mutational burden (TMB)) and glioblas- toma (an immunologically ‘cold’ tumour with a relatively low TMB)6–10.

Despite the unprecedented durable response rates obser ved with cancer immunotherapies in some patients, one of the major obstacles for the broader appli- cability of such therapies is the lack of currently known targetable  tumour- specific antigens (TSAs) for many cancer types1. The selection of appropriate antigens is

critical to ensure the safety and efficacy of immuno- therapy. Melanoma- associated antigen 3 (MAGE- A3) and melanoma antigen recognized by T cells 1 (MART-1) have been two leading target antigens for TCR- T cell- based cancer therapies due to their frequent expression in several tumour types and their restricted/low expres- sion in normal tissues. However, several clinical cases with unexpected severe off- target toxicities have been reported11–13. For example, in patients with melanoma, immunotherapy with T cells engineered with a high- affinity T cell receptor (F5-TCR) targeting MART-1 showed higher clinical efficacy compared with treatment with T cells engineered with a relatively low- affinity TCR (F4-TCR) but also caused uveitis, vitiligo and hearing loss due to MART-1 expression on melanocytes in the eye, skin and middle ear14. TCR- T cells and vaccines that target neoantigens may enable safer and more durable antitumour effects15, although mutational loads vary widely across different tumour types and identifying suitable targets remains a problem16.

In focusing on somatic mutation- derived neoanti- gens in tumour cells, possible neoepitopes derived from mRNA processing events are often overlooked. With respect to cancer, the most well- studied mRNA process- ing event, and the focus of this review, is mRNA splicing. Nonetheless, processing events such as mRNA poly- adenylation and mRNA editing have also been shown to play a role in tumour development and can result in an increased immunotherapy target space (Box 1). The advent of next- generation sequencing technologies has allowed for a wealth of transcriptomic data to be generated. Such data have helped to illuminate the widespread nature of alternative processing in cancer17 and have the potential to be used to identify neoepitopes derived from tumour- specific mRNA processing events, thereby expanding the repertoire of suitable targets for

Major histocompatibility complex (MHC). A set of genes that code for cell surface proteins (most notably the MHC class I and class II glycoproteins) that are responsible for presenting antigens to lymphocytes.

Adoptive cell therapy A type of immunotherapy approach that uses antigen- specific T cells to treat patients with chronic viral infections or various malignancies.

Non- synonymous mutation A nucleotide mutation that changes the amino acid sequence of a protein.

Alternative mRNA splicing in cancer immunotherapy Luke Frankiw1, David Baltimore 1* and Guideng Li 2,3*

Abstract | Immunotherapies are yielding effective treatments for several previously untreatable cancers. Still, the identification of suitable antigens specific to the tumour that can be targets for cancer vaccines and T cell therapies is a challenge. Alternative processing of mRNA, a phenomenon that has been shown to alter the proteomic diversity of many cancers, may offer the potential of a broadened target space. Here, we discuss the promise of analysing mRNA processing events in cancer cells, with an emphasis on mRNA splicing, for the identification of potential new targets for cancer immunotherapy. Further, we highlight the challenges that must be overcome for this new avenue to have clinical applicability.

1Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USA. 2Center of Systems Medicine, Institute of Basic Medical Sciences, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China. 3Suzhou Institute of Systems Medicine, Suzhou, China.

*e- mail: [email protected]; [email protected]

https://doi.org/10.1038/ s41577-019-0195-7

REVIEwS

N AT u R e R e v I e W S | I m m u N o Lo gy v o l u m e 1 9 | N o v e m B e R 2 0 1 9 | 675

Box 1 | Beyond RNA splicing: non- canonical neoepitopes

RNA splicing is just one of the processing steps that occurs in a pre- messenger RNA transcript (pre- mRNA) before the formation of a mature transcript. Although splicing is the best- studied process with respect to cancer, dysregulation of other processing steps is known to occur. In particular, polyadenylation (pA) and RNA editing have the potential to alter the proteome of a cancer cell and, thus, like RNA splicing, the identification of such events might broaden the immunogenic target space.

Polyadenylation involves the cleavage and addition of a stretch of adenosines, termed the poly(A) tail, to the 3ʹ end of the vast majority of eukaryotic mRNAs. Polyadenylation is complicated by the fact that the majority of human genes contain more than one pA site and that mRNA transcripts are frequently alternatively polyadenylated156. The majority of alternative polyadenylation (APA) sites are in the 3ʹ untranslated region (uTR) and can alter the stability, localization and translation of a given transcript157. However, there are many APA events that are located in intronic regions upstream of the last exon which act to generate either non- coding transcripts or transcripts with truncated coding regions158. The classic intronic polyadenylation (IPA) event occurs in the Igm heavy chain mRNA wherein, upon activation, a proximal IPA site is used, resulting in a shift to the secreted form of the antibody from the membrane- bound form159 (see the figure, part a).

The increased transcriptome complexity created through APA carries with it the risk of gene dysregulation, and it is perhaps no surprise that APA has been associated with tumorigenesis160–163. Recent work has focused on IPA, which has been shown to be a common mechanism of tumour- suppressor inactivation in chronic lymphocytic leukaemia162. Further, it was shown that the kinase CDK12, which is a key regulator of transcription elongation, also has a role in regulating genes involved in DNA repair by suppressing IPA163. In CDK12 mutant tumours, loss of suppression of IPA leads to impaired production of full- length (Fl) gene products for several genes involved in DNA repair. With respect to immunotherapy, the identification of IPA events is exciting due to the potential discovery of tumour- specific peptides. When cancer- specific IPA events occur in the coding region, sequences downstream of the nearest 5ʹ splice site (SS) and upstream of the new polyadenylation site will be translated, creating peptides that might be presented on major histocompatibility complex (mHC) molecules and recognized by the immune system. These IPA events commonly occur in genes that are important for disease progression162,163, which makes such peptides excellent immunotherapy targets. However, peptides derived from tumour- specific IPA events that bind to mHC molecules have yet to be identified and it is uncertain how immunogenic such peptides would be. As more data from methods such as 3ʹ seq164, which is used to identify and quantify polyadenylation site usage, become available, we will better understand the extent to which IPA events can alter the immunotherapeutic target space.

Another step in processing of pre- mRNA is RNA editing (see the figure, part b). The most common form of RNA editing involves the conversion of adenosine to inosine (A- to-I), a process catalysed by the adenosine deaminases acting on RNA (ADARs)165,166. Because most cellular machinery interprets inosine as guanine167, A- to-I editing can alter the amino acid sequence coded by a given transcript. like splicing and polyadenylation, RNA editing has been shown to be dysregulated in many types of cancer168–171, and it was recently reported that peptides derived from over- edited transcripts are presented by mHC molecules in a subset of tumour samples172. most prevalent in ovarian cancer, breast cancer and melanoma, it was further shown that effector CD8+ T cells specific for such peptides were present in the respective tumours, indicating that the peptides are indeed immunogenic. It is important to note that these peptides cannot be considered tumour specific. As editing still occurs in healthy tissue, and peptides derived from editing events are present on mHC molecules of healthy tissues, these over- edited peptides can be classified as tumour- associated shared self- antigens. As such, the therapeutic window that offers efficacy with limited toxic effects would first need to be defined for potential therapies targeted at such peptides.

extending the focus beyond RNA processing, recent work has uncovered several other non- canonical neoantigens that promise to greatly expand the immunotherapy target space. For example, a complete response to anti- programmed cell death 1 (PD-1) therapy was reported to have been mediated by an immune response targeted at an immunogenic peptide derived from a gene fusion event173. The authors suggest that the immunodominant epitope underlying regression of the tumour is probably derived from a DEK–AFF2 fusion expressed in tumour cells. expanding the analysis to 30 different cancer types revealed that 24% of cancers that expressed fusion proteins had a fusion- derived neoepitope predicted to bind to patient- specific mHCs173. Finally, in a cohort of patients with melanoma who responded to anti- PD-1 therapy, it was shown that predicted fusion neoantigens were eliminated, likely due to immune evasion173. This implicates gene fusions as a source of immunogenic neoantigens that could serve as a predictive biomarker for checkpoint inhibitor response.

Downstream pA siteIPA site

a

b

IPA isoform FL isoform AAAAAA

Potential IPA-derived neoepitopes

AAAAAA

5′ SS 3′ SS 5′ SS 3′ SS

Antigen processing and presentation

Translation

RNA editing

ADAR

A I

Neoantigens Newly formed antigens that have not been previously recognized by the immune system.

Cancer germline antigens Antigens that are normally exclusively expressed in germline cells but have aberrant expression in tumours, such as NY- ESo-1.

Tumour mutational burden (TMB). Also referred to as the tumour mutational load, this is a measurement of mutations carried by tumour tissue taken from a patient.

Tumour- specific antigens (TSAs). Antigens that are exclusively presented by tumour cells but not by any other cells.

RNA editing A molecular process resulting in alteration of the RNA sequence before translating to protein.

Alternative polyadenylation (APA). An RNA- processing event that generates distinct

3ʹ termini on mRNAs and other RNA polymerase II transcripts.

w w w.nature.com/nri

R e v i e w s

676 | N o v e m B e R 2 0 1 9 | v o l u m e 1 9

immunotherapy. Particularly for cancers such as B cell acute lymphoblastic leukaemia, which has a low preva- lence of somatic mutations and copy number variations but displays widespread mRNA splicing aberrations, the expanded target space could lead to the develop- ment of efficient immunotherapies18. Further, because the somatic mutation- derived neoepitope load has been shown to positively correlate with response to immune checkpoint blockade therapy in many cancer types19–24, uncovering processing- derived neoepitopes might offer clinical utility as a predictive biomarker. In this Review, we explore emerging evidence suggesting that mRNA processing- derived neoantigens can be suitable TSAs for cancer immunotherapy and discuss the major challenges that lie ahead.

Alternative mRNA splicing in cancer The processing of pre- mRNA transcripts (pre- mRNAs) represents an essential step in the ultimate function- ality of a gene product. The vast majority of human genes contain multiple exons, with adjoining intronic sequences that need to be spliced from a transcribed pre- mRNA to form the mature mRNA. Alternative splicing, a process by which a single pre- mRNA can be variably spliced into unique mature transcripts, can contribute to transcriptomic and proteomic diversity25–27 (FIg. 1a). This process is tightly regulated in different tissues, cell types and differentiation stages28–33. Of note, one specific

alternative splicing event, intron retention, can be derived from a regulated process affecting select junctions, or from a lack of processing throughout an entire gene30,32. Although the association between dysregulated splicing events with specific cancers has been known for many years34, the recent transcriptomic characterization of cancers has led to the finding that such events are much more frequent than previously predicted18,35–39.

Although our understanding of the extent to which specific alternative splicing events drive tumorigenesis is still evolving, there are several factors that help to explain the widespread dysregulation of splicing in cancer. First and foremost, a surprising finding from the genomic characterization of different cancers was the recur- rent somatic mutations found in genes encoding core spliceosome components as well as in trans- acting splic- ing factors that are essential to the regulation of alter- native splicing40,41. Frequent mutations in components of the spliceosome were initially detected in patients with myelodysplastic syndrome42–44 and chronic lymphocytic leukaemia45,46 but later also found in a wide variety of solid tumours such as breast cancers47–49, pancreatic ductal adenocarcinoma50, uveal melanoma51,52 and lung adenocarcinoma53. Mutations that affect spliceosomal components can alter splicing efficiency and splice- site selection. For example, a recent transcriptomic analysis of chronic lymphocytic leukaemia cells revealed that mutations in the small ribonuclear protein (RNP) U2

Alternative splicing A regulated process during gene expression that results in a single gene coding for multiple proteins.

Intron retention A form of alternative splicing that results in inclusion of introns in the final protein product.

Spliceosome The multi- megadalton ribonucleoprotein complex responsible for removing introns from pre- mRNA sequences.

Constitutive splicing

Exon skipping/inclusion

Alternative 5′ splice sites

Alternative 3′ splice sites

Intron retention

Mutually exclusive exons

Pre-mRNA

a b Mature mRNA

Peptide–MHC

TAP

T cell

Alternatively spliced protein

Proteasome

ER

Fig. 1 | Alternative splicing and immunotherapy. a | Schematic depicting constitutive splicing, as well as the five common modes of alternative splicing: exon skipping/inclusion, alternative 5′ splice- site selection, alternative 3ʹ splice- site selection, intron retention and mutually exclusive exons. Shown on the right are the mature mRNA transcripts derived from each event. b | Alternative splicing- derived proteins can be processed into 8–11 residue peptides by the proteasome. They are then shuttled into the endoplasmic reticulum (ER) via the transporter associated with antigen processing (TAP), after which they can be loaded onto major histocompatibility complex (MHC) class I. Peptide–MHC complexes can then be recognized by T cells.

N AT u R e R e v I e W S | I m m u N o Lo gy

R e v i e w s

v o l u m e 1 9 | N o v e m B e R 2 0 1 9 | 677

component SF3B1, which is involved in 3ʹ splice- site recognition, resulted in increased levels of alternative 3ʹ splice- site events54. A recent re- analysis of data from The Cancer genome Atlas (TCGA) showed that 119 genes that encode core spliceosome and splicing factors carry putative driver mutations across 33 different tumour types, highlighting the extent to which such mutations affect cancer development55. Beyond the somatic muta- tions found in these splicing- related genes, there is a large body of work implicating the altered expression of genes encoding splicing factors in the widespread dysregulation of alternative splicing in cancer40,56,57. The regulation of alternative splicing is frequently carried out by trans- acting splicing factors, which bind to specific sequence motifs and promote or repress a given splicing event58. Several studies have shown that altered expres- sion of such factors occurs in numerous different cancers and can be linked to malignant transformation59–68. For example, MYC has been shown to upregulate hetero- geneous nuclear RNP (hnRNP) A1, hnRNP A2 and polypyrimidine tract binding protein B (PTB) in glio- mas, promoting the expression of the pyruvate kinase M2 (PKM2) isoform over the PKM1 isoform61,62. PKM2 activity can be regulated through the binding of various allosteric ligands, in turn allowing a cell to shunt glucose carbons towards other biosynthetic processes, which provides a selective advantage for cancer cells.

In addition to the alterations of spliceosome com- ponents and trans- acting splicing factors, mutations found in cis- regulatory elements have also been shown to alter alternative splicing and thereby affect tumori- genesis37,69–72. The conserved cis- regulatory elements of an intron include the 5ʹ and 3ʹ splice sites located at intron– exon junctions, the branch point sequence located near the 3ʹ end of an intron and the polypyrimidine tract located downstream of the branch point73. The dinucleo- tides GT and AG define the 5ʹ and 3ʹ splice sites for ~99% of annotated introns74. The branch point is 15–50 nucleotides upstream of the 3ʹ end of the intron and con- tains an adenine nucleotide that is important for the first transesterification reaction involved in splicing75. Apart from these core elements, numerous motifs exist in both exonic and intronic sequences that act to recruit RNA- binding proteins, which in turn activate or repress splic- ing at neighbouring junctions76. Mutations that create or destroy these cis- regulatory elements change splice- site selection and thus result in multiple RNA isoforms.

One mechanism by which such a mutation can lead to tumour development involves the inactivation of tumour suppressor genes. It has recently been shown that cis mutations leading to unproductive splicing events can act as a mechanism of tumour suppressor inactivation in a range of different cancers37,71. This occurs when an alternative splicing event introduces either a frameshift, which alters the amino acid sequence of a functional protein domain in the gene, or a premature termination codon, which subjects the transcript to degradation via the nonsense- mediated decay machinery77,78. Although splice sites are the most well- studied cis- regulatory ele- ment with respect to cancer- related mutations, intronic and exonic motifs within the pre- mRNA that are bound by RNA- binding proteins have an important role in

splicing regulation and, when mutated in cancer, have been shown to alter splicing decisions72,79,80. It is worth noting that although it may be a good starting point to predict splicing impact based on how a given mutation alters a splicing cis- regulatory element, these elements are not the only factors that constitute the ‘splicing code’. In fact, focusing only on these elements may dramati- cally underestimate how many somatic mutations affect splicing. A recent study analysed 8,656 paired DNA sequencing and RNA sequencing (RNA- seq) samples from the TCGA and found 1,964 somatic mutations that impact splice- site usage37. Such mutations were called splice- site- creating mutations (SCMs). Of the identi fied SCMs, 26% had been previously mis- annotated as mis- sense mutations and 11% had been mis- annotated as silent mutations. An increased wealth of genomic and transcriptomic data, as well as new tools such as the recently published SpliceAI81, allow deeper insights into the “splicing code” and have the potential to dramatically impact our understanding of the relationship between splicing and disease development.

The impact of alternative mRNA splicing on the target space for cancer immunotherapy. Several recent stud- ies have shown that peptides derived from tumour- specific mRNA splicing events have the potential to bind to MHC class I (MHC I) molecules where they serve as neoepitopes37,38,82 (FIg.  1b). One study com- prehensively analysed TCGA data to investigate alter- native splicing across 8,705 patients and found that tumours consistently bear more alternative splicing events compared with healthy tissue38. Restricting the downstream analysis to 63 breast and ovarian cancers that had corresponding mass spectrometry (MS) data from the Clinical Proteomic Tumor Analysis Consortium (CPTAC), it was found that 68% of the tumours con- tained one or more alternative splicing- derived neo- epitopes. In contrast, only 30% of the tumours contained a neoepitope derived from a somatic single- nucleotide variant event. Such findings highlight the vast increase in target space that can be gained by analysing splicing- derived neoepitopes in addition to single- nucleotide variant- derived neoepitopes.

Although this work showed the potential broadened target space associated with the inclusion of splicing- derived events, it may in fact still underestimate the true number of splicing- derived neoepitopes. This is because the study only considered neoepitopes with MS evidence from the CPTAC38. However, trypsin is used to generate peptides from full- length proteins for MS, and it has been shown that there is a cleavage bias towards exon–exon junctions83. Trypsin cleaves lysine and argi- nine, which are encoded by conserved nucleotides at exon–exon boundaries83. This, in turn, can obscure the detection of peptides derived from neojunctions. Further, the analysis focused solely on peptides derived from tumour- specific splice junctions. Thus, peptides that either reside completely within a skipped exon, or within a retained intron, were missed. With respect to the latter, a second study looked specifically at intron retention events and concluded that the inclusion of retained intron- derived neoepitopes roughly doubled

The Cancer Genome Atlas (TCgA). The world’s largest and richest collection of genomic data.

Nonsense- mediated decay A translation- coupled mechanism that degrades mRNAs harbouring premature translation- termination codons.

Splice- site-creating mutations (SCMs). genomic mutations that induce splice- site creation. often mis- annotated as missense and silent mutations.

Clinical Proteomic Tumor Analysis Consortium (CPTAC). The first large- scale project that produced proteomics data sets from the mass spectrometric interrogation of tumour samples previously studied by The Cancer genome Atlas programme.

Neojunctions Novel exon–exon junctions found in tumour samples that are not typically found in healthy tissue.

w w w.nature.com/nri

R e v i e w s

678 | N o v e m B e R 2 0 1 9 | v o l u m e 1 9

the estimate of the total neoepitope load82. This find- ing may be surprising considering the fact that many transcripts that retain an intron incorporate a premature termi nation codon and are considered non- functional due to rapid degradation via the nonsense- mediated decay pathway84. However, degradation occurs following the pioneer round of translation and it has been shown that peptides produced during the pioneer round can bind to MHC I molecules85.

The potential for splicing- derived neoepitopes to expand the immunotherapy target space is further sup- ported by the study of paired DNA and RNA samples from the TCGA as described above37. It was found that, on average, an SCM generated more than two times as many neoepitopes per mutation as compared with an average non- synonymous mutation. Furthermore, a substantial number of these neoantigen events were recurrent (present in three or more samples), including events in important cancer- related genes such as GATA3, TP53 and PTEN. Such events have the potential to make excellent immunotherapy targets, although the extent to which they are immunogenic remains unknown.

Although the aforementioned work directly impli- cates splicing in the generation of tumour- specific neoepitopes37,38,82, another notable study published this year indirectly implicates RNA splicing86. Here, the authors developed a proteogenomic strategy involving MS that was capable of identifying aberrantly expressed TSAs, which are cancer- restricted non- mutated epitopes, in a high- throughput manner. They found that a signi- ficant majority of neoepitopes from two murine can- cer cell lines and seven human primary tumours were derived from the translation of out- of-frame coding exons or non- coding regions, many of which proved to be immunogenic. With respect to the former, it can be speculated that such frameshifts could be derived from dysregulated alternative splicing events. However, there are other potential mechanisms, including indels, which can also generate highly immunogenic tumour neoantigens87. Further, several peptides were derived from intronic sequences, again implicating dysregu- lated splicing86. Other non- coding regions that pro- duced aberrantly expressed TSAs include intergenic sequences, non- coding and untranslated regions, and even endogenous retroelements86.

Finally, splicing- derived neoantigen discovery might also be useful as a predictive biomarker for response to immune checkpoint blockade therapy. Immune checkpoint cascades, such as those controlled by pro- grammed cell death 1 (PD-1)88–90 or cytotoxic T lympho- cyte antigen 4 (CTLA-4)91–93, act as negative regulators of immune activation and blocking these with mono- clonal antibodies revolutionized the treatment of many cancers, resulting in unprecedented rates of long- lasting tumour responses94. As might be expected, it was found that the TMB correlates with an increase in neoanti- gens displayed by MHC I and II molecules95,96. Several studies have shown a correlation between the response to checkpoint inhibitors (CPIs) and the TMB in different tumour types19–24, such as melanoma, urothelial carci- noma, head and neck cancer and non- small-cell lung cancer. However, the data clearly show that the TMB is not

the only factor that dictates the response to CPIs. Some patients with high TMB respond poorly to CPIs, whereas some patients with low TMB respond well19–23. Given the number of potential neoepitopes derived from mRNA splicing events, it is reasonable to hypothesize that the splicing- derived neoepitope load might be of use as a clinical biomarker for response to CPIs. In fact, the study of SCMs described above found a higher expres- sion of the programmed cell death 1 ligand 1 (PD- L1) in tumours with an SCM as compared with tumours with- out an SCM, suggesting PD- L1 immunotherapy might be more effective in tumours with SCMs37. However, no association was found between the neoepitope load derived from intron retention events and the clinical benefit from CPIs using data from two cohorts of mela- noma patients treated with these drugs19,82,97. A poten- tial explanation for this result stems from the fact that only intron retention events were analysed and the data only involved patients with melanoma, a cancer with an extremely high TMB16. It is possible that a correlation of the splicing- derived neoepitope load with clinical effi- cacy is more pronounced in tumours with a low TMB. As more clinical data from CPI- treated patients become available, it will be interesting to determine the extent to which the mRNA splicing- derived neoepitope burden can serve as a predictive biomarker for CPI response.

Technological and biological challenges Recent technological innovations have made it possi- ble to identify tumour- specific mRNA splicing- derived neoantigens (FIg.  2). Such neoantigens can present a new class of cancer immunotherapy targets. However, numerous challenges remain for the development and application of immunotherapies targeting these mRNA splicing- derived neoantigens. These challenges include the identification of tumour- specific mRNA splicing events, the validation of peptide presentation, specificity and crossreactivity, immunogenicity and the prevention of tumour escape arising from tumour heterogeneity and evolution.

Identification of tumour- specific splicing events. The accurate identification of tumour- specific mRNA splic- ing events will be central to the ultimate efficacy of immunotherapies targeting neoantigens. With respect to the analysis of RNA- seq data, there are numerous commonly used computational tools that allow the quantification of alternative splicing (Box 2). Still, given that potential off- target effects of cell- based immuno- therapies have drastic consequences12,13, there are several questions regarding the classification of splicing events as tumour specific. A central question concerns the controls that are needed in the mRNA splicing- derived neoepitope identification process to ensure the safety of a potential therapy directed against it. A comparison with matched adjacent healthy tissue is the obvious first step; however, given the fact that splicing is uniquely regulated in different tissues33, is the comparison with matched adjacent healthy tissue sufficient or would dif- ferent tissue types need to be surveyed? As it is not fea- sible to acquire tissues from multiple vital organs from the same patient, a comprehensive comparison with a

Indels Insertion or deletion of nucleotides into genomic DNA, less than 1 kb in length.

Checkpoint inhibitors (CPIs). Types of drug that block the inhibitory checkpoint molecules.

Crossreactivity The recognition of two or more peptide–major histocompatibility complex complexes by a T cell receptor.

N AT u R e R e v I e W S | I m m u N o Lo gy

R e v i e w s

v o l u m e 1 9 | N o v e m B e R 2 0 1 9 | 679

database of human healthy tissues is likely needed to confidently select a given target. Along the same lines, what level of signal in healthy tissue, if any, would be considered acceptable? This is best highlighted in the case of intron retention, where, despite their transient nature, signals from introns can still be captured in actively transcribed pre- mRNAs that do not contain an intron retention event due to the time it takes for splic- ing to occur. Finally, it may also need to be determined whether a splice variant is expressed at other develop- mental stages, as it would be expected that neoepitopes derived from such events will likely suffer from a lack of immunogenicity.

A technical challenge is presented by the fact that mRNA splicing- derived neoepitope analysis has so far relied on bulk RNA- seq37,38,82. Although bulk RNA- seq has provided important insights, the technique lacks the ability to detect splicing effects at the subclonal level. As described in detail below, it has been shown that individual tumours have intratumoural heterogeneity with respect to individual splicing events98. Therapies that target a specific event present in only a fraction of the tumour will therefore lack efficacy. The development

of single- cell RNA- seq (scRNA- seq) offers the potential to identify splicing events that are present in all cells of a tumour. However, the combination of alternative splicing analysis with scRNA- seq is still technically dif- ficult (reviewed elsewhere99) as scRNA- seq relies on a low amount of starting material, which can restrict the analysis to highly abundant transcripts100. These analyses become even more problematic when studying isoform abundance in non- mutually-exclusive cases as non- dominant isoforms tend to be expressed at low levels and, thus, are susceptible to ‘drop- out’99. Further, the low sequence coverage common to scRNA- seq data makes it difficult to accurately characterize splicing variations in low- abundance transcripts. This problem might be alleviated through the development of machine learn- ing algorithms such as the recently published DARTS101, which offers the ability to better characterize splic- ing variations in transcripts with minimal coverage. In conclusion, although it is currently difficult to accu- rately quantify splicing with scRNA- seq, technological advances in both library preparation and sequencing methods, as well as new computational strategies that are tailored to the challenges of scRNA- seq data (namely

AGTCAGTumour tissue specimens

Normal tissue specimens

Identify antigen- specific TCRs

Express TCR in donor T cells

Vaccination

Adoptive transfer of TCR-T cells

Pools of synthetic peptides

Neoepitopes

Prediction and verification • MHC binding prediction • Mass spectometry

confirmation • T cell assay verification

RNA sequencing

Alternative splicing analysis Normal vs tumour tissues

Fig. 2 | Schematic illustration of the development of potential immunotherapies targeting mRNA processing- derived neoantigens. Tumour tissue and adjacent normal tissue specimens are obtained from a cancer patient and then subjected to RNA sequencing to identify tumour- specific alternative RNA processing events. This is then followed by a comprehensive comparison with a database of human healthy tissues to avoid selection of targets that might be presented in other healthy tissues. Computational tools are then used to predict the potential target neoepitopes derived from alternative RNA processing events, which are likely to be presented by either major histocompatibility complex (MHC) class I or II. Mass spectrometry of eluted peptides from MHCs and functional T cell activation/cytotoxic assays are performed to narrow down the validated immunogenic neoepitopes. These neoepitopes can hypothetically be used for therapeutic vaccination (for example, peptide, DNA and RNA) to elicit potent antitumour T cell responses in the patients. They can also be used to identify antigen- specific T cell receptors (TCRs) by MHC multimer- based screens or functional T cell expansion assays. The resulting TCRs can be used to engineer T cells to treat cancer patients. TCR- T cell, T cell receptor engineered T cell.

w w w.nature.com/nri

R e v i e w s

680 | N o v e m B e R 2 0 1 9 | v o l u m e 1 9

high technical noise, high processing requirements and misquantification of poorly expressed isoforms due to lack of coverage), offer a great deal of promise102–104.

Prediction and validation of peptide presentation. Although there is experimental evidence for tumour- specific splicing events, the extent to which they contri- bute to proteomic diversity remains a topic of debate. It has been suggested that a large percentage of alternatively

spliced transcripts are not translated into proteins105. Moreover, the detection of proteins derived from alter- native splicing events can be challenging as a large number of alternative splicing events are found in low- abundance transcripts106 and the sensitivity of MS is often not high enough to detect peptides derived from such low- abundance transcripts. Additionally, as dis- cussed above, peptides that span splice junctions are under- represented in MS- based proteomics data sets. Finally, MS requires substrate from large numbers of cells107. In general, the usage of 1 g of primary tissue is the minimum requirement for the detection of a few thousand MHC I or II binding peptides. If such an amount is not available, this may further preclude the detection of low- abundance peptides.

Even in cases where there is clear evidence of trans- lation of a particular mRNA, there is no guarantee that peptides derived from the full- length protein will be pre- sented by surface MHC molecules for immune recog- nition. Our understanding of which peptides will be processed and presented on surface MHC molecules is far from complete. Studies have looked at this question in the context of mRNA splicing- derived neoepitopes using MS and machine learning algorithms37,38,82. MS measurements of peptides eluted from MHC mol- ecules allow the direct identification of peptides bound to MHC molecules. Although such experimental evi- dence is the preferred method to confirm presentation, the sensitivity of this approach may not be sufficient. In the last decade, there has been an emergence of computational programs that utilize machine learn- ing frameworks to predict MHC I ligands108 (TABlE 1). However, the reliability of such in silico analyses has been called into question109 because the quality of their predictions relies on the quality of the training set used for machine learning. Classically, this involved in vitro data from the Immune Epitope Database (IEDB)110. More recently, MS- based immunopeptidome data have been integrated into the training sets, with the hope that such data may better reflect endogenous antigen pro- cessing, and such integration has indeed been shown to improve prediction accuracy111–114. For example, a recently developed neural network trained on MS data generated from cell lines expressing a single HLA allele outperformed previous algorithms trained on data from in vitro measured affinities when compared using two external MS- binding data sets111. Another issue with these in silico tools is the inability to accurately predict the binding of peptides to MHC II molecules, which means an entire arm of cellular immunity is over- looked. The binding of peptides to MHC II molecules is extremely promiscuous and there are limited data to train machine learning algorithms to predict MHC II- binding peptides115. Tools that predict the binding affin- ity for peptide–MHC II complexes are very inaccurate and, thus, lack utility116. In conclusion, current predic- tion tools trained on in vitro measured affinities and MS data offer the ability to narrow down the number of candidate neoepitopes in highly mutated tumours or tumours with a large number of tumour- specific splic- ing events, but further improvements are necessary to increase prediction accuracy.

Box 2 | Computational analysis of RNA splicing

There have been significant developments in the computational tools used to analyse and quantify differential splicing using RNA sequencing (RNA- seq) data. Broadly, such tools fall into two main categories: those that analyse full- length transcripts and those that analyse splicing events. For the analysis of transcripts, the transcriptome is computationally reconstructed and the abundance/relative proportion of full- length mRNA isoforms can be estimated by the RNA- seq reads that have been aligned to a given reference genome. more recently, pseudoalignment tools such as kallisto174 have been developed to perform alignment- independent isoform quantification with extraordinary computational efficiency. However, pseudoalignment tools quantify transcript abundance by directly comparing raw sequencing reads with transcript sequences and determining which transcripts a sequencing read is compatible with. Inherently, this relies on the selection of transcript annotations, which is an important consideration with respect to the discovery of cancer- specific splicing events and is discussed in more detail below. Then, tools such as sleuth can be used in conjunction with the data on transcript abundance that have been quantified with kallisto to determine differential transcript expression175. With respect to the analysis of splicing, differential transcript expression between samples, for example tumour and healthy tissues, is of limited utility as a transcript might have altered expression due to splicing changes, transcriptional changes or both. As such, there are also tools that compute the differential transcript usage (DTu), which quantifies the ratio of expression of a given transcript relative to all other transcripts for a given gene. Notable tools that can calculate DTu include RATs176, SuPPA2 (REF.177) and DRImSeq178.

Still, the identification and quantification of full- length transcripts from short reads is not trivial. The second category of computational tools involves those that analyse splicing events using an event- based approach. These tools detect alternative splicing events by comparing reads at a given junction between multiple samples. In general, the readout is the metric- like percent spliced in (Ψ), which represents the percentage of a given gene’s mRNA transcripts that include a specific splicing event. There are various commonly used tools in this category; for example, mISo179, rmATS180, mAJIQ181, leafCutter182, SplAdder183, Jum184 and Whippet185. of note, the aforementioned tool SuPPA2 can employ a hybrid approach, leveraging transcript quantification to produce both the isoform- centric DTu information and event- centric Ψ information. each tool and approach has its own advantages and various factors dictate which tool works best for a given experiment. Further, it has been proposed that multiple tools should be used to identify all possible significant splicing variants177.

With respect to the discussion of cancer- specific splicing events, an important consideration when comparing tools relates to their reliance on predefined transcript annotations because this may mask disease- specific splicing events. With respect to the transcript- level approach, pseudoalignment tools have made isoform quantification extraordinarily fast and thus scalable to large data sets. However, as previously mentioned, there is an inherent reliance on the choice of transcript annotations, and downstream splicing analyses are thus unable to discover or quantify novel alternative splicing events. It is possible to discover novel transcripts with older transcript- level approaches, for example with Cufflinks186, which can perform a reference- based de novo transcriptome assembly. This allows alternative splicing changes to be quantified based on the annotation of the assembled transcriptome. However, such an approach is both computationally difficult and expensive. With respect to the programs that analyse splicing events, these vary in their reliance on a reference annotation. Broadly speaking, mISo, SuPPA and Whippet rely on an annotation; rmATs, mAJIQ and SplAdder use a transcriptome annotation to guide alternative splicing analysis but can extend the analysis to detect novel splicing events; and leafCutter and Jum are annotation free. The importance of novel junction discovery, as compared with analysis of known annotations, plays a role in the selection of the analysis tool.

Immune Epitope Database (IEDB). A database containing detailed information for more than 100,000 unique immune epitopes related to infectious and immune- mediated diseases.

N AT u R e R e v I e W S | I m m u N o Lo gy

R e v i e w s

v o l u m e 1 9 | N o v e m B e R 2 0 1 9 | 681

Specificity and crossreactivity. Antigen specificity and crossreactivity are two major concerns for immuno- therapy. Regarding the issue of specificity, on- target off- tumour toxicities caused by the expression of target antigen on normal tissues have been reported in several clinical trials of TCR- T cell- based immunotherapies targeting MAGE- A3, MART-1, carcinoembryonic antigen (CEA) and glycoprotein 100 (gp100)11–14. Due to the difficulty of identifying tumour- specific public antigens, person- alized immunotherapies targeting neoantigens arising from mRNA splicing events have the potential to be safer and more effective15. However, as previously mentioned, mRNA splicing can be a noisy process as the dynamic nature of the spliceosome can be a source of stochastic fluctuation117. This fluctuation may prove detri mental if it leads to low levels of expression of the splicing isoform in healthy tissue. Thus, adequate controls are necessary during the splicing- derived tumour- specific peptide identification process to ensure specificity, or, at the very least, a significant enrichment of the targeted splicing- derived peptides in a given tumour. Related to specificity is the problem of crossreactivity of TCRs, which has been found for TCRs targeting both public antigens and neoantigens. A high- affinity TCR targeting an epitope (EVDPIGHLY) derived from MAGE- A3 was found to recognize a peptide derived from the muscle protein titin that has a similar sequence (ESDPIVAQY)13. Further, a recent study found that a neoantigen- specific TCR identified from a patient with ovarian cancer showed crossreactivity against the corresponding wild- type peptide118. These findings highlight the need for careful evaluation of epitope similarity between targeted candidate neoepitopes and similar epitopes known to be presented on healthy tissue. If the peptides derived from the processing event are very similar to the peptides generated from unrelated proteins in normal tissues, TCR crossreactivity will be an issue. Several recent TCR- ligand screening technologies based on trogocytosis, signalling and antigen- presenting bifunctional recep- tors, yeast surface display, DNA barcoded multimers,

TetTCR- seq or organoid co- culture methods have been developed and allow for a better understanding of the problem of crossreactivity119–124.

Immunogenicity. Although a subset of expressed tran- scripts is translated, processed and presented on sur- face MHC molecules, only a fraction of this subset will elicit an immunogenic response125. For example, in four high- profile vaccination trials for melanoma and glio- blastoma, only a portion of somatic mutation- derived candidate neoepitopes (51.7–66% of MHC II- restricted epitopes and 16–43% of MHC I- restricted epitopes) elic- ited CD4+ or CD8+ T cell responses in patients6–9. The immunogenicity of an epitope can be impacted by several factors, including antigen abundance, antigen processing efficiency, peptide binding affinity to MHC molecules, peptide–MHC complex stability and central tolerance due to the expression of other peptides with a similar amino acid composition15,126. As such, experimental vali- dation of immunogenicity is crucial to the development of personalized immunotherapies. Over the past 20 years, there have been numerous examples of mRNA splicing- derived neoepitopes with evidence of immunogenicity (see TABlE 2 for experimentally vali dated splicing- derived peptides that are recognized by T cells)127–136. The alter- native splicing of CD20 in B cell lymphomas offers an excellent example133. The B cell lineage marker CD20 is subjected to an alternative splicing event whereby a 168-nucleotide region within exons 3–7 is spliced out. Although absent from normal resting B cells, this alterna- tive splice variant was present in several patient- derived B cell lines. Importantly, the alternatively spliced variant can give rise to HLA- DR1 binding epitopes and vaccina- tion with a CD20-derived peptide (D393-CD2028–47) was able to elicit epitope- specific CD4+ and CD8+ responses in HLA- A2/HLA- DR1 transgenic mice133. Another recent study showed that peptides derived from alter- natively spliced out- of-frame BCR/ABL transcripts are able to elicit a peptide- specific cytotoxic T lymphocyte response, as suggested by the detection of out- of-frame

Table 1 | Commonly used major histocompatibility complex class I binding prediction tools

Name Training data Allele coverage

Access Refs

MixMHCpred Mass spectrometry Allele specific https://github.com/GfellerLab/MixMHCpred 112,114

NetMHCpan4.0 Binding affinity + mass spectrometry

Pan- class I http://www.cbs.dtu.dk/services/NetMHCpan-4.0/ 113

MHCflurry Binding affinity Allele specific https://github.com/openvax/mhcflurry 187

NetMHC4.0 Binding affinity Allele specific http://www.cbs.dtu.dk/services/NetMHC/ 188

PickPocket Binding affinity Pan- class I http://www.cbs.dtu.dk/services/PickPocket/ 189

NetMHCcons Binding affinity Allele specific http://www.cbs.dtu.dk/services/NetMHCcons/ 190

MHCnuggets Binding affinity Allele specific https://github.com/KarchinLab/mhcnuggets-2.0 191

ConvMHC Binding affinity Pan- class I http://jumong.kaist.ac.kr:8080/convmhc 192

HL A- CNN Binding affinity Allele specific https://github.com/uci- cbcl/HL A- bind 193

NetMHCstabpan Binding stability Pan- class I http://www.cbs.dtu.dk/services/NetMHCstabpan/ 194

NetMHCstab Binding stability Allele specific http://www.cbs.dtu.dk/services/NetMHCstab/ 195

SYFPEITHI Binding affinity + mass spectrometry

Allele specific http://syfpeithi.de/0-Home.htm 196

Trogocytosis A biological process where interacting cells share membrane and membrane- associated proteins.

w w w.nature.com/nri

R e v i e w s

682 | N o v e m B e R 2 0 1 9 | v o l u m e 1 9

peptide- specific IFNγ+CD8+ T  cells in patients with chronic myeloid leukaemia and the specific recognition and killing of out- of-frame peptide- pulsed target cells in vitro by these cytotoxic T lymphocytes134.

Although there has been a vast increase in the iden- tification of splicing- derived neoepitopes in the recent past, the percentage of these neoepitopes that are immuno- genic remains unknown. As previously discussed, the process of splicing has some intrinsic noise. This can negatively impact immunogenicity, as low levels of expression of a splicing- derived neoepitope might be expressed in the thymus, inducing central tolerance. Further, a large number of alternative splicing events are found in low- abundance isoforms of proteins106, which may preclude efficient targeting. With respect to immunogenicity, it is important to note that our under- standing of antigen processing and presentation is not complete and, thus, it is difficult to accurately predict which peptides will be immunogenic and which will not. This is especially apparent for class II MHC peptides15,115. Additionally, the measurement of peptide–MHC stabil- ity is the primary experimental method to predict neo- antigen immunogenicity137, but this is not particularly accurate or efficient. It would be more useful to assess antigen immunogenicity using an approach that involves both measuring the T cell response in an in vitro cell co- culture setting and immunizing transgenic mice that express human HLA class I and/or II125,138 molecules.

Tumour heterogeneity and evolution. Immunotherapy can induce long- lasting responses in cancer patients; however, tumour escape mechanisms can significantly impair clinical outcomes. One of the root causes of

resistance is intratumoural heterogeneity139, as shown for chimeric antigen receptor T (CAR T) and TCR- T cell- based adoptive cell therapy140. mRNA splicing- derived neoepitopes are not immune to this obstacle. As pre- viously stated, the identification of such neoepitopes has relied on bulk RNA- seq and, thus, we lack a proper understanding of splicing at the subclonal level37,38,82. This raises the question of what percentage of tumour cells in a given tumour harbour a given splicing- derived neoepitope, and scRNA- seq has shown that there can be significant heterogeneity at the single- cell level98,102,141–143. One of the seminal scRNA- seq studies performed on glioblastomas found considerable cell- to-cell variabil- ity in splicing patterns98. For example, three different variants of the EGFR gene were found to be mutually exclusively expressed across individual cells from the same tumour.

In addition to heterogeneity, tumour evolution poses another challenge to the efficacy of immuno- therapy. Acquired resistance to immunotherapy has been observed in both model systems and clinical tri- als. Several CAR T  cell clinical trials suggested that antigen- negative escape is one of the most common mechanisms of acquired resistance to immunotherapy treatment144–146. For example, epitope loss was observed in B cell acute lymphoblastic leukaemia and diffuse large B cell lymphoma patients receiving anti- CD19 and/or anti- CD22 CAR T cell therapy144,146. Further investi- gation of the anti- CD19 CAR T cell trial revealed that, in some instances, epitope loss involved the selection of clones that splice out exon 2 of CD19, which contains the epitope recognized by the antigen- binding moiety of the anti- CD19 CAR T cell. This alternative splicing

Chimeric antigen receptor (CAR). Recombinant receptor protein that has been engineered to direct T cells to target a specific protein on malignant cells.

Table 2 | Experimentally validated mRNA splicing- derived peptides that are recognized by T cells

Tumour type Antigen Peptide sequence HLA type Ref.

Melanoma AIM2 RSDSGQQARY HL A- A1 127

Melanoma NA17-A • VLPDVFIRC • VLPDVFIRCV

HL A- A1 127

Melanoma GP100 VYFFLPDHL HL A- A24 128

Melanoma TRP-2 EVISCKLIKR HL A- A*68011 and HL A- A*3301 129

Melanoma CAMEL MLMAQEAL AFL HL A- A*0201 130

Melanoma CAMEL RTAACFSCTSRCLSRRPWKRSWS Unknown 131

Melanoma CAMEL CLSRRPWKRSWSAGSCPGMPHL HL A- DR7/HL A- DR11/HL A- DR12 131

Melanoma CAMEL MLMAQEAL AFLMAQGAML AA HL A- DR 132

Melanoma CAMEL QGAML AAQERRVPRAAEVPG HL A- DR3 132

Melanoma CAMEL APRGVRMAVPLLRRMEGAPA HL A- DR 132

Lymphomas CD20 KPLFRRMSSLELVIAGIVEN HL A- DRB1 133

Lymphomas CD20 RMSSLELVI HL A- A2 133

Leukaemia BCR/ABL • QQAHCLWCV • GVRGRVEEI • LLREPLQHP • CLWCVPQLR • RLLREPLQH • RVLERSCSH

HL A- A2 and/or HL A- A3 134

Oral cancer Survivin-2B AYACNTSTL HL A- A24 135

Breast cancer melanoma

NY- ESO-ORF2 L AAQERRVPR HL A- A31 136

N AT u R e R e v I e W S | I m m u N o Lo gy

R e v i e w s

v o l u m e 1 9 | N o v e m B e R 2 0 1 9 | 683

event is mediated by SRSF3, a trans- acting splicing factor that acts to promote exon 2 inclusion, and levels of SRSF3 were lower in splicing- mediated relapsed samples146.

It is likely that mRNA splicing- derived events that are targets for cell- based immunotherapies will be vulnerable to similar splicing- mediated mechanisms of immune escape. For instance, when targeting an intron retention event, potential pre- existing clones that efficiently splice the intron would be able to escape immunotherapy (FIg. 3a). In addition to the selection of pre- existing resistant clones, individual tumour cells that are sensitive to immunotherapy might also develop resistance upon treatment. As an example, in response to TCR- T cell therapy, splicing might be altered in tumour cells such that a given neoepitope is no longer translated, resulting in immune escape of tumour cells lacking this epitope (FIg. 3b). Resistance may be less of an issue if an immunotherapy target is derived from an mRNA splicing event that acts as a driver of tumori- genesis, because altering the splicing event will have a negative impact on cancer cell survival. Yet, still, it is unknown how many of these single splicing- mediated driver events exist. It may be possible to overcome the outgrowth of pre- existing antigen- negative clones or emerging antigen- negative clones by targeting multi- ple tumour- specific events, which has been shown in both in vitro and preclinical studies to offset antigen escape and result in increased antitumour activity147,148. However, beyond splicing- mediated mechanisms of

antigen escape, there are many other mechanisms by which antigen escape can occur. An example includes the loss of MHC I expression (HLA- A, HLA- B, HLA- C and β2M) due to simultaneous molecular defects in both copies of the gene or by loss of heterozygosity in one copy of one chromosome and a mutation/deletion in the other homologous gene8,149–155 (FIg. 3c,d). The development of new technologies that either restore antigen presenta- tion or prevent antigen escape will be important for the continued improvement of durable response rates in patients treated with immunotherapy.

Conclusion Mounting evidence indicates that alternative mRNA splicing- derived neoepitopes can be promising immu- notherapy targets. Most of the work has focused on mRNA splicing- derived events, although other forms of mRNA processing have been shown to be dysregulated in cancer and offer promise with regard to immuno- therapy target expansion. Although progress made to expand the immunotherapy target space using tumour- specific mRNA processing events has been significant, a great deal of work is needed. The first critical hurdle involves identifying the most immunogenic epitopes from the numerous candidates derived from mRNA processing events. Current antigen- presentation predic- tion algorithms need to be optimized for the accurate identification of immunogenic neoepitopes. In the last few years, a significant amount of high- quality MHC

T cella b

c d

Pre-existing resistant cell

Alternative spliced isoform

Pre-existing resistant cell

Pre-existing resistance Acquired resistance

Selection of alternative splicing-derived neoantigen clone Loss of alternative splicing-derived neoantigen expression

Selection of HLA-negative clones Loss of MHC expression

Fig. 3 | Potential mechanism of tumour escape from immunotherapy. a | A tumour might consist of a heterogeneous cell population expressing either constitutive splicing or alternative spliced gene products (shown is an intron retention event). The cells in which the immunogenic epitope is constitutively spliced out will not be recognized by T cell receptor engineered T cells (TCR- T cells) that target the alternative splicing- derived epitope and will be selected for in response to TCR- T cell therapy. b | Splicing- mediated mechanisms of antigen- negative escape. When targeting an epitope derived from an intron retention event, acquired resistance to TCR- T cell therapy might occur through epitope loss that is mediated by splicing out the intron. c | Low expression of HL A molecules (for example, class I major histocompatibility complex (MHC) and β2-microglobulin) or alterations in genes encoding components of the antigen- processing machinery and/or HL A molecules can impair antigen presentation to TCR- T cells and result in relapse of antigen- negative tumours following TCR- T cell therapy. d | Loss of MHC expression following TCR- T cell therapy allows antigen- negative tumour cells to escape from T cell attack, rendering these cells resistant to TCR- T cell- directed therapy151–155.

Loss of heterozygosity A common somatic genome event that results in loss of the entire gene and the surrounding chromosomal region.

w w w.nature.com/nri

R e v i e w s

684 | N o v e m B e R 2 0 1 9 | v o l u m e 1 9

immunopeptidome data has been generated. Sharing the published or even unpublished data, together with genomic information, would be valuable for further improvement in the accuracy of computational predic- tions. Furthermore, current experimental validation methods for immunogenicity and crossreactivity are still laborious, non- robust and of low throughput. Thus, there is a need for new technologies that allow for more rapid, robust and precise identification of tumour- specific immunogenic epitopes so that we can accu- rately assess their therapeutic potential. Finally, with the

expanding clinical data on CPIs, more research needs to be carried out to investigate whether splicing- derived neoantigens could aid the prediction of CPI response. The last decade has witnessed great advances in the field of cancer immunotherapy, and incorporating mRNA splicing- derived neoepitopes as potential targets for cell- based and/or vaccination- based immuno therapeutic anticancer approaches may allow more patients to benefit from such treatments.

Published online 30 July 2019

1. Paucek, R. D., Baltimore, D. & Li, G. The cellular immunotherapy revolution: arming the immune system for precision therapy. Trends Immunol. 40, 292–309 (2019).

2. Hackl, H., Charoentong, P., Finotello, F. & Trajanoski, Z. Computational genomics tools for dissecting tumour– immune cell interactions. Nat. Rev. Genet. 17, 441–458 (2016).

3. Rapoport, A. P. et al. NY- ESO-1-specific TCR- engineered T cells mediate sustained antigen- specific antitumor effects in myeloma. Nat. Med. 21, 914–921 (2015).

4. Robbins, P. F. et al. Tumor regression in patients with metastatic synovial cell sarcoma and melanoma using genetically engineered lymphocytes reactive with NY- ESO-1. J. Clin. Oncol. 29, 917–924 (2011).

5. Robbins, P. F. et al. A pilot trial using lymphocytes genetically engineered with an NY- ESO-1-reactive T cell receptor: long- term follow- up and correlates with response. Clin. Cancer Res. 21, 1019–1027 (2015).

6. Ott, P. A. et al. An immunogenic personal neoantigen vaccine for patients with melanoma. Nature 547, 217–221 (2017). This work, along with the studies by Carreno et al. (2015) and Sahin et al. (2017), provides in- human evidence that vaccines against tumour neoantigens could be safe and effective in treating patients with advanced- stage melanoma.

7. Carreno, B. M. et al. A dendritic cell vaccine increases the breadth and diversity of melanoma neoantigen- specific T cells. Science 348, 803–808 (2015).

8. Sahin, U. et al. Personalized RNA mutanome vaccines mobilize poly- specific therapeutic immunity against cancer. Nature 547, 222–226 (2017).

9. Hilf, N. et al. Actively personalized vaccination trial for newly diagnosed glioblastoma. Nature 565, 240–245 (2019).

10. Keskin, D. B. et al. Neoantigen vaccine generates intratumoral T cell responses in phase Ib glioblastoma trial. Nature 565, 234–239 (2019).

11. Van den Berg, J. H. et al. Case report of a fatal serious adverse event upon administration of T cells transduced with a MART-1-specific T cell receptor. Mol. Ther. 23, 1541–1550 (2015).

12. Linette, G. P. et al. Cardiovascular toxicity and titin cross- reactivity of affinity- enhanced T cells in myeloma and melanoma. Blood 122, 863–871 (2013).

13. Cameron, B. J. et al. Identification of a titin- derived HLA- A1-presented peptide as a cross- reactive target for engineered MAGE A3-directed T cells. Sci. Transl Med. 5, 197ra103 (2013).

14. Johnson, L. A. et al. Gene therapy with human and mouse T cell receptors mediates cancer regression and targets normal tissues expressing cognate antigen. Blood 114, 535–546 (2009).

15. Bethune, M. T. & Joglekar, A. V. Personalized T cell- mediated cancer immunotherapy: progress and challenges. Curr. Opin. Biotechnol. 48, 142–152 (2017).

16. Alexandrov, L. B. et al. Signatures of mutational processes in human cancer. Nature 500, 415–421 (2013).

17. Goodwin, S., McPherson, J. D. & McCombie, W. R. Coming of age: ten years of next- generation sequencing technologies. Nat. Rev. Genet. 17, 333–351 (2016).

18. Black, K. L. et al. Aberrant splicing in B cell acute lymphoblastic leukemia. Nucleic Acids Res. 46, 11357–11369 (2018).

19. Snyder, A. et al. Genetic basis for clinical response to CTLA-4 blockade in melanoma. N. Engl. J. Med. 371, 2189–2199 (2014). This study, in conjunction with Van Allen et al. (2015) and Rizvi et al. (2015), provides evidence

of the correlation between the response to CPI and the TMB.

20. Van Allen, E. M. et al. Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science 350, 207–211 (2015).

21. Rosenberg, J. E. et al. Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum- based chemotherapy: a single- arm, multicentre, phase 2 trial. Lancet 387, 1909–1920 (2016).

22. Seiwert, T. Y. et al. Biomarkers predictive of response to pembrolizumab in head and neck cancer (HNSCC). Cancer Res. 78 (Suppl.), LB–339 (2018).

23. Rizvi, N. A. et al. Mutational landscape determines sensitivity to PD-1 blockade in non- small cell lung cancer. Science 348, 124–128 (2015).

24. Samstein, R. M. et al. Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nat. Genet. 51, 202 (2019).

25. Liu, Y. et al. Impact of alternative splicing on the human proteome. Cell Rep. 20, 1229–1241 (2017).

26. Weatheritt, R. J., Sterne- Weiler, T. & Blencowe, B. J. The ribosome- engaged landscape of alternative splicing. Nat. Struct. Mol. Biol. 23, 1117–1123 (2016).

27. Nilsen, T. W. & Graveley, B. R. Expansion of the eukaryotic proteome by alternative splicing. Nature 463, 457–463 (2010).

28. Kalsotra, A. et al. A postnatal switch of CELF and MBNL proteins reprograms alternative splicing in the developing heart. Proc. Natl Acad. Sci. USA 105, 20333–20338 (2008).

29. Yap, K., Lim, Z. Q., Khandelia, P., Friedman, B. & Makeyev, E. V. Coordinated regulation of neuronal mRNA steady- state levels through developmentally controlled intron retention. Genes Dev. 26, 1209–1223 (2012).

30. Wong, J. J.-L. et al. Orchestrated intron retention regulates normal granulocyte differentiation. Cell 154, 583–595 (2013).

31. Pimentel, H. et al. A dynamic alternative splicing program regulates gene expression during terminal erythropoiesis. Nucleic Acids Res. 42, 4031–4042 (2014).

32. Frankiw, L. et al. Bud13 promotes a type I interferon response by countering intron retention in Irf7. Mol. Cell 73, 803–814 (2019).

33. Baralle, F. E. & Giudice, J. Alternative splicing as a regulator of development and tissue identity. Nat. Rev. Mol. Cell. Biol. 18, 437–451 (2017).

34. Venables, J. P. Aberrant and alternative splicing in cancer. Cancer Res. 64, 7647–7654 (2004).

35. Braun, C. J. et al. Coordinated splicing of regulatory detained introns within oncogenic transcripts creates an exploitable vulnerability in malignant glioma. Cancer Cell 32, 411–426 (2017).

36. Coltri, P. P., dos Santos, M. G. & da Silva, G. H. Splicing and cancer: challenges and opportunities. Wiley Interdiscip. Rev. RNA 10, e1527 (2019).

37. Jayasinghe, R. G. et al. Systematic analysis of splice- site-creating mutations in cancer. Cell Rep. 23, 270–281 (2018). This study focuses on cancer mutations that had evidence of creating specific splicing junctions. These SCMs generated ~2 times as many neoepitopes per event compared with non- synonymous mutations.

38. Kahles, A. et al. Comprehensive analysis of alternative splicing across tumors from 8,705 patients. Cancer Cell 34, 211–224 (2018). This study focuses on cancer- specific neojunctions and shows that peptides derived from such events could significantly increase the target space for immunotherapy.

39. Climente- Gonzalez, H., Porta- Pardo, E., Godzik, A. & Eyras, E. The functional impact of alternative splicing in cancer. Cell Rep. 20, 2215–2226 (2017).

40. Dvinge, H., Kim, E., Abdel- Wahab, O. & Bradley, R. K. RNA splicing factors as oncoproteins and tumour suppressors. Nat. Rev. Cancer 16, 413–430 (2016).

41. Will, C. L. & Lührmann, R. Spliceosome structure and function. Cold Spring Harb. Perspect. Biol. 3, a003707 (2011).

42. Yoshida, K. et al. Frequent pathway mutations of splicing machinery in myelodysplasia. Nature 478, 64–69 (2011).

43. Papaemmanuil, E. et al. Somatic SF3B1 mutation in myelodysplasia with ring sideroblasts. N. Engl. J. Med. 365, 1384–1395 (2011).

44. Graubert, T. A. et al. Recurrent mutations in the U2AF1 splicing factor in myelodysplastic syndromes. Nat. Genet. 44, 53–57 (2012).

45. Wang, L. et al. SF3B1 and other novel cancer genes in chronic lymphocytic leukemia. N. Engl. J. Med. 365, 2497–2506 (2011). This study reveals widespread spliceosomal mutations to the U2 component SF3B1 in chronic lymphocytic leukaemia. It is among the first works showing that such splicing- related mutations are ubiquitous in cancer.

46. Quesada, V. et al. Exome sequencing identifies recurrent mutations of the splicing factor SF3B1 gene in chronic lymphocytic leukemia. Nat. Genet. 44, 47–52 (2012).

47. Ellis, M. J. et al. Whole- genome analysis informs breast cancer response to aromatase inhibition. Nature 486, 353–360 (2012).

48. Stephens, P. J. et al. The landscape of cancer genes and mutational processes in breast cancer. Nature 486, 400–404 (2012).

49. Maguire, S. L. et al. SF3B1 mutations constitute a novel therapeutic target in breast cancer. J. Pathol. 235, 571–580 (2015).

50. Biankin, A. V. et al. Pancreatic cancer genomes reveal aberrations in axon guidance pathway genes. Nature 491, 399–405 (2012).

51. Harbour, J. W. et al. Recurrent mutations at codon 625 of the splicing factor SF3B1 in uveal melanoma. Nat. Genet. 45, 133–135 (2013).

52. Martin, M. et al. Exome sequencing identifies recurrent somatic mutations in EIF1AX and SF3B1 in uveal melanoma with disomy 3. Nat. Genet. 45, 933–936 (2013).

53. Imielinski, M. et al. Mapping the hallmarks of lung adenocarcinoma with massively parallel sequencing. Cell 150, 1107–1120 (2012).

54. Alsafadi, S. et al. Cancer- associated SF3B1 mutations affect alternative splicing by promoting alternative branchpoint usage. Nat. Commun. 7, 10615 (2016).

55. Seiler, M. et al. Somatic mutational landscape of splicing factor genes and their functional consequences across 33 cancer types. Cell Rep. 23, 282–296 (2018).

56. Sebestyén, E. et al. Large- scale analysis of genome and transcriptome alterations in multiple tumors unveils novel cancer- relevant splicing networks. Genome Res. 26, 732–744 (2016).

57. Sveen, A., Kilpinen, S., Ruusulehto, A., Lothe, R. A. & Skotheim, R. I. Aberrant RNA splicing in cancer; expression changes and driver mutations of splicing factor genes. Oncogene 35, 2413–2427 (2016).

58. Fu, X.-D. & Ares Jr, M. Context- dependent control of alternative splicing by RNA- binding proteins. Nat. Rev. Genet. 15, 689–701 (2014).

59. Karni, R. et al. The gene encoding the splicing factor SF2/ASF is a proto- oncogene. Nat. Struct. Mol. Biol. 14, 185–193 (2007). This study shows that slight overexpression of the SF2/ASF splicing factor was pro- tumorigenic.

N AT u R e R e v I e W S | I m m u N o Lo gy

R e v i e w s

v o l u m e 1 9 | N o v e m B e R 2 0 1 9 | 685

60. Karni, R., Hippo, Y., Lowe, S. W. & Krainer, A. R. The splicing- factor oncoprotein SF2/ASF activates mTORC1. Proc. Natl Acad. Sci. USA 105, 15323–15327 (2008).

61. David, C. J., Chen, M., Assanah, M., Canoll, P. & Manley, J. L. hnRNP proteins controlled by c- Myc deregulate pyruvate kinase mRNA splicing in cancer. Nature 463, 364–368 (2010).

62. Clower, C. V. et al. The alternative splicing repressors hnRNP A1/A2 and PTB influence pyruvate kinase isoform expression and cell metabolism. Proc. Natl Acad. Sci. USA 107, 1894–1899 (2010).

63. Anczuków, O. et al. The splicing factor SRSF1 regulates apoptosis and proliferation to promote mammary epithelial cell transformation. Nat. Struct. Mol. Biol. 19, 220–228 (2012).

64. Cohen- Eliav, M. et al. The splicing factor SRSF6 is amplified and is an oncoprotein in lung and colon cancers. J. Pathol. 229, 630–639 (2013).

65. Jensen, M. A., Wilkinson, J. E. & Krainer, A. R. Splicing factor SRSF6 promotes hyperplasia of sensitized skin. Nat. Struct. Mol. Biol. 21, 189–197 (2014).

66. Gallardo, M. et al. hnRNP K is a haploinsufficient tumor suppressor that regulates proliferation and differentiation programs in hematologic malignancies. Cancer Cell 28, 486–499 (2015).

67. Wang, Y. et al. The splicing factor RBM4 controls apoptosis, proliferation, and migration to suppress tumor progression. Cancer Cell 26, 374–389 (2014).

68. Zong, F.-Y. et al. The RNA- binding protein QKI suppresses cancer- associated aberrant splicing. PLOS Genet. 10, e1004289 (2014).

69. Spinelli, R. et al. Identification of novel point mutations in splicing sites integrating whole- exome and RNA- seq data in myeloproliferative diseases. Mol. Genet. Genomic Med. 1, 246–259 (2013).

70. Liu, J. et al. Genome and transcriptome sequencing of lung cancers reveal diverse mutational and splicing events. Genome Res. 22, 2315–2327 (2012).

71. Jung, H. et al. Intron retention is a widespread mechanism of tumor- suppressor inactivation. Nat. Genet. 47, 1242–1248 (2015). This study shows that single- nucleotide variants causing intron retention were enriched in tumour suppressors.

72. Supek, F., Miñana, B., Valcárcel, J., Gabaldón, T. & Lehner, B. Synonymous mutations frequently act as driver mutations in human cancers. Cell 156, 1324–1335 (2014).

73. Lee, Y. & Rio, D. C. Mechanisms and regulation of alternative pre- mRNA splicing. Annu. Rev. Biochem. 84, 291–323 (2015).

74. Parada, G. E., Munita, R., Cerda, C. A. & Gysling, K. A comprehensive survey of non- canonical splice sites in the human transcriptome. Nucleic Acids Res. 42, 10564–10578 (2014).

75. Matera, A. G. & Wang, Z. A day in the life of the spliceosome. Nat. Rev. Mol. Cell. Biol. 15, 108–121 (2014).

76. Wang, Z. & Burge, C. B. Splicing regulation: from a parts list of regulatory elements to an integrated splicing code. RNA 14, 802–813 (2008).

77. Lim, S., Mullins, J. J., Chen, C. M., Gross, K. W. & Maquat, L. E. Novel metabolism of several beta zero- thalassemic beta- globin mRNAs in the erythroid tissues of transgenic mice. EMBO J. 8, 2613–2619 (1989).

78. Popp, M. W. & Maquat, L. E. Nonsense- mediated mRNA decay and cancer. Curr. Opin. Genet. Dev. 48, 44–50 (2018).

79. Singh, B., Trincado, J. L., Tatlow, P. J., Piccolo, S. R. & Eyras, E. Genome sequencing and RNA- motif analysis reveal novel damaging noncoding mutations in human tumors. Mol. Cancer Res. 16, 1112–1124 (2018).

80. Zatkova, A. et al. Disruption of exonic splicing enhancer elements is the principal cause of exon skipping associated with seven nonsense or missense alleles of NF1. Hum. Mutat. 24, 491–501 (2004).

81. Jaganathan, K. et al. Predicting splicing from primary sequence with deep learning. Cell 176, 535–548 (2019).

82. Smart, A. C. et al. Intron retention is a source of neoepitopes in cancer. Nat. Biotechnol. 36, 1056–1058 (2018). This study is the first to show that cancer- specific intron retention events could be a source of neoepitopes.

83. Wang, X. et al. Detection of proteome diversity resulted from alternative splicing is limited by trypsin cleavage specificity. Mol. Cell. Proteomics 17, 422–430 (2018).

84. Wong, J. J.-L., Au, A. Y., Ritchie, W. & Rasko, J. E. Intron retention in mRNA: no longer nonsense: known and putative roles of intron retention in normal and disease biology. Bioessays 38, 41–49 (2016).

85. Apcher, S. et al. Major source of antigenic peptides for the MHC class I pathway is produced during the pioneer round of mRNA translation. Proc. Natl Acad. Sci. USA 108, 11572–11577 (2011).

86. Laumont, C. M. et al. Noncoding regions are the main source of targetable tumor- specific antigens. Sci. Transl Med. 10, eaau5516 (2018).

87. Turajlic, S. et al. Insertion- and-deletion- derived tumour- specific neoantigens and the immunogenic phenotype: a pan- cancer analysis. Lancet Oncol. 18, 1009–1021 (2017).

88. Ishida, Y., Agata, Y., Shibahara, K. & Honjo, T. Induced expression of PD-1, a novel member of the immunoglobulin gene superfamily, upon programmed cell death. EMBO J. 11, 3887–3895 (1992).

89. Freeman, G. J. et al. Engagement of the PD-1 immunoinhibitory receptor by a novel B7 family member leads to negative regulation of lymphocyte activation. J. Exp. Med. 192, 1027–1034 (2000).

90. Baumeister, S. H., Freeman, G. J., Dranoff, G. & Sharpe, A. H. Coinhibitory pathways in immunotherapy for cancer. Annu. Rev. Immunol. 34, 539–573 (2016).

91. Chambers, C. A., Kuhns, M. S., Egen, J. G. & Allison, J. P. CTLA-4-mediated inhibition in regulation of T cell responses: mechanisms and manipulation in tumor immunotherapy. Annu. Rev. Immunol. 19, 565–594 (2001).

92. Walunas, T. L. et al. CTLA-4 can function as a negative regulator of T cell activation. Immunity 1, 405–413 (1994).

93. Leach, D. R., Krummel, M. F. & Allison, J. P. Enhancement of antitumor immunity by CTLA-4 blockade. Science 271, 1734–1736 (1996).

94. Ribas, A. & Wolchok, J. D. Cancer immunotherapy using checkpoint blockade. Science 359, 1350–1355 (2018).

95. Peggs, K. S., Segal, N. H. & Allison, J. P. Targeting immunosupportive cancer therapies: accentuate the positive, eliminate the negative. Cancer Cell 12, 192–199 (2007).

96. Segal, N. H. et al. Epitope landscape in breast and colorectal cancer. Cancer Res. 68, 889–892 (2008).

97. Hugo, W. et al. Genomic and transcriptomic features of response to anti- PD-1 therapy in metastatic melanoma. Cell 165, 35–44 (2016).

98. Patel, A. P. et al. Single- cell RNA- seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344, 1396–1401 (2014). This work provides evidence of intratumoural splicing heterogeneity in glioblastoma.

99. Arzalluz- Luque, Á. & Conesa, A. Single- cell RNAseq for the study of isoforms—how is that possible? Genome Biol. 19, 110 (2018).

100. Stegle, O., Teichmann, S. A. & Marioni, J. C. Computational and analytical challenges in single- cell transcriptomics. Nat. Rev. Genet. 16, 133–145 (2015).

101. Zhang, Z. et al. Deep- learning augmented RNA- seq analysis of transcript splicing. Nat. Methods 16, 307–310 (2019).

102. Song, Y. et al. Single- cell alternative splicing analysis with expedition reveals splicing dynamics during neuron differentiation. Mol. Cell 67, 148–161 (2017).

103. Huang, Y. & Sanguinetti, G. BRIE: transcriptome- wide splicing quantification in single cells. Genome Biol. 18, (123 (2017).

104. Welch, J. D., Hu, Y. & Prins, J. F. Robust detection of alternative splicing in a population of single cells. Nucleic Acids Res. 44, e73 (2016).

105. Tress, M. L., Abascal, F. & Valencia, A. Alternative splicing may not be the key to proteome complexity. Trends Biochem. Sci. 42, 98–110 (2017).

106. Pickrell, J. K., Pai, A. A., Gilad, Y. & Pritchard, J. K. Noisy splicing drives mRNA isoform diversity in human cells. PLOS Genet. 6, e1001236 (2010).

107. Caron, E. et al. Analysis of major histocompatibility complex (MHC) immunopeptidomes using mass spectrometry. Mol. Cell. Proteomics 14, 3105–3117 (2015).

108. Gfeller, D. & Bassani- Sternberg, M. Predicting antigen presentation—what could we learn from a million peptides? Front. Immunol. 9, 1716 (2018).

109. Schmidt, J. et al. In silico and cell- based analyses reveal strong divergence between prediction and observation of T cell–recognized tumor antigen T cell epitopes. J. Biol. Chem. 292, 11840–11849 (2017).

110. Vita, R. et al. The Immune Epitope Database (IEDB) 3.0. Nucleic Acids Res. 43, D405–D412 (2014).

111. Abelin, J. G. et al. Mass spectrometry profiling of HLA- associated peptidomes in mono- allelic cells enables more accurate epitope prediction. Immunity 46, 315–326 (2017). This study uses MS to identify MHC class I- binding peptides from single- HLA-expressing cell lines. Corresponding data were used to train epitope prediction models, which outperform the standard by 2-fold.

112. Bassani- Sternberg, M. et al. Deciphering HLA- I motifs across HLA peptidomes improves neo- antigen predictions and identifies allostery regulating HLA specificity. PLOS Comput. Biol. 13, e1005725 (2017).

113. Jurtz, V. et al. NetMHCpan-4.0: improved peptide– MHC class I interaction predictions integrating eluted ligand and peptide binding affinity data. J. Immunol. 199, 3360–3368 (2017).

114. Guillaume, P. et al. The C- terminal extension landscape of naturally presented HLA- I ligands. Proc. Natl Acad. Sci. USA 115, 5083–5088 (2018).

115. The problem with neoantigen prediction [Editorial]. Nat. Biotechnol. 35, 97 (2017).

116. Backert, L. & Kohlbacher, O. Immunoinformatics and epitope prediction in the age of genomic medicine. Genome Med. 7, 119 (2015).

117. Wan, Y. & Larson, D. R. Splicing heterogeneity: separating signal from noise. Genome Biol. 19, 86 (2018).

118. Matsuda, T. et al. Induction of neoantigen- specific cytotoxic T cells and construction of T cell receptor– engineered T cells for ovarian cancer. Clin. Cancer Res. 24, 5357–5367 (2018).

119. Li, G. et al. T cell antigen discovery via trogocytosis. Nat. Methods 16, 183–190 (2019). This work and that by Joglekar et al. (2019) are the first two studies to develop cell- based TCR ligand screening platforms.

120. Joglekar, A. V. et al. T cell antigen discovery via signaling and antigen- presenting bifunctional receptors. Nat. Methods 16, 191–198 (2019).

121. Gee, M. H. et al. Antigen identification for orphan T cell receptors expressed on tumor- infiltrating lymphocytes. Cell 172, 549–563 (2018).

122. Bentzen, A. K. et al. Large- scale detection of antigen- specific T cells using peptide–MHC- I multimers labeled with DNA barcodes. Nat. Biotechnol. 34, 1037–1045 (2016).

123. Zhang, S.-Q. et al. High- throughput determination of the antigen specificities of T cell receptors in single cells. Nat. Biotechnol. 36, 1156–1159 (2018).

124. Dijkstra, K. K. et al. Generation of tumor- reactive T cells by co- culture of peripheral blood lymphocytes and tumor organoids. Cell 174, 1586–1598 (2018).

125. Vitiello, A. & Zanetti, M. Neoantigen prediction and the need for validation. Nat. Biotechnol. 35, 815–817 (2017).

126. Jensen, P. E. Recent advances in antigen processing and presentation. Nat. Immunol. 8, 1041–1048 (2007).

127. Andersen, R. S. et al. High frequency of T cells specific for cryptic epitopes in melanoma patients. Oncoimmunology 2, e25374 (2013).

128. Robbins, P. F. et al. The intronic region of an incompletely spliced gp100 gene transcript encodes an epitope recognized by melanoma- reactive tumor- infiltrating lymphocytes. J. Immunol. 159, 303–308 (1997).

129. Lupetti, R. et al. Translation of a retained intron in tyrosinase- related protein (TRP) 2 mRNA generates a new cytotoxic T lymphocyte (CTL)-defined and shared human melanoma antigen not expressed in normal cells of the melanocytic lineage. J. Exp. Med. 188, 1005–1016 (1998).

130. Aarnoudse, C. A., Doel, P. B. van den, Heemskerk, B. & Schrier, P. I. Interleukin-2-induced, melanoma- specific T cells recognize camel, an unexpected translation product of LAGE-1. Int. J. Cancer 82, 442–448 (1999).

131. Slager, E. H. et al. CD4+ Th2 cell recognition of HLA- DR-restricted epitopes derived from CAMEL: a tumor antigen translated in an alternative open reading frame. J. Immunol. 170, 1490–1497 (2003).

132. Slager, E. H. et al. Identification of multiple HLA- DR-restricted epitopes of the tumor- associated antigen CAMEL by CD4+ Th1/Th2 lymphocytes. J. Immunol. 172, 5095–5102 (2004).

133. Vauchy, C. et al. CD20 alternative splicing isoform generates immunogenic CD4 helper T epitopes. Int. J. Cancer 137, 116–126 (2015). This study shows that an alternative splice variant of CD20 could give rise to HLA- DR1 binding epitopes and that vaccination with CD20-derived peptide was able to elicit epitope- specific CD4+ and CD8+ responses.

w w w.nature.com/nri

R e v i e w s

686 | N o v e m B e R 2 0 1 9 | v o l u m e 1 9

134. Volpe, G. et al. Alternative BCR/ABL splice variants in philadelphia chromosome- positive leukemias result in novel tumor- specific fusion proteins that may represent potential targets for immunotherapy approaches. Cancer Res. 67, 5300–5307 (2007).

135. Kobayashi, J. et al. Comparative study on the immunogenicity between an HLA- A24-restricted cytotoxic T cell epitope derived from survivin and that from its splice variant survivin-2B in oral cancer patients. J. Transl Med. 7, 1 (2009).

136. Wang, R.-F. et al. A breast and melanoma- shared tumor antigen: T cell responses to antigenic peptides translated from different open reading frames. J. Immunol. 161, 3596–3606 (1998).

137. Strønen, E. et al. Targeting of cancer neoantigens with donor- derived T cell receptor repertoires. Science 352, 1337–1341 (2016).

138. Bräunlein, E. & Krackhardt, A. M. Identification and characterization of neoantigens as well as respective immune responses in cancer patients. Front. Immunol. 8, 1702 (2017).

139. McGranahan, N. & Swanton, C. Clonal heterogeneity and tumor evolution: past, present, and the future. Cell 168, 613–628 (2017).

140. Watanabe, K., Kuramitsu, S., Posey, A. D. J. & June, C. H. Expanding the therapeutic window for CAR T cell therapy in solid tumors: the knowns and unknowns of CAR T cell biology. Front. Immunol. 9, 2486 (2018).

141. Marinov, G. K. et al. From single- cell to cell- pool transcriptomes: stochasticity in gene expression and RNA splicing. Genome Res. 24, 496–510 (2014).

142. Shalek, A. K. et al. Single- cell transcriptomics reveals bimodality in expression and splicing in immune cells. Nature 498, 236–240 (2013).

143. Yap, K. & Makeyev, E. V. Functional impact of splice isoform diversity in individual cells. Biochem. Soc. Trans. 44, 1079–1085 (2016).

144. Fry, T. J. et al. CD22-targeted CAR T cells induce remission in B- ALL that is naive or resistant to CD19-targeted CAR immunotherapy. Nat. Med. 24, 20–28 (2018).

145. O’Rourke, D. M. et al. A single dose of peripherally infused EGFRvIII- directed CAR T cells mediates antigen loss and induces adaptive resistance in patients with recurrent glioblastoma. Sci. Transl Med. 9, eaaa0984 (2017).

146. Sotillo, E. et al. Convergence of acquired mutations and alternative splicing of CD19 enables resistance to CART-19 immunotherapy. Cancer Discov. 5, 1282–1295 (2015). This study shows that resistance to CART-19 immunotherapy could be mediated by alternative splicing of CD19 compromising expression of the CART-19 epitope.

147. Hegde, M. et al. Combinational targeting offsets antigen escape and enhances effector functions of adoptively transferred T cells in glioblastoma. Mol. Ther. 21, 2087–2101 (2013).

148. Bethune, M. T. et al. Isolation and characterization of NY- ESO-1-specific T cell receptors restricted on various MHC molecules. Proc. Natl Acad. Sci. USA 115, E10702–E10711 (2018).

149. d’Urso, C. M. et al. Lack of HLA class I antigen expression by cultured melanoma cells FO-1 due to a defect in B2m gene expression. J. Clin. Invest. 87, 284–292 (1991).

150. Restifo, N. P. et al. Loss of functional beta2- microglobulin in metastatic melanomas from five patients receiving immunotherapy. J. Natl Cancer Inst. 88, 100–108 (1996).

151. Sharma, P., Hu- Lieskovan, S., Wargo, J. A. & Ribas, A. Primary, adaptive, and acquired resistance to cancer immunotherapy. Cell 168, 707–723 (2017).

152. Sade- Feldman, M. et al. Resistance to checkpoint blockade therapy through inactivation of antigen presentation. Nat. Commun. 8, 1136 (2017).

153. Gettinger, S. et al. Impaired HLA class I antigen processing and presentation as a mechanism of acquired resistance to immune checkpoint inhibitors in lung cancer. Cancer Discov. 7, 1420–1435 (2017).

154. Zaretsky, J. M. et al. Mutations associated with acquired resistance to PD-1 blockade in melanoma. N. Engl. J. Med. 375, 819–829 (2016).

155. Chang, C.-C., Campoli, M., Restifo, N. P., Wang, X. & Ferrone, S. Immune selection of hot- spot β2-microglobulin gene mutations, HLA- A2 allospecificity loss, and antigen- processing machinery component

down- regulation in melanoma cells derived from recurrent metastases following immunotherapy. J. Immunol. 174, 1462–1471 (2005).

156. Elkon, R., Ugalde, A. P. & Agami, R. Alternative cleavage and polyadenylation: extent, regulation and function. Nat. Rev. Genet. 14, 496–506 (2013).

157. Tian, B. & Manley, J. L. Alternative polyadenylation of mRNA precursors. Nat. Rev. Mol. Cell. Biol. 18, 18–30 (2017).

158. Singh, I. et al. Widespread intronic polyadenylation diversifies immune cell transcriptomes. Nat. Commun. 9, 1716 (2018).

159. Alt, F. W. et al. Synthesis of secreted and membrane- bound immunoglobulin mu heavy chains is directed by mRNAs that differ at their 3′ ends. Cell 20, 293–301 (1980).

160. Mayr, C. & Bartel, D. P. Widespread shortening of 3′ UTRs by alternative cleavage and polyadenylation activates oncogenes in cancer cells. Cell 138, 673–684 (2009).

161. Ni, T. K. & Kuperwasser, C. Premature polyadenylation of MAGI3 produces a dominantly- acting oncogene in human breast cancer. eLife 5, e14730 (2016).

162. Lee, S.-H. et al. Widespread intronic polyadenylation inactivates tumour suppressor genes in leukaemia. Nature 561, 127–131 (2018). This work reveals that intronic polyadenylation is widespread in leukaemia and is a common mechanism of tumour- suppressor inactivation.

163. Dubbury, S. J., Boutz, P. L. & Sharp, P. A. CDK12 regulates DNA repair genes by suppressing intronic polyadenylation. Nature 564, 141–145 (2018).

164. Lianoglou, S., Garg, V., Yang, J. L., Leslie, C. S. & Mayr, C. Ubiquitously transcribed genes use alternative polyadenylation to achieve tissue- specific expression. Genes Dev. 27, 2380–2396 (2013).

165. Bass, B. L. RNA editing by adenosine deaminases that act on RNA. Annu. Rev. Biochem. 71, 817–846 (2002).

166. Nishikura, K. Functions and regulation of RNA editing by ADAR deaminases. Annu. Rev. Biochem. 79, 321–349 (2010).

167. Speyer, J. F., Lengyel, P., Basilio, C. & Ochoa, S. Synthetic polynucleotides and the amino acid code. II. Proc. Natl Acad. Sci. USA 48, 63–68 (1962).

168. Han, L. et al. The genomic landscape and clinical relevance of A- to-I RNA editing in human cancers. Cancer Cell 28, 515–528 (2015).

169. Fumagalli, D. et al. Principles governing A- to-I RNA editing in the breast cancer transcriptome. Cell Rep. 13, 277–289 (2015).

170. Paz- Yaacov, N. et al. Elevated RNA editing activity is a major contributor to transcriptomic diversity in tumors. Cell Rep. 13, 267–276 (2015).

171. Chen, Y., Wang, H., Lin, W. & Shuai, P. ADAR1 overexpression is associated with cervical cancer progression and angiogenesis. Diagn. Pathol. 12, 12 (2017).

172. Zhang, M. et al. RNA editing derived epitopes function as cancer antigens to elicit immune responses. Nat. Commun. 9, 3919 (2018). This study shows that RNA editing- derived epitopes are immunogenic and can broaden the immunotherapy target space.

173. Yang, W. et al. Immunogenic neoantigens derived from gene fusions stimulate T cell responses. Nat. Med. 25, 767–775 (2019).

174. Bray, N. L., Pimentel, H., Melsted, P. & Pachter, L. Near- optimal probabilistic RNA- seq quantification. Nat. Biotechnol. 34, 525–527 (2016).

175. Pimentel, H., Bray, N. L., Puente, S., Melsted, P. & Pachter, L. Differential analysis of RNA- seq incorporating quantification uncertainty. Nat. Methods 14, 687–690 (2017).

176. Froussios, K., Mourão, K., Simpson, G. G., Barton, G. J. & Schurch, N. J. Identifying differential isoform abundance with RATs: a universal tool and a warning. Preprint at bioRxiv https://www.biorxiv.org/content/ 10.1101/132761v2 (2017).

177. Trincado, J. L. et al. SUPPA2: fast, accurate, and uncertainty- aware differential splicing analysis across multiple conditions. Genome Biol. 19, 40 (2018).

178. Nowicka, M. & Robinson, M. D. DRIMSeq: a Dirichlet- multinomial framework for multivariate count outcomes in genomics. F1000Res. 5, 1356 (2016).

179. Katz, Y., Wang, E. T., Airoldi, E. M. & Burge, C. B. Analysis and design of RNA sequencing experiments

for identifying isoform regulation. Nat. Methods 7, 1009–1015 (2010).

180. Shen, S. et al. rMATS: robust and flexible detection of differential alternative splicing from replicate RNA- Seq data. Proc. Natl Acad. Sci. USA 111, E5593–E5601 (2014).

181. Vaquero- Garcia, J. et al. A new view of transcriptome complexity and regulation through the lens of local splicing variations. eLife 5, e11752 (2016).

182. Li, Y. I. et al. Annotation- free quantification of RNA splicing using LeafCutter. Nat. Genet. 50, 151–158 (2018).

183. Kahles, A., Ong, C. S., Zhong, Y. & Rätsch, G. SplAdder: identification, quantification and testing of alternative splicing events from RNA- Seq data. Bioinformatics 32, 1840–1847 (2016).

184. Wang, Q. & Rio, D. C. JUM is a computational method for comprehensive annotation- free analysis of alternative pre- mRNA splicing patterns. Proc. Natl Acad. Sci. USA 115, E8181–E8190 (2018).

185. Sterne- Weiler, T., Weatheritt, R. J., Best, A. J., Ha, K. C. & Blencowe, B. J. Efficient and accurate quantitative profiling of alternative splicing patterns of any complexity on a laptop. Mol. Cell 72, 187–200 (2018).

186. Trapnell, C. et al. Differential gene and transcript expression analysis of RNA- seq experiments with TopHat and Cufflinks. Nat. Protoc. 7, 562–578 (2012).

187. O’Donnell, T. J. et al. MHCflurry: open- source class I MHC binding affinity prediction. Cell Syst. 7, 129–132 (2018).

188. Andreatta, M. & Nielsen, M. Gapped sequence alignment using artificial neural networks: application to the MHC class I system. Bioinformatics 32, 511–517 (2015).

189. Zhang, H., Lund, O. & Nielsen, M. The PickPocket method for predicting binding specificities for receptors based on receptor pocket similarities: application to MHC- peptide binding. Bioinformatics 25, 1293–1299 (2009).

190. Karosiene, E., Lundegaard, C., Lund, O. & Nielsen, M. NetMHCcons: a consensus method for the major histocompatibility complex class I predictions. Immunogenetics 64, 177–186 (2012).

191. Bhattacharya, R. et al. Evaluation of machine learning methods to predict peptide binding to MHC class I proteins. Preprint at bioRxiv https://www.biorxiv.org/ content/10.1101/154757v2 (2017).

192. Han, Y. & Kim, D. Deep convolutional neural networks for pan- specific peptide–MHC class I binding prediction. BMC Bioinformatics 18, 585 (2017).

193. Vang, Y. S. & Xie, X. HLA class I binding prediction via convolutional neural networks. Bioinformatics 33, 2658–2665 (2017).

194. Rasmussen, M. et al. Pan- specific prediction of peptide–MHC class I complex stability, a correlate of T cell immunogenicity. J. Immunol. 197, 1517–1524 (2016).

195. Jørgensen, K. W., Rasmussen, M., Buus, S. & Nielsen, M. NetMHCstab—predicting stability of peptide–MHC- I complexes; impacts for cytotoxic T lymphocyte epitope discovery. Immunology 141, 18–26 (2014).

196. Rammensee, H., Bachmann, J., Emmerich, N. P., Bachor, O. A. & Stevanovic, S. SYFPEITHI: database for MHC ligands and peptide motifs. Immunogenetics 50, 213–219 (1999).

Acknowledgements Preparation of this review was supported by an endowment provided by the Raymond and Beverly Sackler Foundation, the Parker Institute for Cancer Immunotherapy and the National Cancer Institute (grant 1U54 CA199090-01).

Author contributions The authors contributed equally to all aspects of the article.

Competing interests The authors declare no competing interests.

Peer review information Nature Reviews Immunology thanks Zlatko Trajanoski and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

N AT u R e R e v I e W S | I m m u N o Lo gy

R e v i e w s

v o l u m e 1 9 | N o v e m B e R 2 0 1 9 | 687