Alternative tumour-specific antigens

Abstract

The study of tumour-specific antigens (TSAs) as targets for antitumour therapies has accelerated within the past decade. The most commonly studied class of TSAs are those derived from non-synonymous single-nucleotide variants (SNVs), or SNV neoantigens. However, to increase the repertoire of available therapeutic TSA targets, 'alternative TSAs', defined here as high-specificity tumour antigens arising from non-SNV genomic sources, have recently been evaluated. Among these alternative TSAs are antigens derived from mutational frameshifts, splice variants, gene fusions, endogenous retroelements and other processes. Unlike the patient-specific nature of SNV neoantigens, some alternative TSAs may have the advantage of being widely shared by multiple tumours, allowing for universal, off-the-shelf therapies. In this Opinion article, we will outline the biology, available computational tools, preclinical and/or clinical studies and relevant cancers for each alternative TSA class, as well as discuss both current challenges preventing the therapeutic application of alternative TSAs and potential solutions to aid in their clinical translation.

Figure 1:. Summary of tumour-specific antigen production in the tumour cell.

Mutations and other tumour-specific nucleotide sequences (shown in red) can be observed at the genomic DNA level, where they undergo transcription (1) and splicing to form mRNA (2). Alternative splicing can occur at this step to form splice variant mRNA. Next, translation occurs on variant mRNA, resulting in production of variant proteins (3). Post-translational frameshifts (e.g. ribosomal slippage, among other mechanisms) can occur at this step, resulting in frameshifted protein variants. These proteins then can undergo proteasomal degradation (4) and transport to the endoplasmic reticulum (ER) to subsequently be loaded on major histocompatibility complexes (MHCs) (5). Other forms of post-translational frameshift can occur during these steps (e.g. protein splicing). Lastly, peptides containing variant sequences can be presented at the cell surface in the context of MHC, resulting in T-cell targetable tumour-specific antigens (6).

Figure 2:. Average tumour-specific antigen counts by cancer type.

Plots represent number of unique identified epitopes by The Cancer Genome Atlas (TCGA) cancer type. Insertion or deletion (INDEL)-neoantigen counts demonstrated significant correlation with single nucleotide variant (SNV)-neoantigens among all cancer types (coefficient: 0.81, p < 0.0001). Notable outliers in this correlation were kidney renal clear cell carcinoma (KIRC; commonly known as clear cell renal cell carcinoma (ccRCC)) and kidney renal papillary cell carcinoma (KIRP; commonly known as papillary RCC), where the INDEL-to-SNV ratio was significantly higher than other cancer types (ccRCC: 0.85 and papillary RCC: 0.90; all others: 0.43 – 0.72). Analysis of splice variant antigens demonstrated similar burden to INDEL-neoantigens, with significant correlation with INDEL-and SNV-neoantigen burden.A notable outlier is thyroid cancer (thyroid carcinoma (THCA)), where the average number of splice variant antigens per sample is higher than SNV-neoantigens. Mean burden of fusion-derived neoantigens was highest in sarcomas (sarcoma (SARC): 1.1, uterine carcinosarcoma (UCS): 0.78), with carcinoma fusion burden highest in breast (breast invasive carcinoma (BRCA); 0.70) and prostate (prostate adenocarcinoma (PRAD); 0.58) cancer. Testicular cancer (testicular germ cell tumours (TGCT)) contained substantially greater burden of human endogenous retrovirus (hERV)-derived tumour-specific antigens (TSAs) than any other TCGA cancer type. SNV and INDEL epitopes are derived from Thorsson et al.. Fusion epitopes are derived from Gao et al. (Cell Reports,2018). Splice variant epitopes are derived from Jayasinghe et al. (Cell Reports, 2018). Viral epitopes are derived from Selitsky et al. (mSystems, 2018). hERV epitopes are derived from differentially expressed hERVs (>10-fold tumour-vs-mean normal expression by DESeq2) in Smith et al. (JCI, 2018). All TSA classes represent the average number of predicted class I human leukocyte antigen (HLA) binders (8–11mers, < 500 nM) predicted from NetMHCPan. Stomach adenocarcinoma (STAD) INDEL and SNV calls were absent from Thorsson et al. and esophageal carcinoma (ESCA), acute myeloid leukaemia (LAML), and ovarian serous cystadenocarcinoma (OV) were not included in all original reports. Data shown represents reanalysis of the above reports, with modification of data in order to derive values comparable across TSA groups. ACC, adrenocortical carcinoma; BLCA, bladder urothelial carcinoma; CESC, cervical and endocervical cancers; CHOL, cholangiocarcinoma; COAD, colon adenocarcinoma; DLBC, lymphoid neoplasm diffuse large B-cell lymphoma; GBM, glioblastoma multiforme; HNSC, head and neck squamous cell carcinoma; KICH, kidney chromophobe; LGG, brain lower grade glioma; LIHC, liver hepatocellular carcinoma; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; MESO, mesothelioma; PAAD, pancreatic adenocarcinoma; PCPG, pheochromocytoma and paraganglioma; READ, rectum adenocarcinoma; SKCM, skin cutaneous melanoma; THYM, thymoma; UCEC, uterine corpus endometrial carcinoma; UVM, uveal melanoma. A version of these data with individual numbers of unique TSAs by cancer type is available online (Supplementary Fig. 1).

Figure 3:. Computational workflow for tumour-specific antigen calling.

a) Identification of tumour-specific antigens begins with variant calling. This can be done through comparison of tumour versus normal tissue DNA sequences (single nucleotide variants (SNVs) and insertions or deletions (INDELs)) or RNA sequences (splice variants, fusions, viral sequences and retroelements) to look for tumour-specific variants in the exome or tumour-specific transcripts in the transcriptome, respectively. b) Tumour human leukocyte antigen (HLA)-typing is performed to enable downstream major histocompatibility complex (MHC) binding prediction. c) Peptide enumeration occurs through translation of variant nucleotide sequences into their respective amino acid sequences, filtering for translation incompatible sequences such as those containing intervening stop codons or those with low evidence of RNA expression. These polypeptides are then used to derive 8–11 mer sequences (for MHC class I epitopes) or 15mer sequences (MHC class II epitopes) to allow for d) downstream MHC or HLA binding prediction of each sequence. Binders are typically defined in the literature as those with predicted binding affinity (K_d) of ≤ 500 nM or are selected from those with the highest rank percentile for predicted binding affinity. Other filtering criteria may be performed after this step, such as immunogenicity prediction or filtering away sequences with high homology to self-antigens. e) Lastly, therapies are generated using predicted tumour-specific antigens. These can be either DNA, RNA, or peptide vaccines or cellular therapies such as adoptive T-cell (ACT) therapy.