Microproteins encoded by noncanonical ORFs are a major source of tumor-specific antigens in a liver cancer patient meta-cohort

Abstract

The expression of tumor-specific antigens during cancer progression can trigger an immune response against the tumor. Here, we investigate if microproteins encoded by noncanonical open reading frames (ncORFs) are a relevant source of tumor-specific antigens. We analyze RNA sequencing data from 117 hepatocellular carcinoma (HCC) tumors and matched healthy tissue together with ribosome profiling and immunopeptidomics data. Combining human leukocyte antigen-epitope binding predictions and experimental validation experiments, we conclude that around 40% of the tumor-specific antigens in HCC are likely to be derived from ncORFs, including two peptides that can trigger an immune response in humanized mice. We identify a subset of 33 tumor-specific long noncoding RNAs expressing novel cancer antigens shared by more than 10% of the HCC samples analyzed, which, when combined, cover a large proportion of the patients. The results of the study open avenues for extending the range of anticancer vaccines.

Fig. 1.. HCC transcriptome.

(A) Datasets used. They comprise four cohorts with matched tumor-adjacent tissue RNA-Seq (HCC1 to HCC3 and TCGA) and one cohort with matched tumor-adjacent tissue Ribo-Seq data (HCC4). (B) Biomarkers of HCC in the four datasets. Gene expression was measured as FPKM, in both tumor and adjacent tissue samples (normal). By paired Wilcoxon signed-rank test, we confirmed that the expression of MT1M, TERT, and THBS4 coding genes was significantly different between tumor samples and adjacent tissue following the expected trends (MT1M P value = 1.596 × 10⁻¹⁹; TERT P value = 1.578 × 10⁻¹⁹; THBS4 P value = 9.941 × 10⁻²⁰). (C) Main steps of the computational and experimental pipeline. We gathered RNA-Seq and Ribo-Seq data from matched tumor/normal samples. We quantified gene expression and reconstructed nonannotated transcripts. We then determined the tumor-specific transcriptome in each patient. We predicted the translation rate of lncRNAs and novel transcripts using the HCC Ribo-Seq data. We quantified tumor-specific antigens derived from ncORFs versus other sources and performed experiments to validate HLA-binding and immunogenicity. (D) Distribution of gene expression levels for different types of transcripts. lncRNAs and novel transcripts tended to be expressed at lower levels than protein-coding genes, although there was a considerable overlap in expression levels between the classes. The line at 1 FPKM indicates the expression cutoff used to consider a transcript as expressed. Data shown are for dataset HCC3. (E) Number of exons in different types of transcripts. lncRNAs and novel transcripts tended to have a lower number of exons than coding genes; the data shown are for the HCC3 dataset. (F) Relative abundance of different types of transcripts in tumors. Coding genes were the largest class of expressed transcripts, followed by lncRNAs and novel transcripts.

Fig. 2.. Translation of ncORFs in tumor-expressed lncRNAs.

(A) Prediction of translated ORFS using Ribo-Seq data. From the total predicted noncanonical ORFs, we analyzed translation patterns in ncORFs with at least five mapped Ribo-Seq reads, selecting those that had a RibORF score of at least 0.5. (B) Comparison of lncRNAs containing ncORFs with signatures of translation from different cohorts. The intersection between the sets of translated lncRNAs in the four different transcriptomics cohorts is shown. Translated lncRNAs (124 of a total of 251) were shared across all cohorts. (C) Comparison of ncORFs with signatures of translation from different cohorts. The intersection between the sets of translated ncORFs shown in the different cohorts is shown. Translated ncORFs (524 of a total of 909) were shared across all cohorts. (D) Many lncRNAs contain several putatively translated ncORFs. The graph shows the distribution of the number of translated ncORFs per transcript. From a total of 251 lncRNAs, 79 translated one single ncORF and 172 translated more than one ncORF. (E) ncORFs are significantly smaller than canonical coding sequences. Comparison of the ORF length distribution of micropeptides encoded by ncORFs versus canonical ORFs, with median values of 39 and 456 amino acids (aa), respectively. Differences are significant at a P value of <2.2 × 10⁻¹⁶ (Kolmogorov-Smirnoff test). (F) Frequency of different start codons in canonical coding sequences and ncORFs. ATG as well as ACG, CTG, TTG, and GTG were considered as putative start codons. (G) Translation of ZNF674-AS1. Coverage of RNA-Seq and Ribo-Seq reads and putatively translated ORFs are indicated. The second exon of the mRNA transcript is shortened for visualization purposes. No Ribo-Seq signal was detected in the region not shown. (H) Translation of LINC01419. Coverage of RNA-Seq and Ribo-Seq reads and putatively translated ORFs are indicated.

Fig. 3.. Most tumor-specific transcripts are noncoding.

(A) lncRNAs and novel transcripts tend to be more tumor-specific than coding genes. The number of different types of transcripts per patient and cohort is shown. (B) Tumor-specific versus normal-specific gene expression. By paired Wilcoxon signed-rank test, we confirmed that the tumors are enriched in noncanonical genes with respect to coding ones (HCC1 P value = 4.883 × 10⁻⁰³; HCC2 P value = 7.773 × 10⁻¹⁰; HCC3 P value = 7.813 × 10⁻⁰³; TCGA P value = 2.469 × 10⁻⁰⁹). (C) Shared tumor-specific transcripts. Despite the privacy of most tumor-specific transcripts, a subset is found in several patients. (D) Expression in testis. Proportion of tumor-specific transcripts that are also expressed in testis for different transcript types and datasets. (E) Proportion of lncRNA and novel transcripts overlapping HERVs. Differences between the fraction of lncRNAs overlapping HERVs in the complete transcriptome and in the tumor-specific transcriptome. Differences are significant in all cohorts except for lncRNA-HCC1 (P value < 0.05, Fisher’s exact test). Statistical significance is indicated as follows: ***P < 0.001, ** P < 0.01, *P < 0.05.

Fig. 4.. ncORFs make a substantial contribution to the HCC antigen landscape.

(A) Proportion of shared or private predicted antigens. Antigens derived from mutations are almost all patient-specific, whereas antigens derived from tumor-specific transcripts can be shared across patients. (B) Predicted number of antigens per patient and dataset. Antigen load was predicted by selecting peptides with HLA-binding affinity IC₅₀ < 50 nM as predicted by NetMHCpan, using patient-specific HLA allele information. For lncRNAs and novel transcripts, it was then corrected by the translation index, which is the fraction of ncORF estimated to be translated by the analysis of Ribo-Seq data. (C) HLA-A*02.01 binding assays for ncORFs. Binding affinity expressed as FI ± SEM for each peptide in an in vitro T2 cell binding assay. The FI value shown corresponds to the mean of two different assays (with two replicates each). A line at FI = 1 indicates the expectation under no binding. * indicates P value of <0.05 when comparing the values with the peptide (Wilcoxon-Mann-Whitney test). Information on the transcripts/ncORFs can be found in tables S16 to S18. CONTROL refers to a positive control (peptide 58-66 from influenza matrix protein). (D) IFN-γ ELISPOT assays. The spleens of mice immunized with four peptides were processed to measure the number of IFN-γ secreting cells (IFNγSFC). In the case of the peptide WMSLDWELYV, the measurement was >1000 IFNγSFC per 8 × 10⁵ cells in all four replicates of the experiment. Of the four peptides tested, two yielded highly significant results, WMSLDWELYV and GLFHIYHKI (***P value < 0.001, t test), and the other two were not significant.

Fig. 5.. Tumor-specific transcripts shared by more than 10% of the patients.

Only genes that were tumor-specific in >10% of the HCC tumor samples, expressed at more than 5 FPKM in at least one sample and expressed in less than 1% of the normal liver samples (FPKM cutoff = 1), were considered. The intensity of the color in the cell reflects the level of expression (minimum of 1 FPKM). (A) Protein-coding transcripts. Number of transcripts: 14. The transcripts tend to cluster in the leftmost group of patients. (B) lncRNAs. Number of transcripts: 33. The transcripts are scattered across different patients. (C) Overlap with HERVs for this set of lncRNAs. Overlap is based on genomic coordinates. (D) Proportion of lncRNAs with detected translated ncORFs. Prediction of ncORF translation was performed using Ribo-Seq data from a different HCC cohort of 10 patients (HCC4). (E) Proportion of lncRNAs containing ncORFs with immunopeptidomics evidence in different cancer datasets. (F) List of lncRNAs and peptides with immunopeptidomics evidence. The source of the data is indicated.