MHC-I-presented non-canonical antigens expand the cancer immunotherapy targets in acute myeloid leukemia

Abstract

Identification of tumor neoantigens is indispensable for the development of cancer immunotherapies. However, we are still lacking knowledge about the potential neoantigens derived from sequences outside protein-coding regions. Here, we comprehensively characterized the immunopeptidome landscape by integrating multi-omics data in acute myeloid leukemia (AML). Both canonical and non-canonical MHC-associated peptides (MAPs) in AML were identified. We found that the quality and characteristics of ncMAPs are comparable or superior to cMAPs, suggesting ncMAPs are indispensable sources for tumor neoantigens. We further proposed a computational framework to prioritize the neoantigens by integrating additional transcriptome and immunopeptidome in normal tissues. Notably, 6 of prioritized 13 neoantigens were derived from ncMAPs. The expressions of corresponding source genes are highly related to infiltrations of immune cells. Finally, a risk model was developed, which exhibited good performance for clinical prognosis in AML. Our findings expand potential cancer immunotherapy targets and provide in-depth insights into AML treatment, laying a new foundation for precision therapies in AML.

Fig. 1

Comparison of different characteristics between ncMAPs and cMAPs. (A) The workflow for identification of cMAPs and ncMAPs. (B) Difference of main characteristics between ncMAPs and cMAPs which was calculated as the median of cMAPs minus the median of ncMAPs. Red points indicate difference in favor of ncMAPs, blue points indicate in favor of cMAPs, and grey indicates no difference. (C,D) Boxplot showing the difference of PEP (posterior error probability) and T-cell contact hydrophobicity between ncMAPs and cMAPs. The PEP is a statistical measure that estimates the probability of a peptide being wrongly identified in proteomics analysis with MaxQuant. The Wilcoxon test was utilized for statistical analysis. (E) Pearson correlations between observed retention times and predicted retention time. (F) The length distribution of source ORFs. (G) The percentage distribution of the number of MAPs identified across different numbers of samples. The x-axis represents how many samples MAPs will be detected in, while the y-axis displays the percentage of MAPs appearing in n samples (n = 1, 2,…, 18).

Fig. 2

Comparison of source proteins encoding ncMAPs and cMAPs. (A) Diverse non-canonical ORFs encoding ncMAPs. (B) Compared with reference library, the tendency of peptides MS-identified. (C,D) The expression levels of source genes encoding MAPs. (E–G) The expressions in TCGA samples, the length of source ORFs and the ratio of the length covered with non-coding derived ncMAPs, with protein-coding derived ncMAPs, with protein-coding derived cMAPs, or with protein-coding derived MAPs.

Fig. 3

The distribution of MAPs across patients in AML. (A) The top barplot shows the proportion of MAPs presented by different alleles in the samples. The left barplot shows the proportion of MAPs shared by different samples. The middle heatmap shows the percentage of allele-specific presented MAPs to total MAPs in the patient. The right barplot shows the number of MAPs by HLA alleles. (B) The pairwise similarity between samples. (C) The boxplot shows the distribution of Simpson coefficient with shared 0, 1, 2 or > 2 alleles.

Fig. 4

Alleles Similarity and identified motifs. (A) Pairwise correlations between alleles, based on a vector of frequencies of the 20 amino acids at every peptide position(left), and the Simpson coefficient (right). (B) 2D-visualization of sub-motifs identified across alleles (left), colored according to HLA locus and scaled in size according to the number of underlying peptides making up the sub-cluster. (C) Number of alleles sharing a sub-motif colored according to HLA locus.

Fig. 5

Accurately identify tumor-specific antigens by integrative multi-level omics profiling. (A) Tumor specific antigen identification pipeline at the transcription level. (B) The number of identified antigens is shown according to ncMAP and cMAP. (C) Antigens are distributed in normal tissues and AML, and the multi-omics screened antigens were highlighted in red and bold. (D) Multi-omics screening antigen source gene expression levels are displayed by heatmap. Red is the example shown next. (E) ILPSYQLFL combined with HLA-A02:01 schema diagram. (F) The Scatter plot of neoantigens, x-axis is PEP, y-axis is Andromeda score. (G) ILPSYQLFL mass spectra.

Fig. 6

Neoantigen expressions correlated with immune evasion. (A) Correlation of expression values between genes, and Correlation between gene expression values and various indexes. ImmuneScore, StromaScore and MicroenvironmentScore were calculated by xCell. (B) The correlation between the infiltration proportion of immune cells predicted by xCell and gene expression values in AML. (C) Correlation between gene expression values and immune cell marker gene expression values. (D) ROC curve for predicting OS in AML patients by risk scores. (E) Gene expression levels are displayed according to low, middle and high groups, and the Count value represents the number of genes expressed in each sample. (F) Kaplan-Meier survival analysis between risk groups.