MHC class I-associated peptides derive from selective regions of the human genome

Abstract

MHC class I-associated peptides (MAPs) define the immune self for CD8+ T lymphocytes and are key targets of cancer immunosurveillance. Here, the goals of our work were to determine whether the entire set of protein-coding genes could generate MAPs and whether specific features influence the ability of discrete genes to generate MAPs. Using proteogenomics, we have identified 25,270 MAPs isolated from the B lymphocytes of 18 individuals who collectively expressed 27 high-frequency HLA-A,B allotypes. The entire MAP repertoire presented by these 27 allotypes covered only 10% of the exomic sequences expressed in B lymphocytes. Indeed, 41% of expressed protein-coding genes generated no MAPs, while 59% of genes generated up to 64 MAPs, often derived from adjacent regions and presented by different allotypes. We next identified several features of transcripts and proteins associated with efficient MAP production. From these data, we built a logistic regression model that predicts with good accuracy whether a gene generates MAPs. Our results show preferential selection of MAPs from a limited repertoire of proteins with distinctive features. The notion that the MHC class I immunopeptidome presents only a small fraction of the protein-coding genome for monitoring by the immune system has profound implications in autoimmunity and cancer immunology.

Figure 1. The immunopeptidome presented by 27 HLA allotypes.

(A) Total number of nonredundant MAPs and their source genes in the immunopeptidome of 18 B-LCLs compared with an expected binomial distribution. The curve depicts the expected number of source genes if all genes had a similar ability to generate MAPs. The black diamond shows the actual number of source genes (n = 6,195) observed for 25,270 MAPs (P < 1 × 10^-250, binomial test). (B) Histogram showing the number of MAPs generated per MAP source gene (range = 1–64). (C) The number of unique identifications of MAPs (left panel) and MAP source genes (right panel) was counted for various numbers of randomly selected HLA allotypes. Results show the average of 1,000 simulations. (D) The promiscuity of antigen presentation for MAPs (left panel) and their source genes (right panel). Histograms show the number of allotypes associated with each peptide or gene.

Figure 2. Spatial distribution of MAPs along source proteins.

(A) Distribution of overlap types for 3,682 pairs of overlapping MAPs formed by 5,046 individual peptides: pairs with any overlapping residues and no common ends; pairs with a common C terminus (C term); pairs with a common N terminus; and pairs with 1 peptide contained within the other. (B) Proportion of overlapping MAP pairs presented by the same allotype or different allotypes. For MAP pairs presented by different allotypes, whether the 2 allotypes belong to the same superfamily is indicated (34). (C) Distances between MAP start sites along proteins generating more than 1 MAP compared with a matched, random distribution. Distances are shown up to 30 residues. Distances are significantly shorter in the actual distribution (Wilcoxon rank sum test, P = 7 × 10^-52). (D) Exome coverage by the immunopeptidome. A window of 50 or 25 amino acids (left and right panel, respectively) was moved residue by residue along proteins of the transcribed exome of B-LCLs. Histograms show the number of MAPs found in each window; the proportion of windows containing 0 versus at least 1 MAP is indicated.

Figure 3. Features of MAP source and nonsource genes, transcripts, and proteins.

Error bars represent a 95% CI based on bootstrapping for Cliff’s d value, a nonparametric measurement of effect size. P values derived from 2-sided Wilcoxon tests; 6,195 source and 4,380 nonsource genes and gene products were studied for each comparison. * indicate features that were normalized for the respective transcript, UTR, or protein lengths. See Methods for details of how each feature was calculated. miR, microRNA; TS, TargetScan software; Ub, ubiquitination site.

Figure 4. GO analysis of source and nonsource genes.

Enrichment in source (A) and nonsource (B) groups was calculated on a background of both groups using the topGO algorithm to eliminate redundancies (60). The top 15 most enriched functions are shown for each group including all 3 ontology categories. For all GO terms significantly enriched in source and nonsource gene categories, see Supplemental Tables 3 and 4. PR, positive regulation; RNP, ribonucleoprotein; CC, cellular component; MF, molecular function; BP, biological process.

Figure 5. A logistic regression model to predict whether or not a gene will generate MAPs.

(A) Prediction scores for each gene grouped by experimentally defined source classification. (B) Prediction scores for each gene and the number of MAPs generated. (C) Model performance measured by a ROC plot of sensitivity (the rate of true positives) as a function of specificity (the rate of true negatives); the AUC is 0.81 ± 0.02 (95% CI). (D) Frequency of input variable selection in a logistic regression model using recursive feature elimination; frequencies above 0.05 are shown. (E) The relative weight of all input variables in the 2-class logistic regression model. Variables normalized by the length of the corresponding UTR, transcript, or protein are denoted by * and GO terms denoted by #. EC, extracellular; IC, intracellular; Mem., membrane; MFE, minimum free energy; MM, macromolecular; NR, negative regulation of; PR, positive regulation of; TS, TargetScan software. All metrics are averaged over 1,000 models (see Methods for details).

Figure 6. Evaluation of model performance with independent data sets on human cancer cell lines.

(A) Overlap in source gene identifications between the present study and 2 independent studies of JY B-LCLs using different MS techniques: JY (C.) and JY (B-S.). (B) Distribution of prediction scores for MAP source genes in B-LCLs and cancer cell lines (details in Table 1); median value is shown with whiskers extending to the extremes of the interquartile range x 1.5; outliers are hidden. (C) Proportion of MAP source genes captured as a function of prediction score threshold.