The neoepitope landscape of breast cancer: implications for immunotherapy

Abstract

Background: Cancer immunotherapy with immune checkpoint blockade (CKB) is now standard of care for multiple cancers. The clinical response to CKB is associated with T cell immunity targeting cancer-induced mutations that generate novel HLA-binding epitopes (neoepitopes).

Methods: Here, we developed a rapid bioinformatics pipeline and filtering strategy, EpitopeHunter, to identify and prioritize clinically relevant neoepitopes from the landscape of somatic mutations. We used the pipeline to determine the frequency of neoepitopes from the TCGA dataset of invasive breast cancers. We predicted HLA class I-binding neoepitopes for 870 breast cancer samples and filtered the neoepitopes based on tumor transcript abundance.

Results: We found that the total mutational burden (TMB) was highest for triple-negative breast cancer, TNBC, (median = 63 mutations, range: 2-765); followed by HER-2(+) (median = 39 mutations, range: 1-1206); and lowest for ER/PR(+)HER-2(-) (median = 32 mutations, range: 1-2860). 40% of the nonsynonymous mutations led to the generation of predicted neoepitopes. The neoepitope load (NEL) is highly correlated with the mutational burden (R² = 0.86).

Conclusions: Only half (51%) of the predicted neoepitopes are expressed at the RNA level (FPKM≥2), indicating the importance of assessing whether neoepitopes are transcribed. However, of all patients, 93% have at least one expressed predicted neoepitope, indicating that most breast cancer patients have the potential for neo-epitope targeted immunotherapy.

Keywords: Breast cancer; Epitopes; Immunotherapy; Mutation burden; Neoepitope prediction; TNBC.

Fig. 1

EpitopeHunter: Pipeline to identify clinically relevant neoepitopes. Proposed pipeline to generate and identify clinically relevant neoepitopes from the landscape of somatic mutations from tumor and normal exome sequencing data

Fig. 2

Mutational burden, potential neoepitopes, and expressed neoepitopes in breast cancer determined using Epitopehunter. (a) The mutational load is highest in triple negative breast cancer (TNBC), followed by the HER-2(+) breast cancer subtype, and least for the ER/PR(+)HER-2(−) subtype. The median and range of non-synoymous mutations per cancer type are: 32 (1–2860) in ER/PR(+)HER-2(−), 39 (1–1206) in HER-2(+) and 63 (2–765) in TNBC. The number of samples in each subtype are: 630 (ER/PR(+)HER-2(−)), 141 (HER-2(+)), 99 (TNBC). (b) The range of potential binding neoepitopes (IEDB score ≤ 500 nM) is highest for the TNBC subtype, followed by HER-2(+); and lowest for the ER/PR(+)HER-2(−) subtype. The median and range of high affinity binding neoepitopes are as follows: 10 (0–864) in ER/PR(+)HER-2(−), 15 (0–717) in HER-2(+) and 26 (0–237) in TNBC. The number of samples in each subtype are: 586 (ER/PR(+)HER-2(−)), 138 (HER-2(+)), 93 (TNBC). (c) The median and range of predicted neoepitope with expression (FPKM ≥5) across breast cancer subtypes are: 3 (0–230) in ER/PR(+)HER-2(−), 4 (0–226) in HER-2(+) and 8(0–82) in TNBC. The number of samples in each case are: 583(ER/PR(+)HER-2(−)), 138(HER-2(+)), 92(TNBC). Significant differences between subtypes of cancer are computed pairwise for each breast cancer subtype using a Wilcox rank sum test, *** P < 0.001 ** P < 0.01

Fig. 3

Correlations between predicted neoepitopes, mutational burden, and expressed neoepitopes. (a) The number of predicted binding neoepitopes (IEDB score ≤ 500 nM) is highly correlated (R² = 0.86, p < 0.0001) with the number of nonsynonymous mutations across all breast cancers. (b) The number of predicted binding neoepitopes (IEDB score ≤ 500 nM) is highly correlated (R² = 0.94, p < 0.0001) with the number of expressed neoepitopes (FPKM≥5) in all breast cancers. A fitted line from a linear regression is shown in red, with 95% CI levels shown in the grey shaded areas

Fig. 4

Expression analysis for the high affinity neoepitopes (FPKM ≥5). (a) The number of the expressed neoepitopes (normalized by total number of samples in each breast cancer subtype) is shown for each FPKM range. 35% (6098/17518) of the neoepitopes are expressed with an FPKM threshold of ≥5. (b) The number of neoepitopes (normalized by the number of samples in each breast cancer subtype) with the highest expressed neoepitope for each patient is shown. 87% (709/815) of patients have at least one potential binding epitope. < 1 includes neoepitopes with expression less than 1.0 FPKM (not including 1.0 FPKM), 1–2 includes neoepitopes with expression equal to or greater than one and less than 2 FPKM, and so on, for all categories