Purpose: Next-generation sequencing has opened the door to precision cancer therapies targeting mutations expressed by tumor cells. However, most neo-epitopes selected by traditional T cell epitope prediction algorithms prove to be non-immunogenic. Poor predictive performance may partially be due to inclusion of mutated epitopes cross-conserved with self-epitopes recognized by the T cell receptor of regulatory (Treg), anergic or deleted T cells. Vaccination with self-epitopes can lead to weak effector responses, active immune suppression, and toxicity due to immune-mediated adverse effects. In addition, most cancer vaccine studies specifically focus on the selection of CD8 neo-epitopes due to an apparent lack of robust and accurate CD4 epitope prediction tools.
Methods: We have developed Ancer, an advanced cancer T cell epitope identification and characterization tool, that streamlines the selection of both CD4 and CD8 T cell neo-epitopes. Ancer leverages EpiMatrix and JanusMatrix, state-of-the-art predictive algorithms that have been extensively validated in prospective vaccine studies for infectious diseases [Moise et al., Hum. Vaccines Immunother 2015; Wada et al., Sci. Rep. 2017]. Distinctive features of Ancer are its ability to accurately predict Class II HLA ligands, or CD4 epitopes, with EpiMatrix and its 82% positive predictive value, as estimated in previous prospective studies.
Furthermore, the application of JanusMatrix, a tool for identifying tolerated or Treg epitopes, allows for the prioritization of neo-epitopes with reduced potential for Treg induction, which could be detrimental to cancer vaccine efficacy. Screening candidate sequences with JanusMatrix also enables to the removal of neo-epitopes that may trigger off-target events, which have in some cases abruptly halted the development of promising cancer therapies.
Results: Here, we performed two retrospective analyses to confirm the accuracy of Ancer. First, we validated Ancer’s predictive accuracy using datasets of HLA-Class-I-bound peptides detected by mass spectrometry, which are independent of training sequence data used in model development. Analysis of sequences published in Abelin et al., Immunity 2017 showed that the majority of eluted peptides were specifically identified by EpiMatrix as strong (top 1%) HLA ligands for common HLA class I alleles, whereas this was not true for predictions using various versions of NetMHCpan (Figure 1). In addition, 95% of eluted peptides were correctly predicted by EpiMatrix, while only 88% of these sequences were accurately predicted by NetMHCpan. These results suggest that higher quality candidate targets are retrieved by Ancer, as compared to other in silico tools, which is of critical importance for patients with low mutational burden
Second, we performed a retrospective analysis of a cancer immunogenicity study [Strønen et al., Science 2016] where HLA A2-restricted neo-epitopes were validated in T cell assays. In our analysis of the same data, we selected top 1% epitopes that were not conserved with self, using EpiMatrix and JanusMatrix. Ancer classified HLA Class I neo-epitopes with 72% accuracy, as compared to 21% accuracy when using public in silico prediction tools or 65% accuracy when using lengthy and costly in vitro characterization techniques (Figure 2).
Conclusion: These two sets of results demonstrate that Ancer may focus neo-epitope candidate selection on higher value sequences than conventional algorithms. CD4 and CD8 neo-epitopes with low Treg activation potential may then be used to support the development of safer and more effective personalized cancer vaccines. Future steps include the design of prospective studies to test the efficacy of Ancer-derived vaccines in the CT26 and GL261 murine cancer models.
Guilhem Richard– Computational Immunologist, EpiVax, Inc., Rhode Island
Matthew Ardito– Principal Application Developer, EpiVax, Inc.
Lenny Moise– Scientific Director, Vaccine Research, EpiVax, Inc., Providence, Rhode Island
Frances Terry– Director of Analysis, EpiVax Inc, Providence, Rhode Island
Gad Berdugo– CEO, EpiVax Oncology, Inc.
William Martin– CIO & COO, EpiVax, Inc., Providence, Rhode Island
Anne De Groot– CEO, EpiVax, Inc., Providence, Rhode Island
Jane Healey– EpiVax, Inc., Providence