Description

Genome-scale neoantigen survey using ATLAS^TM in a non-small cell lung cancer patient identified neoantigens not predictable by algorithms

Lila Ghamsari¹, Emilio Flano¹, Biao Liu¹, Judy Jacques¹, Zheng Yan¹, Aula Alami¹, Christine Kelley¹, Johanna Kaufmann¹, Theresa Zhang¹, Pamela Carroll¹, Wendy Broom¹, Jonathan Havel², Vladimir Makarov², Taha Merghoub³, Jedd Wolchok⁴, Matthew Hellman⁴, Timothy Chan², Jessica Flechtner^1*

¹Genocea Biosciences, Cambridge, MA; ²MSKCC, New York, NY; ³Ludwig Collaborative Laboratory, MSKCC, New York, NY; ⁴Department of Medicine, MSKCC, New York, NY

*Corresponding author

Despite the unprecedented efficacy of immune checkpoint blockade (ICB) therapy in treating some cancers, many patients still fail to respond. Recent data indicate the combination of ICB therapy and neoantigen vaccines may prolong survival in some patients. Current approaches to neoantigen prioritization involve deep sequencing of tumors, followed by selection of epitopes based on MHC class I focused prediction algorithms. ATLAS^TM is a platform in which putative antigens are expressed as individual clones that can be processed by any subject’s antigen presenting cells and presented to autologous CD4⁺ or CD8⁺ T cells for measurement of recall responses. In this study, somatic mutations were identified from a tumor biopsy of a non-small cell lung cancer patient successfully treated with pembrolizumab. DNA fragments spanning each of 201 tumor-specific mutations were cloned and expressed; CD8⁺ and CD4⁺ T cell responses were measured pre- and post-therapy. Responses to 17 and 44 neoantigens were identified by CD8⁺ and CD4⁺ T cells, respectively; only two by both subsets. The number of CD4⁺ neoantigens increased upon treatment but the CD8⁺ neoantigen number did not change. In both subsets, we also noted “inhibitory” antigens, which reduced cytokine secretion below the negative control, a functionality that cannot be predicted by in silico tools. Half of the ATLAS-identified CD8⁺ neoantigens were not predicted by algorithms; only 6.5% of the predicted antigens were confirmed. Our data reinforce the need to develop T cell immunotherapies with biologically relevant neoantigens.