Abstract 1334: A diagnostic platform for precision cancer therapy enabling composite biomarkers by combining tumor and immune features from an enhanced exome and transcriptome

• Published in: Cancer research (Chicago, Ill.) Ill.), 2020-08, Vol.80 (16_Supplement), p.1334-1334
• Main Authors: Power, Robert Peter, Bartha, Gabor, Harris, Jason, Boyle, Sean M., Levy, Eric, Milani, Pamela, Tandon, Prateek, McNitt, Paul, Lee, Mandy, Morra, Massimo, Desai, Sejal, Salvidar, Sebastian, Clark, Michael J., Haudenschild, Christian, Jang, Sekwon, West, John, Chen, Richard
• Format: Article
• Language: English
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2020-1334

There is an increasing need for more advanced, composite biomarkers that can model the complex systems biology driving response and resistance to cancer therapy. However, many cancer diagnostic platforms to date, with their focus on mutational changes in a relatively small panel of genes, provide limited data to support integrative, multidimensional biomarkers that can better predict immunotherapy response. To enable the identification of composite biomarkers that combine tumor- and immune-related information from both DNA and RNA, we have developed ImmunoID NeXT, an enhanced exome/transcriptome-based diagnostic platform that can simultaneously profile the tumor and immune system from a single FFPE sample, across all of the approximately 20,000 genes. By co-optimizing assay and analytics design, we enable sensitive evaluation of clinically-relevant cancer biomarkers from >=25ng of co-extracted DNA/RNA, while also providing a broader evaluation of neoantigens, HLA typing and LOH, antigen processing machinery (APM), TCR/BCR repertoire, immune expression signatures, tumor-infiltrating lymphocytes (TILs), oncoviruses, and germline variants. Leveraging this expansive feature set, we developed methods that combine individual analytes to construct composite biomarker scores that correlate with immunotherapy response. Validation of ImmunoID NeXT demonstrated high sensitivity and specificity to somatic and structural variants across ~20,000 genes at allelic fractions as low as 5%, with clinical diagnostic reporting on actionable mutations (SNVs, indels, CNAs, fusions) in 248 cancer-driver genes that have been boosted further for higher sensitivity, as well as reporting on TMB and MSI status. For neoantigen prediction, immuno-peptidomic data from monoallelic HLA-transfected cell lines were used to train neural networks to predict pMHC binding with higher precision than public tools. For TCRα/β analysis in FFPE tumor samples, strong correlation with targeted TCR kit results was shown (R^2>0.9 and >0.94). For TILs, we developed signatures for eight immune cell types, demonstrating concordance with orthogonal immunofluorescence methods. We achieved genotyping accuracy of 99.1% for HLA Class I, and 95% for HLA Class II, and have developed and verified the performance of a tool for HLA LOH detection. In a cohort of 55 late-stage melanoma patients, the integration of neoantigen burden, HLA LOH, and APM mutational data formed a composite neoantigen score that more accurate