HERV-K Expression as a Biomarker for Predicting Immunotherapy Response

Background & Unmet Need

• Immune checkpoint blockade (ICB) therapies, such as anti-PD-1 and anti-CTLA antibodies, have demonstrated great success in a variety of cancers
• Despite eligibility for immune checkpoint blockade (ICB) therapy increasing over 30x, response rates continue to barely exceed 12%
• Biomarkers for predicting patient responsiveness currently in use, including PD-L1 expression, tumor mutational burden, and microsatellite instability fail consistently classify patients with high accuracy
• A variety of host factors influencing ICB efficacy have been proposed, including human endogenous retroviruses (HERVs), a type of transposable element contributing to oncogenesis
• Unmet Need: Additional biomarkers that enable more accurately prediction of patient responsiveness to ICB therapies

Technology Overview

• The Technology: HERV-K/HML2 provirus expression as a biomarker for predicting response to ICB therapy
• The Discovery: An extreme boosted gradient machine learning model (XGBOOST) was developed to identify bacterial and viral signatures associated with ICB response among RNA from 139 melanoma, NSCLC, RCC, and GBM tumor samples
• Support vector machine framework and recursive feature elimination were used to develop a separate model (SVM) based on DNA from 385 melanoma, NSCLC, and RCC tumor samples were used to train had available TME-derived DNA
• PoC Data: The models identified identified HERV-K/HML2 expression as a key factor for ICB response
• The XGBoost model identified 8 proviruses which classify participants by ICB response (AUC-ROC=0.801)

Technology Applications

• Method to identify patients who are likely to respond to ICB therapy in melanoma, NSCLC, RCC, and/or GBM
• Use as a biomarker for patient selection or stratification for ICB clinical trials

Technology Advantages

• Broad applicability across multiple solid tumor types
• HERV-K expression may provide an independent biomarker to complement existing biomarkers for ICB therapy, such as PD-L1 expression, microsatellite instability, and tumor mutational burden

Figure: Graphical abstract of XGBOOST and SVM model development