Incorporating long non-coding RNA (lncRNA) into genome-wide biomarker screening for prognostic gene signatures of immunotherapy outcomes

2617 Background: Emerging data have shown that long non-coding RNA (lncRNAs) have the potential to become biomarkers in cancers with potential roles in regulating cell proliferation, apoptosis, migration, invasion and maintenance of stemness during cancer development. In this research, we aimed for a genome-wide biomarker screening approach incorporating lncRNA in order to identify key molecules and pathways involved in ICI response and resistance. Methods: We utilized clinical and transcriptomic data collected under the Total Cancer Care Protocol (NCT03977402) and Avatar project within the Oncology Research Information Exchange Network (ORIEN). We analyzed RNA-seq data from a total of 875 ICI treated pts. A novel resampling-based technique with elastic net regression was utilized for genome-wide biomarker screening, in which Cox regression was used for survival outcomes and logistic regression for a 2-year-based dichotomized outcome. All search was done through 200 resampling steps, each consisting of randomly selecting 80% of pts and 80% of all gene signatures. Moreover, we searched the genes that are most complementary to the established immune cell scores (i.e., Ecotyper and xCell) by incorporating these scores into the regression adjustment. Finally, we applied the same approach to identify prognostic long non-coding RNA (lncRNA) signatures. All analyses were carried out on all ICI pts, as well as on subgroups stratified by cancer type. Results: Of 875 pts treated with ICI, 125 had melanoma, 143 had kidney cancer, 126 had non-small lung cancer, and 110 had head and neck cancer. The genome-wide selection method successfully identified a panel of top prognostic genes and pathways for all cancer types included. In melanoma, the top five prognostic genes were ZDHHC2, CBX4, MET, NTN4, and HLA-DQA1, with 80%, 78%, 77.5%, 77.5%, 76.5% resampling-based selection probability respectively; the top five lncRNAs were AC073655.3, AC116348.3, PITPNA-AS1, AC011447.3, and AL021978.1. When adjusting for the immune scores in the selection model, the top five prognostic protein-coding genes in melanoma were TENM3, POLR2J3, SCCPDH, MET and SIAH1, and the top five lncRNAs were AC007620.2, PITPNA-AS1, HCG11, SLC25A5-AS1, and LINC02258, respectively. Conclusions: Our analysis has successfully identified gene panels of mRNA and lncRNA molecules with potential roles related to ICI mechanisms of action. Ongoing analyses are further investigating the potential mechanisms and pathways linking these molecules and will be presented at the meeting. [Table: see text]