Abstract 3095: Combination of spatial protein data with bulk transcriptional profiling of the same cohort shows relationships between RNA and protein and facilitates combined predictive signatures

Background: Although there are several mRNA signatures with some ability to predict outcomes in melanoma patients treated with immunotherapy, more robust predictive models are needed to optimize therapeutic selection. Here, we integrated spatially-resolved protein information acquired by the NanoString® GeoMx® Digital Spatial Profiler (DSP) with information from bulk mRNA gene expression acquired using NanoString® nCounter® PanCancer IO 360™ panel on the same cohort of immunotherapy treated melanoma patients to create predictive models associated with clinical outcomes. Methods: We assessed pretreatment tumor samples from 59 immunotherapy treated melanoma patients. RNA for gene expression was extracted from formalin-fixed paraffin-embedded whole tissue sections, then hybridized to the 770-plex PanCancer IO 360™ panel and measured on the nCounter platform. The same cases were represented in a tissue microarray, where 32 protein targets were quantified in three different compartments (tumor [s100+], leukocyte [CD45+] and macrophage [CD68+]) using NanoString9s GeoMx DSP platform (Toki et al, CCR 2018). The combined dataset of IO 360 gene expression and GeoMx DSP protein data was compared to patient overall survival (OS), objective response (OR), or clinical benefit (CB) using univariate cox or logistic regression models, as well as receiver operator characteristic (ROC) analysis. For each outcome, we used either Lasso (L1 Regularization) or Elastic Net (L0.5 Regularization) approaches to select analytes that were initially identified as nominally significant univariate analysis (P Results: Hierarchical clustering of both data sets showed that the DSP data generally clustered away from bulk RNA profiling data. PD-L1 mRNA showed weak correlation with spatially-derived protein (R2=0.04 in the s100+, 0.11 in the CD68+ and 0.12 in the CD45+ compartment). IDO1 mRNA was moderately correlated with protein (R2=0.39 in the s100+, 0.21 in the CD68+ and 0.23 in the CD45+ compartment). A total of 228 variables (including 191 mRNA and 37 protein) that were statistically significant in univariate analysis were identified. Models were constructed to predict OR, OS and CB with individual and mixed modality markers with AUCs above 0.7. Conclusions: This work supports the integration of spatially-derived protein data with bulk mRNA gene expression data for the construction of predictive models for melanoma patients receiving immune checkpoint inhibition and lays the groundwork for validation in an independent dataset. Citation Format: Ioannis A. Vathiotis, Jason Reeves, Maria Toki, Pok Fai Wong, Harriet Kluger, Thazin Nwe Aung, Konstantinos N. Syrigos, Sarah Warren, David L. Rimm. Combination of spatial protein data with bulk transcriptional profiling of the same cohort shows relationships between RNA and protein and facilitates combined predictive signatures [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3095.