Advanced CT imaging features reflect distinct tissue immune profiles and exhibit high prognostic impact on NSCLC

Abstract

Background: Intersecting genetic, biologic and clinico-pathological features with highthroughput imaging may pave the way to precision oncology. We advanced the hypothesis that the tumor immune microenvironment (TIME) may imprint on CT scan parameters in a qualitative and quantitative (radiomics) fashion, providing a non-invasive approach to identify new prognostic factors in NSCLC patients. Methods: In this study, we enrolled sixty (31 Adenocarcinoma, 29 Squamous Cell Carcinoma) surgically resected patients. We defined TIME by the quantitative assessment of PD-L1 expression and a detailed morphometric evaluation of Tumor Infiltrating Lymphocytes (TILs). Next, from each tumor associated images we extrapolated 841 CT radiomic features through an open-source (3d Slicer) software. Results: We observed high levels of tissue PD-L1 in radiologic lesions displaying a solid texture and any effect on the surrounding parenchyma (p<0.05), while well defined CT margins were seen in TILs-rich cases (p<0.05). The combined analysis of predetermined risk factors from TIME and CT texture had a striking impact on clinical outcome. Patients with low PD-1 expression on CD8+ TILs and CT evidence of tumor effect on parenchyma had significantly increased (p<0.001) OS with respect to their counterpart (median 50 vs 30 months, HR=16.82). We also documented prolonged survival (p<0.05) in cases with well defined CT margins and high CD8-to-CD3 TILs (46 vs 35 months, HR=2.66). Intriguingly, when an unsupervised hierarchical clustering model was applied to radiomics data, we identified two clusters (A and B) with oppositely regulated features: the first of 57 cases (A), further branching into two continuous different clusters, the second (B) comprised only three patients sharing a mutual genetic (EGFR and KRAS mutations), immunologic (PD-L1, CD3+ and CD8+ TILs, PD-1/CD8 ratio), radiologic (shape, effect, texture and structure) and clinical (relapse and death) profile. Conclusions: A highly significant prognostic score can be obtained in NSCLC by integrating TIME with CT features. Distinct tissue immune backgrounds may entail imaging textures potentially able to portray a radiologic signature of lung cancer.