Abstract 1668: Longitudinal dynamics of circulating tumor DNA and plasma proteomics predict clinical outcomes to immunotherapy in non-small cell lung cancer

Abstract

Abstract Given the success of immunotherapy (IO), multiple IO-based options exist for advanced non-small cell lung cancer (NSCLC). Currently used biomarkers do not fully predict clinical outcomes and response assessment remains limited to radiological evaluation. Dynamic biomarkers that evaluate both tumor and host immune responses to IO are needed. We studied patients with NSCLC that received IO and chemo-IO to identify predictive and longitudinal biomarkers of clinical response using circulating tumor DNA (ctDNA) and plasma proteomic dynamics. We conducted deep targeted error-correction sequencing (TEC-Seq) of plasma cell-free DNA (cfDNA) and matched white blood cell (WBC) genomic DNA (gDNA) to identify ctDNA variants. We performed multiplexed antibody-based proximity extension assays of serial plasma samples to detect immune-related proteins. From separate training (n=31) and validation (n=29) cohorts, a total of 288 plasma cfDNA and 52 WBC gDNA specimens underwent TEC-Seq. A total of 260 variants were detected in plasma cfDNA and 78 variants in WBC gDNA. Almost a third of the plasma cfDNA variants (32%, n=61 of 188) were also found in WBC gDNA and filtered out. ctDNA variants were identified in 82% of patients (n=50). In the training cohort, longitudinal decreases in ctDNA variant levels were observed in patients with durable clinical benefit (DCB). Molecular response, defined as the loss of detectable ctDNA, was associated with longer progression-free (PFS; p=0.0004, log-rank) and overall survival (OS; p=0.017, log-rank). We incorporated on-treatment ctDNA dynamics including molecular response, recrudescence, and emergence of new variants into a logistic regression model to predict clinical benefit. This integrative model predicted DCB at a sensitivity of 84%, specificity of 76%, and with an area under the curve (AUC) of 0.90, better than baseline tumor PD-L1 expression alone (AUC 0.70, p=0.044, bootstrap method). In the validation cohort, molecular response was associated with longer PFS (p=0.00098, log-rank) and OS (p=0.0037, log-rank) and the integrative model predicted DCB at a sensitivity of 100%, specificity of 78%, and AUC of 0.89. In a subset (n=28 of 31) of the training cohort, plasma proteomics analysis revealed that increased baseline levels of IL15 and DCN were independently associated with DCB (p<0.05, Mann-Whitney), suggesting that pre-existing IL-15-mediated T-cell activation and DCN-mediated TGF-beta signaling may enable IO response. Sustained longitudinal increases in CCL17 were associated with DCB (p=0.037, Mann Whitney), which may reflect T-cell chemotaxis to promote on-treatment IO response. In summary, we validated longitudinal ctDNA alongside exploratory plasma proteomics dynamics to characterize on-treatment anti-tumor responses to IO, and enable prediction of clinical responses in NSCLC. Citation Format: Joseph C. Murray, Karlijn Hummelink, Lamia Rhymee, Alessandro Leal, Leonardo Ferreira, Mara Lanis, James R. White, Noushin Niknafs, Kristen Marrone, Jarushka Naidoo, Benjamin Levy, Samuel Rosner, Christine Hann, Josephine Feliciano, Vincent Lam, David Ettinger, Qing K. Li, Peter Illei, Kim Monkhorst, Hatim Husain, Julie R. Brahmer, Victor Velculescu, Patrick Forde, Robert B. Scharpf, Valsamo Anagnostou. Longitudinal dynamics of circulating tumor DNA and plasma proteomics predict clinical outcomes to immunotherapy in non-small cell lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 1668.