Methylated circulating tumor DNA (cfMeDIP) as a predictive biomarker of clinical outcome in pan-cancer patients (pts) treated with pembrolizumab (P).

2550 Background: Bespoke mutation-based circulating tumor DNA (ctDNA) predicts response to P but relies on availability of tumor tissue and presence of mutations. Cell-free methylated immunoprecipitation and high-throughput sequencing (cfMeDIP-seq) may overcome these limitations and be applied to more pts. Methods: Pts with mixed solid tumors divided into 5 cohorts received P 200 mg Q3wks in the investigator-initiated INSPIRE trial (NCT02644369). cfMeDIP-seq was performed at baseline (B), pre cycle 3 (C3) and later cycles. Methylation probability was inferred from read depths in 300 bp bins. cfMeDIP score was the probability-weighted sum of 3270 pan-cancer differentially-methylated regions in the TCGA PanCanAtlas. ctDNA concentration was assayed using tissue-informed bespoke targeted NGS (Signatera). ΔctDNA and ΔcfMeDIP denote the change in ctDNA or cfMeDIP between B and C3, respectively. Association with OS or PFS was assessed using Cox proportional hazards model, adjusting for cohort (aHR). Multivariable analysis (MVA) also included tumor mutation burden and PD-L1 status. Results: 194 plasma samples from 87 pts were analysed with cfMeDIP-seq (84 at B, 55 at C3, 55 at later cycles). Demographics: male 33%; median age = 61 yrs (34–82); Cohorts: triple negative breast (26%), ovarian (25%), head & neck (21%), melanoma (12%), others (15%). Median follow-up = 10.6m (0.6–64.4); Median PFS = 1.9m; Median OS = 10.6m. cfMeDIP at B below median was associated with better OS (aHR = 0.51, 95%CI 0.29-0.91; p = 0.02) in MVA. ΔcfMeDIP was evaluable in 53 pts; any decrease in ΔcfMeDIP was predictive for OS (aHR = 0.36, 95%CI 0.18-0.72; p < 0.01) and PFS (aHR = 0.42, 95%CI 0.22-0.82; p = 0.01). Both ΔctDNA and ΔcfMeDIP were evaluable in 51 pts; decrease in ΔctDNA and ΔcfMeDIP predicted for longer OS (aHR = 0.45, 95%CI 0.23-0.86; p = 0.02 vs aHR = 0.39, 95%CI 0.19-0.80; p = 0.01); and PFS (aHR = 0.44, 95%CI 0.23-0.83; p = 0.01 vs aHR = 0.5, 95%CI 0.25-0.99; p = 0.04), respectively. When both ΔctDNA and ΔcfMeDIP are integrated in MVA, ΔcfMeDIP was predictive for OS (aHR = 0.48, 95%CI 0.23-1; p = 0.05). A decrease in ΔcfMeDIP and/or ΔctDNA was associated with longer OS (aHR = 0.2, 95%CI 0.09-0.45) and PFS (aHR = 0.27, 95%CI 0.13-0.58) compared to an increase in both assays (p < 0.01) (Table). Conclusions: We applied for the first time cfMeDIP-seq and mutation-based ctDNA analysis concurrently in pan-cancer pts treated with checkpoint blockade. ΔcfMeDIP correlated strongly with OS and PFS, representing a promising plasma-based predictive epigenetic biomarker in pts treated with P. ΔctDNA and ΔcfMeDIP can complement each other to predict outcomes, demonstrating that they may capture different biological changes. Clinical trial information: NCT02644369. [Table: see text]