211 - Prognostic Implications of Cerebrospinal Fluid Circulating Tumor DNA (CSF ctDNA) Following Radiation Therapy in Patients with Leptomeningeal Metastasis: Analysis of a Randomized Phase II Trial

A. G. Goglia¹, J. T. Yang^1,2, L. B. Kratochvil¹, D. Yerramilli¹, D. S. Higginson¹, A. Miller^1,2, M. Berger¹, S. N. Powell¹, A. Boire¹, and N. A. Wijetunga^1,3; ¹Memorial Sloan Kettering Cancer Center, New York, NY, ²New York University School of Medicine, New York, NY, ³University of North Carolina School of Medicine, Chapel Hill, NC

Purpose/Objective(s): We previously demonstrated that baseline and post-treatment summary measures of CSF ctDNA, including variant allele fraction (VAF), were predictive of outcomes following proton craniospinal irradiation (pCSI) for leptomeningeal metastasis (LM). Here, we report the analysis of CSF ctDNA from a randomized phase II trial evaluating pCSI and involved-field radiotherapy (IFRT) for patients with LM (NCT04343573). Materials/

Methods: Patients with non-small lung cancer (NSCLC) or breast cancer (BC) LM were randomized 2:1 to pCSI or IFRT. We performed targeted sequencing on CSF ctDNA collected before and at 3- and 6-months (mo) after protocol treatment in a subset of patients. We examined recurrent mutations, mean baseline VAF, and mean change in VAF at first follow-up measurement. Associations between mutations and CNS progression free survival (PFS) were assessed using Kaplan-Meier analysis and log-rank testing.
Results: Baseline CSF ctDNA was available for 24 randomized pCSI patients and 9 IFRT patients. At least one follow-up measurement was obtained for 11 and 3 patients, respectively. The median CNS PFS was 8 mo for pCSI and 2 mo for IFRT in the subset of patients with available baseline CSF ctDNA. Median baseline VAF was not significantly different between the two cohorts (0.43 vs. 0.35, p=0.56), or between NSCLC and BC (0.43 vs. 0.38, p=0.63). Improved CNS PFS was observed in pCSI patients with higher baseline mean VAF (median 13 mo, n=13) compared to those with lower mean VAF (4 mo, n=11, p<0.001). By histology, higher baseline VAF was associated with improved CNS PFS in NSCLC (median 16 mo vs. 3 mo, p=0.05), but not in BC (14 mo vs. 8 mo, p=0.60). Baseline VAF was not associated with CNS PFS for IFRT. A decrease in VAF was seen in 8 (73%) patients after pCSI. Improved CNS PFS was observed with decreased mean VAF after pCSI (15 mo) versus patients with increased VAF (8 mo, n=3, p<0.001). The most detected variants in baseline CSF ctDNA were TP53 (41%), EGFR (18%), MGA (12%), KMT2C (12%), FAT1 (10%), KMT2D (10%), ANKRD11 (10%), and PIK3CA (10%). For NSCLC, EGFR mutation was associated with improved PFS (median 16 mo, n=6) compared to EGFR wild type after pCSI (7mo, n=9, p=0.005), but not IFRT (EGFR mutation 3 mo, n=3, EGFR wild type 2 mo, n=3).
Conclusion: We identified genomic features in CSF ctDNA that correlate with CNS PFS for patients with NSCLC or BC LM treated with pCSI vs. IFRT. While both baseline CSF ctDNA VAF and decreased VAF at first follow up were prognostic of CNS PFS for pCSI, they were not for IFRT in this limited cohort, likely related to consistently poor outcomes across IFRT patients. The unexpected association observed between increased baseline VAF and improved PFS highlights the need to refine CSF ctDNA mutational analysis to make it a more consistent biomarker. Other summary measures incorporating mutation clonality may be more reliable and warrant further study.