Abstract 765: Early diagnosis of brain metastases using cerebrospinal fluid cell-free DNA-based breakpoint motif and mutational features in lung cancer

Abstract Background: Brain metastases (BM) are associated with high mortality rate in lung cancer patients. Despite advancements that have led to improved survival of lung cancer patients, the prognosis of lung cancer brain metastases (LCBM) is still dismal. Brain MRI is currently the method of choice for BM assessment. However, it is limited due to a lack of surveillance. Cytosolic evaluation of cerebrospinal fluid (CSF) is generally unsatisfactory due to its poor sensitivity. Therefore, exploring a sensitive and accurate method for early detection of BM in lung cancer patients is needed to improve disease management. Methods: CSF circulating tumor DNA (ctDNA) of 81 lung cancer patients confirmed or suspected of BM who were screened between June 2019 and October 2021 were analyzed using targeted next-generation sequencing. Patients were classified into three subgroups according to their BM status and the relationship with sampling time, including 61 POS patients (patients whose BM status was already positive at CSF sampling), 10 NEG patients (patients whose BM status was negative at CSF sampling and remained unchanged during the follow-up), and 9 NTP patients (patients whose BM status turned from negative at CSF sampling to positive during the follow-up). The training dataset recruited 70 patients whose BM status at CSF sampling was definitive, including 62 POS and 8 randomly selected NEG patients. The testing dataset consisted of 11 patients, including 9 NTP and 2 randomly selected NEG patients. A robust machine learning model was developed based on the 6bp breakpoint motif (BPM) features in CSF ctDNA. The mutational profile of CSF ctDNA was also assessed to reveal potential prognostic-related genetic alterations in LCBM. Results: The BM predictive model built on BPM features demonstrated a remarkable capacity to detect LCBM, with an Area Under the Curve (AUC) of 0.929 and 0.944 in the training and testing datasets, respectively. Model performance was not significantly improved by combining CSF ctDNA status features. Our model showed consistently high performance in identifying BM regardless of the patient’s clinical characteristics, such as histologic type, smoking or treatment history, and Karnofsky performance status. The risk score of patients in the testing cohort computed by our model showed a trend of negative correlation with the BM detection time (rho=-0.4, P=0.2861, Spearman’s rank correlation). We also identified RB1 variants, EGFR amplification, and Fanconi anemia (FA) pathway mutations as promising BM-related prognostic biomarkers in patients with lung cancer. Conclusions: We established a robust BM predictive model using BPM features in CSF ctDNA and profiled genomic alterations associated with BM aggressiveness, which provided insights into the potential use of CSF ctDNA sequencing for early detection of LCBM and disease management. Citation Format: Xueting Qin, Yujun Bai, Shizhen Zhou, Hongjin Shi, Xiaoli Liu, Song Wang, Xiaoying Wu, Jiaohui Pang, Xi Song, Xiaojun Fan, Qiuxiang Ou, Yang Xu, Hua Bao, Li Li, Haimeng Tang, Yang Shao, Shuanghu Yuan. Early diagnosis of brain metastases using cerebrospinal fluid cell-free DNA-based breakpoint motif and mutational features in lung cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 765.