CSF proteomics for predicting response to treatment in patients with primary and secondary central nervous system lymphoma.

e19083 Background: Despite exquisite responses to initial therapy in patients with CNS lymphoma (CNSL), relapse is common and overall responses are disappointing compared to other non-CNS diffuse large B-cell lymphomas. Incorporation of multi-agent intraventricular chemotherapy (MAIVC) alongside systemic therapy is a novel strategy. To maximize MAIVC benefit, identification of minimally-invasive biomarkers for guiding patient management is necessary. This study sought to identify cerebrospinal fluid (CSF)-based proteomic predictive biomarkers for MAIVC treatment response at baseline and for longitudinal monitoring of tumor burden in CNSL. Methods: A retrospective study of patients with primary or secondary CNSL that were treated with MAIVC at Penn State Health (2015-2021) was conducted and reported following REMARK guidelines. Each MAIVC cycle was administered via Ommaya reservoir and included a 2-drug combination comprised of methotrexate, rituximab, cytarabine, etoposide, topotecan, gemcitabine, or thiotepa. Patient-matched CSF was sampled at pre- and post-therapeutic endpoints and profiled using shotgun proteomics. Patients were grouped by treatment response status. Early responders were defined as patients achieving malignant-cell-free CSF within 6 MAIVC cycles, whereas never responders were those patients that i) did not achieve malignant-cell-free CSF by endpoint or ii) required salvage surgery or radiation during MAIVC. Results: 59 patients were included (21 primary; 38 secondary CNSL) with a median age of 66 years (range 25-90) and median number of MAIVC cycles of 8 (range 2-23). Only 1 required Ommaya removal secondary to infection. A proteomic-based classifier, comprised of SGCE, LCP1 and AGRN, discriminated between early and never responders with AUROC of 0.95. Comparison of CSF at baseline (<3 cycles MAIVC) vs. post-treatment (>12 cycles MAIVC) identified H3F3A, YWHAE, LCP1, CA1 and GAPDH as tumor burden biomarkers. Conclusions: Here we developed a proteomic signature capable of predicting CNSL patient response to MAIVC, with potential to improve clinical decision making. Furthermore, the biomarkers associated with tumor burden represent important indicators of treatment response. Ongoing efforts are underway to evaluate these candidates in independent validation cohorts.