NONINVASIVE DETECTION, CLASSIFICATION, AND RISK STRATIFICATION OF PRIMARY CNS LYMPHOMAS BY CTDNA PROFILING

Introduction: Circulating tumor DNA (ctDNA) has great potential as a noninvasive biomarker in diverse systemic lymphomas (Huet et al, J Clin Oncol 2020). Noninvasive access to tumor-derived DNA is particularly appealing for patients with primary CNS lymphoma (PCNSL), since tumor material otherwise requires invasive surgical procedures. Here, we explored the value of ctDNA in PCNSL patients for disease classification, MRD detection, and early prediction of relapse. Methods: We applied Cancer Personalized Profiling by Deep Sequencing (CAPP-Seq) to 65 tumor biopsies, 101 blood plasma specimens, and 43 cerebrospinal fluid (CSF) samples from 68 subjects with PCNSL, targeting 580 distinct genomic regions. We separately used Phased variant Enrichment Sequencing (PhasED-Seq, Kurtz et al, Nat Biotech 2021; in press) for ultrasensitive ctDNA monitoring. Levels of ctDNA were correlated with radiological measures of tumor burden and tested for associations with clinical outcomes. Finally, we developed a novel machine learning classifier to noninvasively distinguish CNS lymphomas from other CNS tumors based on their mutation patterns in plasma and CSF, using supervised training of a random forest followed by its independent validation. Results: We identified genetic aberrations in 100% of PCNSL tumor specimens (n = 65), with a median of 378 mutations per patient. Pretreatment plasma ctDNA was detectable in 82% of patients and in 100% of CSF samples (Fig. 1A), with ctDNA concentrations ranging from 0.013-1038 hGE/mL (median: 0.97) in plasma and 0.043-4342 hGE/mL (median: 3.55) in CSF (Fig. 1B). While ctDNA levels were significantly correlated with total tumor volume in MRI (p = 0.004, Fig. 1C), we did not observe significant associations between ctDNA levels and MSKCC score or concurrent steroid treatment (Fig. 1D-E). Pretreatment ctDNA was significantly associated with PFS (p = 0.0005, HR 3.6) and OS (p = 0.019, HR 3.1), both as continuous and binary variable (Fig. 1F). Furthermore, ctDNA positivity during curative intent induction therapy accurately predicted clinical outcomes (Fig. 1G). Finally, we applied our novel machine learning classifier to 129 specimens from an independent validation cohort. We observed high specificity (100%) and positive predictive value (100%) for noninvasive diagnosis of CNSL, with moderate sensitivity (50% for CSF, 20% for plasma), suggesting that a significant subset of CNSL patients might be able to forego invasive biopsies. Conclusions: We demonstrate robust and ultrasensitive detection of ctDNA at various disease milestones in PCNSL. Our findings suggest that ctDNA could serve as a valuable clinical biomarker for tumor burden assessment, outcome prediction, and biopsy-free lymphoma classification. We envision an important future role of ctDNA for personalized risk stratification and guiding therapies in clinical trials and in routine PCNSL management. (A) Sensitivity and specificity of ctDNA monitoring in blood plasma and CSF using PhasED-Seq. Grey bars on the left: sensitivities copmpared to previous NGS-based technologies (Fontanilles et al., Oncotarget 2017, Yoont et al., ASH Annual Meeting 2019, Montesinos-Rongen et al., J Mol Diagn 2020, Hattori et al., Cancer Sci 2018). Grey bars on the right show sensitivities of flow cytometry (FC) and Cytopathology (CP) in our cohort. (B) Scatter plot showing the comparision of ctDNA concentrations in pretreatment blood plasma and CSF. (C) Correlation of Association between pretreatment ctDNA plasma concentrations and total radiographic tumor volume (TRTV) measured by MRI. (D) ctDNA concentrations and steroid treatment at blood draw. (F) Kaplan Meier analysis of PFS and OS in patients with detactabe and non-detectable pretreatment ctDNA at diagnosis or progression. (G) Kaplan Meier analysis of PFS in patients with positive or negative ctDNA during curative intent induction treatment The research was funded by: the Else Kröner-Fresenius-Stiftung (to FS, 2018_A83), the Fördergesellschaft Forschung Tumorbiologie (to FS), and the Clinician Scientist Program of the Deutsche Gesellschaft für Innere Medizin (to FS) Keywords: Diagnostic and Prognostic Biomarkers, Liquid biopsy, Extranodal non-Hodgkin lymphoma Conflicts of interests pertinent to the abstract D. M. Kurtz Consultant or advisory role: Roche Molecular Diagnostics, Genentech Other remuneration: Ownership equity in Foresight Diagnostics. Dr. Kurtz has patents pending related to methods for analysis of cell free nucleic acids and methods for treatment selection based on statistical frameworks of clinical outcome. M. Diehn Consultant or advisory role: Roche, AstraZeneca, RefleXion and BioNTech Research funding: Varian Medical Systems, Illumina Other remuneration: Ownership interest in CiberMed, patent filings related to cancer biomarkers A. A. Alizadeh Consultant or advisory role: Genentech, Roche, Chugai, Gilead, Celgene Other remuneration: Ownership interest in CiberMed and FortySeven Inc, patent filings related to cancer biomarkers.