P1209: peripheral blood ctdna-sequencing enables prediction of outcomes in patients with primary central nervous system lymphoma using a dynamic risk model

Topic: 20. Lymphoma Biology & Translational Research Background: Primary central nervous system lymphoma (PCNSL) is an aggressive B-cell lymphoma of immune-privileged sites. Although advances in treatment steadily improve outcomes, fatal toxicities and relapsed/refractory disease remain a significant burden to PCNSL patients. Current clinical challenges include a lack of tools for baseline risk stratification and insufficient discriminatory power of response evaluation by magnetic resonance imaging (MRI). Aims: To develop noninvasive strategies for baseline risk profiling and refined response evaluation to ultimately improve outcomes of PCNSL patients. Methods: We applied circulating tumor (ct)DNA sequencing to a training (n=101 samples) and a subsequent validation cohort (n=42 samples) of PCNSL patients consecutively recruited at the University Hospitals Cologne and Essen, Germany. Peripheral blood was collected before, during, and after treatment. In the current version, our ctDNA sequencing workflow shows improved sensitivity and specificity over our previously published version (Sobesky et al., Med, 2021), allowing for application to diseases with low peripheral ctDNA. Results: In the training cohort, the mean concentration of ctDNA at baseline was 1.7 (range 0.0-2.8) log hGE/ml plasma. Compared to our experience with peripheral lymphomas (Sobesky et al., Med, 2021 and unpublished), the detection of mutations causing changes to the amino acid sequence of the coding genome was challenging. Nevertheless, we detected mutations in most patients, including recurring mutations in PCNSL-associated genes like MYD88, PIM1, CD79. To assess a potential influence of the peripheral ctDNA burden on outcomes, we correlated the average mutated allele frequency (mAF) at baseline with progression-free and overall survival (PFS/ OS). Interestingly, higher average peripheral mAF were associated with impaired PFS and OS. Most important, no events were observed in patients with an average peripheral mAF < 0.5% (17% of all patients). Next, we investigated mutation-based tracking of peripheral residual disease (PRD) as a response biomarker in PCNSL. Of note, MRI-based response evaluation indicated partial response in most patients, thus proving to be insufficient for outcome prediction (Fig. 1). In contrast, persistence of PRD during treatment was associated with impaired PFS (median 3.7 vs. 26.9 months, p < 0.0001) and OS (median 4.7 months vs. not reached, p = 0.0009). Next, we hypothesized that dynamic detection of PRD might improve baseline risk stratification by average peripheral mAF alone. We created a dynamic, molecular risk score consisting of a low- and high-risk group: DyMoRisk-PCNSL. Patients were first assigned to a group according to their average peripheral mAF at baseline (<1% vs. >1%). Integrating PRD results, low risk patients became high risk when PRD was positive, and high risk patients became low risk when PRD was negative. Strikingly, DyMoRisk-PCNSL was highly predictive of both PFS (median 9.3 vs. 26.9 months, p = 0.0004) and OS (median 10.3 months vs. not reached, p = 0.0010). To validate our findings, we investigated our approach keeping all parameters constant in a validation cohort. The final results of this validation will be presented at the conference. Summary/Conclusion: Here, we present a fully noninvasive ctDNA sequencing approach in PCNSL. We were able to identify risk factors at baseline (average peripheral mAF), during treatment (PRD), and ultimately combined these findings into a dynamic risk score - DyMoRisk-PCNSL. This model greatly outperformed the current standard of clinical risk stratification and response evaluation in PCNSL.Keywords: Non-Hodgkin’s lymphoma, CNS lymphoma, ctDNA, liquid biopsy