Clinical Implications of CSF-Ctdna Positivity in Newly Diagnosed Diffuse Large B Cell Lymphoma in a Large Cohort

Background: CNS involvement (CNSi) is an uncommon and often fatal event, occurring in around 5% of patients with diffuse large B-cell lymphoma (DLBCL) during primary treatment or shortly after completion (SCNSL). The importance of CSF-ctDNA as a noninvasive biomarker in detecting and monitoring of CSF infiltration and predicting CNS in 100 newly diagnosed (ND) DLBCL has been reported in our previous work (Jin-Hua Liang, ASH 2023; Jin-Hua Liang, Leukemia). In this study, we want to validate the previous findings in a large cohort of 222 ND DLBCL patients and explore the potential mechanisms promoting CNS dissemination in single cell level.

Methods: December April 2021 and December 2023, 222 ND DLBCL patients with matched pretreatment tissue/plasma and CSF samples in our center were enrolled in the present study. Patients were identified as CSF-ctDNA positive (CSF(+)) based on the following two criteria: (1) if tissue is available, at least one mutant variant detected in CSF-ctDNA was shared with tissue; (2) if tissue is not available, at least one mutant variant detected in CSF-ctDNA was canonical driver mutation as previously reported in DLBCL (Jin-Hua Liang, Leukemia). Patients were diagnosed as CNSi positive (CNSi(+)) according to conventional methods (the definition of CNS disease was diagnosed via presence of typical symptoms, radiologic findings, and/or detection of lymphoma cells in the CSF). Single-cell RNA sequencing was performed on tumor cells of the 12 DLBCL patients including 6 CSF(+) and 6 CSF(-) patients. The first-line treatments of all the ND-DLBCL patients were RCHOP-like regimens or R-CGVP regimens and the strategy of CNS prophylaxis were according to NCCN guidelines.

Results: Fifty-three patients (23.9%) were diagnosed with CSF(+) while 43 patients (81.1%) were according to the first criteria and 10 patients (18.9%) were according to the second criteria. Among the 53 CSF(+) patients, only 5 patients (9.4%) have evidences of CNSi according to conventional methods. The unique mutations detected in CSF were rare compared with plasma and tissue, suggesting early CNS infiltration without clone evolution. The mean variant allele frequency was significantly higher in tumor tissue than CSF in CSF(+) DLBCL. As regard to the distributions of lymphGen subtypes, similarly only 21 patients (46.7%) were MCD subtype (about 75% was MCD subtype for PCNSL patients) which were similar to our previous result. Univariate analyses showed that CSF(+) was significantly associated with high-risk clinical varivables, including advanced stage (P<0.001), CNSi high-risk sites involvement (P<0.001), elevated serum lactate dehydrogenase levels (P=0.003), high IPI score (P<0.001), and high-risk CNS-IPI score (P<0.001).

In the three-risk group model of CSFi-IPI (elevated CSF protein levels, high plasma ctDNA burden, and high-risk CNSi sites) developed in our previous study, the low-risk, intermediate-risk, and high-risk groups showed CSF(+) rate of 8.4%, 42.4%, and 56.0% in the whole cohort (P<0.001). Patients with mutations of CARD11 (P=0.013) and PLCG2 (P=0.031) were still more predominant with CSF(+) in the whole group. The down-regulated genes of CNS(+) DLBCL patients significantly enriched in PI3K-AKT signaling pathway, focal adhesion, regulation of action cytoskeleton and tight junction pathwayswhich might be the potential mechanisms promoting CNS dissemination in ND DLBCL have been validated at single cell level.

Furthermore, pretreatment CSF(+) was also significantly associated with poor outcomes both for progression free survival (P=0.002) and overalls survival (P=0.010). The sensitivity and specificity of pretreatment CSF-ctDNA to predict CNS relapse were 100.0% and 78.2%. From our data of dynamic monitoring of CSF-ctDNA burden, traditional CNS prophylaxis could clear CSF-ctDNA in a subset (62.8%) of patients, while it did not work for the other 37.2% patients for whom additional CNS treatment should be provided.

Conclusion: Taken together, we validate the prevalence and the genomic landscape for CSF(+) DLBCL and highlight the importance of CSF-ctDNA as a noninvasive biomarker in detecting and monitoring of CSF infiltration and predicting CNS relpase in a large cohort of ND DLBCL.