CSF ctDNA Molecular Clearance as a Prognostic Biomarker for Orelabrutinib‐Treated Primary Central Nervous System Lymphoma: Insights From Comparison With PET‐CT

Abstract

This prospective biomarker study evaluated serial CSF ctDNA monitoring in newly diagnosed PCNSL treated with R‐MO induction therapy. Mid‐treatment ctDNA clearance was significantly associated with complete response and superior progression‐free survival, showing stronger predictive value than interim PET‐CT in this cohort. These findings support CSF ctDNA clearance as a promising molecular biomarker for early response assessment in PCNSL.

To the Editor,

1.

Primary central nervous system lymphoma (PCNSL) is an aggressive subtype of diffuse large B‐cell lymphoma localized to the central nervous system (CNS) [1]. Although high‐dose methotrexate (HD‐MTX)‐based polychemotherapy remains the standard of care [2], clinical outcomes vary due to early relapse and chemoresistance [3]. Contrast‐enhanced magnetic resonance imaging (MRI), the reference standard for response evaluation according to International Primary CNS Lymphoma Collaborative Group (IPCG) criteria, is confounded by treatment‐induced enhancement changes and pseudoprogression, limiting its reliability for interim assessment [4]. Fluorine‐18‐fluorodeoxyglucose positron emission tomography‐computed tomography (¹⁸F‐FDG PET‐CT) provides complementary metabolic assessment that may detect treatment response earlier than that achieved based on conventional anatomical changes. Recent evidence indicates that post‐treatment metabolic response on PET‐CT significantly predicts survival, whereas MRI‐based anatomic response does not [5]. However, PET‐CT shows limited sensitivity for small, diffuse, or low‐metabolism lesions [6, 7] and lacks standardized response criteria for PCNSL [8], highlighting the need for reliable prognostic tools to inform mid‐treatment decision‐making [9].

Circulating tumor DNA (ctDNA) in the cerebrospinal fluid (CSF) has emerged as a minimally invasive biomarker for CNS lymphomas, with higher sensitivity than that of plasma for detecting residual disease and high concordance with tumor tissue mutations [10, 11, 12]. However, it remains unconfirmed whether early ctDNA dynamics can inform risk stratification or guide therapeutic adaptation. In our multicenter phase II R‐MO trial combining rituximab, HD‐MTX, and the CNS‐penetrant Bruton's tyrosine kinase inhibitor (BTKi) orelabrutinib, although high response rates were achieved, early response biomarkers were not validated [13]. To address this gap, we conducted a prospective biomarker study to evaluate the association of mid‐treatment ctDNA clearance with tumor response and survival outcomes. Because MRI‐defined response based on IPCG criteria was the primary outcome measure, MRI could not simultaneously serve as an independent comparator without introducing circular reasoning. Therefore, interim PET‐CT was selected as the exploratory comparator owing to its demonstrated prognostic value [5] and its capacity to provide functional assessment independent of anatomical criteria. This single‐site translational study was performed within the R‐MO trial (ChiCTR2100049483). S1 and Supporting Information Tables S1–S3 show details of trial eligibility criteria, treatment regimens, next‐generation sequencing protocols, and statistical methods.

Between December 2021 and December 2024, 22 enrolled patients completed ≥ 4 cycles of R‐MO and were included (Figure 1A). Most patients (63.6%) were aged ≥ 60 years, 90.9% had deep parenchymal lesions, and 40.9% had an Eastern Cooperative Oncology Group performance status of 2–3 (Supporting Information Table S4). Pathogenic somatic mutations or copy number alterations were identified in all tumor tissues and 95% of matched baseline CSF samples, supporting the feasibility of CSF ctDNA profiling (Figure 1B).

FIGURE 1.

Longitudinal molecular profiling of tumor tissues and CSF reveals genomic correlates of the best treatment response in PCNSL. (A) Timeline of sample collection and radiological assessments. (B) Comparison of the prevalence of detection between tumor tissues and CSF. (C) Exposure duration and the best response for each patient. (D) Genomic alterations in tumor tissues and CSF. (E) CSF ctDNA detection status at baseline and C5D1, and BOR during induction. (F) Complete response rates stratified by ctDNA status at C5D1 (left) and by ctDNA clearance at C5D1 (right). (G) ROC curve of ctDNA clearance and PET‐CT response at C5D1 for predicting CR. (H) Diagnostic performance of CSF ctDNA clearance compared to that of PET‐CT in predicting optimal treatment response. (I) CR rates by CSF ctDNA MYD88 clearance at C5D1. (J) ROC analysis for predictive accuracy of C5D1 CSF MYD88 clearance. ASCT, autologous hematopoietic stem cell transplantation; AUC, area under the curve; BOR, best overall response; C5D1, cycle 5 day 1; CR, complete response; CSF, cerebrospinal fluid; ctDNA, circulating tumor DNA; EOT, end of treatment; NPV, negative predictive value; PD, progressive disease; PET‐CT, positron emission tomography‐computed tomography; PPV, positive predictive value; PR, partial response; ROC, receiver operating characteristic; SD, stable disease.

After four cycles (cycle 5 day 1, C5D1), nine (40.9%) patients achieved complete response (CR), 10 (45.5%) exhibited partial response (PR), two (9.1%) had stable disease (SD), and one (4.5%) showed disease progression (Figure 1C; Supporting Information Table S5). MRI and PET‐CT exhibited 90.9% concordance in response assessment (Supporting Information Tables S6 and S7). Following two additional cycles, the PR status of seven patients (including two with complete metabolic response detected in PET‐CT at C5D1) improved to CR, two patients died from infection, and two patients with SD showed disease progression. At the end of treatment (EOT), 15 (78.9%) patients achieved CR, and 2 (10.5%) remained in PR status, yielding an overall response rate (ORR) of 89.5% (95% confidence interval [CI]: 66.9–98.7; Supporting Information Figure S1 and Supporting Information Table S5). The best ORR during induction was 86.4% (95% CI: 65.1–97.1), with the best complete response rate (CRR) of 72.7% (95% CI: 49.8–89.3; Supporting Information Table S5).

Of all detected mutations, 93 (50.5%) mutations were shared between tumor tissues and the CSF, while 43 mutations appeared exclusively in the CSF, suggesting that ctDNA captures additional sub‐clonal or spatially heterogeneous events (Supporting Information Figure S2A). Mutational landscape of tumor tissues and the CSF was largely concordant, with recurrent alterations in MYD88 (85.7% vs. 84.2%), PIM1 (71.4% vs. 52.6%), KMT2D (47.6% vs. 31.6%), CD79B (38.1% vs. 26.3%), and CARD11 (33.3% vs. 36.8%; Figure 1D). Most cases (57.1%) were classified as the MCD subtype, consistent with typical PCNSL features (Supporting Information Figure S2B), without significant response difference between MCD and non‐MCD patients (Supporting Information Figure S2C). Mean variant allele frequencies (VAFs) were higher in tumor tissues (Supporting Information Figure S2D). Patients with CR had lower baseline VAFs than those with PR/SD (p = 0.020) and showed a significant decline in VAF at C5D1 (p < 0.001; Supporting Information Figure S2E,F).

Among 20 patients with paired CSF samples, 19 were ctDNA‐positive at baseline, and 5 remained ctDNA‐positive at C5D1 (Figure 1E). Additionally, among nine patients who achieved MRI‐confirmed CR at C5D1, 8 with detectable baseline CSF‐ctDNA (100%) demonstrated ctDNA clearance at interim assessment (Supporting Information Table S6). Consistently, C5D1 ctDNA negativity and dynamic ctDNA clearance from baseline to C5D1 correlated with a higher best CRR (p < 0.05; Figure 1F). Receiver operating characteristic analysis showed that ctDNA clearance at C5D1 demonstrated a stronger association with CR than interim PET‐CT (area under the curve [AUC]: 0.862 vs. 0.762), with numerically higher sensitivity and negative predictive value (Figure 1G,H), although these estimates are likely unstable given the small sample size and should be considered hypothesis‐generating. Additionally, CSF MYD88 mutation clearance also predicted higher CR probability (Figure 1I,J), while KMT2D mutations showed a trend toward reduced efficacy (Supporting Information Table S8). Representative cases further supported these findings (Supporting Information Figure S2G,H): in patient #17, CSF ctDNA clearance occurred 8 weeks before radiographic CR, while patient #21 showed persistent ctDNA and subsequent disease progression (Supporting Information Figure S2G,H).

On the data cutoff date (February 12, 2025), the median follow‐up period was 18.9 months (range: 4.0–46.7 months). The estimated 2‐year progression‐free survival (PFS) and overall survival (OS) rates were 62.5% (95% CI: 44.8–87.2; Figure 2A) and 75.2% (95% CI: 58.3–97.0; Figure 2B), respectively. The median time to response was 2.6 months, and the 2‐year duration of response rate was 72.4% (95% CI: 54.3–96.5; Supporting Information Figure S3A). Patients achieving CR at the EOT showed significantly longer PFS than those without CR (p = 0.009; Supporting Information Figure S3B,C). CSF ctDNA (p = 0.024, AUC = 0.740) and MYD88 clearance (p = 0.014; AUC = 0.786) at C5D1 correlated with improved PFS, whereas interim PET‐CT response showed no association with patient survival (Figure 2C,E; Supporting Information Figure S3D,F). Grade ≥ 3 adverse events occurred in five patients (22.7%), with infections (18.2%) and infectious pneumonia (9.1%) as the common events; however, no treatment‐related deaths or treatment discontinuation were reported (Supporting Information Table S9).

FIGURE 2.

Survival outcomes and prognostic impact of mid‐treatment molecular response versus interim PET‐CT. (A) Kaplan–Meier curves for PFS. (B) Kaplan–Meier curves for OS. (C) Kaplan–Meier curves for PFS and OS stratified by ctDNA clearance at C5D1. (D) Kaplan–Meier curves for PFS and OS based on MYD88 clearance at C5D1. (E) Kaplan–Meier curves for PFS and OS stratified by interim PET‐CT response at C5D1. C5D1, cycle 5 day 1; CI, confidence interval; CR, complete response; ctDNA, circulating tumor DNA; HR, hazard ratio; OS, overall survival; PET‐CT, positron emission tomography‐computed tomography; PFS, progression‐free survival; PR, partial response; SD, stable disease.

This prospective biomarker study provides the first systematic comparison of CSF ctDNA dynamics and interim PET‐CT for response assessment during BTKi‐based induction in patients with newly diagnosed PCNSL. Prior landmark studies established ctDNA as a prognostic biomarker in CNSL. Mutter et al. demonstrated that pretreatment plasma ctDNA concentrations were independently prognostic of PFS and OS [14]. Heger et al. developed the MOP‐C risk model integrating clinical factors, radiographic response, and peripheral residual disease [12]. However, these studies primarily utilized plasma ctDNA, which has lower sensitivity than the CSF for detecting CNS‐localized diseases. The present study extends this paradigm to CSF‐based ctDNA monitoring within a BTKi‐containing regimen, where the pharmacological properties of the CNS‐penetrant orelabrutinib may produce distinct molecular response kinetics. The orelabrutinib‐based R‐MO regimen demonstrated robust efficacy and durable benefit, comparable to prior BTKi‐based induction regimens in a higher‐risk cohort [13, 15, 16, 17].

MYD88 L265P clearance by droplet digital PCR (ddPCR) is already used at some centers for CSF monitoring and offers practical advantages, including rapid turnaround (~80 min), low cost, and high sensitivity [18]. In this cohort, mid‐treatment MYD88 clearance correlated with improved CRR and PFS, whereas baseline MYD88 status was not prognostic, indicating that dynamic clearance rather than static detection drives prognostic discrimination. However, the prognostic signal was not attributable to MYD88 alone: KMT2D alterations showed a trend toward reduced efficacy, and 43 CSF‐exclusive mutations were undetectable by single‐gene approaches, suggesting that multi‐gene profiling captures subclonal heterogeneity and potential resistance mechanisms. Additionally, approximately 15%–33% of PCNSL cases lack MYD88 mutations, rendering targeted assays uninformative in this subset. Hence, a prospective study balancing genomic comprehensiveness against cost, turnaround time, and DNA input requirements is needed to confirm whether multi‐gene NGS profiling materially outperforms MYD88 ddPCR for routine serial monitoring.

This exploratory, single‐center study is limited by its small sample size and observational design. Despite these constraints, mid‐treatment CSF ctDNA clearance showed a stronger association with CR and PFS than interim PET‐CT, suggesting that molecular response dynamics may capture treatment efficacy beyond imaging alone; however, these findings do not establish ctDNA as a superior assessment modality. Although persistent ctDNA positivity may identify molecular non‐responders, the actual clinical outcome remains unproven without interventional data. Hypothetically, ctDNA‐guided early intensification could be considered for molecular non‐responders, although such strategies remain speculative without interventional data. In systemic LBCL, ultrasensitive ctDNA‐MRD has demonstrated prognostic utility beyond PET‐based criteria [19], and prospective trials are currently assessing ctDNA‐guided treatment adaptation [20]. Extending this paradigm to CSF‐based monitoring in PCNSL represents a logical next step. Biomarker‐stratified trials are required to determine whether molecular response dynamics can guide consolidation and improve clinical outcomes.

Author Contributions

L.S., H.H., and Y.L. contributed to the conceptualization, data curation, formal analysis, investigation, methodology development, and drafting of the original manuscript; S.W. was responsible for pathological diagnosis and treatment outcome assessment, and contributed to formal analysis as well as manuscript review and editing; D.W. conducted biomarker sequencing analysis, performed bioinformatics interpretation, and provided software support; S.T., K.X., Y.S., P.Z., B.L., and L.C. participated in patient recruitment, clinical management, and provided essential clinical resources; L.H. contributed to data analysis; S.N. and Y.P. contributed to conceptualization, supervision, and critical revision of the manuscript; G.O., L.F., and X.G. led conceptualization, secured funding, supervised the project, and contributed to manuscript review and editing. All authors contributed substantively to the article, critically revised it for important intellectual content, and approved the final version for submission.

Funding

This study was supported by a grant from the Zhejiang Basic Public Welfare Research Project (Grant No. LBY24H080002), the Natural Science Foundation of Ningbo City (Grant Nos. 2023J017 and 2022J210), the Ningbo Science and Technology Plan Project (Grant No. 2022S032), and the Central Nervous System Drugs Key Laboratory of Sichuan Province (Grant No. 230042‐01SZ).

Ethics Statement

The study was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. Ethical approval was obtained from the Ethics Committee of the First Affiliated Hospital of Ningbo University (approval number: 2021‐R166‐02).

Consent

All patients provided written informed consent.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information: Table S1. Inclusion criteria for patient enrollment.

Supporting Information: Table S2. Exclusion criteria for patient enrollment.

Supporting Information: Table S3. List of 55 Genes in the targeted sequencing panels.

Supporting Information: Table S4. Patient demographics and clinical characteristics.

Supporting Information: Table S5. Response in evaluable patients.

Supporting Information: Table S6. Association between imaging response (PET‐CT/MRI) and CSF ctDNA at C5D1 and EOT.

Supporting Information: Table S7. Concordance and discordance of complete response between MRI and PET at C5D1 and EOT.

Supporting Information: Table S8. Analysis of baseline tissue gene mutations and tumor response.

Supporting Information: Table S9. Adverse events.

Supporting Information: Figure S1. Study flow, tumor responses, and changes in target lesion size.

Supporting Information: Figure S2. Genomic profiling and dynamic ctDNA tracking in PCNSL.

Supporting Information: Figure S3. Survival outcomes and predictive value of mid‐treatment biomarkers.

Data Availability Statement

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.