Circulating Tumor DNA Profiling for Detection, Risk Stratification, and Classification of Brain Lymphomas

Abstract

Purpose: Clinical outcomes of patients with CNS lymphomas (CNSLs) are remarkably heterogeneous, yet identification of patients at high risk for treatment failure is challenging. Furthermore, CNSL diagnosis often remains unconfirmed because of contraindications for invasive stereotactic biopsies. Therefore, improved biomarkers are needed to better stratify patients into risk groups, predict treatment response, and noninvasively identify CNSL.

Patients and methods: We explored the value of circulating tumor DNA (ctDNA) for early outcome prediction, measurable residual disease monitoring, and surgery-free CNSL identification by applying ultrasensitive targeted next-generation sequencing to a total of 306 tumor, plasma, and CSF specimens from 136 patients with brain cancers, including 92 patients with CNSL.

Results: Before therapy, ctDNA was detectable in 78% of plasma and 100% of CSF samples. Patients with positive ctDNA in pretreatment plasma had significantly shorter progression-free survival (PFS, P < .0001, log-rank test) and overall survival (OS, P = .0001, log-rank test). In multivariate analyses including established clinical and radiographic risk factors, pretreatment plasma ctDNA concentrations were independently prognostic of clinical outcomes (PFS HR, 1.4; 95% CI, 1.0 to 1.9; P = .03; OS HR, 1.6; 95% CI, 1.1 to 2.2; P = .006). Moreover, measurable residual disease detection by plasma ctDNA monitoring during treatment identified patients with particularly poor prognosis following curative-intent immunochemotherapy (PFS, P = .0002; OS, P = .004, log-rank test). Finally, we developed a proof-of-principle machine learning approach for biopsy-free CNSL identification from ctDNA, showing sensitivities of 59% (CSF) and 25% (plasma) with high positive predictive value.

Conclusion: We demonstrate robust and ultrasensitive detection of ctDNA at various disease milestones in CNSL. Our findings highlight the role of ctDNA as a noninvasive biomarker and its potential value for personalized risk stratification and treatment guidance in patients with CNSL.

[Media: see text].

FIG 1.

Tumor-informed ctDNA detection/quantification and CSF-ctDNA genotyping in patients with CNSL. (A) ctDNA monitoring in pretreatment plasma and CSF using PhasED-seq. Left: specificity (blue) and sensitivity (purple) of ctDNA monitoring in pretreatment plasma, compared with previous reports using high-throughput sequencing technologies (gray).^,,, Right: specificity (blue) and sensitivity (teal) of ctDNA monitoring in pretreatment. (B) Comparison of ctDNA concentrations (in hGE/mL) between pretreatment plasma and CSF in ctDNA-positive cases. (C) Comparison of pretreatment plasma and CSF ctDNA concentrations in CNSL with pretreatment plasma ctDNA concentrations in patients with DLBCL, normalized to TRTV. TRTVs and ctDNA concentrations in patients with DLBCL were derived from a previously published study reported by Kurtz et al. (D) Comparison of ctDNA levels in pretreatment plasma samples between patients receiving corticosteroid treatment and patients without corticosteroid treatment before sample collection. (E) Comparison of TRTV between ctDNA-positive and ctDNA-negative pretreatment plasma samples. (F) Correlation between TRTV and ctDNA concentrations in ctDNA-positive pretreatment plasma samples. (G) Comparison of ctDNA concentrations in pretreatment CSF between patients with periventricular involvement and patients with no sign of periventricular lymphoma localization. Statistical comparisons in (B), (C), (D), (E), and (G) were performed using the Mann-Whitney U test. Medians and ranges are indicated. (H) Bar plots depicting the monitoring performance of individual SNVs in pretreatment CSF samples, contrasting patients with (left) or without periventricular involvement (right). Dotted lines show the mean fraction of SNVs detected. Squares below the bar plots show whether ctDNA was detected (teal) or not detected (orange). (I) Biopsy-free genotyping from pretreatment CSF. Individual dots represent the percentage of patients with at least one SNV detected by tumor-agnostic genotyping from pretreatment CSF, ordered by increasing DNA input mass. Dotted lines indicate thresholds at 7 ng and 33 ng DNA input. ng, nanogram. CNSL, CNS lymphoma; ctDNA, circulating tumor DNA; DLBCL, diffuse large B-cell lymphoma; hGE/mL, haploid genome equivalents per milliliter; HTS, high-throughput sequencing; ND, not detected; NS, not significant; PhasED-seq, Phased Variant Enrichment and Detection Sequencing; r, Spearman correlation coefficient; SNV, single-nucleotide variant; TRTV, total radiographic tumor volumes.

FIG 2.

Prognostic value of ctDNA in pretreatment plasma samples. Bar graphs showing the percentage of cases with (A) progressive disease within 1 year or (B) death within 2 years after ctDNA analysis, stratified by positive and negative pretreatment ctDNA. Cases censored before 1 year in (A) or 2 years in (B) were not considered for this analysis. Kaplan-Meier analysis of (C) PFS and (D) OS in patients with detectable (red) and undetectable pretreatment plasma ctDNA (blue). Forest plots showing standardized hazard ratios for (E) PFS and (F) OS estimated by univariate and multivariate Cox proportional-hazards regression outcome analyses, incorporating ctDNA concentrations, TRTV, and key clinical and radiographic indices. ctDNA, circulating tumor DNA; HR, hazard ratio; NA, not assessed (ie, not considered for multivariate analysis); NS, not significant; OS, overall survival; PD, progressive disease; PFS, progression-free survival; TRTV, total radiographic tumor volumes.

FIG 3.

On-treatment plasma ctDNA as a prognostic biomarker. (A) Detection of ctDNA during curative-intent treatment as a function of time. Patient-level data demonstrating ctDNA detection during either intended four-cycle (left) or two-cycle induction therapy (right, blue horizontal bars). Relapses/progression were confirmed either clinically (black rhombus) or radiographically (black rectangle). Black arrows indicate further treatment following disease progression during treatment. Triangles show the ultimate clinical outcome, either progression/death (black triangle, always accompanied by radiographic or clinical disease progression) or CR (open triangle). Red bars represent HD-CTx followed by auto-SCT. Purple circle, ctDNA detected; open circle, ctDNA not detected; black rectangle, PD by MRI/CT scan; dark gray rectangle, SD by MRI/CT scan; light gray rectangle, PR by MRI/CT scan; open rectangle, CR by MRI/CT scan. (B) Kaplan-Meier analysis of PFS (left) and OS (right) in patients with at least one ctDNA-positive plasma sample at any time point during induction therapy (red) compared with patients without detectable ctDNA during induction therapy (blue). (C) Kaplan-Meier analysis of PFS (left) and OS (right) in patients with positive ctDNA within the first two cycles of induction therapy (red) compared with patients without detectable ctDNA during the first two cycles of induction treatment (blue). Significance in (B) and (C) was assessed using the log-rank test. auto-SCT, autologous stem-cell transplantation; CR, complete remission; CT, computed tomography; ctDNA, circulating tumor DNA; HD-CTx, high-dose chemotherapy; HR, hazard ratio; MRI, magnetic resonance imaging; OS, overall survival; PD, progressive disease; PFS, progression-free survival; PR, partial remission; SD, stable disease.

FIG 4.

Noninvasive brain lymphoma classification. (A) Classifier scores are shown for each tumor, CSF, and plasma sample of the independent validation cohort (CNSL, left; non-CNSL, right), ordered by decreasing scores within each group (left y-axis). Brain cancer entities or diseases are depicted in different colors at the bottom. True-positive cases are shown in green, false-negative cases in red, and true-negative cases in blue. Black triangles represent the number of SNVs identified by Cancer Personalized Profiling by Deep Sequencing genotyping (right y-axis). The dashed line highlights the threshold for CNSL classification. (B) Sensitivities of correct CNSL diagnosis from either tumor (teal), CSF (green), or plasma samples (purple) by detection of MYD88 L265P hotspot mutation alone or by the classification algorithm developed in this study. In addition, sensitivity for CNSL detection by CSF FC and CP is shown in light gray; and specificities (blue) and PPV (dark gray) of the classification algorithm are displayed for tumor, CSF, and plasma specimens. CNSL, CNS lymphoma; CP, cytopathology; FC, flow cytometry; GBM, glioblastoma; iSCNSL, isolated secondary CNSL; ND, not detected; NSCLC, non–small-cell lung cancer; PCNSL, primary CNSL; PPV, positive predictive values; SNV, single-nucleotide variant.