Patient cohort and methodological feasibility

Extraction of cfDNA was conducted from 33 CSF samples obtained from 30 patients, including one patient with duplicate samples for technical control (revealing identical results) and two patients with serial CSF samples. While the CSF input volume varied from 0.8 to 3 ml, successful cfDNA extraction was achieved in 26 samples from 23 patients. Seven CSF samples (one control patient and six patients with histologically confirmed CNS tumors) did not yield sufficient cfDNA (> 3 ng) for NGS (Fig. 1, Table 1, Supplementary Table 1). Genomic coverage ranged from 0.19 × to 1.1x (Table 2). NGS-based cfDNA profiling enabled the detection of SCNAs, a hallmark of tumor-derived cfDNA, in CSF samples. This profiling also allowed for the comparison of SCNA profiles in cfDNA from CSF with those from matching tissue samples. Additionally, we evaluated the diagnostic relevance of each SCNA profile detected in CSF cfDNA and tissue, incorporating criteria for essential and desirable SCNAs based on the current WHO classification of CNS tumors for the patients with primary brain tumor diagnoses (overview in Fig. 2A, Supplementary Table 1).

Table 2 Sequencing metrics of CSF samplesFig. 2

Somatic copy number aberrations in cell-free DNA from cerebrospinal fluid of patients with CNS cancers. A Overview clinical diagnosis and SCNA parameters. CSF cfDNA SCNA positivity vs. negativity and comparison of SCNA profiles in CSF and corresponding tissue (shared SCNAs) are depicted. B Frequency of detection of somatic copy number aberrations (SCNAs) across various CNS cancer types. SCNA positive samples are shown in green, negative samples in red. C SCNA parameters (CNI score, aberrant bin count, tumor cfDNA fraction) across the tumor patients. D Circos plots showing the copy number profile of cfDNA from cerebrospinal fluid of three exemplarily tumor patients with a high or medium CNI score in contrast to a patient without SCNAs

Sensitivity and specificity of ctDNA detection in CSF

Ten out of 12 CSF samples from CNS cancer patients scored SCNA-positive (83%) and all control samples from patients with benign diagnoses (n = 6) or unclear CNS lesions (n = 5) were SCNA-negative (Table 1, Supplementary Fig. 1, Supplementary Table 1). SCNAs in CSF cfDNA were observed in seven out of nine patients with primary CNS tumors and in all three patients with secondary CNS tumors (Fig. 2, Table 3, Supplementary Fig. 1). SCNAs were detected in six out of 12 (50%) patients at the time of first diagnosis and in 6 out of 12 (50%) patients during progressive diseases (PD) stages (Supplementary Fig. 2). Among the ten SCNA-positive tumor patient samples we observed a mean CNI score of 58 (range 6 to 150), a mean count of aberrant bins of 185 (range 51 to 421), and a mean tumor fraction score of the cfDNA of 0.34 (range 0.11–0.71) (Fig. 2, Table 4, Supplementary Table 1).

Table 3 Patient characteristics of tumor patients with SCNA positive and SCNA negative cfDNA profiles in next-generation sequencingTable 4 SCNA characteristics of all tumor patients with SCNA positive cfDNA in next-generation sequencing

Next, our aim was to investigate the potential impact of sampling and clinical variables on SCNA status, CNI score, aberrant bin count, and estimated tumor cfDNA fraction. Specifically, we compared the timing of CSF sample collection in relation to tissue sample collection, as well as the CSF collection timepoint in comparison to the disease stage (at the time of diagnosis, during PD, or during surveillance). The total count of abnormal genomic regions (aberrant bin count) and the CNI scores tended to be higher in CSF samples collected after tissue preservation, despite similar fractions of tumor cfDNA. However, there was also significant variation observed across the samples (Fig. 3).

Fig. 3

Correlation of CSF liquid biopsies with sampling and clinical variables. Box-plots depict the distribution of the NGS parameters (CNI score, aberrant bin count, tumor cfDNA fraction) considering defined sampling and clinical variables. Significant p-values < 0.05 are shown in the charts, otherwise no statistically significant differences were observed. LB = liquid biopsy; PD = progressive disease, LMD = leptomeningeal disease, cytology-confirmed; ref. = reference; CSF = cerebrospinal fluid.

Regarding the rates of detecting SCNAs per se in CSF, seven out of 12 (58%) were observed after tissue collection for tumor diagnosis, and five out of 12 (42%) were detected prior to tissue collection (Supplementary Fig. 2). Concerning the potential impact of disease progression, there was a tendency towards higher CNI scores and abnormal genomic region counts in patients with progressive tumors compared to those with initial diagnoses. However, these differences did not reach statistical significance in our relatively small cohort (Fig. 3).

SCNA profiling of CSF cfDNA augments the opportunities of CSF cytopathology

Moreover, we examined the correlation of NGS parameters with cytology confirmed LMD, defined by the detection of tumor cells in the CSF sample during routine cytopathological assessment. While SCNAs were detectable in CSF cfDNA from all five patients with cytology-confirmed LMD, we additionally identified SCNAs in the CSF cfDNA from five out of seven patients without cytology-confirmed LMD (Fig. 3, Supplementary Fig. 2). The five patients with confirmed LMD demonstrated significantly higher CNI scores and fractions of tumor cfDNA compared to patients without LMD in cytology. Changes observed in routine CSF parameters, such as cell count and lactate levels, did not exhibit a significant correlation with the NGS parameters (Fig. 3).

Utility of SCNA profiling of cfDNA from CSF for diagnostic classification and disease monitoring of patients with CNS tumors

Ultimately, we assessed the effectiveness of our CSF LB approach using a recently introduced set of quality criteria tailored to assess the benefits of LB tools in patients with brain tumors. Our evaluation focused on key aspects, including: (i) establishing a diagnosis and/or identifying diagnostically relevant genomic alterations (including copy number alterations incorporated as essential or desirable criteria in the current WHO classification [11]), (ii) monitoring tumor response to therapy, and (iii) tracking tumor evolution [8]. To address these inquiries, we complemented NGS of matching tissue samples, allowing for a direct comparison of SCNA profiling of CSF cfDNA and tissue DNA. Tissue was available for seven tumor patients (patients 2, 4, 5, 13, 15, 17, and 23), one patient with meningitis and a history of ependymoma (patient 14), and one patient with IgG4-associated orbital inflammation (Figs. 1, 2A and 4, Supplementary Fig. 3).

Fig. 4

Diagnostic value of SCNA profiling of cell-free DNA from cerebrospinal fluid of patients with CNS cancers. A Concordance analyses between CSF and tumor tissue. Depicted is the fraction of SCNAs private to CSF (blue), private to tissue (gray) or shared between the two (red). B Circos plots with copy number profiles of CSF cfDNA compared to tissue DNA exemplifying the usefulness of SNCA profiling for minimal invasive detection of CNS cancer, molecularly informed diagnostic assessment, mapping of tumor heterogeneity and tracking tumor evolution as well as surveilling patients with a previous cancer diagnosis

First, we aimed to determine whether tissue SCNAs could be traced in cfDNA from CSF, given the essential value of SCNA profiling in neuro-oncology [11]. A concordance analysis, comparing SCNA profiles directly between CSF and tissue, was conducted on six tumor patients with matched tissue/CSF pairs exhibiting SCNAs in their CSF cfDNA (patients 2 (two CSF samples), 4, 5, 13, 17, and 23). Shared SCNAs between CSF cfDNA and tumor tissue DNA were observed in all patients, as particularly notable in patients 4, 5, 17 and 23 (Fig. 2A, Fig. 4A, Supplementary Fig. 3).

To refine the diagnostic utility of our approach, we assessed defined diagnostic genomic alterations, and thereby also considered the criteria of the current WHO classification of CNS tumors for essential and desirable SCNAs in particular brain tumor subtypes, as outlined with positive results in several patients (Supplementary Table 2).

Besides providing relevant molecular pathological information, our data suggests that tracing SCNAs in CSF samples offers an avenue for expediting and facilitating the diagnostic process in patients through a less invasive approach as especially evident in patients with CNS lymphomas (patients 4, 6, 13, 17), a CNS tumor for which nonsurgical treatment is inherently preferred (Fig. 4B, Supplementary Fig. 1).

Beyond that, SCNA profiling of CSF holds the potential for mapping tumor heterogeneity and tracking tumor evolution (Fig. 4B, Supplementary Fig. 3). In fact, the concordance analysis also unveiled distinct SCNAs exclusive to either CSF or tissue in all patients, albeit generally to a lesser extent than shared SCNAs between CSF and tissue (Fig. 4A, Supplementary Fig. 3). SCNA variations between CSF and tissue can be attributed either to spatial heterogeneity within a tissue sample or tumor evolution throughout the disease. Evolution of the SCNA profile could be traced in patient 2 with glioblastoma where copy number profiling of the tissue biopsy at first diagnosis in comparison to the CSF LBs at PD overspun a disease course of approximately 1.5 years including multimodal glioblastoma treatment. The two serial CSF LBs at PD collected within only one week did not reveal major differences compared with each other (Fig. 4). As an example for tracing stable disease (SD) over a longer period, serial CSF LBs of patient 19, with brain and leptomeningeal metastases from lung adenocarcinoma, showed a stable CSF profile over a 10-month interval (Supplementary Fig. 1B). To assess the tool’s applicability in monitoring disease activity or recurrence after complete remission, we sequenced CSF cfDNA from patients under surveillance due to a previous history of CNS cancer and uncertainty regarding the differentiation between tumor recurrence and other potential causes of clinical deterioration (patients 12 and 14): The CSF LB supported the recurrence of pleomorphic xanthoastrocytoma in patient 12 (Supplementary Fig. 3), whereas favoring the diagnosis of a postoperative infection over tumor recurrence in patient 14 after curative resection of posterior fossa ependymoma (Fig. 4, Supplementary Fig. 3).

Of note, SCNA differences between CSF and tissue can also be attributable to methodological aspects: Patient 15, diagnosed with glioblastoma and displaying typical copy number alterations, had no detectable SCNAs in CSF cfDNA collected 1.5 weeks prior to surgical tissue resection. This suggests a deficiency of tumor-derived cfDNA at relevant levels in this CSF sample (Supplementary Fig. 3). This scenario highlights that a SCNA-negative CSF sample does not fully rules out the diagnosis of CNS cancer. Notably, in terms of the diagnostic value of SCNA-negativity, no SCNAs were detected in non-cancer CSF samples, as also evidenced by the matching tissue and CSF samples from a patient with confirmed histology of IgG4-associated orbital inflammation (Supplementary Fig. 3). From a technical perspective, it is noteworthy that the SCNA data obtained through methylation-based profiling of tumor tissue from four glioma patients showed no significant disparities compared to SCNA profiling via NGS (Supplementary Fig. 4). Additionally, a technical replicate of independently processed CSF aliquots from patient 7 exhibited no SCNA deviations between the aliquots, as expected (Supplementary Table 1).