LM is one of the underdiagnosed complications of cancer despite now being a treatable disease (12, 17–21). Establishing the diagnosis of LM based on standard cerebrospinal fluid (CSF) cytologic analysis or MRI findings is often difficult, particularly at the early stages of leptomeningeal dissemination when treatment interventions could be more effective in limiting spread and triggering additional symptoms. Brain and spine MRIs have the advantage of being non-invasive, but their findings may be nonspecific and equivocal for LM (22). CSF cytology examination is a “gold standard” minimally invasive diagnostic test that provides confirmation of LM but has low diagnostic sensitivity, often requiring multiple lumbar punctures to establish a diagnosis of LM (23–25). The interpretation of CSF cytology results in a dichotomous qualitative variable, positive or negative for malignant cells, and can be challenging if reported as suspicious or atypical. The current EANO/ESMO LM group consensus interprets those results as equivocal (26) and thus, negative results could be false negative as a result of low disease burden in the CNS compartment.

Recent development of new technologies and studies showing the feasibility to identifying tumor- derived cell-free circulating tumor DNA (ctDNA) in CSF and capturing and quantifying CSF circulating tumor cells (CTC) using rare cell capture technologies. This provides a unique opportunity to improve LM diagnosis and to characterize the molecular landscape of the CNS metastases, with the hope that “liquid biopsy” of CSF will be diagnostic, prognostic and therapeutic biomarkers ( Figure 2 ).

The validated CellSearch system (Menarini Silicon BioSystems), utilizing an immunomagnetic CTC selection method based on an rare cell capture technology (RCCT) and anti-epithelial cell adhesion molecule (EPCAM) antibody-conjugated ferroparticles, is an FDA-approved methodology for enumerating CTC from blood in patients with breast, prostate and colon cancers (27–31). The CellSearch system has been deployed to evaluate CSF-CTCs in patients with LM and has demonstrated potential as a diagnostic marker ( Figure 3 ). A recent prospective study enrolled 95 patients with solid tumors (36 patients with breast cancer, 31 with lung cancer, 28 patients with other solid tumors) presenting with neurologic symptoms suspicious for LM or with MRI findings suspicious for LM, and referred them for clinical CSF examination (32). All patients underwent MRI of the brain and/or spine, and CSF cytopathology. Additionally, 3 ml of CSF from the same lumbar puncture were submitted for CSF-CTC analysis. Based on ROC analysis, presence of ≥ 3 CSF-CTC/per 3 ml was defined as the optimal cut-off for diagnosis of LM achieving a sensitivity of 93%, specificity 95%, positive predictive value 90%, negative predictive value 97%. Other studies reported similar findings of sensitivity above 80% and specificity above 95% (33–38). Additionally, several retrospective studies have demonstrated that LM patients with higher CSF-CTC counts at the time of diagnosis would have worse survival rate compared to those with lower counts, indicating a prognostic role for CSF-CTCs (39, 40).

Advances in CSF biomarkers analysis allowed to further incorporate the CSF-CTC analysis as an exploratory endpoint in multiple IRB -approved clinical trials (19, 41, 42). For instance, in a phase I/II trial of intrathecal trastuzumab for LM in human epidermal growth factor receptor 2 positive (HER2+) breast carcinoma (42) aimed at characterizing changes in CSF CTCs over time as a potential biomarker of treatment response. Of the 15 patients, 9 had greater than 1 cycle of treatment and were evaluated: a decrease in CSF-CTC count with treatment, and a later increase in CTCs predating radiographic LM progression, were seen in several patients, which led to hypothesis that CSF-CTC analysis may serve as a platform to assess quantitative treatment response and should be further investigated.

The largest prospectively collected case series of 58 patients (41) treated with proton cranio-spinal irradiation (pCSI) reported statistical significance association (HR: 1.05 per 10 cells (95%CI: 1.01–1.10, P = .01) of pre-pCSI CSF-CTCs with CNS progression free survival (PFS) but not associated with overall survival. Moreover, in 31/58 patients with both a pre- and post-pCSI, no increase in CSF-CTCs was observed immediately post-pCSI. It is probably too early to conclude whether CSF-CTC results should influence the decision to discontinue treatment as the relationship between number of CSF-CTCs and radiographic disease CNS burden needs to be investigated further.

Currently the RCCT immunomagnetic platform for enumerating CTC in CSF could be available in a few but not all highly specialized cancer centers, complicating the widespread implementation of this approach, and the validation of its clinical utility across other institutions in the US and worldwide. Other limitations of the CellSearch platform are that it only detects cells from tumors of epithelial origin and have limited ability to provide comprehensive genomic characteristics of the cells. Thus far genomic sequencing of isolated CSF-CTCs has been achieved in patients with LM from breast cancer expressing human epidermal growth factor receptor 2 (HER2) and demonstrated similarities between CSF-CTCs and tumor cells from primary site in addition to several genetic alterations private to CSF-CTC (38, 42–44).

Detection of ctDNA in plasma has showed promising potential as a source of “liquid biopsy” to reflect extracranial treatment response and outcomes in a variety of cancers (45–47). However recent studies have shown that plasma ctDNA may not accurately reflect intracranial disease burden in patients with BM (48, 49). Less than a decade ago several studies showed the feasibility of CSF cfDNA sequencing (16, 48–50). Several retrospective studies have demonstrated a positivity rate of CSF ctDNA consistently above 50% for patients with primary or metastatic brain tumors, including detection of tumor-associated mutations in the CSF of 63% of patients with brain metastases (16); in patients with glioma, CSF ctDNA can be found at a much higher positivity rate than plasma and correlates with prognosis (51, 52). CSF is relatively acellular and has direct anatomical contact with BM, making it easier to detect ctDNA in this fluid, potentially with a better correlation with prognosis than plasma ctDNA. Since than many groups reported results which suggested that CSF sequencing may offer a window into the identification of drug resistance mechanisms in patients whose primary tumor responded to genotype-directed targeted cancer therapy but then developed tumor progression in the CNS ( Table 1 ). Additional work is needed to understand to what extent CSF cfDNA reflects the genetic alterations of the primary tumor and/or CNS metastases as a small study showed that some paired BM specimens had unique alterations in TP53 and KRAS, but there were notably very few unique mutations in the CSF specimens (65). Based on current data many questions need to be answered in a clinical setting: how to use CSF ctDNA to monitoring genetic evolution of the tumor in response to treatment, what the role of clonal and subclonal mutation in the CSF is, and whether CSF ctDNA presence or absence, or the percentage of variant allele frequency (VAL) in the sample are the marker of response and correlation with survival and might thus be clinically useful as a “surrogate” source of tumor-derived DNA.

Molecular profiling of CSF in patients with brain tumors to identify potential micro RNAs (miR) associated to brain or leptomeningeal dissemination is of great interest, as it could provide a tool to predict and anticipate metastatic disease. Due to technical limitations in discerning tumor and immune cell miR profiles due to the rarity of CTCs, results are heterogeneous and require cautious interpretation. Despite heterogeneous findings, some studies report associations of specific miRs found in the CSF with the presence of brain tumors. An early study in patients with primary and metastatic brain tumors found, using a candidate approach, that miR10b, miR-21 and miR-200 were overexpressed in samples from patients with primary and metastatic brain tumors compared to controls{Teplyuk, 2012 #299}. Follow-up studies conducting miR profiling found differential expression of miR-30e, miR-140, let-7b, mR-10a and miR-21-3p in patients with brain tumors in a cohort of 175 patients that included 13 patients with metastasis; these miRs were validated in an external cohort of 105 patients {Kopkova, 2019 #298}. More recent approaches study miR signatures, which could be more informative as they aggregate global miR profile changes; in a cohort of 65 patients (including 11 patients with LM and 6 patients with BM), miR-335-5p and miR-34b-3p were linked to brain metastasis{Im, 2021 #297}. Although provocative, these studies are limited by the small number of patients, and most of them are not tested in prospective settings; besides these low numbers, these cohorts are heterogeneous as metastases from different types of primary tumors are often grouped to compensate the small cohort samples in granular analyses. Thus, future studies in this line should focus on larger, homogeneous cohorts and incorporate to the extent possible prospective testing of miR profiles.

Paralleling the advances in plasma and other non-CNS bodily fluids, considerable progress has been achieved after almost a decade of CSF-based liquid biopsy: from its conceptual inception and feasibility to its clinical validation, including the potential to influence clinical trial design, several roadblocks have been cleared out regarding the utility of CSF liquid biopsy for CNS disease. There are a few additional questions that must be answered to better understand the prognostic implications of detectable CTCs and ctDNA in CSF, the role of the molecular profile of CSF cfDNA from patients with CNS metastases and its correlation with the molecular profile of blood cfDNA/primary tumors or other sites of metastases to identify mutations or clonal populations that may predispose to CNS metastases. Most prominently, randomized prospective studies are sorely needed to provide sufficient evidence of feasibility in clinical pipelines, and a thorough benchmarking of molecular profiling methods (including cell detection and capture methods) has to be conducted to circumvent technical limitations of these approaches.