Total cfDNA concentration is the level/amount of DNA per volume of biofluid (blood, CSF, or urine). Because tumor cell turnover is higher than that of normal tissue, research has found that glioma patients tend to have higher absolute amounts of cfDNA in glioma than healthy patients (24, 33–35). However, total cfDNA can be impacted by many other factors unrelated to tumor burden (e.g. inflammation) reducing its sensitivity to detect disease without supplementary analysis (36, 37). Furthermore, concentration as a biomarker lacks molecular information that might inform targeted treatments and clinical management. Methylation of cf-DNA and recovery via methylation-specific PCR (27) or sequencing (38) has also been proposed as method of detecting disease via measurement of hypo/hyper-methylation at various genomic loci but is not discussed in this review.

qPCR (quantitative polymerase chain reaction) is a quantification method that uses sequence specific primers or fluorescent probes to detect and quantify tumor-specific somatic mutations (39). However, qPCR suffers from a variety of shortcomings that limit its sensitivity and specificity (40–42). Droplet digital PCR (ddPCR) is a modification of qPCR that improves precision and limit-of-detection (42). By dividing a typical qPCR reaction into many isolated droplets with ~1 template copy, a precise VAF can be computed from the ratio of mutant positive to wildtype droplets, with a VAF limit of detection around 0.001% (43). ddPCR has supplanted itself as a highly-accurate technique, and is considered a gold standard approach to quantify VAFs from liquid samples (36, 44). However, accurate and reproducible ddPCR assays require careful development and optimization of input template concentration, target-specific primers, and fluorescent probes and are restricted for use on a limited set of known hotspot mutations (44).

Next-generation sequencing (NGS) refers to a group of massively parallel sequencing technologies including Ion Torrent (45, 46), PacBio (47), and Illumina (48, 49), with the latter used most often for liquid biopsy due to its accuracy (50). Illumina sequencing uses synthesis of fluorescent dNTPs to clusters of template strands to recover the original template sequence (45). Unlike ddPCR, sequencing of cf-tDNA does not rely on sequence specific probes or any prior knowledge of mutations. Various library preparations enable sequencing of the whole genome (WGS), whole exome (WES), targeted hybridization capture, or amplicons from targeted panels with each technique trading genome coverage for read depth. For example, WGS can provide 20x-50x coverage over the entire genome, which enables detection of high-frequency somatic mutations and copy number variation but is too shallow to precisely measure cf-tDNA allele fractions below ~2-5%. Targeted amplicon sequencing can generate >10,000x coverage of specific genomic loci, improving VAF limit of detection but reducing the number of analyzed loci. Illumina sequencing has proven highly-accurate, with an error rate ranging between 0.5%-1% (50–52). However, this approach is relatively expensive, slow, and due to its error rate, cannot reliably detect allele fractions below the sequencer error rate without further assay modifications (53–56).

Nanopore sequencing is a relatively new technology (57) that has been used for liquid biopsies (58, 59). Nanopore sequencers work by feeding DNA strands through small pores embedded in a membrane. As they flow through the pore, each DNA base-pair creates a unique electrical disturbance that can be measured and used to call each base. Nanopore devices are low-cost, can sequence any length DNA strand, have a small form factor, and offer a rapid time to result making them ideal devices for liquid biopsy. However, the device’s error rate has traditionally prevented it from being applied to liquid samples where allele frequencies are less than ~2% (58). Our group previously analyzed CSF samples from 12 pediatric high grade glioma patients and found that nanopore had a 85% sensitivity and 100% specificity in CSF samples (58), which compared favorably to Illumina-based targeted sequencing. More recent improvements to basecaller accuracy, and also the use of circular consensus sequencing (59–61) have improved accuracy to <0.05%, comparable with ddPCR-based approaches (59).

While not perfect, these cf-tDNA detection methods have been used to demonstrate a variety of potential uses for glioma diagnosis and monitoring. In the next section, we highlight translational research that attempts to utilize these instruments to improve disease management.

Identifying tumor-specific molecular information in cf-tDNA that provides an accurate diagnosis, prognosis, and predicts response of a particular treatment is a “holy grail” clinical application for liquid biopsies. There is some work linking molecular markers (e.g. SNVs or CNVs) to the predicted response to radiation or chemotherapy in gliomas [reviewed in Birko et al. (75)]. However, minimal work has explored cf-tDNA diagnostics to personalize immunotherapy treatment. Studies looking at other solid tumors have identified several cell free DNA biomarkers as predictors of immunotherapy response (76), most notably increased tumor mutational burden (TMB) (77–79) and reduced copy number variations (80, 81). Pepe et al. showed feasibility for assessment of TMB in cytological samples from patients with NSCLC using a NGS platform (82). Studies have aimed to identify specific hotspot mutations that predict response to immune checkpoint inhibition or other immunotherapies. Guibert et al. identified that mutations in KRAS or TP53 without PTEN loss lead to increased PD-L1 expression and increased tumor mutational burden, increasing response to PD-1 immune checkpoint inhibition (77). This work was done in lung cancer, but the aforementioned mutations are also commonly present in gliomas, raising a potential opportunity to use glioma cf-tDNA to predict immunotherapy efficacy.

Another important clinical application of liquid biopsy for gliomas is the monitoring of tumor evolution. It has been shown that tumors undergo considerable evolution over the course of treatment, resulting in genetic changes that might suggest a new diagnosis and an adjustment to disease management (69). Several studies have investigated tumor evolution in glioma, with estimates ranging from 33-73% of genetic mutations at recurrence matching with alterations at biopsy (83, 84). However, as previously discussed, serial biopsies are discouraged due to the increased chance of morbidity. Miller et al. used CSF-derived cf-tDNA to monitor tumor evolution in adult gliomas (17) and was able to identify cf-tDNA in 42 of 85 patients who underwent CSF collection and NGS sequencing. In patients with hypermutated tumors, the median percentage match between the CSF derived cf-tDNA mutations and the initial tissue alterations was only 19.6%, while non-hypermutated tumors had an 81.7% match. These results indicate that CSF-based liquid biopsies can capture tumor heterogeneity and used to monitor tumor evolution over time. While this work was not applied to immunotherapy-based treatment, tumor evolution monitoring could be used to identify increased TMB and PD-1 sensitivity (77), or acquired resistance markers (85).

In addition to tumor evolution, several studies have shown the utility of serial cf-tDNA sampling for treatment response monitoring and resolution of pseudoprogression in gliomas (19, 37, 58, 62, 68, 69, 74, 86). One of the largest studies thus far from Panditharatna et al. collected serial CSF samples and concordant MRI from 22 patients (74). They found that cf-tDNA decreased in response to radiotherapy in 83% of patients, which was corroborated by a decrease in tumor size on MRI. The first prospective high grade glioma clinical trial with serial liquid biopsy was recently published by our group (19). We collected serial CSF and plasma samples from 24 patients and found that patients with decreased H3K27M CSF and plasma cf-tDNA VAF had prolonged progression free survival (19). A similar trend was identified by Jensen et al. while tracking response to immunotherapy over a variety of cancers (20).

We also compared serial ct-DNA levels with corresponding radiographic imaging. In individual cases, they were able to identify instances of suspected pseudoprogression, where radiographic progression was accompanied by a decrease in cf-DNA VAF. In another patient, a large increase in cf-tDNA VAF (>25%) preceded radiographic progression in many patients, suggesting that cf-tDNA VAF changes may act as an earlier warning sign of tumor progression versus radiographic imaging. For immunotherapy-based response, Jensen et al. used shallow WGS (0.3x) of cf-DNA to identify copy number alterations (CNAs) and report a metric of “genome instability” over a variety of cancers (20). This study demonstrated that dynamic changes in CNAs could track immunotherapy response and were able to resolve pseudoprogression, but specific use in gliomas has yet to be demonstrated. A common theme among these studies is relative changes in cf-tDNA signals—rather than absolute values—are better indicators of tumor response. Taken together, these results reaffirm the potential clinical utility of serial liquid biopsies for improved molecular profiling and effective therapeutic monitoring for gliomas.

Our current understanding of glioma cf-tDNA release and dissemination to various biofluids is still limited. Research has highlighted variability depending on a tumor’s proximity to CSF reservoirs in the brain (17, 69). This raises concerns about the ability of liquid biopsies to accurately track tumor burden if disease spreads. Blood-brain barrier permeability can also vary highly case-to-case, further complicating efforts to correlate tumor burden with cf-tDNA levels in blood. It is also unclear how various treatments (e.g. radiation, chemotherapy, and immunotherapy) impact both normal cfDNA and cf-tDNA release over time, which could bias cf-tDNA levels and VAF if not properly accounted for (10, 12). Future work might incorporate variables such as tumor ventricle proximity, tumor biology, and treatment type to better understand these patterns.

Even though there is a general consensus that the sensitivity and specificity of CSF-based assays are superior to plasma (18, 28, 70, 87, 88), plasma- and urine-based liquid biopsies are still highly desirable due to the ease of sample collection. It is likely that the large disparity between CSF- and plasma-derived results are partly due to limitations of current gold standard detection methods. As an example of the difficulty of implementing precise detection methods, Li et al. measured the performance of multiple ddPCR-based assays for the detection of H3.3K27M mutations in matched tissue, CSF, and plasma samples across three independent labs using two commercially available ddPCR machines (44). Results indicated that ddPCR was capable of precisely measuring small VAFs from plasma-derived cfDNA, but discovered high-variation among replicates, and statistically significant differences across assays and ddPCR instrument vendors. Significant protocol optimizations were required to improve the sensitivity, repeatability, and reliability of the assay (44). When considering NGS detection methods, even the most accurate NGS instruments have an established raw error rate of ~0.1%, which is most likely too high to precisely resolve plasma-derived cf-tDNA levels that fluctuate between 1% and 0.05%. These results highlight the need for highly optimized and standardized versions of current approaches, as well as improved detection methods to enable proper translation to clinic.

The current limitations with ddPCR, NGS, and Nanopore-based liquid biopsy approaches are not easy to solve, but progress is being made via improved assay design and bioinformatic error correction. For example, the use of universal molecular identifiers (UMIs) during targeted amplification can help resolve sequencing errors targeted amplification, enabling detection of 1 mutant molecule in 10,000 (89), with other assay design techniques further improving detection sensitivity by several orders of magnitude (53–56). Nanopore, long-read sequencing offers the ability to sequence single-molecule tandem-repeats constructed from small cf-tDNA fragments using rolling circle amplification. Even though the native error rate for Nanopore sequencing is relatively high (58), its ability to sequence long DNA strands with multiple redundant copies of a single cf-tDNA template allows for accuracy beyond any available NGS or ddPCR approach (59–61). Some of these methods are so accurate, that they are limited by polymerase error rather than sequencer error rate (59). Furthermore, continual improvements to basecalling software, library preparation methods, and assay design are certain to further reduce false positive rates. Because of these factors we expect long-read, consensus sequencing approaches to become a gold standard liquid biopsy approach for plasma cf-tDNA in the future.

Recent work has explored the effectiveness of CAR-T cell therapy in diffuse midline gliomas, administered serially into CSF via Ommaya reservoirs (6, 90). Ommaya reservoirs are used for intra-cranial administration of immunotherapies as well as frequent, minimally invasive recovery of CSF without the need for lumbar punctures. Ommaya reservoirs would allow for more practical, and frequent use of CSF to monitor disease and apply liquid biopsy techniques more frequently. More frequent sequencing-based liquid biopsies might add undue cost to treatment. This motivates use of lower-cost techniques such as ddPCR as well as investigation of cost-effective sequencing approaches like single-use Oxford Nanopore flow-cells (91).