CtDNA analysis relies on the detection of somatic mutations that are not present in normal DNA. Therefore, prior knowledge of the genetic landscape of each patient’s underlying tumor is used for the detection through targeted assays, namely PCR methods (Forthun et al., 2019). Third generation PCR, also known as droplet digital PCR (ddPCR) has become increasingly popular in liquid biopsy for its ability to measure mutations, amplifications and deletions in DNA (Subhash et al., 2021). However, unlike traditional RQ-PCR methods, it does not rely on a standard curve for sample quantification (Taylor et al., 2017) and the DNA input is minimal. This is because ddPCR employs absolute quantitation and end-point measurement of lower concentrations of starting material sample input of 1 ng/reaction is possible with varying levels of multiplexing (Fitarelli-Kiehl et al., 2018) and Poisson statistics. In a study by Klega et al., ddPCR paired with NGS in pediatric Ewing’s sarcoma showed that ctDNA levels correlate with treatment response and disease-specific biomarkers (Klega et al., 2018) as ddPCR alone is not sufficient to identify all clinically relevant mutations and it is recommended to complement untargeted approaches (Andersson et al., 2021).

NGS is a valuable tool for identifying more comprehensive mutations across wider genomic regions (Kotrova et al., 2015). Currently, there are NGS protocols optimized for < 10 ng of cfDNA input, which is important as yields of cfDNA vary depending on the type of childhood cancer and tumour burden. It has been reported that the proportion of ctDNA to normal cfDNA can range from 0.01% to over 90% (Christensen et al., 2017). Moreover, NGS addresses the limitations of PCR methods, namely low throughput mutational profiling and the necessity for previous knowledge of ctDNA targets for detection (Shu et al., 2017). Sensitivities for NGS techniques depend on the depth of sequencing coverage and are expressed as percentage of MAF as the limit of detection. Most notably, molecular barcoding technology for ctDNA analysis has reduced error rate and improved sensitivity to ensure the valid consensus of sequences (Sato et al., 2019).

Knowledge of the original tumor is essential for serial disease monitoring and detection of resistance mutations with the potential for detection of subclonal disease. For the detection of low frequency clinically relevant somatic alterations (0.004%), Newman et al. used cancer-specific profiling by deep sequencing (CAPP-Seq) as an economical and ultrasensitive method with broad patient coverage (Newman et al., 2014). However, these novel assays are yet to be validated for clinical applicability. More recently, targeted NGS panels of known cancer mutations have shown to be useful for monitoring treatment response, with sensitivities in detecting mutations at MAFs of 0.1% (Gao et al., 2019).

The ability to sequence individual patient tumors is attractive in the field of personalized medicine as it is an unbiased approach for the detection of targets that were not previously associated with the individual’s tumor type. These methods typically involve sequencing of the patient’s tumor, from which the data are then used to design targeted assays, pending validated tumor-specific data. (Belic et al., 2016). With blind bias-corrected NGS analysis of NSCLC patient tumors, Paweletz et al. demonstrated detection of actionable driver and resistance mutations at 0.4% MAF. This assay detected mutations with no false positives, highlighting the potential of NGS ctDNA analysis for monitoring response and subclonal disease (Paweletz et al., 2016). Few studies have investigated NGS for high-throughput screening (HTS) for minimally invasive disease detection in childhood cancers (Applebaum et al., 2020; Peneder et al., 2021; Subhash et al., 2021). However, while significant progress has been made in validating tumor-specific NGS, more prospective and longitudinal ctDNA analyses from the timepoint of diagnosis will contribute to implementation of ctDNA for precision medicine.

In childhood cancers, where the ctDNA burden is expected to be different across underlying cancer types and where tumor burden is below 0.01% of total cfDNA, the available ctDNA is likely to contain less than one cancer genome (Fiala and Diamandis, 2018). Therefore, ultrasensitive NGS technologies will still depend on the absolute amount of ctDNA in every sample. Liang et al. showed that ctDNA detection via deep methylation sequencing paired with machine learning identified more patients with cancer compared to mutation sequencing analysis, highlighting detection of low tumor-derived signals (Liang et al., 2021). However, in another study by Lin et al., solid tissue-NGS detection of clinically relevant genomic aberrations and therapeutic targets was more reliable than plasma-NGS (Lin et al., 2021), thereby discouraging clinicians from making treatment decisions based on plasma-NGS alone. A previous limitation of high-throughput screening as a routine MRD diagnostic method in ALL is the tendency to amplify non-specific targets. However, in each MRD assay designed, background amplification is accounted for by the use pooled peripheral blood from healthy individuals (van der Velden et al., 2007), therefore, such controls are important for ctDNA assays. Although NGS technologies are providing a unique insight into ctDNA dynamics in cancer, the cost currently remains challenging for routine analysis (Gutowska-Ding et al., 2020). However, it is to be expected that these costs will decrease over time with increased depth of coverage as they did for traditional sequencing techniques, allowing this technology to be accessible for routine patient care (Chen and Zhao, 2019).

As solid tumor sampling and low levels of tumor burden are ongoing challenges for the assessment of disease in children with cancer, it is becoming more apparent that ultrasensitive and low-cost techniques for ctDNA detection can improve these limitations. Biosensing technology presents a unique alternative approach to conventional techniques of ctDNA detection, in part due to the lack of DNA amplification, that is, traditionally required for PCR methods. Biosensors are tools that produce a biological signal when detecting target analytes such as ctDNA, which is then converted into a quantifiable signal response (Wu et al., 2019). Electrochemical biosensors are more widely used for ctDNA detection due to their potential for clinical validity and commercialization due to portability and low cost. A review by Li et al. describes in detail all biosensor variations along with advantages, limitations and clinical values. Like PCR methods, biosensors can detect known aberrations, therefore an area for improvement is for biosensors to detect ctDNA in an unbiased manner at unknown mutated regions (Li et al., 2019).

Despite the challenges remaining to be overcome with liquid biopsy analysis, several commercial entities have developed ctDNA detection platforms. The most widely used liquid biopsy test is non-invasive prenatal testing (NIPT) which screens for fetal aneuploidy in pregnant individuals (Hartwig et al., 2017). To date, there are several FDA-approved liquid biopsy tests for ctDNA analysis on EGFR detection in NSCLC – the QIAGEN therascreen EGFR RGQ PCR kit and the cobas EGFR Mutation Test. These tests have been approved as companion diagnostics for solid tumor patients and have shown sensitivities of 0.81 and 3%, respectively, calculated as true positives and false negatives defined in accordance to the tissue-based test (Kwapisz, 2017). The QIAGEN kit is notable as a complementary screening test for patients with breast, NSCLC, colorectal and urothelial cancers, and detection of somatic mutations including EGFR, BRAF and KRAS in whole blood (Colomer et al., 2020). So far, it has been used to select patients in several Phase III clinical trials; however, more work is required to implement this in routine practice.

More recently, an NGS-based FoundationOne Liquid CDx test was approved as a companion diagnostic technology for mutations in NSCLC and prostate cancer to indicate eligibility for treatment (Kwapisz, 2017). Guardant360 CDx Liquid Biopsy was recently approved as another companion diagnostic for NGS detection of EGFR mutations which identified patients who would benefit from Osimertinib (Gupta et al., 2020). While NGS-based methods from the two tests could detect 300 and 60 genetic mutations, respectively, they could only cover a few genetic changes that match a patient to a specific therapy (Ignatiadis et al., 2021). Whether multi-gene panel assays for different cancers or targeted tests are superior for determining clinical outcome is yet to be confirmed. It is also noted that a negative result from these tests does not equate to negativity for the corresponding gene mutation, meaning that conventional tests should be performed in parallel (Ignatiadis et al., 2021). While preclinical studies of ctDNA have rapidly increased for adult cancer in recent years, implementation in routine disease monitoring pediatric oncology is yet to be established.

At initial presentation, children with neuroblastoma typically present with a high tumor burden, therefore, it is no surprise that biomarkers are routinely used for diagnosis, prognosis and treatment response evaluation. In several studies, MYCN amplifications which are prevalent in neuroblastoma patients and are correlated with poor prognosis were detected in peripheral blood by qPCR (Combaret et al., 2002; Van Roy et al., 2017). In a study by Van Roy et al., copy number profiles showed strong concordance with conventional genomic hybridization from the primary tumor in plasma collected from neuroblastoma patients. NGS and ddPCR detection of ALK mutations associated with poor survival (Wang et al., 2013) demonstrated high sensitivity and specificity (Combaret et al., 2009). In a recent review by Van Paemel et al., all of these studies detected mutations in ctDNA that were previously not detected in the primary tumor, underlining the potential of ctDNA as a diagnostic tool for profiling tumor heterogeneity in neuroblastoma and indicating micrometastases (Van Paemel et al., 2020).

Whole-exome and targeted sequencing of tumor and plasma samples obtained at various timepoints were used by Chicard et al. to track SNVs and copy number variations in cfDNA collected from neuroblastoma (Chicard et al., 2016). The subclonal components of neuroblastomas, as well as the accumulation of genomic abnormalities (such as those in the MAPK pathway) in treatment-resistant disease, were investigated in this study. While the available evidence points to the importance of ctDNA in therapeutic decision-making and response evaluation, this is yet to be conclusive. For risk stratification, a prospective study of 13 patients revealed that ctDNA was a successful surrogate marker for molecular diagnosis of segmental chromosomal aberrations, MYCN amplifications and ALK variants, thereby presenting a relevant model for ctDNA analysis (Kahana-Edwin et al., 2021b).

The amount of ctDNA that enters the blood is greatly limited by the blood-brain barrier (De Mattos-Arruda et al., 2015). To date, the presence of cfDNA in blood plasma has only been examined in pediatric patients with medulloblastomas, where it was detected in 40% of patients (Bettegowda et al., 2014). By analyzing only seven genes in cfDNA from cerebrospinal fluid (CSF), Martinez-Ricarte et al. were able to identify 17 of 20 patients (including two children) with diffuse gliomas (Martínez-Ricarte et al., 2018). Plasma from pediatric patients with diffuse midline gliomas was found to have tumor-derived histone H3 gene mutations (Huang et al., 2017). CSF, which has been shown by Seoane et al. to contain ctDNA to some level in adult patients (Seoane et al., 2019), provides another source of ctDNA for brain tumors. Since CSF is collected by lumbar puncture and routinely tested in children with brain tumors, ctDNA is yet to be fully investigated as a complementary marker for predicted CNS tumors. CtDNA in CSF was indicative of tumor response in two glioblastoma and two diffuse midline glioma patients (Izquierdo et al., 2021). More recently, MRD was detected in retrospective CSF samples from medulloblastoma patients who relapsed, earlier than disease progression detected by imaging techniques. The data from this study strongly support prospective analysis of ctDNA in future medulloblastoma clinical trials (Liu A. et al., 2021).

Plasma from 201 pediatric patients with diverse lymphoma subtypes (Mussolin et al., 2013) and 155 children with Hodgkin lymphoma (Primerano et al., 2016) had significantly higher cfDNA concentrations than plasma from healthy controls. Pathogenic NPM-ALK fusion genes were detected in plasma collected from patients with anaplastic large cell lymphoma by Mussolin et al. (Mussolin et al., 2013). In the same study, high cfDNA levels were concordant with poor prognosis in patients with Hodgkin lymphoma (Mussolin et al., 2013), and they were present at diagnosis in plasma from B cell non-Hodgkin lymphoma patients, but decreased after treatment (Machado et al., 2010).

Despite leukemia being the most common childhood cancer, ctDNA analyses have mainly been studied for pediatric solid tumors. In adult blood cancers, the evidence supports concordance between plasma cfDNA levels and disease progression in acute myeloid leukemia (AML) (Mueller et al., 2006; Fleischhacker and Schmidt, 2007; Jiang et al., 2012). Nakamura et al. demonstrated the prognostic value of ctDNA in AML, with ctDNA reflecting clonal dynamics in the bone marrow, predicting relapse and survival post-allogeneic stem cell transplant and showing concordance with MRD status (Nakamura et al., 2019). By strong correlation with conventional MRD testing in the bone marrow, the results suggest that ctDNA analysis in peripheral blood may eventually replace MRD monitoring in the bone marrow. Although this study was performed retrospectively, the evidence suggests the potential for prospective investigation. This potential was recently realized in a Phase 1b study involving 40 AML patients. The results showed that MAFs in plasma ctDNA were comparable to those in bone marrow and that lower ctDNA MAFs were an accurate marker of clinical response after the first cycle of treatment (Zeidan et al., 2020).

To date, ctDNA analyses in children with ALL have identified assays that are not sufficiently sensitive to inform prognosis (Schwarz et al., 2009; Cheng et al., 2013). However, ctDNA analysis could be used as a complementary test to bone marrow MRD measurement, as a representation of overall tumor burden in the patient (Cescon et al., 2020). In other pediatric hematological malignancies, studies have focused on the amount of total cfDNA as opposed to specific biomarkers in ctDNA (Mussolin et al., 2013; Primerano et al., 2016). While total cfDNA levels correlated significantly with treatment response and prognosis, they are not reliable as they lack the high specificity of ctDNA. Likewise, in children with different types of non-Hodgkin lymphoma, the total cfDNA levels were significantly higher than in healthy controls (Machado et al., 2010; Mussolin et al., 2013). In the plasma of patients with B-cell non-Hodgkin lymphoma, EBV-DNA was detected in the plasma of all EBV-positive patients. Plasma EBV levels decreased after treatment, then increased, suggesting tumor relapse (Machado et al., 2010). These findings show the potential of EBV as an alternative biomarker for ctDNA in children with EBV-positive tumors. For other pediatric blood cancers, further ctDNA studies are required to elucidate the value of ctDNA as an alternative biomarker in the clinic.

A retrospective study by Jimenez et al. investigated plasma samples collected at the time of diagnosis of different types of kidney tumors in 18 patients. The results demonstrated sensitive detection of plasma tumor-specific copy number and/or single-nucleotide changes (Jiménez et al., 2019). As renal tumor biopsies are not routinely performed due to risk of rupture, genomic profiling of kidney tumor markers in the blood may be useful for guiding cancer-specific therapies (Van Paemel et al., 2020). Data from a prospective observational study of three patients showed that ctDNA in longitudinal samples could predict treatment response, however, the authors did not conclude whether ctDNA negativity was due to a fully resected disease or eradication of MRD (Kahana-Edwin et al., 2020). Therefore, further prospective studies that include patients who have relapsed or present with low levels of MRD are required to investigate the potential impact of ctDNA analysis for these clinical applications.

Rearrangements of the EWSR1 gene, most typically EWSR1-FLI1 and EWSR1-ERG rearrangements, are the diagnostic hallmark for Ewing sarcoma. The specificity of key fusion genes such as EWSR1 allows for robust detection in ctDNA assays, however whether detection in blood can inform clinical decisions remains to be demonstrated. Recurring or increasing levels of the EWSR1 fusion gene copy numbers were previously shown to be associated with relapse in 234 blood samples from 20 patients (Krumbholz et al., 2016). Patient-specific primers for ddPCR were developed in another study to detect tumor-specific EWS-ETS fusion gene breakpoint fragments in plasma samples from three Ewing sarcoma patients. In this study by Hayashi et al., plasma fusion gene fragments were detectable in two of these patients when the disease was radiographically undetectable, implying that monitoring tumor-specific EWS-ETS fusion gene breakpoint fragments in the blood could be a suitable personalized biomarker for early relapse (Hayashi et al., 2016). These studies highlight the notion that patients with Ewing sarcoma may benefit from therapy follow-up utilizing ctDNA for MRD analysis.

The vitreous fluid has been investigated retrospectively in 26 patients with retinoblastomas, even though the procedure is not minimally invasive or the site easily accessible. With shallow whole-genome sequencing, eye enucleation was predicted with tumor-specific copy number variations and RB1 mutations discovered by Berry et al. in the vitreous fluid. In future trials, these biomarkers may guide the critical decision of whether to enucleate or rescue the eye (Berry et al., 2018). A recent prospective study by Xu et al. performed longitudinal molecular profiling ctDNA in aqueous humor samples, demonstrating clinical utility for monitoring treatment response (Xu et al., 2021). These conclusions were limited to small sample size, however, warrant further prospective validation.

It is important to understand the goal to be achieved with ctDNA analysis, in the context of diagnosis or risk stratification; treatment response monitoring or identification of treatment resistance or subclonal disease. The clinical question being asked would subsequently influence the method for ctDNA analysis, whether it be single-target assays, generic tumor panels or patient-specific panels. For example, whole exome sequencing analysis in 13 patients performed by Escudero et al. investigated common medulloblastoma germline and somatic driver mutations which could facilitate subtype classification and risk stratification. In the same patient cohort, ddPCR assays were valuable for MRD identification in CSF samples, while NGS revealed subclonal disease (Escudero et al., 2020). CtDNA could also be applied in other solid tumor treatment assessments where poor response is usually observed, such as end of induction therapy for high-risk neuroblastoma. In children with ALK mutant or ALK amplified high-risk neuroblastoma, ctDNA monitoring of this marker could identify patients who may benefit from alternative treatment strategies such as the ALK inhibitor lorlatinib, as demonstrated by Kahana-Edwin et al. (Kahana-Edwin et al., 2021b). There is strong evidence for the importance of profiling tumors at relapse, including RAS-MAPK pathway mutations known to occur in relapsed/refractory disease (Eleveld et al., 2015). Several pediatric studies have achieved this without prior knowledge of tumor-derived biomarkers in neuroblastoma (Cimmino et al., 2020) and more recently, in a panel of pediatric solid tumors (Stankunaite et al., 2021). These studies have shown the value of ctDNA for identifying potential mutations associated with drug resistance. While ctDNA has the potential for informing diagnosis, treatment response monitoring and identification of treatment resistance (Table 1), these studies were performed retrospectively in small cohorts. Prospective validation with comparison to standard-of-care at predefined timepoints are needed to translate ctDNA analysis into the clinical setting, where it could improve treatment response monitoring, detection of emerging resistance and MRD detection. However, for implementation into clinical care, ctDNA analyses require standardization.

There is currently no consistent method for liquid biopsy sampling, ctDNA extraction or analysis. A liquid biopsy workflow normally involves collection, biomarker isolation or detection and biomarker analysis, each of which requires procedure standardization. Compared to metastatic malignancies, localized cancers have reduced detectability of ctDNA (Bettegowda et al., 2014). In most solid tumor malignancies, ctDNA levels are < 10% in advanced metastatic disease and <1% in locally advanced non-metastatic disease (Bettegowda et al., 2014). Differences in ctDNA levels are attributed to the variability among patients with the same cancer and patients with different tumor types, stages, responses to treatment, tumor burden and cell replication (Bettegowda et al., 2014), all of which complicate ctDNA as a routine biomarker. Additionally, large volumes of plasma are required for sufficient amounts of ctDNA. Pantel et al. found that ctDNA concentrations in early-stage lung cancer patients can be as low as one genome equivalent in 5 ml of blood (Pantel, 2016). The age and weight of children are limitations that should be considered when collecting large volumes of blood for informative ctDNA data analysis. This is further confounded by the volume of sample, that is, clinically feasible to collect in children (Kahana-Edwin et al., 2021a). Detection and quantification of ctDNA therefore require highly optimal and sensitive methods which should be tailored for individual cancer types (Yi et al., 2017). This includes consideration of pre-analytical factors that affect the stability and yield of cfDNA.

There is no universal standard for pre-analytical procedures (sample collection and storage) and standard operating procedures (total cfDNA isolation and quantification) for ctDNA analysis. However, there are agreed conditions that have been shown to affect sample quality and stability of ctDNA. The impact of various cell-stabilizing blood collection tubes on cfDNA extraction was investigated by Alidousty et al., who demonstrated that cfDNA collection tubes containing preservatives stabilized total cfDNA and intact cells for up to 1–2 weeks at room temperature, which is more feasible for shipping, storage and batched processing protocols (Alidousty et al., 2017). This was contradictory to Parpart-Li et al., who showed cfDNA stability at 4°C (Parpart-Li et al., 2017). The optimal storage conditions for cfDNA sampling have not been standardised however, the general consensus is to store plasma at room temperature for a maximum of 4 h, followed by centrifugation at 2,500 × g at 4°C and long-term storage at low temperatures (Lommen et al., 2020).

Furthermore, lack of standardization for the type of sample (serum, plasma, urine and other bodily fluids) collected for ctDNA analysis is an issue often overlooked in many studies. However, studies performed by Thierry and Bronkhorst have since demonstrated that levels of ctDNA are much higher in plasma (Thierry et al., 2010; Bronkhorst et al., 2015). Lee et al. found higher cfDNA levels in serum and proposed that this occurred due to red blood cell clotting which then resulted in genomic DNA release, leading to dilution of ctDNA (Lee et al., 2001). Isolation of cfDNA for downstream analysis was investigated by Sorber et al., who concluded varying results in ctDNA recovery from pancreatic cancer plasma samples using a range of cfDNA isolation kits (Sorber et al., 2017). Limited yields of ctDNA may therefore affect assay sensitivity and specificity, regardless of whether they are broad coverage (e.g., NGS) or targeted detection (e.g., ddPCR). Thus, it is integral that methods for storage and processing of total cfDNA are optimal for sensitive downstream ctDNA analysis.

The sensitivity of ctDNA detection techniques should always be considered for clinical validity in an MRD or early detection setting. For acute leukemias, MRD for treatment response assessment is most beneficial at early timepoints, when adequate numbers of tumor cells are present (Paganin et al., 2008). Sensitive ctDNA detection is especially challenging with low levels of tumor material, therefore, it is highly unlikely that ctDNA analyses can be used for early cancer diagnosis (Fiala and Diamandis, 2018). Sensitivities can be limited by the dilution of fragmented ctDNA with normal germline cfDNA (Sherwood et al., 2016). Andersson et al. discussed the potential for the lysis of normal blood cells, which may contaminate the sample (Andersson et al., 2021). Plasma samples must typically be centrifuged and separated within 1–4 h of collection. This alone presents with challenges for pre-analytical variability with cfDNA concentrations and purity with varying processing times (El Messaoudi et al., 2013). Comparison with genomic DNA from granulocytes was suggested to distinguish true mutations from underlying clonal hematopoiesis (Rossi et al., 2019), highlighting the importance of robust controls for ctDNA detection. Synthetic cfDNA may be useful as an external control to ensure consistent DNA input. There are cfDNA reference standards that are cell-line derived and commercially available in buffer or plasma with average lengths of 160-170bp. Specifically, these controls are fragmented DNA that mimics the fragmentation pattern of cfDNA (Zhang et al., 2017).

For current MRD measurement, routine analysis requires a threshold to determine the MRD status of the patient to determine the sensitivity and specificity of the MRD assay, which may be adjusted for every patient depending on the clinical question. For every cancer, excluding ALL, AML and chronic myeloid leukemia (CML), there is no gold standard for such a threshold. In the past, sensitivities were established by a serial dilution of cells in a sample from a healthy individual (Björklund et al., 2003) to determine the lowest number of cancerous cells that gives an MRD positive result. This has since been renamed the limit of detection (LOD) rather than sensitivity as sensitivity is defined by probability of a positive test result in a truly positive patient (Steinbach and Debatin, 2008). The threshold for negativity in ctDNA analysis is further complicated by uncertainty of undetectable tumor burden or robustness of the assay when there is limited material input. With a range of technologies available for ctDNA analysis in pediatric solid tumors: 1) the clinical value should take precedence; and 2) the sensitivity of novel tests should be compared to established methods for their translation into routine patient care.

Guidelines for the most suitable unit of measurement for ctDNA have not been established and this is partly due to the purpose and type of ctDNA detection method. CtDNA concentration determined by ddPCR is expressed as ng/mL or haploid genome equivalents (copies)/mL. With NGS, Abbosh et al. measured genomic alterations in ctDNA as a fraction of total cfDNA (variant allele frequency or VAF) (Abbosh et al., 2017) whereas the studies previously described by Bettegowda et al. analyzed copies per volume of plasma (Bettegowda et al., 2014). The choice of technique for ctDNA analysis is confounded by the amount of input sample, which can be subject to intra- and inter-patient variability. A study on ctDNA in solid tumors by Bos et al. demonstrated better concordance between VAFs and mutant molecules per volume of plasma with ddPCR than NGS (Bos et al., 2021). However, the authors demonstrated that the variability between both units of measurement could have been partially explained by pre-analytical and analytical variables, as well as low and variable molecular coverage. To this end, appropriate sequencing quality controls to enable correction and better molecular coverage were suggested for better interpretation of ctDNA analyses.

In general, while targeted approaches were shown to have better analytical sensitivity than untargeted, both methods have been used extensively in clinical studies of pediatric cancer in adult cancers. They are required for: 1) investigating ctDNA biomarkers and their biological and clinical significance; and 2) designing personalized assays for ultrasensitive ctDNA detection. It is evident that standardization of units for ctDNA measurement in cancer, irrespective of adult or childhood, is challenging. Ultimately, the choice of ctDNA of biomarker and units of ctDNA analysis should always answer the clinical question at hand.