Liquid Biopsy to Detect Minimal Residual Disease: Methodology and Impact

Simple Summary The field of liquid biopsy is rapidly evolving. Techniques that improve accuracy are constantly being developed, and clinicians increasingly use liquid biopsy as a tool to guide their clinical practice. The assessment of minimal or microscopic residual disease (MRD) after oncological treatment with curative intent, however, remains challenging. Therefore, the implementation of liquid biopsy to determine the presence of MRD is not yet standardized. In this review, we focus on the detection of MRD through liquid biopsy in solid cancers, highlighting currently available methodologies and ongoing challenges. Abstract One reason why some patients experience recurrent disease after a curative-intent treatment might be the persistence of residual tumor cells, called minimal residual disease (MRD). MRD cannot be identified by standard radiological exams or clinical evaluation. Tumor-specific alterations found in the blood indirectly diagnose the presence of MRD. Liquid biopsies thus have the potential to detect MRD, allowing, among other things, the detection of circulating tumor DNA (ctDNA), circulating tumor cells (CTC), or tumor-specific microRNA. Although liquid biopsy is increasingly studied, several technical issues still limit its clinical applicability: low sensitivity, poor standardization or reproducibility, and lack of randomized trials demonstrating its clinical benefit. Being able to detect MRD could give clinicians a more comprehensive view of the risk of relapse of their patients and could select patients requiring treatment escalation with the goal of improving cancer survival. In this review, we are discussing the different methodologies used and investigated to detect MRD in solid cancers, their respective potentials and issues, and the clinical impacts that MRD detection will have on the management of cancer patients.


Minimal Residual Disease
Minimal or microscopic residual disease (MRD) is defined as 'a very small number of cancer cells that remain in the body during or after treatment' [1]. Tumor burden varies with disease and treatment. When there is a response to treatment, tumor burden decreases, and the tumor eventually becomes undetectable in imaging or clinical examination. However, if tumor cells remain in the patient's body after treatment, they may induce relapse, either locally or through distant metastases (Figure 1). The importance of MRD detection was first highlighted in hematological cancers and allowed clinicians to evaluate the success of therapy and to predict short-and long-term relapse [2]. In solid tumors, the detection of MRD has enabled patients to be classified as having a high or low risk of relapse [3].
tumors, the detection of MRD has enabled patients to be classified as having a high or low risk of relapse [3]. Hence, several research strategies are now being investigated to accurately detect MRD in solid tumors and integrate it into treatment strategies. The ability to detect MRD may prove beneficial for patients and clinicians. After completion of standard therapy with curative intent, treatment could theoretically be adapted based on liquid biopsy results. Treatment intensification in MRD-positive patients has the potential to improve disease-free survival and overall survival if MRD is efficiently treated. At the opposite end, de-intensification strategies omitting adjuvant treatment (for example, systemic chemotherapy) in MRD-negative patients could reduce patient burden, decrease treatment related side effects, improve the quality-of-life of the patient, and reduce financial costs for society without impairing survival rates.
New clinical trials are now being designed or initiated on diverse tumor types in order to establish the role of MRD detection in the selection of the optimal treatment after curative local treatment [4][5][6]. This strategy is limited by the reliable detection of MRD, as it cannot be assessed by standard imagery or clinical examination. Liquid biopsies have become a possible means to overcome this limitation. Although no consensus has yet been reached on the appropriate methodology to detect MRD through liquid biopsy, research has drastically increased, yielding technological and analytical improvements over recent years [3,7,8]. In this review, we discuss the methodologies that have been developed to detect MRD through liquid biopsy in non-hematological cancers, as well as the potential impact on future trials and clinical practice.

Liquid Biopsy
Liquid biopsy is defined as 'a test done on a sample of blood to look for cancer cells from a tumor that are circulating in the blood, or for pieces of DNA from tumor cells that are in the blood' [9]. It was first mentioned by Pantel and Alix-Panabières [10] to describe the use of a blood test to assess the presence and characteristics of a solid tumor. More generally, liquid biopsy refers to all biomarkers that can be detected in the blood: circulating deoxyribonucleic acid (DNA), proteins such as carcinoembryonic antigen (CEA) or cancer antigen 15.3 (CA15.3), tumor-associated cells, different types of ribonucleic acids (RNAs), and exosomes, among others. In this review, we focus on liquid biopsy The ability to detect MRD may prove beneficial for patients and clinicians. After completion of standard therapy with curative intent, treatment could theoretically be adapted based on liquid biopsy results. Treatment intensification in MRD-positive patients has the potential to improve disease-free survival and overall survival if MRD is efficiently treated. At the opposite end, de-intensification strategies omitting adjuvant treatment (for example, systemic chemotherapy) in MRD-negative patients could reduce patient burden, decrease treatment related side effects, improve the quality-of-life of the patient, and reduce financial costs for society without impairing survival rates.
New clinical trials are now being designed or initiated on diverse tumor types in order to establish the role of MRD detection in the selection of the optimal treatment after curative local treatment [4][5][6]. This strategy is limited by the reliable detection of MRD, as it cannot be assessed by standard imagery or clinical examination. Liquid biopsies have become a possible means to overcome this limitation. Although no consensus has yet been reached on the appropriate methodology to detect MRD through liquid biopsy, research has drastically increased, yielding technological and analytical improvements over recent years [3,7,8]. In this review, we discuss the methodologies that have been developed to detect MRD through liquid biopsy in non-hematological cancers, as well as the potential impact on future trials and clinical practice.

Liquid Biopsy
Liquid biopsy is defined as 'a test done on a sample of blood to look for cancer cells from a tumor that are circulating in the blood, or for pieces of DNA from tumor cells that are in the blood' [9]. It was first mentioned by Pantel and Alix-Panabières [10] to describe the use of a blood test to assess the presence and characteristics of a solid tumor. More generally, liquid biopsy refers to all biomarkers that can be detected in the blood: Circulating deoxyribonucleic acid (DNA), proteins such as carcinoembryonic antigen (CEA) or cancer antigen 15.3 (CA15.3), tumor-associated cells, different types of ribonucleic acids (RNAs), and exosomes, among others. In this review, we focus on liquid biopsy performed specifically on blood samples but note that liquid biopsy also includes analyses of other body fluids such as cerebrospinal fluid (CSF), saliva, urine, etc.
Liquid biopsy has several advantages over tumor biopsy. First, it is usually less invasive, dangerous, and painful. Second, it may provide a more comprehensive molecular overview of tumor heterogeneity, possibly reflecting the characteristics of different cancer locations in a particular patient. This cannot be achieved by a single tumor biopsy [11]. Finally, it is easier to obtain repeated blood samples over time to understand the dynamics of response to a specific treatment.
We will briefly review the main advances in the field of liquid biopsy over past years, focusing on the detection of plasma circulating-tumor DNA (ctDNA), one of the most advanced and frequently investigated technologies used to detect MRD [6].

Circulating Nucleic Acids
• ctDNA ctDNA is characterized by tumor-specific genomic alterations. The presence of ctDNA in the blood was first assessed in 1948 [12] and is a proportion of circulating free DNA (cfDNA). Every cell is able to shed DNA into the bloodstream through necrosis, apoptosis, or active secretion [3]. Although cfDNA biology is not fully understood, the quantity of cfDNA depends on several parameters: it can be increased during pregnancy, in case of infection or cell death (i.e., crush syndrome or major physical exercise), or after organ transplantation [13]. In healthy individuals, most cfDNA is derived from hematopoietic cells [14] and is fragmented to a mean length of 166 bp [15].
In the 1970s, Leon et al. showed that the quantity of cfDNA was greater in cancer patients [16]. Tumor cells have the same intrinsic capacity as normal cells to shed DNA, albeit with a higher probability due to higher cellular turnover. DNA with tumor-specific alterations can thus be detected in blood [17]. The DNA shed by tumor cells is more fragmented than that of healthy cfDNA, with the mean length of ctDNA molecules being around 143 bp [15]. The proportion of ctDNA is highly variable and can range from 0.1% to over 10% of the total cfDNA depending on the type of tumor, its size and localization, and tumor staging [18]. For example, ctDNA is detected in more patients with colon adenocarcinoma than in those with glioblastoma, and the amount of ctDNA is higher in the case of metastatic spread compared to localized disease [18]. The methodology of MRD detection through ctDNA and the results of current clinical trials will be discussed in detail in a later paragraph.

•
Mitochondrial ctDNA Mitochondrial circulating tumor DNA (mtDNA) is also shed by tumor cells and could potentially be a biomarker of cancer. The quantity of mtDNA may be elevated in some cancers, although not in all [19]. Until now, MRD detection through mtDNA mutations has not been extensively investigated in solid tumors, but promising results have been reported in leukemia patients [20].

Methylation patterns
Epigenetic modification could be divided into three main categories: methylation abnormalities, histone modification, and non-coding RNAs.
DNA methylation is a very powerful mechanism to regulate gene expression [21]: a methyl-group is added to the cytosine on CpG sites by proteins belonging to the family of DNA methyltransferases (DNMT). DNMT3a and DNMT3b are mostly involved in de novo methylation and thus in embryonic differentiation. DNMT1 is responsible for the maintenance of methylation during cell division. In normal cells, methylation occurs in a pattern specific to each cell type and does not occur in promotor regions, as they contain regulatory elements. In cancer cells, methylation patterns are disrupted, with hyper-methylation of promotor regions of tumor suppressor genes and hypo-methylation of promotor regions of oncogenes [21]. This leads to a down-regulation of the expression of tumor suppressor genes and an increased expression of oncogenes. Colon cancer is a seminal example of the role of methylation in tumorigenesis. The methylation of the MLH1 gene promotor leads to the repression of MLH1 protein expression, resulting in the inactivation of the mismatch repair machinery [22]. Several clinical trials are ongoing with therapies targeted at the mis-methylation process that occurs in cancer cells [23].
Detection of cancer-associated methylation patterns in the blood after surgery in Stage II-III colon cancer was found to be an independent predictive marker of high risk of relapse [24]. However, methylation pattern-based MRD detection did not show high specificity in this study, with 62% of negative patients eventually relapsing after completion of treatment. Detection of methylated-ctDNA (met-ctDNA) may also be a potential biomarker for MRD in breast cancer [25]: the absence of met-ctDNA decrease during neoadjuvant chemotherapy and the presence of met-ctDNA after treatment was correlated with a higher risk of relapse. One concern is that the proportion of patients in whom met-ctDNA was detected prior to treatment was low (20%). Sensitivity could be increased by looking for more than one methylation pattern [26,27]. In these studies, a multiple-gene methylation pattern correlated with treatment response and was associated with worse progression-free survival. Met-ctDNA detection could thus be a powerful biomarker of MRD, but detection methods are not yet sensitive enough to be implemented in clinical practice [28]. Studies are currently ongoing to assess for methylation patterns in blood as a cancer detection tool (NCT04814407, NCT04511559) but also for to monitor disease and determine the presence of MRD (NCT03634826, NCT03737539). The detection of met-ctDNA might be a more powerful tool if combined with other biomarkers.

•
Non-coding RNA RNA is composed of a single strand of ribonucleic acids and can be divided into coding RNAs, also called messenger RNA (mRNA), and non-coding RNA (ncRNA). ncRNA is a family of RNA molecules defined by their function and length. ncRNA is subdivided into housekeeping RNA, which includes transfer-RNA and ribosomal-RNA, and regulatory-ncRNA. Regulatory-ncRNA is also subdivided into multiple categories based on structure: long-ncRNA (>200 nucleotides), circular-RNA, and functional ncRNA including micro-RNA, small-ncRNA, small nuclear-ncRNA, and Piwi-interacting-RNA (piRNA) [29]. As these types of RNA are found in the nucleus and cytoplasm of cells, they can also be detected in the bloodstream or in other body fluids, via the same mechanisms as ctDNA [30]. The most investigated are miRNA and lncRNA. miRNAs modulate the expression of genes by binding to mRNA and silencing translation [31]: cancer cells express specific miRNA signatures that can be detected in tumoral tissue and in the bloodstream [30].
Data have shown that it is possible to detect cancer through the detection of cancerspecific miRNAs in testicular germ cell cancer [32], colon cancer [33], ovarian cancer [34,35], gastric cancers [36], and breast cancer [37]. Multiple studies are currently ongoing to assess the potential of miRNA as an MRD biomarker in several cancers, such as breast cancer (NCT04720508), pancreatic cancer (NCT04406831), prostate cancer (NCT04835454), head and neck cancer (NCT04305366), and many others (source: clinicaltrials.gov, accessed on 26 April 2021).
Long non-coding RNA (lncRNA) has also been investigated as a potential biomarker of tumor cells in many cancers [38][39][40][41][42]. LncRNA can be detected in the bloodstream and correlated with tumor stage and treatment response [43][44][45]. Despite showing good sensitivity (~70-80%), the specificity of lncRNA diagnostic tests is too low (~60%) to be applicable in routine practice [43,46]. Most ongoing clinical studies investigating lncRNA focus on its use as a cancer diagnostic tool (NCT03830619, NCT03469544, and NCT04269746). One ongoing clinical trial in triple-negative breast cancer assigns patients to different treatment strategies based on an assessment of their risk of relapse as determined by lncRNA (NCT02641847). To our knowledge, there are no current trials investigating lncRNA to detect MRD (source: clinicaltrials.gov, accessed on 28 April 2021).

• Exosomes
Extracellular vesicles (EVs) have a phospholipid bilayer membrane containing bioproducts derived from cells that secrete them via an active process. Exosomes are a subset of EVs released by living cells [47,48], ranging in size from about 40 to 160 nm (median 100nm) [49]. Apart from cytosolic and transmembrane proteins, they contain single and double stranded DNA [50], RNA (mRNA, miRNA, lncRNA, cRNA, etc. [49]), and cytosolic metabolites [48]. Exosomes are key players in inter-cellular communication. Their secretion allows cells to induce genetic, epigenetic, and protein transformation of other cells. This method of communication has been investigated in conditions including pregnancy [51], auto-immune disease [52], and cancer [49,53]. Studies have shown that cancer cells may produce up to 20-fold more exosomes than non-cancerous cells [47,54]. Exosomes play a role in tumorigenesis by allowing cell-to-cell communication to induce, for example, epithelial-to-mesenchymal transition (EMT), invasion and migration of cancerous cells, and a tolerant immune environment [49]. Cancer-derived exosomes can be isolated using cancer-specific proteins expressed at their surface [49]. Cancer-specific DNA and RNA modifications are also found in exosomes [49,55,56]. MRD could therefore theoretically be detected through cancer exosomes [55]; however, standardized isolation and quantification methods are lacking [57]. While clinical trials are ongoing to assess the diagnosis of tumors and the detection of MRD through exosomes, also under investigation is the use of the modulating power of exosomes to deliver systemic therapies to specific cells [58].

Circulating Tumor Cells
Circulating tumor cells (CTCs) were first described in 1869 when Ashworth detected malignant cells in the bloodstream of a metastatic cancer patient [59]. It would be more than a century, however, before CTCs were defined and isolated. By definition, CTCs are tumor cells. They can be isolated from other circulating cells based on tumor-specific alterations (immuno-chemistry, size, or density [59]). The tumor-specific genomic and epigenetic alterations that they carry can also be harnessed to detect CTCs [60], with the advantage that their DNA, in contrast to ctDNA, is not fragmented [61].
Although CTCs are indeed shed into the blood from primary tumors or secondary lesions [62], their presence does not necessarily imply the presence, nor predict the development, of micro-metastases. For a cell to be able to metastasize in a specific organ, it needs to undergo MET (mesenchymal to epithelial transition) in order to breach the vascular endothelium and colonize tissue [63]. CTCs with metastatic potential would have to express specific proteins on their surface in order to be recognized by endothelial cells. The FDA-approved CellSearch ® [64] isolation technique is based on the detection of one such cell-surface protein: the epithelial cell antigen EpCAM, on the surface of CTCs. This technique would, however, miss any CTCs without an epithelial phenotype [65].
Notwithstanding their limitations, numerous clinical studies have investigated the prognostic role of CTCs. These studies are either quantitative, counting CTCs in the bloodstream, or qualitative, analyzing CTCs for their genome, transcriptome, and/or proteome. The presence of CTCs after neoadjuvant chemotherapy was found to be associated with worse disease-free survival (DFS) in early breast cancer [66,67], nasopharyngeal carcinoma [68], and lung cancer [69], as well as a worse overall survival (OS) in melanoma [70,71]. Similarly, the detection of CK-19 mRNA in CTCs after curative intent treatment was found to be associated with worse DFS in early breast cancer [72,73]. No doubt due to the difficulty of immediately processing CTCs and the lack of standardized methods for isolation, there is, to our knowledge, only one interventional trial that evaluated the use of CTCs as a marker of MRD [74]: this randomized trial has shown the potential benefit of detecting CTCs to guide adjuvant treatment in hormone-dependent, HER2-negative metastatic breast cancer. There is to our knowledge no interventional trial based on the detection of CTCs ongoing at the moment (source: clinicaltrials.gov, accessed on 15 May 2021). With the support of the European Liquid Biopsy Society, which is working on defining technical guidelines on the collection, processing, and detection of CTCs, we expect more trials in this field to open in the coming years [75].

Circulating Proteins
Today, the most validated MRD-biomarkers remain cancer-specific antigens such as CA125, CA15.3, carcinogen embryonic antigen (CEA), alpha-feto protein (αFP), and prostate-specific antigen (PSA) [76]. Although these are currently used in daily practice to follow patients throughout the course of their disease, they are not specific or sensitive enough to be used alone as biomarkers for MRD [77][78][79][80][81].

Methodology to Detect Minimal Residual Disease with ctDNA
ctDNA detection and characterization are the most frequently used and investigated methods to detect MRD, based on the presence of tumor-specific genomic alterations. Some applications have already been implemented in daily clinical practice [82]. Next generation sequencing (NGS) refers to several different techniques of high-throughput nucleotide sequencing (sometimes referred to as massively parallel sequencing), based on the extreme miniaturization, parallelization, automation, and digitalization of polymerase chain reaction (PCR)-based "reading while copying" of DNA [83,84]. These methods allow scientists and clinicians to obtain increasingly large quantities of sequencing data within a few hours [83]. The main limiting factor of these sequencing tools is the error rate, which varies from 1% to 0.01% depending on the brand and type of sequencer [85]. High-confidence detection of variants present at a fraction below 1% and therefore requires sufficient depth of coverage (i.e., the number of sequences that "read" any given nucleotide position) in patient as well as control samples, and the use of bioinformatic analyses, and algorithms that allow for the application of adequate quality control criteria.
Other PCR-based techniques such as droplet digital PCR (ddPCR) allow for the detection of DNA variants up to 0.001% [10], although they are suited to the assessment of limited numbers of pre-determined variants, rather than the discovery or determination of large numbers of different variants.
Below, we describe the sequencing methodologies and techniques used to detect ctDNA in the context of MRD application.

Personalized Methods
Tumor-personalized methodologies to detect ctDNA are methods based on prior knowledge of the genomic landscape of the tumor.

Tumor-Customized Based Panels
One of the most validated methodologies to detect ctDNA is the specific interrogation of plasma-derived cfDNA for genomic alterations previously identified in the corresponding tumor. The tumor biopsy or surgical sample is sequenced (e.g., whole exome sequencing (WES) or a large gene panel) to identify somatic mutations (absent, or present at significantly lower fractions in non-tumoral tissue). A custom-made panel is designed accordingly to sequence the altered positions in the plasma. This custom panel, specific to the patient's tumor, can employ PCR-amplification using specific primers, or hybridization-based capture using specific probes, in order to enrich the regions of interest for NGS [86,87], with a barcoding method typically incorporated to allow for sample multiplexing.
This design has been implemented in several clinical studies in the context of MRD detection [87][88][89][90] and has shown interesting results. The techniques used in these studies, and their main conclusions, are detailed in Table 1. The main advantage of this methodology is the comprehensive view of the genomic landscape of the tumor that is achieved. This provides multiple targets for follow-up in the plasma, thereby optimizing the sensitivity and robustness of detection. By decreasing target size (i.e., the number of nucleotides to be sequenced) as compared to WES or a large gene panel, use of a patient-specific panel can exploit the considerable capacity of NGS sequencers to increase the depth of sequencing. This can allow for the detection of variants present in very low quantities (for example, variant allele frequency (VAF) < 0.01%, based on a high pretest probability of detecting the variant), extremely useful in detecting MRD. Some technologies that combine deep sequencing with unique molecular identifier (UMI) barcoding and bioinformatic pipelines such as Safe-Seq [90] or TARDIS [89] have been shown to detect tumor variant allele frequency as low as 0.002%. However, the choice of which tumor variants will make up the tumor-personalized panel may be subjective or arbitrary, often based on disease-pertinence bio-informatic algorithms, and can omit driver mutations while selecting passenger mutations. The threshold for positivity of ctDNA is also a subjective decision with some authors setting the limit at one mutation per sample [86], while others require the detection of at least two mutations per sample [88].
This technique requires a tumor sample, and this itself creates several issues. First, obtaining a good-quality biopsy can be challenging and sometimes carries risks (bleeding and pain) for the patient depending on the location (e.g., close to blood vessels). Second, patient-specific panels based on sequences from a particular tumor sample fail to capture genomic alterations in other parts of the same tumor, as well as those in distant metastases, due to tumor heterogeneity. Third, the genomic landscape of the primary tumor and its metastases can change over time as treatment and the natural course of the disease select specific tumor clones. The use of a tumor-guided panel therefore carries the risk of missing (potentially informative) alterations in the plasma. Despite these drawbacks, this method has become the gold standard for detecting ctDNA for MRD due to its ability to detect tumor variants even when present in very low quantities [95].

Custom-Based PCR Assay
Droplet Digital PCR (ddPCR) is a highly specific technique that allows for the detection of individual pre-defined genomic variants, down to a VAF of 0.01%, with promising results [93]. ddPCR has the benefit of giving quantitative results and, unlike quantitative PCR, also presents a proportion of variant allele frequency within the same assay. It also offers a high-sensitivity compared to other techniques, as very low quantity of DNA (~1 ng) can be used to detect a variant [96].
ddPCR probes are designed to detect genomic alterations already identified in the tumor; this method of MRD detection therefore shares many of the same advantages and disadvantages as the custom-based NGS panel technology described above. The use of ddPCR to detect MRD has been demonstrated in hematological neoplasia, when the carcinogenesis is due to a specific mutation (for example BRAF V600E in hairy cell leukemia [97]). However, data, although still scarce in solid tumors, show promising results in tumors where hotspot mutations are well-established as driver mutations [98].
PCR has the very useful potential to reliably detect known single alterations such as the presence of an oncogenic virus. Chronic infections by viruses have been described as a cause of several hematologic and non-hematologic neoplasms [99]. The presence of the virus is usually determined by immunohistochemistry [100], immune-fluorometry assay [101], or PCR on the tumor biopsy [100]. It can also be detected either by assaying for circulating viral antigens or antibodies directed against the virus [102] or the presence of the viral genome in the bloodstream [103]. Studies have mainly focused on detecting specific viral oncogenes in the plasma using PCR-derived methods, for example, quantitative PCR (qPCR) [103], digital PCR (dPCR) [104,105], and droplet digital PCR (ddPCR) [106,107]. Detection of the viral genome to assess MRD has shown interesting results, especially in Human Papilloma Virus (HPV)-derived and Epstein-Barr (EBV)-derived cancers. In HPV-related oropharyngeal cancers, Chera and colleagues [105] have demonstrated the prognostic value of detecting HPV-DNA in the plasma of patients after curative-intent treatment. The detection of EBV after curative-intent treatment has in turn also been shown to be predictive of disease relapse in EBV-related nasopharyngeal carcinoma [103,108]. NGS-based methods can also be used for viral detection [109,110], with the added benefit of assessing for multiple subtypes of virus in a single assay. Nevertheless, the sensitivity of this assay in the MRD setting is yet to be determined.

Non-Personalized Methods
To address the potential of ctDNA as a biomarker, many have chosen a nonpersonalized or agnostic approach, i.e., one that is not guided by the specific genomic alterations present in a particular patient's tumor. Gene panels are typically the method of choice, due to their potential to detect any mutations in the neoplasm that are located in the targeted (i.e., sequenced) region of its genome, in one go. As we previously discussed, however, NGS is not the most sensitive method to detect variants present at very low frequency [111,112], necessitating the use of supplementary methods to improve accuracy. For example, to separate real tumor variants from sequencing errors, the use of molecular barcodes (often termed UMIs, for unique molecular identifiers) has been implemented, to lower the error rate of variant-calling and increase the accuracy of VAF estimation [113]. Statistical and bio-informatic tools are used to further separate very low-frequency variants from possible sequencing errors [114].
To detect MRD, two main approaches have been studied: the use of (i) NGS gene panels, either comprehensive or specific to one tumor type, and (ii) PCR for mutations in very specific tumor subtypes.
A synthetic view of the main studies performed using gene panels can be found in Table 2. As shown, it is feasible to use a large panel with over 1000 genes [115], but sophisticated bio-informatic tools and barcoding are essential to detect tumor variants with high confidence. One of the most widely-used strategies is cancer personalized profiling by deep sequencing (CAPP-Seq) [116]. It combines the deep sequencing of several genes (usually >300) implicated in carcinogenesis by the amplification of the target regions, the use of molecular barcoding, and the integrated digital error suppression (iDES) bio-informatic tool [117], specifically designed to detect ctDNA at a very low allele frequency. This technique has shown very promising results when it comes to detecting MRD in specific cancer types. Even without prior knowledge of the genomic mutations in the tumor, this method can detect ctDNA before treatment in >90% of localized NSCLC [116,118]. It is, however, less effective in localized esophageal cancer, detecting ctDNA pre-treatment in only 60% of patients [119]. This may be explained by differences in the genomic landscape and the fact that esophageal cancer seems to shed less tumor DNA than NSCLC [119]. In the post-curative treatment setting, the success rate of the technique drops. Although it shows a good positive predictive value, the negative predictive value still has room for improvement: when ctDNA is detected post-treatment, patients are significantly more likely to experience relapse in the following months [87,120,121]. However, due mainly to technical limitations, not detecting ctDNA after curative-intent treatment does not predict the tumor's definitive elimination (cf/ Table 2). Despite these drawbacks, this technology provides a much more comprehensive view of genomic heterogeneity than tumor-customized methods. It also enhances the understanding of the evolution of the disease as it can detect the emergence of certain clones. Based on these observations, several companies have developed either pan-cancer or cancer-specific panels (Roche Avenio ctDNA targeted kit, Foundation One ® LiquidCDx, Guardant360 ® [122]) to implement this technology in clinical practice. These commercial panels were designed for diagnostic purposes but their use in the detection of MRD still needs to be proven. Other studies have used large panels, with or without UMIs, with interesting results [115,120,123,124]; however, the detection rate of ctDNA undeniably decreases as panel size increases ( Table 2). Eight patients (40%) were positive for preoperative ctDNA. Four patients (20%) were positive for postoperative ctDNA, and this was significantly correlated with histological grade (3 vs. 1 or 2, p = 0.032). Postoperative positivity for ctDNA also predicted shorter recurrence-free survival (RFS). To increase assay sensitivity in particular populations of interest, several groups have designed smaller gene panels, restricting target size in order to divert data capacity towards maximizing sequencing depth, with promising results in disease detection by ctDNA. Reducing panel size increases coverage depth, as the coverage depth depends on the capacity of the sequencing machine to generate a defined quantity of data per run, the number of samples per run, and the target size, i.e., the number of bases to be sequenced. By reducing the panel-target size, coverage depth increases, and tumor variants may be detected at a much lower VAF. For example, Mes and colleagues detected ctDNA in 67% of patients with head and neck squamous cell carcinomas using a 12-gene panel [127]. For MRD detection, however, no studies to date have used a panel of fewer than 127 genes [123].
PCR techniques may also be used in a non-personalized setting, but their use is limited to specific known, recurrent genomic alterations and clinical situations. For example, the only FDA-approved PCR test, cobas ® EGFR mutation test v2 (Roche), detects the presence of 42 EGFR mutations, including the hotspot T790M mutation, in patients with NSCLC previously treated with tyrosine kinase inhibitors [128]. This methodology can thus only be used in a specific group of patients and is not a suitable option for most cancers.

Other Methods
Carcinogenesis can be driven by focal mutations (point mutations, small insertions/ deletions) in the nucleotide sequence and/or by copy number changes in genes or groups of genes [129]. Tumor copy number alterations (CNA) can be detected by different types of sequencing: regular or shallow WGS [130,131], deep WES [132], or panel sequencing [133]. Detection of CNA in the bloodstream has been used for many years in the prenatal diagnosis of fetal abnormalities [134]. Although several bioinformatic tools have been designed to detect CNA in ctDNA [135][136][137], a consensus on methodology is still lacking. Furthermore, the detection of tumor CNA perhaps does not faithfully reflect the persistence of tumor cells, having been found in treated breast cancer patients without evidence of relapse [138].
One hypothesis that has been tested in squamous cell carcinoma of the head and neck (SCCHN) is that the combined assessment of mutations and CNA [127] may improve ctDNA detection. This has not, however, been tested for MRD detection.

Clinical Impact of Detecting MRD with ctDNA and Perspectives
As discussed previously, clinicians wish to detect MRD for two main reasons: (i) to offer treatment escalation when MRD is detected after curative-intent therapy with the hope of improving the cancer outcome and (ii) to de-escalate treatment when MRD is not detected with the goal of decreasing treatment toxicity. Therefore, accurate MRD assessment is likely to play an important role in cancer treatment personalization.
Multiple studies in various cancer types (Tables 1 and 2) have demonstrated a satisfactory positive predictive value of ctDNA detection, as the presence of MRD identified through ctDNA is associated with worse DFS. Nevertheless, several hurdles maintain the negative predictive value of the technique at a low level: detection methods are not sensitive enough to detect ctDNA at very low proportions, new clones not captured by the selected technique can emerge, and the timing of post-treatment sampling can be inappropriate.

Issues
Detecting MRD with ctDNA is still a complex issue as there are no standardized methods ready for implementation in daily practice.
The technical issues have been discussed above. While personalized methods use the most sensitive technologies and can accurately detect tumor variants, sequencing the tumor and then designing patient-specific NGS or PCR assays is costly and time-consuming. Nonpersonalized methods would therefore theoretically be more practical in clinical practice, as one technique would serve multiple patients suffering from the same tumor type and potentially even be used across different tumor types. Until recently, however, sequencing techniques have not been sensitive enough even with barcoding methodology. The risk of not correctly identifying MRD could negatively impact a patient's outcome. Moreover, large panels can yield high numbers of variants, and sorting through which to correctly identify driver variants can be unreliable. To exclude non-pathogenic variants, sequenced reads would ideally be aligned not only to reference genomes but also to the germline DNA of the patient using dedicated pipelines, in addition to excluding variants that are frequent in germline variant databases. This would require germline DNA sequencing for all patients.
Clonal hematopoiesis of indeterminate potential (CHIP) [139] might also be a confounding factor-these variants arise from hematopoiesis and might incorrectly be called tumor variants by bioinformatic tools. One way to separate CHIP variants from tumor variants would be to align the sequencing reads on germline DNA collected from circulating white blood cells, such as lymphocytes or peripheral blood mononuclear cells (PBMC), which would also carry the former. Nevertheless, accurately differentiating nontumor from tumor variants can remain difficult [140][141][142]. Finally, the treatment itself (e.g., radio(chemo)therapy) may induce variants, either in hematopoietic cells (CHIPs) or in cells of the exposed tissue. Separating these variants from true tumor variants is difficult because treatment-induced variants will, like tumor variants, appear at low variant allele frequency and might also occur in genes implicated in carcinogenesis (e.g., TP53). Consortia on guidelines to discriminate tumor variants from other variants are emerging [143], and more databases are expanding their pipelines to sort variants, e.g., Varsome Clinical [144]. However, a systematic bioinformatic pipeline has yet to be proven useful in the clinical setting.
The timing of post-treatment sampling is also crucial when trying to adequately detect MRD. ctDNA has a very short half-life (~30 min) [3] and is dependent on active secretion by tumor cells and cell death. During the course of treatment, a decrease in ctDNA has been observed but a precise and universal wash-out time has not been defined [87]. Circulating DNA (cfDNA) can be released into the bloodstream after surgery, mainly due to tissue trauma, and can remain elevated up to four weeks later [145]. Adjuvant therapies usually start 6-8 weeks after primary curative treatment. Assessing for MRD within one to two weeks of curative treatment is, at least for the moment, not technically feasible, and delaying adjuvant therapy until results are available may lower patient prognoses.
Studies are warranted across different cancer types to determine the most effective means of harnessing ctDNA as a biomarker of MRD. Indeed, a patient can be ctDNApositive several months [146] or just a few weeks before relapse. Some studies have also shown that two positive ctDNA samples collected at two different time points can be more specific than a single positive sample obtained at a unique time point [146].
One of the obstacles to prospective multi-center trials is to perform reproductible assays to detect MRD. Numerous pre-analytical biases exist and include liquid biopsy processing (blood drawing, tubes used, plasma separation, etc.), DNA extraction methodology, sample conservation, equipment used (NGS sequencer, PCR assay), etc.
Despite the issues observed in retrospective studies, prospective trials based on ctDNA as marker of MRD have begun to establish the feasibility of the process.

Current Prospective Trials
Ongoing clinical trials relying on ctDNA as a marker of MRD use four main designs as schematized in Figure 2: prognosis (A), intensification (B), de-escalation (C), and combined (D).
Despite the issues observed in retrospective studies, prospective trials based on ctDNA as marker of MRD have begun to establish the feasibility of the process.

Current Prospective Trials
Ongoing clinical trials relying on ctDNA as a marker of MRD use four main designs as schematized in Figure 2: prognosis (A), intensification (B), de-escalation (C), and combined (D). The first, (A), is a non-interventional design that evaluates the prognostic impact of ctDNA detection. Although ctDNA has already been shown to be a prognostic marker in some cancers, trials in tumors where ctDNA has been underexplored, at least until now, are warranted. Such trials are also necessary to evaluate the accuracy of the ctDNA detection assay under investigation. The optimal time to detect ctDNA following curative-intent therapy, and the impact of this timing on prognosis, can also be evaluated in these trials.
The second trial design (B) is an interventional trial that randomizes patients who are ctDNA-positive after receiving standard of care (SOC) treatment ito adjuvant intensification therapy or placebo. This design investigates whether adjuvant treatment can improve OS and/or PFS if ctDNA is detected. Although this trial design is elegant and can potentially improve survival rates, in the event of a negative outcome, it can be difficult to The first, (A), is a non-interventional design that evaluates the prognostic impact of ctDNA detection. Although ctDNA has already been shown to be a prognostic marker in some cancers, trials in tumors where ctDNA has been underexplored, at least until now, are warranted. Such trials are also necessary to evaluate the accuracy of the ctDNA detection assay under investigation. The optimal time to detect ctDNA following curative-intent therapy, and the impact of this timing on prognosis, can also be evaluated in these trials.
The second trial design (B) is an interventional trial that randomizes patients who are ctDNA-positive after receiving standard of care (SOC) treatment ito adjuvant intensification therapy or placebo. This design investigates whether adjuvant treatment can improve OS and/or PFS if ctDNA is detected. Although this trial design is elegant and can potentially improve survival rates, in the event of a negative outcome, it can be difficult to evaluate whether it is the patient selection that is ineffective (ctDNA may not be a good marker to select those in need of adjuvant treatment) or the adjuvant treatment.
The third trial design (C) is the de-escalation of SOC adjuvant treatment. In some tumor types, adjuvant treatment is routinely administered following curative-intent treatment (surgery), but adjuvant treatment may not be required for ctDNA-negative patients. If these trials show non-inferiority in terms of survival between cohorts, they might redefine SOC adjuvant treatment guidelines and spare a subset of patients the burden of additional treatment.
Finally, it is possible to combine these trial designs (D) so that the prognostic impact of ctDNA detection and the added value of adjuvant treatment in the event of post-treatment ctDNA positivity can both be evaluated in a single study.
A non-exhaustive list of the ongoing prospective and interventional trials can be found in Table 3. The majority of ongoing trials are still 'prognosis' trials designed to prove the value of ctDNA as a biomarker of MRD. However, 'escalation interventional' trials are being launched. One of the first trials to explore these issues is the Australian DYNAMIC trial (ACTRN/12615000381583), in which adjuvant chemotherapy is administered to MRDpositive Stage II colorectal cancer patients after surgery. The COBRA trial (NCT04068103) has almost the same design in Stage II colon cancer and is recruiting at the time of writing. Other ongoing randomized trials are the IM-VIGOR 011 trial (NCT04660344) in invasive urothelial cancer and the Mermaid-1 trial (NCT04385368) in NSCLC, with immunotherapy as adjuvant treatment. The c-TRAK-TN clinical trial (NCT/03145961) in early-stage triple negative breast cancers, currently recruiting in the UK, is a non-randomized trial where adjuvant therapy is given to MRD-positive patients outside SOC. There is no doubt that more trials based on these models will soon be launched. De-escalation trials may soon become available but require careful design due to the risk to patient outcomes.  : non-small cell lung cancer; HNSCC: head and neck squamous cell carcinoma; ctDNA: circulating tumor DNA; MRD: minimal residual disease; PFS: progression-free survival; OS: overall survival. This table has been established using the following keywords on ClinicalTrials.gov: "ctDNA", "Liquid biopsy" with recruitment status: "not yet recruiting", "recruiting", "enrolling by invitation", "active, not recruiting", and "unknown status".

Conclusions
The detection of minimal residual disease (MRD) through circulating tumor DNA (ctDNA) is an important potential application of liquid biopsy that could help personalize a patient's adjuvant treatment. However, several technical issues need to be resolved before these assessments can be implemented in clinical practice. We now need to perform carefully designed randomized interventional clinical trials specifically powered to evaluate if treatment escalation improves cancer outcome in MRD-positive patients and if treatment de-escalation is safe in MRD-negative patients. These future trials will determine the positioning of liquid biopsy in clinical guidelines. For some tumors, there is no doubt that ctDNA will, in the coming years, become a regular biomarker in daily practice.