Ben Fangman, MD, and colleagues provide an overview of the use of circulating tumor DNA levels to detect minimal residual disease in colorectal cancer.
Colorectal cancer (CRC) remains the third most common cancer and second most common cause of cancer death in the United States.1 Traditionally, treatment and prognostication have been based on American Joint Committee on Cancer tumor–node-metastasis staging, relying on the extent of radiographically evident disease; locoregional disease is treated primarily with surgery, with or without adjuvant chemotherapy, and metastatic disease is primarily treated with systemic therapy. The clinical decision regarding use of adjuvant therapy has been based on lymph node status and other clinic pathological risk factors (eg grade or presence of obstruction or perforation).2 However, this risk stratification is incomplete. With increasingly widespread availability of genomic sequencing methods and subsequent greater sensitivity in detecting microscopic disease in peripheral blood, significant research has been focused on utilizing this capability to improve outcomes in CRC, particularly after curative resection.
A term initially coined in the context of hematologic malignancies after induction therapy, molecular residual disease (MRD)—sometimes also referred to as “minimal residual disease “in solid tumors—is defined as molecular evidence of disease detected via cell-free circulating tumor DNA (ctDNA) or another tumor-derived moiety, in the absence of radiographically evident disease.3,4 ctDNA is shed from either a primary tumor or metastatic site via secretion, apoptosis, or necrosis into the bloodstream, where it can subsequently be collected by venous sampling.5 ctDNA has been widely studied for MRD detection in CRC to aid in clinical decision-making.3,5-9
The ability to use ctDNA to identify patients with MRD has important clinical relevance in CRC for several reasons. First, it identifies those who have micrometastatic disease and are therefore at highest risk for relapse after curative-intent therapy.10,11 Second, ctDNA shares the same somatic and epigenetic variants of the tumor from which it was shed, thus providing dynamic information regarding acquired resistance to therapy.12 Lastly, given that ctDNA allows for higher sensitivity of disease detection, it provides the ability to tailor systemic therapy to a greater degree than do standard imaging modalities. In patients without radiographic evidence of disease, MRD detection allows for earlier intervention, potentially when minimal disease, without a substantial tumor protective microenvironment, may still be curable; this is thereby an extension of a risk-stratified adjuvant approach. In patients with known metastatic disease who are undergoing systemic therapy with palliative intent, MRD may be used as a guide for treatment deescalation or treatment holidays, sparing patients from cumulative therapy-related toxicities.
The utilization of ctDNA in the management of patients with CRC has evolved over time. Advancements in methods of polymerase chain reaction (PCR) and genome-wide sequencing have improved sensitivity of detection, and standardization of practices has allowed for widespread access to genomic profiling capabilities. A key challenge that continues to impede the broad clinical applicability of ctDNA in clinical practice, particularly in the MRD setting, is related to false negative results. This is influenced by a variety of factors, most notably low levels of circulating ctDNA; these low levels can occur because of either low tumor burden or, in some cases, decreased shedding of ctDNA, as is seen in peritoneal disease.13
Today, multiple different assays are used clinically, each with differing methodology. There are 2 broad categories of platforms, tumor-informed and tumor-uninformed, each with unique roles, advantages, and disadvantages, as we outline below.
Tumor-informed assays involve sequencing the patient’s tumor specimen, utilizing either biopsy or resection sample, to create a tumor-specific signature of detectable unique aberrations such as single nucleotide variants, short indels, chromosomal breakpoints, and others. Depending on the platform, multiple strategies exist for initial profiling, but they are largely based either on next-generation sequencing (NGS) through whole exome sequencing or whole-genome sequencing (WGS) of the surgical tumor specimen. Once the unique alterations have been identified, PCR-based or personalized NGS panels are typically utilized for serial monitoring of ctDNA. If ctDNA harbors a detectable quantity of the same mutations as the tumor-informed probe, the sample is said to be MRD positive.14 Assays differ in the threshold of number and quantity of ctDNA alterations required to be considered MRD positivity, which impacts the specificity and sensitivity of the resulting assay. The focus on the previously identified alterations in the tumor allows increased depth of coverage of the alterations of interest, which makes these assays more sensitive. Additionally, by using a tumor-informed probe, this approach increases specificity by reducing the impact of sequencing errors or naturally occurring clonal populations not derived from the tumor of interest, such as clonal hematopoiesis of indeterminant potential from white blood cells.
Although a thorough description of available assays is not within the scope of this review, we will describe several of the more commonly utilized assays to provide a perspective on the methodologic differences. There are not yet robust comparisons of the performance of these techniques, although such studies are planned.
The Signatera assay was the first to offer MRD testing commercially, and it is currently the most utilized clinical assay. After whole exome or large targeted panel sequencing, 16 unique alterations are identified and multiplex PCR primer pairs are generated to probe for the identified alterations.14 The test results are either negative or positive for MRD, based upon whether 2 or more of the alterations are present within the circulating plasma; if the results are positive, quantification of the level of detected ctDNA is provided. The RaDaR assay expands the panel to 48 tumor-informed alterations for amplicon-based sequencing.15
SafeSeqS (and the newer methodologic variant, SaferSeqS) is a PCR-based approach that utilizes double-stranded molecular barcoding and hemi-nested PCR for duplex sequencing. This decreases the sequencing error rate and detects mutations at frequencies of 10–5 as well.16 Asaf Zviran, PhD, MSc, and colleagues, created a machine-learning integrated WGS approach using a tumor-informed prior which increased the sensitivity for single nucleotide variations (SNVs) and copy number alterations (CNAs) down to approximately 10–5 in the setting of low tumor burden. One notable limitation to this approach is limited ability to confidently identify driver mutations, which would decrease the sensitivity for mutational profiling for treatment decisions. Thus, the utility of this approach in CRC is largely confined to the MRD detection setting.17
Targeted error correction sequencing (TEC-Seq) and Cancer Personalization Profiling by Deep Sequencing (CAPP-Seq) are hybrid capture–based NGS methods that target multiple regions of the genome known to be associated with driver mutations in a variety of malignancies. These are subsequently deep sequenced. With this approach, the sensitivity of detection of SNV, indels, CNAs, and fusions is greatly increased.18 Phased variant enrichment and detection sequencing (PhasED-Seq) is another approach that utilizes WGS and has reported the lowest limits of detection of a plasma-based approach at 10–6.19
As a group, tumor-informed assays have important limitations. First, there is an approximate 4-week turnaround time for construction of the tumor-specific probe. However, once the tumor-informed probe is created, the initial and subsequent plasma testing can be completed with rapid turnaround time, typically on the order of 1 to 2 weeks. Second, the creation of tumor-specific probes relies on adequate tissue for genomic sampling, which may not always be available. Currently, the theoretical performance benefit of tumor-informed testing justifies this longer testing timeframe, but this will be a key consideration as clinical experience with tumor-agnostic assays accumulates.
Tumor-uninformed assays are plasma-based approaches that do not need to sequence the primary tumor. These assays largely use targeted NGS-based approaches, and each has specific features to improve sensitivity and specificity. We outline the general approaches to these assays below.
To improve its sensitivity, the Reveal assay (Guardant) leverages a fixed gene panel of known CRC-specific mutations with the fact that CRC commonly has aberrant methylated DNA, particularly at C5 positions on cytosine. Aparna Parikh, MD, and colleagues demonstrated a sensitivity of 55% with a single data point, which was improved with serial monitoring to 91%; its specificity was 100% in predicting recurrence after definitive therapy with this approach.9 Collections of larger prospective data sets are underway.
Other methylation-specific strategies, such as the evaluation of the methylation of WIF1 and NPY, genes commonly hypermethylated in CRC, have been attempted, with evidence of prognostic effect in CRC cohorts.20 Although this approach benefits from low cost and simplicity of implementation, neither the sensitivity nor specificity of these assays is as exemplary as those of NGS-based assays.
The advantages of tumor-uninformed approaches include quicker turnaround time, lower costs, and lack of need to rely on tissue and biopsy integrity. An added benefit of these approaches is detecting epigenetic information, which tumor-informed or plasma-based genomic panels cannot do.21
For the applications envisioned with MRD testing, both high sensitivity and specificity are required. High sensitivity is especially critical for deescalation strategies in which patients may be offered less intense therapy than would otherwise be considered, such as shorter duration or delayed adjuvant therapy, or nonoperative management. In contrast, high specificity is required particularly in studies where escalation would be considered, especially when the escalation is to experimental therapy with less well-defined toxicity. Many of the ongoing trials of novel therapeutics, including unapproved agents and cellular therapies, are possible because of the high specificity of the assays, which reduces the risk of treating patients on the basis of false positives when in fact no MRD is present. Several variables (too numerous to examine in depth in this review) impact the performance of these assays. The need to integrate these outcomes into clinical management mandates their timely turnaround, which represents another variable in consideration of the optimal assay.
The effective utilization of ctDNA for MRD detection begins with the optimal timing of sample collection, storage, and processing. Ideally, samples should be collected at least 2 weeks postoperatively (Figure 1) to allow for the large levels of host cell-free DNA (cfDNA) that may be seen postoperatively to decrease, as increased levels can negatively impact sample sensitivity via ctDNA dilution.22 Figure 2 illustrates the optimal process of collecting ctDNA. Samples should be collected in either K2 ethylenediaminetetraacetic blood collection tubes, if white cells are to be able to be separated from plasma within 1 to 2 hours, or Streck cell-free DNA tubes, in which plasma can be stored for days without significant ctDNA degradation.13 Two sequential centrifugations to isolate plasma are recommended to further reduce the level of germline DNA contamination. Although serum was utilized in earlier studies, it is not recommended for current methodologies because serum contains increased cfDNA from hematologic cells due to lysis from clotting; this dilutes ctDNA in the sample, decreasing assay sensitivity. Thus, current methodologies utilize plasma for ctDNA detection.
The rapid progress in the development of ctDNA assays provides substantial opportunities and challenges for integrating them into clinical practice. Several clinical trials are ongoing, as described more fully within this issue; trial results will ultimately define the clinical utility of these assays. Although the clinical oncology community will likely ultimately favor assays with the highest level of clinical data, the rapidly changing field means that studies to understand the relative performance of new assays will be necessary. As the performance of these assays improves, it will be a challenge to integrate their results into clinical practice, as some questions, such as when patients may benefit from adjuvant therapy, may be very dependent on the threshold of detection of the assays. The efforts recently announced by the National Institutes of Health to compare ctDNA assays will be a welcome addition to the landscape and will help define performance more rigorously for future studies.23 Novel methodologies—including integration of fragmentation patterns (fragmentomics), nucleosomic compartment, and methylomics—will supplement existing approaches and further drive the amount of information discernible from ctDNA. Ultimately, the focus on conducting high-quality trials with rigorously validated assays will be required to unlock the full potential of this technology to benefit patients.
DISCLOSURES: BF has nothing to disclose. SK has served as a consultant or paid advisory board member for AbbVie, Amal Therapeutics, AstraZeneca/MedImmune, Bayer Health, Bicara Therapeutics, Boehringer Ingelheim, Boston Biomedical, Carina Biotechnology, Daiichi Sankyo, Eli Lilly and Company, EMD Serono, Endeavor BioMedicines, Flame Biosciences, Genentech, Gilead Sciences, GlaxoSmithKline, HalioDx, Holy Stone, Inivata, Ipsen, Iylon, Jacobio, Jazz Pharmaceuticals, Johnson & Johnson/Janssen, Lutris, Merck, Mirati Therapeutics, Natera, Novartis, Numab Pharma, Pfizer, Pierre Fabre, Redx Pharma, Repare Therapeutics, Servier, and Xilis; received grants from Amgen, Array BioPharma, Biocartis, Daiichi Sankyo, Eli Lilly and Company, EMD Serono, Guardant Health, Genentech/Roche, MedImmune, Novartis, and Sanofi; and has ownership stake in Iylon, Lutric, and MolecularMatch. KR has served as a consultant or paid advisory board member for and receives honoraria from Bayer Health, Daiichi Sankyo, Eisai, and Seagen.
Ben Fangman, MD1; Kanwal Raghav, MD1; and Scott Kopetz, MD, PhD1
1Division of Cancer Medicine and Department of Gastrointestinal Medical Oncology, The University of Texas MD Anderson Cancer Center. Houston, TX
Scott Kopetz, MD, PhD