A Practical Approach to Interpreting Circulating Tumor DNA in the Management of Gastrointestinal Cancers

Abstract

Background

There is accumulating evidence supporting the clinical use of circulating tumor DNA (ctDNA) in solid tumors, especially in different types of gastrointestinal cancer. As such, appraisal of the current and potential clinical utility of ctDNA is needed to guide clinicians in decision-making to facilitate its general applicability.

Content

In this review, we firstly discuss considerations surrounding specimen collection, processing, storage, and analysis, which affect reporting and interpretation of results. Secondly, we evaluate a selection of studies on colorectal, esophago-gastric, and pancreatic cancer to determine the level of evidence for the use of ctDNA in disease screening, detection of molecular residual disease (MRD) and disease recurrence during surveillance, assessment of therapy response, and guiding targeted therapy. Lastly, we highlight current limitations in the clinical utility of ctDNA and future directions.

Summary

Current evidence of ctDNA in gastrointestinal cancer is promising but varies depending on its specific clinical role and cancer type. Larger prospective trials are needed to validate different aspects of ctDNA clinical utility, and standardization of collection protocols, analytical assays, and reporting guidelines should be considered to facilitate its wider applicability.

Background

First discovered in 1948 (1), cell-free DNA (cfDNA) is fragmented DNA released from both normal and diseased cells into plasma. In 1977, Leon et al. found high concentrations of cfDNA in the serum of cancer patients (2). Circulating tumor DNA (ctDNA) is a subset of cfDNA that ranges typically between 90 to 160 base pairs in length and is thought to be released from cellular apoptosis or necrosis of tumor cells from primary or metastatic sites (3, 4).

ctDNA carries specific genomic and epigenetic information reflective of the original tumor, and with advances in detection methods, it is fast emerging as a promising biomarker with clinical applicability across various cancer types. Detection methods range from single-locus detection to whole-exome and whole-genome sequencing. Polymerase chain reaction (PCR)- based assays, such as droplet digital PCR, quantitative PCR, or “Beads, Emulsion, Amplification, and Magnetics” (BEAMing) PCR, are typically used to identify allele-specific changes with a detection threshold down to 0.01% mutant allele fraction (percentage of total copies of DNA being of the mutant form) (5). Next-generation sequencing (NGS) enables broader gene panel testing through amplicon-based sequencing and hybrid capture sequencing with comparable detection sensitivity to digital PCR if error-correction methods are employed (6). In contrast, whole-exome sequencing (WES) or whole-genome sequencing (WGS), although able to provide global coverage of the genome, comes at the cost of a lower detection sensitivity (5). Detection assays are generally either via tumor-informed or tumor-agnostic platforms. Tumor-informed assays require prior identification of specific genomic changes present in an individual patient’s tumor specimen followed by plasma DNA sequencing or PCR to look for these patient-specific mutations, vs tumor-agnostic assays where a predetermined target or panel of targets characteristic of the tumor type of interest is utilized for ctDNA analysis without prior tumor tissue testing.

In gastrointestinal cancer, the potential applications of ctDNA can be seen across different disease stages and along multiple time points of the cancer treatment pathway. However, most data to date are derived from observational studies while evidence on clinical utility from randomized trials demonstrating the benefit of using ctDNA results to guide management is only just emerging. In this review, we aim to discuss factors that influence results and interpretation, and current limitations surrounding the clinical applicability of ctDNA in different types of gastrointestinal cancer, to provide a foundation for informed decision-making for clinicians.

Pre-Analytical and Analytical Considerations and Challenges of ctDNA Testing

To better understand and interpret results of ctDNA clinically, one must appreciate the challenges in the work flow of tissue and blood retrieval, processing, storage, analysis, and interpretation of ctDNA results. In the context of gastrointestinal cancer, confirmation of the histological diagnosis is typically obtained via endoscopic examination and biopsy of the primary tumor, prior to surgical resection of the primary tumor with or without neoadjuvant treatment. Tumor specimens from either the diagnostic biopsy or the resected specimens can be retrieved for genomic analyses. Tumor specimens are typically stored in formalin and less frequently kept as fresh-frozen as part of a research or biobanking protocol. Subject to the assay being used, the amount of tumor and tumor-content from a biopsy specimen required for successful analysis can be variable. WES or WGS require a much larger quantity of tumor tissue than that used for targeted assays. In the case of early-stage cancer, small primary tumors, or specimens with low tumor content (e.g., following good response after neoadjuvant treatment), obtaining sufficient tissue for analysis may be a challenge. This is particularly true for tumors such as pancreatic cancer where tissue biopsy can be difficult to obtain—a scenario where tumor-agnostic assay may be more feasible.

Blood collection for ctDNA analysis can be performed utilizing tubes containing K2 ethylenediaminetetraacetic acid (EDTA), though plasma separation should be carried out within 2 h of sample collection to minimize lysis of white blood cells (WBC) and contamination with leukocyte DNA, which could dilute plasma ctDNA concentration and confound results (7, 8). Alternative options of DNA- and WBC-stabilizing blood tubes such as Streck^® or PAXgene^® cfDNA tubes allow delayed processing of up to 10 to 14 days (7, 9), but are more costly and are not universally available. The amount of plasma required is dependent on the clinical application and assay type with MRD detection for rare mutant molecules requiring larger volumes than genotyping or disease burden assessment in metastatic disease where mutant molecules are more abundant. Germline mutations and baseline somatic mutations are typically gathered via analysis of buffy coat or normal tissue specimens and are used as a reference for identification of tumor-specific somatic mutations. Separated plasma, tumor specimens, and buffy coat or normal tissue are typically stored in a −80°C freezer until DNA extraction.

Another pre-analytical variable stems from timing of blood collection(s), which should be carefully determined and protocolized. Several factors can affect cfDNA release from normal and cancer cells, including anticancer therapies, surgery, radiation, and trauma (10, 11). Studies have demonstrated a relatively short half-life of ctDNA of between 16 min to 2.5 h (12). While surgical resection of the tumor decreases ctDNA levels dramatically, surgery itself stimulates cfDNA release with a lasting effect up to 4 weeks (10). Similarly, initiation of radiotherapy or systemic therapies including chemotherapy, targeted therapy, and immunotherapy may induce a transient spike in ctDNA before a more consistent and steady decline over weeks to months is observed in treatment responders (11, 13). Thus, collection time points should be carefully designed to address the scientific question posed in the clinical context whilst avoiding potential confounders.

The decision on the analytical assay chosen is heavily influenced by the tumor type, treatment plan, clinical question, available specimen(s), assay turnaround time, and cost. Certain tumor types may be better represented by a selection of common driver gene mutations, such as Kirsten rat sarcoma virus (KRAS) mutations in pancreatic ductal adenocarcinoma (PDAC) (14). In fact, many studies evaluating ctDNA in PDAC utilized predetermined gene panels without identifying tumor-specific mutations from biopsies (15–17), as biopsy of the primary tumor can be practically challenging to obtain due to the location of the pancreatic tumor. Cancer types lacking common driver gene mutations require sequencing using a larger gene panel, which is often more costly and time-consuming. Both tumor-agnostic and tumor-informed assays are widely described in the literature. Practically, tumor-agnostic assays are more straightforward with a shorter turnaround time (approximately 1 week) as no prior evaluation of the tumor specimen is required, but consequently they have a lower sensitivity and specificity rate (18). Potential clinical applications of tumor-agnostic assays include cancer screening and in the setting of metastatic disease to allow noninvasive genotyping and detection of emerging resistant clones. However, acquired somatic mutations in the absence of an underlying malignancy can result in false positives. This is typically seen in myeloid cells with a variant allele frequency (VAF) around 2%, and is termed clonal haematopoiesis of indeterminate potential (CHIP) (19). CHIP is a well-described phenomenon, and disease-, age-, and environmental-related factors all contribute; therefore, genes such as TP53, ATM, and CHEK2 with known overlapping associations should be carefully evaluated (20). The effect of CHIP can be minimized with tumor-informed assays whereby a personalized panel is constructed via prior tumor sequencing, and subsequent plasma analysis is performed based on concordant mutations, i.e., a mutation in the plasma is concordant with that identified in the matched tumor. Although more costly and laborious at the beginning, with turnaround times for tumor sequencing and analysis of 4 to 5 weeks, this approach offers significantly greater sensitivity and is therefore preferred for MRD detection, monitoring for disease recurrence, and detecting early response to anticancer therapies. However, the potential shortfall of a tumor-informed approach is the underrepresentation of the spatial and temporal heterogeneity in the limited biopsy material. Thus, there is a potential for false negativity, especially when monitoring patients on active treatment where minor clones not detectable in the tumor specimen exhibit treatment resistance and eventually outgrow treatment-sensitive major clones. At present there is no single ctDNA assay that addresses all potential clinical scenarios, and appropriate consideration of the type and individual assay selected should be given.

In order to produce accurate and meaningful ctDNA results, the reporting method utilized to quantify and interpret results is crucial. There is currently no consensus on reporting guidelines. Allelic mutations are commonly reported in VAF or mutant copies per mL. VAF is a ratio defined by the number of variant reads over the total reads of a particular gene locus, whereas mutant copies per mL is a concentration that describes the absolute level of mutant DNA. While some argue ctDNA should be reported as a concentration, similar to that of tumor markers currently in use, neither of these units is perfect. A study comparing both units noted that assays with a low molecular coverage would underestimate the absolute level of mutant DNA, whereas VAF is less affected (21). On the other hand, VAF is obscured if there is an abundance of wild-type alleles (10, 21). Either way, the measurement of ctDNA is a continuum and a clinical report providing ctDNA quantification in addition to a binary result (positive vs negative) will provide more clinically meaningful information especially for disease monitoring.

Colorectal Cancer

Detection of molecular residual disease post curative-intent treatment

Amongst gastrointestinal cancer types, the most comprehensive clinical evidence for ctDNA utility is in colorectal cancer (CRC). There is now sufficient evidence to demonstrate that ctDNA detection post curative treatment is indicative of MRD in early-stage CRC and is prognostic and correlates with disease recurrence. In a prospective study of 230 patients with stage II colon cancer, there was a marked difference in recurrence rates between the ctDNA-positive (ctDNA+) and ctDNA-negative (ctDNA−) groups. In patients not treated with adjuvant chemotherapy (ACT), postoperative ctDNA+ patients experienced poorer recurrence-free survival (RFS) compared with ctDNA− patients (hazard ratio [HR] 18; 95% CI, 7.9–40; P = 2.6 × 10⁻¹²) (22). In patients who received ACT, ctDNA+ post chemotherapy completion was associated with inferior RFS (HR 11; 95% CI, 1.8–68; P = 0.001), demonstrating that ctDNA detection post curative-intent treatment in stage II colon cancer provides evidence for MRD (22). Similarly, in another study of 96 patients with stage III colon cancer post resection and ACT, postoperative ctDNA+ patients had a significantly poorer RFS (HR 3.8; 95% CI, 2.4–21; P < 0.001). In post ACT ctDNA+ patients, the estimated 3-year recurrence-free interval was only 30% compared with 77% in ctDNA− patients (HR 6.8; 95% CI, 11–157; P < 0.001) (23). In another prospective study evaluating 150 patients with stage I–III colon cancer, detection of ctDNA after surgery was associated with worse RFS (HR 7.0; 95% CI, 2.6–18.9; P < 0.001), and ctDNA was the only significant independent predictor of RFS in the multivariate analysis (24). In the large observational GALAXY (UMIN000039205) study, 1039 patients with stage II–IV resectable CRC with a median follow-up of 16.74 months were analyzed for post-surgery ctDNA and ctDNA+ was strongly associated with recurrence risk (HR 10.0; 95% CI, 7.7–14.0; P < 0.001) and ctDNA+ was the most significant prognostic factor associated with recurrence risk in patients with stage II or III CRC (HR 10.8; 95% CI, 7.1–16.1; P < 0.001). Additionally, postsurgical ctDNA+ identified patients with stage II or III CRC who derived benefit from ACT (HR 6.6; 95% CI, 3.5–12.3; P < 0.001) (25). In rectal cancer, similar results were demonstrated. Studies have found post-surgery ctDNA was an independent prognostic marker for disease recurrence (HR 7.7; 95% CI, 1.6–42.0; P = 0.013) (26), and significantly worse RFS was found in ctDNA+ patients post neoadjuvant chemoradiotherapy (HR 6.6; 95% CI, 2.6–17; P < 0.001) or post-surgery (HR 13; 95% CI, 5.5–31; P < 0.001) (27). Collectively, these results support the role of ctDNA in detecting MRD in localized CRC which is prognostic for early recurrence.

Prediction of pathological response post neoadjuvant therapy

In locally advanced rectal cancer (LARC), guidelines recommend treatment via neoadjuvant radio-/chemoradio-therapy or total neoadjuvant therapy followed by total mesorectal excision with or without further ACT. There has been increasing advocacy towards a “watch-and-wait” organ preservation approach to spare patients from the morbidity of major rectal surgery. However, current tools for predicting pathological complete response (pCR), such as magnetic resonance imaging, positron emission tomography imaging, and colonoscopy, lack accuracy. Murahashi et al. explored the clinical utility of ctDNA in predicting pCR post neoadjuvant therapy (NAT) found that the change in ctDNA status between baseline and post NAT was an independent predictor of pCR (odds ratio [OR] 7.4; 95% CI, 1.2–144; P = 0.028) (26). However other studies found inconsistent evidence to support this. A study of advanced rectal cancer of 159 patients found that ctDNA− patients post neoadjuvant chemoradiation had a higher percentage of pCR than ctDNA+ patients (21% [28/132] vs 8% [1/12]; P = 0.46), yet 79% of ctDNA− patients failed to achieve pCR (27). In the STELLAR (NCT02533271) trial, patients with a clearance of ctDNA post NAT from baseline (post NAT ctDNA/baseline ctDNA <2%) were more likely to achieve pCR or clinical complete response than the ones without (44% [7/16] vs 5% [1/19]; P = 0.013), however, ctDNA− itself was not associated with pCR (28). Currently, evidence suggests that ctDNA− patients post NAT may have a higher likelihood of achieving good pathological response than ctDNA+ patients; however, ctDNA− does not predict pCR. Findings from these observational studies may provide a foundation for designing future interventional studies.

Guiding adjuvant therapy

Based upon the evidence in detecting MRD, the utility of ctDNA in risk stratifying patients and guiding ACT in colon cancer has been explored. The randomized phase II multicenter DYNAMIC (ACTRN12615000381583) II trial, designed to assess whether a ctDNA-guided approach in stage II colon cancer could reduce the use of ACT without compromising recurrence risk, achieved its primary aim. Of the 455 patients randomized in a 2:1 ratio to ctDNA-guided management vs standard management, fewer patients in the ctDNA-guided cohort received ACT (15% vs 28%; relative risk 1.8; 95% CI, 1.3–2.7) compared with the standard management cohort. Importantly, the 2-year RFS of the ctDNA-guided cohort was noninferior to the standard management cohort (93.5% vs 92.4%, absolute difference 1.1% points; 95% CI, −4.1–6.2 [non-inferiority margin, −8.5% points]) (29). Eighty-seven percent of postoperative ctDNA+ patients effectively treated with ACT were converted to a ctDNA− status with a significantly improved RFS (HR 55.7; 95% CI, 5.8–532.2; P < 0.001), whilst the median time to disease recurrence for patients who remained ctDNA+ post ACT was only 5.3 months (30). Further trials of ctDNA in early-stage colon cancer are currently underway and will provide further evidence regarding selection, duration, and intensity of ACT in ctDNA+ vs ctDNA− patients.

Early detection of disease recurrence during surveillance

Serial ctDNA monitoring can be used during surveillance as an additional marker for disease relapse. A prospective study of 150 stage I–III colon cancer patients evaluated serial ctDNA at 4-monthly intervals during patient follow-up. Detection of ctDNA by mutation tracking was associated with poorer disease-free survival (DFS) (HR 8.0; 95% CI, 1.8–36.0; P = 0.006). Of the 17 patients who experienced disease relapse, 8 (47.1%) were initially postoperatively ctDNA+, and with serial ctDNA tracking, an additional 6 patients (35.3%) became ctDNA+ over time. Detection of ctDNA also preceded radiological recurrence with a median lead time of 11.5 months (range 3 to 18 months) (24). In another study of 168 patients with resected stage III CRC who underwent serial ctDNA surveillance, ctDNA was able to detect recurrence before standard-of-care computed tomography (CT) imaging with a median lead time of 9.8 months (31). Authors also reported on follow-up of 13 patients who were ctDNA+ post-surgery and received ACT with serial ctDNA testing. At 3 years’ post-treatment the only patients who had not experienced disease recurrence were the 3/13 (23%) patients who achieved persistent ctDNA clearance with effective ACT (31). A cautionary note with earlier detection of molecular disease by ctDNA during surveillance in these retrospective observation studies is the timing of ctDNA analysis relative to standard-of-care radiological assessments. The MD Anderson INTERCEPT program observed that of 86 patients who were ctDNA+ during surveillance reflex imaging revealed concomitant new metastases in 46 (53%) (32).

Assessment of therapy response for advanced disease

Treatment response may also be assessed by monitoring ctDNA levels. This idea is supported by a study of 53 patients with metastatic colorectal cancer (mCRC), where ctDNA were measured at 3 time points: before and 3 days after commencement of chemotherapy treatment, and immediately prior to cycle 2 of chemotherapy. A significant reduction (≥10- fold) in ctDNA pre-cycle 2 correlated with CT responses at 8 to 10 weeks (OR 5.3; 95% CI, 1.4–19.9; P = 0.016), and this was also associated with a trend towards improved progression-free survival (PFS) (median PFS 14.7 vs 8.1 months; HR 1.9; 95% CI, 0.6–5.6; P = 0.266) (13). Similarly, van’t Erve and colleagues defined molecular response in mCRC as elimination of more than 98% of ctDNA upon treatment based on the highest mutant allele fraction (and molecular nonresponse as patients with an increase in ctDNA or elimination of less than 98% of ctDNA), and found molecular responders showed a significantly longer overall survival compared to molecular nonresponders (median 59 vs 27 months; HR = 2.4; 95% CI, 0.9–6.1; P = 0.039) (33). Lueong et al. found that circulating KRAS mutations and demonstrated ctDNA clearance 15 to 22 days after treatment commencement conferred better disease control (P = 0.008), improved overall survival (OS) (log-rank P = 0.002), and PFS (log-rank P = 0.002) and that ctDNA+ at this time point was the most significant independent prognostic factor on OS in multivariable analysis (HR 2.3; 95% CI, 1.4–3.3; P < 0.001) (34). These findings demonstrate that ctDNA dynamics, in addition to absolute results, carry important prognostic information and have the potential to complement current methods of evaluating treatment response and better personalize clinical decision-making.

Detection of treatment resistance and guiding rechallenge therapy

Right- and left-sided CRC are slightly different in their histology and morphology (35). Anti-EGFR therapy is more beneficial in left-sided primary, RAS- and RAF-wildtype mCRC than in right-sided disease (36). Resistance to therapy typically develops due to treatment selection pressure and clonal evolution. Here, serial ctDNA testing can be used to track the early development of treatment-resistant clones and to guide a rechallenge strategy. Clonal populations harboring KRAS mutations, which are resistant to anti-EGFR therapy, can be detected via ctDNA 5 to 10 months prior to radiological evidence of disease progression (37, 38). Interestingly, the level of mutant KRAS clones declines upon withdrawal of anti-EGFR therapy (39), thus providing a window of opportunity to rechallenge with similar therapies. Supporting this hypothesis, 2 phase II trials demonstrated that a ctDNA-guided rechallenge with anti-EGFR therapy led to further objective responses exceeding that of standard third-line treatment options (40, 41).

Esophago-Gastric Cancer

Disease screening

Compared to colorectal cancer, the clinical utility of ctDNA in esophago-gastric cancer is less established. The majority of evidence resides in the role of ctDNA in detecting MRD and guiding therapy in stage IV disease. Its role in screening and surveillance, on the other hand, is supported by only a limited number of studies. Current literature demonstrates inconsistent evidence in disease screening of esophago-gastric cancer using ctDNA. One study that examined all stages of gastric cancer proposed the use of cfDNA as a screening biomarker (42). By using a concentration quantified by a branched DNA-based Alu assay, Qian et al. found a significant difference in cfDNA concentrations between patients with gastric cancer, benign gastric disease, and healthy controls (1475.92, 244.42, and 181.90 ng/mL respectively, P < 0.05). Moreover, this significant difference remained when only stage I disease was compared to the control group. Their reported sensitivity of 79.0% and specificity of 91.8% in diagnosing gastric cancer appeared promising. However, other studies have reported baseline ctDNA detection rates of 21% in locally advanced nonmetastatic gastric cancer (43) and 36% in all stages of esophageal adenocarcinoma (44). The variability between the 3 studies can be explained by the different assays utilized and the relative sensitivity of each, the first study evaluating cfDNA vs the latter 2 studies measuring ctDNA with a tumor-informed assay. However, in the absence of further studies, screening is not a validated use of ctDNA in esophago-gastric cancer in high-risk patients.

Detection of minimal residual disease post curative-intent treatment

The use of ctDNA in detecting MRD in esophago-gastric cancer is supported by multiple studies. The largest cohort study of 1630 patients, including gastroesophageal adenocarcinoma of all stages, demonstrated that in the subset of patients treated with curative-intent, ctDNA detection postoperatively translated to an inferior median DFS (HR 0.1; 95% CI, 0.01–1.1; P = 0.03) (45). In another study of localized esophageal adenocarcinoma and squamous cell carcinomas, Azad et al. demonstrated that patients who were ctDNA+ post chemoradiation experienced a shorter disease-specific survival (HR 23.1; 95% CI, 1.9–273.5; P < 0.001) and ctDNA+ preceded radiological disease relapse by an average of 2.8 months (46). Yang et al. examined 46 stage I–III gastric cancer patients and again demonstrated that a postoperative ctDNA+ status was strongly associated with increased risk of relapse (100% recurrence in the ctDNA+ vs 32% in the ctDNA− group, P = 0.002), and poorer DFS (HR 6.6; 95% CI, 8.3–208.5; P < 0.001) and OS (HR 6.0; 95% CI, 3.8–138.1; P < 0.001). The sensitivity and specificity of postoperative ctDNA in predicting 30-month disease recurrence were 39% and 100%, respectively (47).

Early detection of disease recurrence during surveillance

In current clinical practice, a sensitive and specific plasma-based tumor marker for esophago-gastric cancer is lacking to help survey disease relapse and recurrence. ctDNA could potentially be applied for this purpose. In the aforementioned study, Yang and colleagues also performed longitudinal ctDNA evaluation in the same cohort of gastric cancer patients. They found that of patients who experienced disease recurrence, 41% (7/17) were ctDNA+ postoperatively, and 84% (16/19) had ctDNA detected in at least one serial longitudinal sample. In this cohort, detection of ctDNA preceded radiographic recurrence by a median of 6 months. In patients who remained disease-free, 100% (21/21) and 96% (24/25) were ctDNA− at the initial postoperative ctDNA time point, and any serial longitudinal time point respectively (47). Another study of esophageal adenocarcinoma showed patients who became ctDNA+ during surveillance had a much higher risk of relapse (HR 7.9; 95% CI, 1.2–50.1; P = 0.02) (44). These studies highlight the importance of serial measurement and evaluation of ctDNA dynamics, and raise the opportunity to incorporate ctDNA evaluation into current clinical surveillance modalities.

Guiding targeted therapy

Systemic therapies are often used in advanced esophago-gastric cancers, and molecular biomarkers, if identified, can guide therapeutic decision-making. For example, in human epidermal growth factor receptor 2 (HER2) positive metastatic gastric cancer, the addition of anti-HER2 to standard chemotherapy improved PFS and OS (48). Studies have found that ctDNA analysis of HER2 status demonstrates high concordance with tumor tissue testing (49). Furthermore, ctDNA analysis allows for detection of HER2 copy number changes (49), and clonal evolution of HER2 amplification (50), which may herald resistance to therapy. The phase II clinical expansion-platform type-II PANGEA (NCT02213289) trial evaluated personalized antibodies in the treatment of stage IV gastroesophageal adenocarcinoma, utilizing a combination of ctDNA-based NGS and immunohistochemistry to personalize therapy at initial diagnosis, and then serially in up to 3 lines of anticancer treatment. Sixty-eight patients received tailored therapy against programmed cell death protein 1 (PD-1), epidermal growth factor receptor (EGFR), HER2, fibroblast growth factor receptor 2 (FGFR2) or vascular endothelial growth factor receptor 2 (VEGFR2), and the study met its primary end point of an improved 1-year survival of 66% (45/68 patients; 95% CI, 54%–76%; one-sided P = 0.002) exceeding the 50% historical control rate (51). This study highlighted the unique utility of ctDNA for biomarker profiling both at baseline and over time to accurately reflect spatial tumor heterogeneity and allow matching of personalized molecular therapies to improve survival outcomes. Currently, however, immunohistochemistry for HER2 testing remains the standard of care and more evidence is needed to support the use of ctDNA molecular testing in personalized treatment.

Pancreatic Ductal Adenocarcinoma (PDAC)

Utility of ctDNA In the locally advanced PDAC

There is clinical equipoise in the definition and management of locally advanced PDAC with significant variations across countries and institutions. The traditional treatment is to offer upfront surgery followed by ACT, however, the use of perioperative (neoadjuvant) chemotherapy in borderline resectable or resectable disease is being advocated (52), and trials evaluating the efficacy of NAT in these clinical setting are ongoing (53). In this context, ctDNA could be used to identify those patients who would derive the most benefit from NAT. In a study of 42 patients who underwent upfront resection of their PDAC, KRAS-mutated ctDNA was detected in 62% (23/37) of patients preoperatively. Preoperative ctDNA+ was associated with increased risk of recurrence (HR 4.1; 95% CI, 1.8–9.0; P = 0.002) and poorer OS (HR 4.1; 95% CI 1.6–10.5; P = 0.015) (15). This suggests that preoperative ctDNA+ predicts for a high-risk cohort and that ctDNA status could be utilized to guide NAT selection.

Postoperative ctDNA status is also indicative of MRD and strongly predicts for disease recurrence in PDAC. Lee et al. evaluated the prognostic value of postoperative ctDNA status, and demonstrated that PDAC patients who were postoperative ctDNA+ had a 100% positive predictive value at a median follow-up of 38.4 months (all 13 patients who were postoperatively ctDNA+ experienced disease recurrence), with a specificity and sensitivity of 100% and 57%, respectively (15). This has also been observed by several other groups (17, 54–56). Kitahata et al. evaluated the use of ctDNA in 27 borderline resectable PDAC patients and found that postoperative ctDNA status was associated with poorer OS (HR 5.0; 95% CI, 1.2–20.5; P = 0.025), whereas this association was not observed according to pretreatment or post NAT ctDNA status (54). In a study of 27 patients with early and locally advanced PDAC, postoperative ctDNA+ was associated with worse DFS (HR 5.2, P = 0.019), which was not observed with preoperative ctDNA status (56). Those studies collectively support the prognostic utility of postoperative ctDNA status to identify patients at high risk of disease relapse.

Prediction and assessment of therapeutic response

ctDNA in PDAC may also be used to assess treatment response. In a study of 28 patients with borderline resectable PDAC undergoing modified FOLFIRINOX NAT, authors used ctDNA to detect mutations in specific DNA damage repair (DDR) genes and found that patients with mutated DDR genes experienced better PFS than those without DDR gene alterations detected (median PFS 26.6 vs 13.5 months, P = 0.004). Moreover, detection of KRAS mutations at baseline correlated with worse OS (median 8.5 months vs not reached, P = 0.003) (57). In another study, Kruger et al. evaluated mutated KRAS ctDNA in combination with tumor markers of cancer antigen 19-9 (CA 19-9), carcinoembryonic antigen (CEA) and cytokeratin 19 fragment (CYFRA 21-1) in 54 patients with advanced PDAC and showed that a decrease in ctDNA levels during chemotherapy was an early indicator of response. In contrast, the other tumor markers did not demonstrate such correlation, suggesting that mutated KRAS ctDNA is a superior biomarker in assessing therapeutic response and disease progression (16).

ctDNA: Is It Ready for Prime Time?

Table 1 summarizes the current clinical utility of ctDNA in gastrointestinal cancers. ctDNA has multiple potential clinical applications including use in population/high-risk cohort screening, detection of MRD after definitive local treatment, surveillance after curative-intent treatment to predict early disease relapse, genotyping advanced disease to personalize therapy, early assessment of treatment efficacy, and identifying mechanisms of resistance to therapy. At present, the areas with the strongest evidence to consider incorporating ctDNA analysis into clinical practice are MRD assessment post definitive treatment for early-stage disease, and for genotyping in advanced disease.

There is strong evidence that ctDNA detects MRD post curative-intent treatment, and therefore has the potential to be an adjunct biomarker to risk stratify patients and allow for tailored management pathways including treatment escalation in ctDNA+ patients vs de-escalation in ctDNA− patients. As shown in the DYNAMIC II trial, utilization of ctDNA to risk stratify stage II colon cancer significantly reduced the use of unnecessary adjuvant chemotherapy without compromising RFS outcomes (29). Several phase II/III trials are currently underway to assess ctDNA guided adjuvant therapy in different stages of colorectal cancer, such as the DYNAMIC (ACTRN12617001566325; ACTRN12617001560381) and CIRCULATE (NCT04089631) trials (59–61). Similarly, in a post resection PDAC cohort, the DYNAMIC-pancreas study aims to assess the use of ctDNA in guiding adjuvant therapy (ACTRN12618000335291). Given that DFS is a well-validated surrogate end point for overall survival, especially in early-stage colon cancer, all of these ctDNA-guided MRD studies are using DFS as their primary efficacy end point, which should be sufficient to change practice. Results from these and further prospective randomized trials will confirm the clinical utility in different gastrointestinal cancers as well as various stages of the disease.

The next logical consideration is the utility of ctDNA analysis for MRD monitoring and surveillance in early-stage cancers that have undergone definitive treatment. Multiple studies in colorectal and gastric cancer support the use of serial ctDNA measurement during surveillance post curative-intent treatment, with persistent positive or conversion of negative to positive ctDNA status preceding clinical and radiological disease relapse (24, 31, 47). However, it is unknown whether earlier treatment based upon detection of MRD during surveillance changes the natural history of the disease and ultimately survival outcomes. The Danish IMPROVE-IT2 (NCT04084249) trial is a randomized controlled trial investigating ctDNA-guided surveillance in stage II and III CRC to determine whether incorporating ctDNA analysis results in a higher fraction of patients with recurrent disease receiving curative-intent or oligo-metastatic–directed treatment as compared to standard-of-care Danish surveillance (62). Contrastingly, in PDAC and esophago-gastric cancers, where the role of endoscopic and radiologic surveillance post definitive treatment has not prospectively been proven to improve survival outcomes, the role of incorporating ctDNA analysis into surveillance is even less clear.

Multiple prospective studies have validated the high accuracy of ctDNA for tumor genotyping in advanced gastrointestinal cancers when compared with tissue-based PCR and NGS testing (63). As such, validated and adequately sensitive ctDNA testing may be considered in clinical practice for advanced disease genotyping where tissue testing is not feasible or where significant delays will affect therapeutic outcomes. However, local restrictions on the availability of targeted treatments requiring tissue rather than ctDNA confirmation of genomic aberrations may limit its clinical use. The opportunity for ctDNA analysis in advanced gastrointestinal cancers to more accurately reflect intra-patient spatial and temporal tumor heterogeneity supports its use in longitudinal monitoring for response to treatment as well as the detection of emergent resistant mutations. Despite promising early results, there is currently insufficient evidence to support the routine use of ctDNA analysis for monitoring of advanced gastrointestinal cancer treatment. In particular, whether changing patients’ treatment based upon the early detection of changes in ctDNA dynamics or a resistant mutation through ctDNA analysis affects survival outcomes is yet to be confirmed.

Although much of cancer medicine focuses upon treatment of cancer, screening and earlier diagnosis in asymptomatic and/or high-risk populations may result in higher disease cure rates. As discussed, the role of ctDNA in cancer screening has been explored with the hope of developing a sensitive and noninvasive test, especially in esophago-gastric cancers and PDAC, which lack an effective community-based screening protocol in the Western population. The challenge remains to find a multicancer screening test with very high specificity and clinically meaningful sensitivity to detect early-stage disease (especially stage I and II cancers) for asymptomatic population-based screening to reduce mortality. There are ongoing large-scale efforts to both improve multicancer early detection methods and demonstrate the feasibility and utility of these tests.

Our review mainly focuses on genomic mutational analysis, but it is notable that epigenetic markers such as DNA methylation patterns are becoming a powerful tool to expand the application of ctDNA analysis as a cancer detection tool. Additionally, given there is little or no validation of a negative ctDNA result and its clinical prediction, i.e., negative prediction of cancer, the combination of genetic and epigenetic information may increase the sensitivity of using mutation-based ctDNA testing in clinical practice. For example, Parikh et al. reported that integrating epigenomic signatures with genomic alterations increased test sensitivity by 25% to 36% for colorectal cancer treated with curative intent (64).

Greater Health Economic Considerations

The health economic impact of ctDNA analysis in cancer management is another consideration, and the incorporation of cost-effective analyses is an integral part to ongoing research on the clinical utility of ctDNA in gastrointestinal cancer management. The obvious theoretical advantage of ctDNA is to personalize therapeutic decisions to maximize the chance of choosing an effective treatment whilst avoiding unnecessary treatment toxicity and improving health outcomes. However, to do so, confirmation of the clinical validity of ctDNA analysis guiding therapeutic decision-making resulting in improved health outcomes is required. This then raises the question of cost of the ctDNA assay and its funding. To ensure equity of a validated predictive diagnostic biomarker, reimbursement is important. The downstream financial benefits of more efficient treatment strategies and improved health outcomes should also be evaluated to understand the health economic impact of ctDNA analysis.

Conclusion

There is now strong evidence to support the predictive validity of ctDNA in gastrointestinal cancers. As a noninvasive blood test, ctDNA evaluation has the potential to be a sensitive companion biomarker in disease detection, testing for MRD, assessment of treatment response, and guiding systemic therapies. However, further clinical trials are required to confirm its clinical utility in disease screening, early detection of recurrence, monitoring, and guiding therapy, especially with regards to improved outcomes. Additionally, health economic cost-effective analyses and standardized reporting strategies are required before it can be incorporated into routine practice. Future studies should aim to explore the utility of ctDNA in those under-evaluated clinical aspects and to incorporate discussion of health economics to facilitate earlier implementation of ctDNA into everyday practice.

Nonstandard Abbreviations

ctDNA, circulating tumor DNA; MRD, molecular residual disease; cfDNA, cell-free DNA; PDAC, pancreatic ductal adenocarcinoma; VAF, variant allele frequency; CRC, colorectal cancer; ACT, adjuvant chemotherapy; RFS, recurrence-free survival; HR, hazard ratio; pCR, pathological complete response; NAT, neoadjuvant therapy; DFS, disease-free survival; PFS, progression-free survival; OS, overall survival.

Human Genes

KRAS, Kirsten rat sarcoma virus; HER2, human epidermal growth factor receptor 2; PD1, programmed cell death protein 1; EGFR, epidermal growth factor receptor; FGFR2, fibroblast growth factor receptor 2; VEGFR2, vascular endothelial growth factor receptor 2; TP53, tumor protein 53; ATM, ataxia-telangiectasia mutated; CHEK2, checkpoint kinase 2.

Author Contributions

The corresponding author takes full responsibility that all authors on this publication have met the following required criteria of eligibility for authorship: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; (c) final approval of the published article; and (d) agreement to be accountable for all aspects of the article thus ensuring that questions related to the accuracy or integrity of any part of the article are appropriately investigated and resolved. Nobody who qualifies for authorship has been omitted from the list.

Zexi Allan (Conceptualization-Equal, Data curation-Equal, Formal analysis-Equal, Methodology-Equal, Validation-Equal, Writing—original draft-Equal, Writing—review & editing-Equal), David ShiHao Liu (Conceptualization-Equal, Investigation-Equal, Methodology-Equal, Resources-Equal, Supervision-Equal, Writing—original draft-Equal, Writing—review & editing-Equal), Margaret Lee (Conceptualization-Equal, Investigation-Equal, Methodology-Equal, Resources-Equal, Supervision-Equal, Writing—original draft-Equal, Writing—review & editing-Equal), Jeanne Tie (Conceptualization-Equal, Data curation-Equal, Formal analysis-Equal, Methodology-Equal, Resources-Equal, Supervision-Equal, Validation-Equal, Writing—review & editing-Equal), Nicholas Clemons (Conceptualization-Equal, Supervision-Equal, Validation-Equal, Writing—review & editing-Equal)

Authors’ Disclosures or Potential Conflicts of Interest

No authors declared any potential conflicts of interest.

References

Author notes