Purpose: We investigated whether detection of ctDNA after resection of colorectal cancer identifies the patients with the highest risk of relapse and, furthermore, whether longitudinal ctDNA analysis allows early detection of relapse and informs about response to intervention.
Experimental Design: In this longitudinal cohort study, we used massively parallel sequencing to identify somatic mutations and used these as ctDNA markers to detect minimal residual disease and to monitor changes in tumor burden during a 3-year follow-up period.
Results: A total of 45 patients and 371 plasma samples were included. Longitudinal samples from 27 patients revealed ctDNA postoperatively in all relapsing patients (n = 14), but not in any of the nonrelapsing patients. ctDNA detected relapse with an average lead time of 9.4 months compared with CT imaging. Of 21 patients treated for localized disease, six had ctDNA detected within 3 months after surgery. All six later relapsed compared with four of the remaining patients [HR, 37.7; 95% confidence interval (CI), 4.2–335.5; P < 0.001]. The ability of a 3-month ctDNA analysis to predict relapse was confirmed in 23 liver metastasis patients (HR 4.9; 95% CI, 1.5–15.7; P = 0.007). Changes in ctDNA levels induced by relapse intervention (n = 19) showed good agreement with changes in tumor volume (κ = 0.41; Spearman ρ = 0.4).
Conclusions: Postoperative ctDNA detection provides evidence of residual disease and identifies patients at very high risk of relapse. Longitudinal surveillance enables early detection of relapse and informs about response to intervention. These observations have implications for the postoperative management of colorectal cancer patients. Clin Cancer Res; 23(18); 5437–45. ©2017 AACR.
This article is featured in Highlights of This Issue, p. 5323
This study evaluated the potential clinical implications of monitoring of circulating tumor DNA (ctDNA) for 3 years after resection of colorectal cancer. The clinicopathologic relapse risk markers currently used to inform adjuvant treatment decisions indicate only a propensity for relapse, whereas postoperative detection of ctDNA provided direct evidence of residual disease and identified patients with very high risk of relapsing. ctDNA was detected in all relapsing but not in any nonrelapsing patients. Serial ctDNA analysis predicted clinical relapse with a lead time of 9.4 months compared with standard-of-care imaging, which might be sufficient to change patient management. Potentially, a combination of ctDNA and radiologic assessments may increase the fraction of relapse patients eligible for curative intervention. Changes in ctDNA levels induced by relapse intervention correlated with changes in tumor volume and may prove clinically useful, particularly for patients with nonmeasurable disease by CT imaging, for whom an assessment of response would greatly assist in decision making.
About 1.3 million cases of colorectal cancer are diagnosed annually worldwide (1). Approximately two-thirds of patients present with localized and potentially curable disease at diagnosis (stages I-III by TNM; refs. 2, 3). Despite curatively intended surgery, a large fraction of the patients (30%–50%) experience relapse (4). Identifying the patients at high risk of relapse, and accordingly who should be offered adjuvant therapy, remains a clinical dilemma. The current approach for estimating relapse risk is based on TNM staging. Overall survival (OS) benefit of adjuvant therapy has been documented for the stage III patients (node positive). But also a small subset of stage I and larger subset of stage II patients relapse and could potentially benefit from adjuvant therapy (5). In practice, however, it has been difficult to document a benefit of adjuvant therapy for stage I and II patients, even when TNM classification was refined by additional relapse-risk markers, such as T4 extension, bowel perforation, or obstruction, inadequate nodal sampling, poorly differentiated histology, and lymphovascular invasion (6, 7). Consequently, benefit remains to be conclusively demonstrated. The challenge is in part due to overall low risk of relapse in stages I and II and that the added markers only modestly affects relapse risk. Better markers of relapse-risk would allow a high-risk subgroup to be identified, the selection of which could enrich studies designed to demonstrate adjuvant therapy benefit.
A better relapse-risk marker may potentially also identify the many stage III patients that are being overtreated today as it is well documented that even without adjuvant therapy a large fraction of stage III patients (∼40%) become long-term survivors (8). The development of a novel sensitive approach for assessing whether occult residual disease is present after surgery is highly desirable, and could potentially be used to identify those patients, who should be offered adjuvant therapy.
Regardless of whether patients have received adjuvant therapy, early detection of relapse during follow-up is associated with improved survival in patients with early-stage colorectal cancer (9–11). However, the biomarker now used for standard of care follow-up, carcinoembryonic antigen (CEA), has limited sensitivity and specificity (12–15). CT imaging is also part of standard follow-up programs and do improve detection of relapse, but is associated with radiation exposure, has a limited sensitivity for small lesions, and a high rate of false positives (16, 17). Accordingly, there is a clinical need for an improved approach for longitudinal noninvasive monitoring of the tumor burden, such an approach would expectedly also benefit the assessment of response to relapse intervention. The current gold standard for assessing the response to intervention is the CT-imaging based Response Evaluation Criteria in Solid Tumors (RECIST). RECIST limitations include poor inter- and intraobserver reproducibility and limited categorization (18). In addition, a minority of patients does not have imaging-based “measurable disease.”
Sequencing of tumor DNA has shown that nearly all colorectal cancer genomes are characterized by somatic structural variants (SSV) and somatic point mutations (SPM; refs. 19, 20). Emerging results have shown that tumor-specific DNA mutations can be detected in body fluids, including plasma, allowing for noninvasive molecular characterization of tumors (21–24). In addition, the short half-life of approximately 2 hours for circulating tumor DNA (ctDNA) indicates that ctDNA is a useful real-time marker of changes in tumor burden (25). However, further work is needed to determine how well the level of ctDNA correlates with tumor burden.
In the current work, we identified both SSVs and SPMs and applied these as ctDNA markers for analysis of serial postsurgery plasma samples from colorectal cancer patients treated with curative intend. Our ctDNA analyses demonstrate the potential for early detection of colorectal cancer, assessment of postsurgery residual disease in two independent cohorts, early detection of relapse, and for assessing response to therapeutic intervention. In summary, our analyses provide predictors of clinical outcome in colorectal cancer and have promising implications for personalized patient management during postsurgery follow-up.
Materials and Methods
This prospective observational study recruited colorectal cancer patients undergoing surgical treatment at the Department of Surgery, Aarhus University Hospital (Aarhus, Denmark). Fresh frozen (FF) tumor specimens and germline specimens (from peripheral blood) were collected at inclusion. Blood samples were collected at day 0 (pre-op), 8, 30, and every three months until death, patient withdrawal from the study, or month 36. The patients received treatment and follow-up according to Danish National Guidelines. The study was approved by the Committees on Biomedical Research Ethics in the Central Region and the Capital Region of Denmark (1-10-72-223-14 and H-15010930) and was performed in accordance with the Declaration of Helsinki. All participants provided written informed consent.
Isolation and quantification of DNA
Prior to isolation of tumor DNA the cancer content in each biopsy was assessed by evaluation of hematoxylin and eosin–stained sections cut before and after those used for extraction. Tumor samples had a median cancer cell content of 79% (range 60%–95%). DNA was extracted from fresh frozen tissue using the Puregene DNA purification kit (Gentra Systems), from formalin-fixed paraffin-embedded (FFPE) tissue using QiAamp DNA FFPE tissue kit (Qiagen), or from 2–8 mL plasma by QIAamp DNA Blood Midi Kit or QIAamp Circulating Nucleic Acid Kit (Qiagen). Assessment of plasma DNA purification efficiency, lymphocyte DNA contamination, and quantification of the cell-free DNA (cfDNA) content was carried out using digital droplet PCR (ddPCR) as described previously (21). The ddPCR assays are listed in Supplementary Table S1.
Identification of somatic structural variants in fresh frozen tumor tissue
We have previously described the establishment of a low-pass mate-pair sequencing approach for identification of SSVs from tumor tissue (21). Briefly, SSVs were identified by identification of discordant mapping mate-pair sequencing reads. Somatic variants were distinguished from germline variants by comparing mate-pair data derived from matched tumor and germline samples. For summary statistics of the sequencing, see Supplementary Table S2.
Identification of hotspot mutations in KRAS in fresh frozen tumor tissue
High-resolution melting analysis (HRMA) was used for mutational screening of primary tumor tissue for the presence of SPM in KRAS codon 12, 13, and 61. PCR was carried out in LightScanner master mix (Idaho Technology) containing 0.2 μmol/L forward and reverse primer and 20 ng DNA from FF tumor samples in a 10-μL reaction volume. Samples that deviated from the wild-type control were validated by Sanger sequencing.
Identification of tumor-specific mutations by whole-exome sequencing
Whole-exome sequencing (WES) was performed on matched FFPE tumor DNA (1,000 ng) and germline DNA (100–500 ng from leucocytes). Library preparation, sequencing, and data analysis were carried out as described previously (26).
Development of ddPCR assays for analysis of SSVs and SPMs
ddPCR assays targeting the identified SSVs and SPMs were designed using “Primer 3” (27). Amplicon lengths were kept below 100 nucleotides. Linearity and sensitivity of the assays were assessed using a 6-point dilution series of tumor DNA [4,000 to 4 genome equivalents (GEs) in pool of 20,000 GEs of matched germline DNA]. All assays consistently detected 0.08% tumor DNA and showed excellent linearity (R2 > 0.99) across three orders of magnitude. In total, 110 assays were designed (Supplementary Table S3). The assays target 100 patient-specific SSVs, six recurrent hotspot SPMs in KRAS, and four patient-specific SPMs identified by WES. Primer and probe sequences for the KRAS assays are listed in Supplementary Table S1, and the patient-specific assays are available from the authors upon request. A panel of at least 24 negative control samples from healthy donors was used to confirm the mutation specificity of the assay.
DNA quantification by ddPCR
Serial plasma DNA samples were analyzed on a QX200 Droplet Digital PCR system according to the manufacturer's instructions (Bio-Rad). Each analysis included matched tumor DNA (positive control), germline DNA (negative control), and a nontemplate control.
Carcinoembryonic antigen analysis
Carcinoembryonic antigen (CEA) analyses were performed on a Cobas e601 (Roche), according to the manufacturer's recommendations using 500 mL serum, at the Department of Clinical Biochemistry, Aarhus University Hospital (Aarhus, Denmark).
Radiologic assessment of tumor volume
CT scanning was done using multi-slice-CT scanners (Philips Medical Systems). All abdominal and pelvic scans were contrast enhanced. Tumor volume was assessed using a tumor-tracking application (Intellispace 6.0.4.02700, Philips Medical Systems). A lesion was regarded as malignant, if it fulfilled the criteria for nontarget lesion or progressed in size during the study period. In cases with multiple lung or liver metastases, or in case of lymph node bulks, a composite measure of tumor volume in the lobe, segment, or lymph node area in question was reported.
Wilcoxon rank-sum tests were used to compare differences between groups. Cohens Kappa analysis was used to assess agreement between different types of ctDNA markers, as well as between tumor burden changes measured by ctDNA analysis and CT imaging. The correlation between changes in tumor volume and ctDNA levels was assessed using Spearman rank correlation coefficient. Curves for 5-year OS and relapse-free survival (RFS, measured as the interval between surgery and disease recurrence and censored at the last follow-up) were constructed using the Kaplan–Meier method and compared using the log-rank test. Cox proportional hazards regression analysis was used to assess the impact of ctDNA in plasma on RFS and 5-year OS. All P values were based on two-sided testing and differences were considered significant at P ≤ 0.05. Fisher exact test was used to compare response and nonresponse groups and to compare SSVs to SPMs and WES mutations. Statistical analysis was performed using STATA IC/12.1 and R Statistical software, version 2.4 for Windows.
Between January 2005 and December 2007, 118 colorectal cancer patients were prospectively enrolled. Of these 27 were randomly selected aiming at a 1:1 ratio of relapsing and nonrelapsing patients and included in the present investigation (Cohort 1). Twenty-six were treated with curative intent (14 relapsing and 12 nonrelapsing) and one with palliative intend. From July 2015 to December 2016 a validation group (Cohort 2) consisting of 18 colorectal cancer patients with metachronous (15 patients) or synchronous (3 patients) liver metastases were enrolled. These were all treated with curative intent. Patient enrolment and study overview are presented in Fig. 1 and for each patient the clinicopathologic tumor characteristics, follow-up, and number of blood samples available are given in Supplementary Tables S4 and S5. To identify SSVs or SPMs, we subjected FF tumor tissue to mate-pair sequencing, as previously described (21), and/or to KRAS hotspot mutational profiling using a combination of HRMA and Sanger sequencing. For a subset of patients, WES was conducted on FF or FFPE tumor tissue. At least one SSV or SPM was identified in all tumors and used for profiling of plasma for ctDNA. On average 4.2 assays were designed for each patient in cohort 1. For cohort 2, only one mutation was used for ctDNA profiling. A total of 371 prospectively collected pre- and postsurgery plasma samples from the 45 colorectal cancer patients were analyzed to examine the diagnostic and prognostic value of ctDNA as a surrogate marker for the presence of cancer. Part of the data presented for 11 of the patients have been reported previously (21).
ctDNA in preoperative plasma samples has diagnostic value
To test whether ctDNA quantification can be used in a diagnostic setting, we examined the ability to detect ctDNA in plasma samples drawn prior to surgery. To achieve a lower limit of detection of minimum 0.5% ctDNA at least of 200 GEs of cfDNA were analyzed per ctDNA marker. The median GEs analyzed were 950 (Supplementary Table S6). ctDNA was detected in 20 of 27 preoperative plasma samples from cohort 1, resulting in a detection rate of 74%. The detection rate increased with stage of disease (Supplementary Fig. S1A), which is consistent with previous findings (22). There were no significant difference in the number of cfDNA GEs analyzed for ctDNA-positive versus ctDNA-negative plasma samples (P = 0.53, Wilcoxon rank-sum test; Supplementary Fig. S1B). Preoperative CEA measurements from the patients showed elevated levels in 15 of 26 patients (for one patient no pre-op serum was available, and therefore CEA could not be measured for this patient) giving a detection rate of 55.5%. The level of CEA was not associated with stage of disease (Supplementary Fig. S1C).
Prognostic potential of ctDNA in early postoperative plasma samples
Subsequently, postoperative plasma samples were analyzed (cohort 1). No ctDNA was detected after surgery for the relapse-free patients (n = 12). In contrast, the relapsing patients (n = 14) all had multiple postoperative plasma that were ctDNA-positive (Fig. 2A). Accordingly, ctDNA analysis of postoperative plasma predicted impending relapse with 100% sensitivity and 100% specificity. In comparison, postoperative elevated CEA levels were found only in 11 of 14 relapse patients (78.6%), while none of the relapse-free patients showed elevated CEA levels. Next, we examined whether ctDNA measurements performed shortly after surgery (within 3 months) were predictive of future relapses and correlated to RFS and 5-year OS for patients diagnosed with localized colorectal cancer, stages I–III (n = 21; Fig. 2B). ctDNA was detected 6 of the 21 patients. They all relapsed and had significant shorter RFS (HR 37.7; 95% CI, 4.2–335.5; P < 0.001) and 5-year OS (HR 6.7; 95% CI, 1.6–28.7; P = 0.01) than those with no ctDNA detected (Fig. 2B). Kaplan–Meier estimates of RFS at 3 years were 0% for the ctDNA-positive and 73% for the ctDNA-negative patients. The four patients that were ctDNA negative at three months, but later recurred, all had ctDNA detected later in multiple consecutive plasma samples (Fig. 2A). Supporting that detection of ctDNA (residual disease) is associated with very high relapse risk. As more cfDNA were analyzed for nonrelapsing than relapsing patients in cohort 1, it is unlikely that the results are confounded by DNA input (median = 1030 GEs and 837.5 GEs, respectively, P = 0.04, Wilcoxon rank sum test, Supplementary Fig. S1D). Furthermore, the number of ctDNA markers (ddPCR assays) analyzed per patient in cohort 1 were not different between relapsing and nonrelapsing patients (median 4.5 and 4 respectively, P = 0.77, Wilcoxon rank-sum test).
To confirm that ctDNA detected within 3 months after surgery is associated with a very high relapse-risk, we analyzed an independent group of 23 liver metastasis patients (18 from cohort 2 and 5 from cohort 1) treated with curative intent, including 15 relapse cases. Differently from cohort 1 only a single plasma sample (3 months after surgery) and a single ctDNA marker was available for the cohort 2 patients (Supplementary Tables S4 and S5). Nevertheless, the analysis confirmed that ctDNA-positive patients have a very high risk of relapsing (Fig. 2C, HR 4.9; 95% CI, 1.5–15.7; P = 0.007). The relapse rate was 100% for the ctDNA-positive patients (n = 7) and 50% for the ctDNA-negative. The higher fraction of ctDNA-negative patients with relapses, compared with the former analysis (Fig. 2B), may potentially be explained by only one ctDNA marker being profiled for the cohort 2 samples. For the patients in Fig. 2B, we analyzed an average of four markers, which increases the likelihood that at least one is informative. To add further support for the predictive power of ctDNA we searched the literature for colorectal cancer studies reporting ctDNA analysis (within 3 months after surgery) and RFS. Two studies encompassing a total of 244 patients, including 40 relapse events were identified (28, 29). A meta-analysis combining these two studies with the present confirmed that detection of ctDNA after curative surgery was indeed a very strong predictor of future relapse (HR 9.5; 95% CI, 5.8–15.5; P < 0.001; Fig. 2D).
Detection of relapse by CT imaging, ctDNA, and CEA
In all 14 relapsing patients from cohort 1, where we have longitudinal blood samples, we observed ctDNA in plasma prior to relapse being detected by standard-of-care CT imaging (Fig. 2A; Supplementary Table S5). Serial ctDNA analysis revealed that the ctDNA level generally increased several folds from the first detection until radiologic relapse (Fig. 3A). On average the relapses were detected after 8.2 months by ctDNA and 16.9 months by CT imaging (P = 0.001, Wilcoxon signed rank test; Fig. 3B). The median ctDNA lead time, defined as time from relapse detected by ctDNA quantification to relapse detected by standard-of-care CT imaging, was 9.4 months, ranging from 0.4 to 14.9 months (95% CI, 7.8–11). In contrast, CEA only detected relapse earlier than standard-of-care CT imaging in eight cases. In the remaining six cases, CEA either did not detect relapse at all, or detected it later or at the same time as CT imaging (Supplementary Fig. S2A; Supplementary Table S5). CEA on average detected relapse after 12.3 months which was not significantly different from the 16.9 months for CT imaging (P = 0.14 Wilcoxon signed rank test; Supplementary Fig. S2A). The median lead time for CEA relapse detection relative to CT imaging was 3.3 months, which was significantly shorter than the median 9.4 months of lead time for ctDNA (P = 0.02, Wilcoxon signed rank test; Supplementary Fig. S2B).
ctDNA changes after treatment reflect changes in tumor volume.
During follow-up, ctDNA levels were found to change after local or systemic interventions, such as surgery or chemotherapy (Supplementary Fig. S3). To characterize the relationship between ctDNA levels and tumor volume, we retrospectively searched the patient medical files (cohort 1) for relapse interventions with paired “before and after” CT images that could be used to assess the impact on tumor volume. Nineteen interventions were identified covering the following modalities: chemotherapy (n = 7), radiotherapy (n = 1), radiofrequency ablation (RFA; n = 7), and surgery (n = 4). On average, the ctDNA and tumor volume measurements were based on blood samples drawn 1.6 and 1.3 months and CT images recorded 1.4 and 1.3 months before and after intervention. A case example, showing an RFA intervention with concomitant reduction in ctDNA level and tumor volume is shown in Fig. 3C. Of the 19 interventions 13 lead to a reduced ctDNA level (median 5.2-fold, P = 0.002 Wilcoxon signed rank test) and six to an increased ctDNA level (median 4.5-fold, P = 0.03 Wilcoxon signed rank; Fig. 3D). Overall, a good agreement between the change in ctDNA level and tumor volume was observed, (Kappa = 0.41, P = 0.028; Fig. 3D). Inconsistency was observed for four interventions: in all four cases ctDNA indicated increased tumor burden, while but CT imaging indicated reduced tumor volume. Intriguingly, in three of the cases the subsequent CT scan showed progression or relapse of disease, consistent with the ctDNA measurement, indicating that only one case was truly inconsistent (agreement = 94,7%). When excluding these three cases a positive correlation between percentage change in tumor volume and ctDNA fold-change was observed (Spearman ρ = 0.4; Fig. 3E).
Concordance between somatic structural variants and somatic point mutations.
While FFPE tissue is generally available in the clinic, the fresh frozen tissue used in this study is not. Therefore, to enable translation of ctDNA surveillance into the clinic it may be necessary to use ctDNA markers that can be readily identified from FFPE tissue, for example, SPMs. To assess the feasibility of this, we compared ctDNA monitoring by SPM and SSV for 9 patients, 3 nonrelapsing and 6 relapsing (Supplementary Fig. S4). A total of 111 plasma samples were profiled. The SPMs were identified by KRAS hotspot mutation profiling using HRMA or WES (Supplementary Table S3) and produced results similar to the SSV assays. ctDNA was detected after surgery in all relapsing patients and in none of the nonrelapsing patients (Supplementary Fig. S4). For all six relapse cases, the SPM assays detected relapse before standard-of-care CT imaging. Taken together, the data indicate that SPM assays also can be used for relapse surveillance.
Identification of the patients, who will benefit from adjuvant chemotherapy after curative intended treatment, remains to be one of the most challenging tasks in colorectal oncology. The histopathologic and molecular tumor characteristics associated with higher relapse risk (including the TNM classification) only indicate a propensity for metastasis, and not whether metastatic cells has been seeded at the time of surgery. Detection of postoperative ctDNA by contrast is a direct indication that occult tumor cells remain in the patient. The achieved results show that patients treated with curative intend for localized disease (stages I–III) and who were ctDNA-positive within the first postoperative trimester had a very high risk (100%) of relapsing (HR, 37.7; 95% CI, 4.2–335.5; P < 0.001). Conversely, patients that were ctDNA negative in the first postoperative trimester were at a low risk of relapsing (3-year RFS of 75%). We chose to evaluate ctDNA within the first postoperative trimester, because this is the clinical relevant time-frame for assessing relapse risk and deciding whether or not to offer adjuvant therapy.
Using an independent cohort of liver metastasis patients treated with curative intent, we confirmed that detection of ctDNA within the first postoperative trimester was associated with a high relapse risk (HR 4.9; 95% CI, 1.5–15.7; P = 0.007). Furthermore, a meta-analysis based on the current study and two independent studies (28, 29), covering a total of 288 cases, including 65 relapse events, confirmed that patients positive for ctDNA postoperatively are at very high risk of relapsing (HR 9.5; 95% CI, 5.8–15.5; P < 0.001).
Relapse cases were also observed among the ctDNA-negative patients (e.g., Fig. 2B); we did not detect ctDNA within the first 3 months for four patients. However, as we demonstrate in Fig. 3A the ctDNA level increases from first detection until radiologic relapse, indicating that if residual tumor cells are left in the patient after surgery, then their number (tumor burden) continue to increase during follow-up. For the patients in Fig. 2B we only had available 4 mL of plasma, and obviously this was too little to detect the residual tumor burden for the four relapsing patients already at 3 months after surgery. However, all four patients became ctDNA-positive in later blood samples, most likely because the tumor burden had grown to a point where 4 mL of plasma was sufficient to sample the released tumor DNA (Fig. 2A, patients 8, 13, 14, and 16). Another important finding from the result presented in Fig. 2A is that when a blood draw tested positive for ctDNA then all subsequent blood draws from the same patient remained positive unless relapse intervention was performed. This indicates that ctDNA measurements are stable and have the potential to be used in clinical settings. Although, our findings are promising, they need to be validated in a larger number of patients, and in addition, the benefit of treating the ctDNA-positive patients with adjuvant therapy needs to be documented. Importantly, enrolling only postoperative ctDNA-positive patients in such a study would mean that the cohort would be highly enriched for patients with high risk of relapse. Accordingly, the sample size required to document an effect on outcome would be substantially reduced given the expected high event rate. Furthermore, refrainment of enrolling ctDNA-negative patients would mean that patients with low relapse-risk and similarly low chance of benefitting of adjuvant therapy would not be exposed to unnecessary treatment-related adverse effects. Optimally, ctDNA-negative patients should be offered participation in a surveillance program based on longitudinal ctDNA analysis.
The longitudinal analysis of postoperatively collected plasma samples for changes in ctDNA levels revealed a positive relationship between ctDNA changes and changes in tumor burden and proved to be efficient for detecting relapse. All 14 relapse patients were ctDNA-positive before the time of radiologic relapse. Furthermore, the serial ctDNA measurements showed that the ctDNA level continued to increase from first detection until radiologic relapse, indicating that had the radiologic examination been performed earlier then a smaller relapsing tumor burden would potentially have been detected. Conversely, none of the postoperative samples (n = 162) from the relapse-free patients (n = 12) were ctDNA-positive confirming that the cancer was eradicated by surgery in these patients, and the very high specificity of ctDNA analysis as previously reported (28).
The median lead time for ctDNA in this study was 9.4 months, which might be sufficient to change the clinical management and impact patient morbidity and mortality. It is important to note however, that the presented lead times have to be interpreted with some caution. While ctDNA quantification and CEA testing were done on samples from the same blood draw, the timepoints for CT scans and blood collections were not coordinated. Consequently, many of the positive ctDNA measurements were from timepoints without a matching CT scan, which may cause lead time bias. Nevertheless, consistent with other studies, we detected ctDNA on numerous occasions where simultaneous CT scans where negative, indicating that ctDNA analysis may be more sensitive than CT imaging (28). Summarized, these findings prompt for an evaluation of whether a combination of radiologic and ctDNA assessments could result in earlier detection of relapsing disease, potentially increasing the fraction of patients eligible for curative intervention.
It has previously been reported that changes in ctDNA levels after local or systemic interventions may occur; such changes indicate that ctDNA may carry a potential as a marker of response to therapy (21, 30). To evaluate whether changes in ctDNA levels predict radiologic response, a direct comparison of changes in ctDNA levels to changes in tumor volume, as assessed by CT imaging, were performed for 19 relapse interventions. Consistent with previous reports (30, 31), an agreement between changes in tumor volume and ctDNA was generally observed. Inconsistency occurred in four cases, all showing reduced tumor volumes by CT scanning but increased ctDNA levels. However, in three out of the four cases a subsequent CT scan confirmed progression of disease indicating that ctDNA changes may be a more direct measure of tumor response.
It has to be noted that the assessment of treatment response by CT scanning requires that all relevant tumor lesions can be and are measured. Accordingly, the observed inconsistencies may be explained by the progressing lesion not being measured, or not dominating the overall tumor volume change. The use of serial ctDNA measurements to assess and monitor the response to therapeutic intervention may have several potential benefits, for example, if serial ctDNA analysis enables early and reliable detection of a nonresponse, then it allows an earlier switch to an alternate therapy which may benefit the patient. At least it minimizes the side-effects of the ineffective therapy. In principle, a ctDNA assessment measures all lesions. Therefore, the greatest impact of serial ctDNA quantification may be for patients with nonmeasurable disease by RECIST criteria. For these patients, a reliable measurement of treatment response would greatly assist clinical decision making.
The results of the current study indicate that postoperative detection of ctDNA is a marker of residual disease and very strong predictor of future relapse risk. Furthermore, it indicates that serial ctDNA assessments during patient follow-up may allow early detection of relapse and enable assessment of the response to relapse intervention. The limitations of the study include the retrospective nature and the limited number (n = 45) of patients included. Further validation is needed and incorporation of serial ctDNA assessment in future clinical trials will be essential to confirm and further clarify the clinical utility of the promising findings.
Disclosure of Potential Conflicts of Interest
The authors declare no potential conflicts of interest.
Conception and design: L.V. Schøler, T. Reinert, A.R. Knudsen, H.J. Nielsen, C.L. Andersen
Development of methodology:L.V. Schøler, T. Reinert, S. Vang, D. Andreasen, K. Sivesgaard, P. Mouritzen, C.L. Andersen
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.V. Schøler, T. Reinert, M.-B.W. Ørntoft, C.G. Kassentoft, S.S. Árnadóttir, I. Nordentoft, F.V. Mortensen, A.R. Knudsen, K. Stribolt, K. Sivesgaard, H.J. Nielsen, S. Laurberg, T.F. Ørntoft, C.L. Andersen
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.V. Schøler, T. Reinert, S. Vang, I. Nordentoft, M. Knudsen, P. Lamy, A.R. Knudsen, S. Laurberg, C.L. Andersen
Writing, review, and/or revision of the manuscript: L.V. Schøler, T. Reinert, M.-B.W. Ørntoft, C.G. Kassentoft, S.S. Árnadóttir, S. Vang, I. Nordentoft, P. Lamy, A.R. Knudsen, K. Stribolt, K. Sivesgaard, H.J. Nielsen, S. Laurberg, T.F. Ørntoft, C.L. Andersen
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.V. Schøler, T. Reinert, C.L. Andersen
Study supervision: T. Reinert, H.J. Nielsen, C.L. Andersen
The Danish Cancer Biobank is acknowledged for the tissue material. Finally, the Danish Research Group on Early Detection of Recurrence of Colorectal Cancer is thanked for access to the sequentially collected plasma samples.
This research is supported by grants to C.L. Andersen from The Danish Council for Independent Research (11–105240), The Danish Council for Strategic Research (1309-00006B), The Novo Nordic Foundation (NNF14OC0012747), and the Danish Cancer Society (R107-A7035, R133-A8520 and R40-A1965-11-S2). L.V. Schøler was supported by The Danish Cancer Research Foundation and the Faculty of Health at Aarhus University. H.J. Nielsen was supported by The Kornerup Fund, The Aase and Ejnar Danielsen Fund, and The Aage and Johanne Louis-Hansen Fund.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.