Circulating tumor DNA for prognosis assessment and postoperative management after curative‐intent resection of colorectal liver metastases

Abstract The recurrence rate of colorectal liver metastases (CRLM) patients treated with curative intent is above 50%. Standard of care surveillance includes intensive computed tomographic (CT) imaging as well as carcinoembryonic antigen (CEA) measurements. Nonetheless, relapse detection often happens too late to resume curative treatment. This longitudinal cohort study enrolled 115 patients with plasma samples (N = 439) prospectively collected before surgery, postoperatively at day 30 and every third month for up to 3 years. Droplet digital PCR (ddPCR) was used to monitor serial plasma samples for somatic mutations. Assessment of ctDNA status either immediately after surgery, or serially during surveillance, stratified the patients into groups of high and low recurrence risk (hazard ratio [HR], 7.6; 95% CI, 3.0‐19.7; P < .0001; and HR, 4.3; 95% CI, 2.3‐8.1; P < .0001, respectively). The positive predictive value (PPV) of ctDNA was 100% in all postoperative analyses. In multivariable analyses, postoperative ctDNA status was the only consistently significant risk marker associated with relapse (P < .0001). Indeterminate CT findings were observed for 30.8% (21/68) of patients. All patients (9/21) that were ctDNA positive at the time of the indeterminate CT scan later relapsed, contrasting 42.6% (5/12) of those ctDNA negative (P = .0046). Recurrence diagnoses in patients with indeterminate CT findings were delayed (median 2.8 months, P < .0001). ctDNA status is strongly associated with detection of minimal residual disease and early detection of relapse. Furthermore, ctDNA status can potentially contribute to clinical decision‐making in case of indeterminate CT findings, reducing time‐to‐intervention.


| INTRODUCTION
Colorectal cancer (CRC) is the third most common type of cancer worldwide. 1 Liver involvement is the primary cause of death for CRC patients with approximately half of the patients developing colorectal liver metastases (CRLM). 2 The last decade has seen an increasingly aggressive approach to management of CRLM, often with the intention to cure. 3 Nevertheless, the recurrence rate after curative intent treatment for CRLM remains high, with nearly 50% recurring within 2 years [4][5][6] and a 5-year disease-free survival of only 27.9%. 7 Most CRLM surveillance programs involve frequent computed tomographic (CT) scans. 8,9 A common challenge related to this, is indeterminate CT findings in the liver and lungs. 10,11 Often, they represent benign lesions, but results in additional investigations and increased patient unease. 11 In other instances, the indeterminate findings represent malignant lesions, but the requirement for further investigations before the final diagnosis often leads to delayed intervention. 11,12 Currently, there are no validated biomarkers of patient recurrence risk that could inform and personalize the use of chemotherapy and help guide and resolve the issues related to CT-imaging based surveillance. Detection of circulating tumor DNA (ctDNA) by the use of tumor-specific DNA mutations is an emerging biomarker, which in the setting of localized CRC has been shown to have potential to change the fields of postoperative prognostication and surveillance. [13][14][15][16][17][18][19] Recent studies have indicated a similar potential for clinical application also in the setting of CRLM. 20,21 However, further prospective studies confirming and expanding on these findings are needed.
Here, we report findings from a prospective and observational biomarker study in the setting of CRLM, aiming to assess the clinical value of serial ctDNA analysis for postoperative prognostication and guidance of CTimaging in patients treated with curative intent.  (Tables 1 and Table S1).

| Blood collection and plasma isolation
Blood samples were collected in K2-EDTA 10 mL tubes (Becton Dickinson) and processed to plasma and buffy coat within 2 hours of collection by double centrifugation at room temperature. First the blood sample was centrifuged for 10 minutes at 3000g and then the plasma was centrifuged for 10 minutes at 3000g. Plasma was aliquoted into 5 mL cryotubes and stored at À80 C.

| Tumor screening for mutations in KRAS codon 12 and 13 and BRAF codon 600
To identify a tumor specific clonal mutation to be used as a marker of tumor DNA in droplet digital PCR (ddPCR) based plasma cell free DNA (cfDNA) analyses, the liver metastases were initially screened by the ddPCR KRAS G12/G13 Screening Kit (Biorad). KRAS positive patients were subsequently profiled by individual KRAS mutation assays to identify the specific mutations. KRAS negative samples were screened with a BRAF V600E mutation assay (Table S2).  (Table S2). The MSK IMPACT panel targets 468 cancer genes and includes probes targeting single-nucleotidepolymorphisms to enable the concordance between matched samples to be assessed. 22 In brief, tumor and germline DNA was sheared using the Covaris E200 instrument (Covaris, Woburn, Massachusetts).  Table S3.

| Mutational profiling by targeted duplex sequencing
Another subset of tissue samples (N = 23) and two preoperative plasma samples from patients, where no liver metastasis tissue was available were analyzed by ultra-deep targeted duplex sequencing to identify the clonal mutation to be used in plasma ddPCR (Table S3) of the cfDNA content was carried out using ddPCR as previously described. 17 The primer and probe sequences of the used ddPCR assays are provided in Table S5.

| Droplet digital PCR assays for quantification of ctDNA
The selection of mutations for ctDNA analysis was restricted to mutations in the genes APC, BRAF, KRAS, NRAS, PIK3CA and TP53. Mutations with variant allele frequencies below 25% of the histology estimated tumor fraction were judged to be subclonal, and not selected for ctDNA analysis. For each patient matched white blood cell DNA was used to exclude variants arising from clonal hematopoiesis. When more than one clonal somatic mutation was identified in a patient's tumor tissue, the mutation with the highest variant allele frequency (VAF) was selected. For each patient, only one mutation identified in the tumor tissue was assessed in the plasma. ddPCR assays were designed to the selected mutations (Thermo Fisher Scientific).
Before being used for ctDNA quantification, the performance of each assay was assessed as previously described. 16 In brief, the linearity and technical sensitivity of the assays were assessed using a 6-point To assess specificity and determine the assay specific limit of detection (LOD), 25 each assay was applied to 94 control samples from healthy donors. The primer and probe sequences of the designed ddPCR assays are provided in Table S5.

| Carcinoembryonic antigen analysis
Carcinoembryonic antigen (CEA) analysis was performed on a Cobas e601 platform (Roche), according to the manufacturer's recommendations using 500 μL serum. The threshold levels were set to 4.0 and 6.0 μg/L for nonsmokers and smokers, respectively. A person who had not smoked for 8 weeks before sample collection was considered a former smoker and thresholds levels set to 4.0 μg/L.    Figure 2A). The presence of ctDNA was associated with a markedly reduced RFS as compared to ctDNA negative patients (HR, 7.6; 95% CI, 3.0-19.7; P < .0001; Figure 2A). Univariate analysis identified the postoperative ctDNA status at day 30 (P < .0001) and primary tumor N stage (P = .023) as significant prognostic factors. In a multivariable Cox regression model, including only the significant prognostic factors, ctDNA status was the only significant prognostic factor (P < .0001; Table S8).  Figure 3A).

| Paired ctDNA and CEA status and association to recurrence
For both ctDNA and CEA the pre-OP positive rate was 75% (12/16). In the postoperative setting, 27 paired ctDNA and CEA measurements were available from samples collected longitudinally from end of definitive treatment (surgery or surgery/ACT) and until diagnosis of recurrence.
Comparing the sensitivities of ctDNA and CEA we found that the sensitivity for detecting recurrence during was 77.3% (17/22) and 54.5%  To increase the number of recurrence events, we extended the analysis to include also recurrence events occurring beyond first recurrence. By this, we reached 36 and 15 recurrence events located to the liver and lungs, respectively. All other recurrence sites were infrequent, and consequently excluded from further analysis due to lack of statistical power. Analysis of blood samples collected at the time of recurrence revealed detection of ctDNA concurrent with 86% (31/36) of the liver metastases and 40% (6/15) of the lung metastases (Fisher's exact test, P = .0016; Figure 3B). To investigate if the false negative observations could be related to the selected ddPCR markers, we procured tumor DNA from four lung lesions not detected by ddPCR. In all four cases, the tumor specific mutations selected as ddPCR markers were confirmed to be present in the lung metastases ( Figure S3). These results indicate that the inability to detect ctDNA in the plasma was not due to the metastatic lesions not harboring the mutation, but more likely a result of the ctDNA level in these patients being below our LOD, despite the high plasma volume analyzed (8 mL).  Figure 4A). Nine patients were ctDNA positive and 100% (9/9; PPV = 100%) of these were later diagnosed with recurrence. The remaining 12 patients were ctDNA negative and only 4/12 (33.3%; NPV = 66.7) recurred (Fisher's exact test, P = .0046, Figure S4). Patients that were ctDNA positive at the time-point of an indeterminate CT finding had a significantly increased recurrence risk compared to patients that were ctDNA negative (HR, 9.0; 95% CI, 2.3-34.4; P = .0005; Figure 4A).

| Association between ctDNA status and outcome of follow-up after indeterminate CT findings
Extending the analysis to include indeterminate findings occurring after the diagnosis of the first recurrence, 37 indeterminate CT findings with a concomitant blood sample were identified. We observed a positive predictive accuracy of 100% and a negative predictive accuracy of 50% (Fisher's exact test, P = .0009; Figure 4B). Patients that were ctDNA positive at the time-point of an indeterminate CT finding had a significantly increased recurrence risk (HR, 3.5; 95% CI, 1.6-8.0; P = .0022; Figure 4B). Longitudinal ctDNA plasma analyses in CRC cohorts have previously shown that ctDNA was detected 8.7 to 10 months ahead of radiological recurrence. [16][17][18] In the current study, we found that ctDNA detected recurrence with median lead time of 2.5 months compared to CT imaging. This finding in a high-risk cohort with standard of care CT imaging every 3 months show that ctDNA analyses have potential to improve the postoperative management of patients treated for CRLM.

| DISCUSSION
We found that the ability to detect ctDNA postoperatively was associated with the metastatic site. Lung metastases were 15.1 times more likely to be ctDNA negative than single liver metastases (P = .0004). Most probable, this discrepancy is an effect of better radiographic resolution in the lung than in the liver, resulting in lung metastases being smaller when detected. The mean diameter of lung metastases in our study were 8. and cfDNA-fragment lengths. Recently, we reported how cfDNA fragment length analysis also can provide a ctDNA discriminatory signal. 28 The findings suggest that until a ctDNA approach with improved sensitivity for detection of lung metastases is identified, radiographic imaging of the lungs should remain a central modality in CRLM surveillance.
Postoperative surveillance after CRC and CRLM resection include frequent CT imaging and clinical challenge is the frequent occurrence of indeterminate CT findings. 11,12 In the present study, nearly one third (31%) of the 68 patients eligible for ctDNA analyses had one or more indeterminate findings. In agreement with previous reports, the consequence was additional radiographic work-up and a delay in diagnosis and intervention for the patients who eventually were diagnosed with recurrence. 12 The consequence of the delay may be an increased tumor burden and a reduced efficacy of the intervention.
This line of thinking is supported by a recent study, where serial ctDNA measurements were used to estimate the growth of colorectal metastases. Growth was extensive ranging from 25% to 143% per month. 29  There are several limitations to our study, including a limited sample size from a single hospital and the potential risk for false positive findings related to subset analyses. Furthermore, the lead-time of ctDNA compared radiographic imaging for detection of recurrence may be an overestimate as blood sampling was slightly more frequent than imaging. Moreover, while our tumor-informed single marker ctDNA detection strategy is highly specific in detecting molecular residual disease, there remains a window of opportunity for improving the analytical sensitivity. The sensitivity of our serial analysis was 83.3%, that is, in 8 of 48 recurrence patients our approach was unable to detect ctDNA. A promising approach for increasing sensitivity is to increase the number of ctDNA markers, though even with improved ctDNA detection strategies, false negative results can still occur due to low shedding tumors. Moreover, as we have exemplified for lung metastases, other modalities may in some instances have better sensitivities than ctDNA. Location of the false negative results and the sample size was limited. Ultimately, the optimal surveillance approach will have to balance sensitivity, specificity, cost and throughput. In