Tissue-free circulating tumor DNA assay and patient outcome in a phase III trial of FOLFOX-based adjuvant chemotherapy (Alliance N0147)

Abstract

Purpose:

Detection of molecular residual disease using circulating tumor DNA (ctDNA) may enable postoperative risk stratification and guide adjuvant therapy. We evaluated the prognostic value of a tissue-free, epigenomic ctDNA assay in patients with stage III colon cancer enrolled in a phase 3 adjuvant chemotherapy trial.

Methods:

Plasma samples were collected after surgery and before adjuvant FOLFOX alone or combined with cetuximab. ctDNA was analyzed using a tissue-free assay; in ctDNA-positive cases, tumor fraction (TF) was quantified and genotyping performed with a 739-gene panel. Associations with disease-free survival (DFS), time-to-recurrence (TTR), and overall survival (OS) were assessed and included multivariable Cox models adjusted for covariates.

Results:

Among 2,260 evaluable patients, 461 (20.4%) were ctDNA positive with significantly higher detection in advanced T/N stage, high grade, obstruction/perforation, and BRAF^V600E tumors. At a median follow-up of 6.1 years, ctDNA positivity was independently associated with shorter TTR (HR 5.96; 95% CI 5.11–6.96), DFS (HR 5.03; 4.36–5.81), and OS (HR 4.45; 3.76–5.27) (all P < 0.0001). Five-year DFS was 27.7% (95% CI 23.8–32.2) vs 77.1% (95% CI 75.1–79.1) in ctDNA-positive vs negative patients, adverse prognostic impact was greater in lower T/N stage, low-risk, and dMMR subsets (interaction P = 0.0012–0.041). Among ctDNA-positive patients, TF was nearly double in those who recurred or died (P = 0.0002), and stratified patients for TTR, DFS, and OS (all adjusted P < 0.002). Genotyping identified mutations in FLT1 (OR 8.99) and PREX2 (OR 7.73) genes that were most strongly associated with recurrence (p <0.03).

Conclusion:

In the largest study evaluating ctDNA in resected stage III colon cancer, a tissue-free assay provided robust and independent prognostic value. Higher ctDNA burden, dMMR, and specific mutations defined poor prognostic groups.

INTRODUCTION

Standard of care treatment for patients with stage III colon cancer (CC) is surgical resection followed by adjuvant chemotherapy, typically with three or six months of CAPOX or FOLFOX.¹ At present, only TNM staging data are used for adjuvant treatment decisions based on categorization of tumors as low (T1-3N1) or high risk (T4 and/or N2) for recurrence.^1,2 Adjuvant chemotherapy is given to reduce the risk of recurrence, presumably by effectively clearing micrometastatic residual disease following surgery.^3,4 Among stage III patients, there is considerable variability in outcomes as shown in the International Duration Evaluation of Adjuvant chemotherapy (IDEA) collaboration where 5-year disease-free survival (DFS) varied from 89% for resected T1N1a tumors to 31% for T4N2b tumors.⁵ These survival data indicate that some patients have post-surgical minimal residual disease (MRD), defined as detectable circulating tumor DNA (ctDNA), which has been strongly associated with poor prognosis in colorectal cancer (CRC).^6-9 Adjuvant chemotherapy reduces the absolute recurrence rate by only 10–15% despite its routine use in stage III cancers where it is associated with significant treatment-related side effects.^10,11

A tissue-free MRD assay can offer a logistical advantage by avoiding the burden of tissue retrieval and shortening the turnaround time for the initial result compared with a tumor-informed, tissue-based assay.^13,14 Tissue-free assays may also be the only option for patients with limited tumor specimens. Given that large studies evaluating tissue-free ctDNA assays with extended follow-up are lacking, we evaluated post-surgical plasma from a phase 3 adjuvant trial in stage III CC patients (NCCTG N0147) using two tissue-free ctDNA assays (Guardant Reveal and Guardant 360) to determine the prognostic significance of ctDNA detection, tumor fraction (TF), and genomic alterations in relation to clinicopathologic features and long-term patient outcome. Quantifying the ctDNA TF can serve as an approximate measurement of tumor burden¹² that along with MRD, may further refine recurrence risk estimation.

METHODS

Study Design and Participants

NCCTG N0147 (Alliance) is a phase 3 trial that enrolled 3084 patients with stage III CC randomized (1:1) to receive FOLFOX (n=1402) alone or combined with cetuximab (n=1350) for 12 cycles (6 months).³ Eligible patients were age 18 or older, had histologically confirmed adenocarcinoma, and began adjuvant chemotherapy within 10 weeks of curative-intent surgery. Patients were initially enrolled regardless of KRAS mutation status but inclusion criteria were later amended when it was reported that only wild-type KRAS tumors respond to cetuximab.^16,17 Thus, there was a third non-randomized arm for patients with KRAS mutant tumors (n=332) who were treated with FOLFOX (12 cycles) as standard of care. The trial showed no benefit for its primary endpoint of DFS with the addition of cetuximab to FOLFOX.³ Recurrence, site of first recurrence, and survival data were collected and recorded by the study sites as mandated by the study protocol.

Written informed consent was obtained from all patients. The clinical protocol was approved by the Institutional Review Board of Mayo Clinic. The study (NCT00079274) was conducted in accordance with the Declaration of Helsinki.

Study Population for ctDNA Analysis

The protocol included a postsurgical blood collection for correlative studies obtained within 10 weeks after surgery and prior to adjuvant chemotherapy.³ Plasma was isolated from blood and stored at −80 C in the Mayo Biobank. Of 3084 participants in the N0147 trial, 824 were excluded from the post-surgical ctDNA analysis based on the following factors: no blood collected or unknown collection date (n=726), blood collected but withdrew prior to protocol treatment (n = 44); blood obtained after starting treatment (n = 15); and technical failure of the ctDNA assay (n = 39). A total of 2,260 patients were available for ctDNA analysis. The status of ctDNA by study arm is shown in Appendix Figure 1. For ctDNA analysis, data from the study arms were combined since the primary DFS endpoint for N0147 did not differ significantly by arm.

Tissue-free ctDNA analysis in plasma

In cell-free DNA extracted from plasma, a tissue-free epigenomic assay (Guardant Reveal^™) was performed which has shown high clinical and analytical sensitivity and >98% specificity for recurrence prediction.¹⁸ This assay does not use genomic data for ctDNA detection. Among > 20,000 differentially methylated regions (DMRs) evaluated, a bioinformatic model focuses on ~2,000 DMRs that are most strongly associated with CRC. Values over a predefined threshold are considered ctDNA positive and are reported along with an estimated TF, determined as the ratio of methylated molecules in DMRs compared to the total methylated molecules in control regions.

Genomic profiling of ctDNA positive patients

All ctDNA positive cases were evaluated with the Guardant360 assay that analyzes 739 cancer-related genes, including copy number variants (CNV), gene fusions, tumor mutation burden (TMB) and microsatellite instability (MSI) status. The assay detects >95% of single nucleotide variants (SNV) found in samples with ≥0.20% TF with specificity of >99.99%. To assess whether somatic mutations were significantly enriched in recurred vs non-recurred patients, the proportion of patients with non-synonymous alterations (after excluding canonical mutations associated with clonal hematopoiesis) was compared for each gene based on 3-year DFS status using the two-sided Fisher’s exact test, and the corresponding odds ratio was calculated. Patients with < 3 yrs of follow-up (n=27) were excluded from genotyping.

Tissue Biomarker Data from N0147 Tumor Specimens

Tumor tissue KRAS and BRAF^V600E mutational status was analyzed centrally during the N0147 trial using polymerase chain reaction (PCR) that analyzed seven KRAS mutations in codons 12 and 13, and evaluated the BRAF^V600E point mutation.^3,19 Following trial completion, mismatch repair (MMR) protein expression was evaluated by immunohistochemistry in all participants where tissue was available.²⁰

Statistical Analysis

Associations between ctDNA status and patient demographics, clinicopathological variables, and biomarkers were determined using the chi-squared test or the Wilcoxon Rank Sum test for categorical and continuous variables, respectively. Study endpoints were measured from date of randomization to first documented recurrence (time-to-recurrence, TTR), first documented recurrence or death due to any cause (disease-free survival, DFS) or to date of death due to any cause (overall survival, OS). Patients without an event were censored at the last follow-up visit. Time-to-event endpoints were estimated by the Kaplan–Meier method and compared using the log-rank test. Cox proportional hazards models were adjusted for age, gender, ECOG PS, grade, number of positive nodes, T-stage, primary tumor location, surgery type (open/laparoscopic/conversion), bowel obstruction or perforation, tumor adherence, and tissue biomarkers (KRAS, BRAF^V600E, and MMR). Subgroup analyses were performed if the interaction p-value was < 0.05. Survival outcomes by ctDNA status were examined by treatment arm. All analyses were performed using SAS software v9.4; two-sided P values < 0.05 were considered statistically significant.

RESULTS

Patient characteristics and ctDNA detection

The 2,260 patients with post-surgical ctDNA results were similar to the overall N0147 cohort.³ Plasma samples were collected at a median of 42 days (± 7 days) post-surgery and prior to adjuvant chemotherapy [Appendix Figure 1]. ctDNA was detected, i.e.,positive, in 461 (20.4%) patients of which 397 (89.0%) were pMMR and 49 (11.0%) were dMMR. The highest ctDNA positivity rate occurred in patients with bowel perforation (36.8%); the lowest was in patients with T1-2 tumors (9.1%) [Table 1, Appendix Figure 2]. Rates of ctDNA positivity were 31.0% for T4 tumors, 31.6% for N2, 24.4% for high grade, and 25.7% for those with BRAF^V600E [Table 1]. A higher median number of involved nodes was observed in ctDNA positive vs negative patients [median 5.0 (1.0, 31.0) vs 2.0 (0.0, 32.0); p<0.0001].

Table 1.

Patient characteristics by post-operative ctDNA status

	ctDNA Status			P-value
	Detected (N=461)	Not Detected (N=1799)	Total (N=2260)
Age				0.67461
Mean (SD)	58.1 (11.47)	58.0 (11.01)	58.0 (11.10)
Median (Range)	59.2 (25.0, 84.2)	58.5 (19.4, 87.0)	58.5 (19.4, 87.0)
Sex, n (%)				0.47602
Female	209 (45.3%)	849 (47.2%)	1058 (46.8%)
Male	252 (54.7%)	950 (52.8%)	1202 (53.2%)
Race, n (%)				0.54142
Asian	16 (3.5%)	86 (4.8%)	102 (4.5%)
Black	33 (7.2%)	122 (6.8%)	155 (6.9%)
Other	13 (2.8%)	39 (2.2%)	52 (2.3%)
White	399 (86.6%)	1552 (86.3%)	1951 (86.3%)
BMI				0.79521
Mean (SD)	28.8 (6.31)	28.7 (6.20)	28.7 (6.22)
Median (Range)	27.5 (15.4, 50.8)	27.7 (14.5, 60.2)	27.7 (14.5, 60.2)
ECOG Performance Status, n (%)				0.02212
0	334 (72.6%)	1410 (78.6%)	1744 (77.3%)
1	121 (26.3%)	373 (20.8%)	494 (21.9%)
2	5 (1.1%)	12 (0.7%)	17 (0.8%)
Missing	1	4	5
Primary Tumor Location, n (%)				0.31662
Distal	212 (46.0%)	898 (49.9%)	1110 (49.1%)
Proximal	243 (52.7%)	878 (48.8%)	1121 (49.6%)
Both	6 (1.3%)	22 (1.2%)	28 (1.2%)
Missing	0	1	1
Histology Grade, n (%)				0.00612
High	138 (29.9%)	427 (23.7%)	565 (25.0%)
Low	323 (70.1%)	1372 (76.3%)	1695 (75.0%)
Number of Positive Nodes				<.00011
Mean (SD)	6.1 (4.91)	3.6 (3.49)	4.1 (3.95)
Median (Range)	5.0 (1.0, 31.0)	2.0 (0.0, 32.0)	3.0 (0.0, 32.0)
Missing	1	0	1
N-Stage, n (%)				<.00012
N1	163 (35.4%)	1157 (64.3%)	1320 (58.4%)
N2	297 (64.6%)	642 (35.7%)	939 (41.6%)
Missing	1	0	1
T-Stage, n (%)				<.00012
1	6 (1.3%)	95 (5.3%)	101 (4.5%)
2	25 (5.4%)	215 (12.0%)	240 (10.6%)
3	347 (75.3%)	1303 (72.5%)	1650 (73.0%)
4	83 (18.0%)	185 (10.3%)	268 (11.9%)
Missing#
Clinical Risk Group, n (%)				<.00012
High Risk	324 (70.4%)	751 (41.8%)	1075 (47.6%)
Low Risk	136 (29.6%)	1047 (58.2%)	1183 (52.4%)
Missing	0	1	1
KRAS Status, n (%)				0.28422
Mutant	167 (37.7%)	612 (35.0%)	779 (35.5%)
Wildtype	276 (62.3%)	1138 (65.0%)	1414 (64.5%)
Missing	18	49	67
BRAF Status, n (%)				0.02272
Mutant (V600E)	66 (15.2%)	191 (11.2%)	257 (12.0%)
Wildtype	368 (84.8%)	1511 (88.8%)	1879 (88.0%)
Missing	27	97	124
MMR Status, n (%)				0.73882
pMMR	397 (89.0%)	1547 (88.5%)	1944 (88.6%)
dMMR	49 (11.0%)	202 (11.5%)	251 (11.4%)
Missing	15	50	65
Adherence, n (%)				<.00012
Yes	99 (21.5%)	225 (12.5%)	324 (14.3%)
No	362 (78.5%)	1574 (87.5%)	1936 (85.7%)
Bowel Obstruction, n (%)				0.00042
Yes	100 (21.7%)	268 (14.9%)	368 (16.3%)
No	361 (78.3%)	1531 (85.1%)	1892 (83.7%)
Bowel Perforation, n (%)				<.00012
Yes	43 (9.3%)	74 (4.1%)	117 (5.2%)
No	418 (90.7%)	1725 (95.9%)	2143 (94.8%)
Surgery Type, n (%)				0.01342
Open	290 (63.0%)	1046 (58.2%)	1336 (59.2%)
Laparoscopic	132 (28.7%)	639 (35.5%)	771 (34.1%)
Conversion	38 (8.3%)	113 (6.3%)	151 (6.7%)
Missing	1	1	2

¹Wilcoxon rank sum p-value

²Chi-Square p-value

Post-operative ctDNA Detection is Associated with Inferior Patient Outcomes

At a median patient follow-up of 6.1 yrs. (95% CI, 6.0 to 6.2 yrs), ctDNA positivity was significantly associated with poorer 5-year DFS [hazard ratio (HR) 5.03, 95% confidence interval [CI] 4.36-5.81, P < 0.0001], shorter TTR (HR 5.96, 95% CI 5.11-6.96, P <0.0001) and OS (HR 4.45, CI 3.76-5.27, P < 0.0001) [Figure 1A-C]. Among ctDNA positive vs negative patients, 5-yr DFS rates were 27.7% (95% CI 23.8-32.2%) vs 77.1% (95% CI 75.1-79.1%), respectively; and for OS were 50.4% (95% CI 45.9-55.3%) vs 86.8% (95% CI 85.2-88.4%), respectively. By multivariable analysis, ctDNA positivity remained significantly associated with adverse outcomes (DFS: adjusted HR 3.74, 95% CI 3.18-4.39, P <0.0001), (TTR: adjusted HR 4.33, 95% CI 3.65-5.13, P <0.0001), (OS: adjusted HR 3.17, 95% CI 2.63-3.83, P<0.0001) [Table 2].

Figure 1:

5-year Disease-free Survival (A), Overall Survival (B) and Time-to-Recurrence (C) by postsurgical ctDNA status in patients with stage III colon cancer treated with FOLFOX-based adjuvant chemotherapy.

Table 2:

Multivariable analyses comparing patient clinical outcomes based on postsurgical ctDNA status (positive vs. negative) and further stratifying ctDNA positive cases by TF values

Endpoint	ctDNA Status*	Hazard Ratio (95% CI)	p-value (likelihood ratio)
DFS	Positive vs Negative	3.74 (3.18 - 4.39)	<0.0001
TTR	Positive vs Negative	4.33 (3.65 - 5.13)	<0.0001
OS	Positive vs Negative	3.17 (2.63 - 3.83)	<0.0001
Endpoint	Tumor Fraction ( high vs low)*	Hazard Ratio (95% CI)	p-value (likelihood ratio)
DFS	High vs. Low	1.52 (1.21 - 1.92)	0.0004
TTR	High vs. Low	1.48 (1.17 - 1.88)	0.0011
OS	High vs. Low	1.579 (1.206 - 2.068)	0.0009

^*Adjusting for age, gender, ECOG PS (0 vs. 1/2), grade, number of positive nodes, T-stage (1/2 vs. 3 vs 4), primary tumor location (left vs. right), surgery (open vs. lap vs. conversion), bowel obstruction (y vs n), bowel perforation (y vs. n), adherence (y vs. n), KRAS, BRAF^V600E, and MMR status.

^**Relative to median value.

DFS, Disease Free Survival; TTR, Time to Recurrence; OS, Overall Survival.

A greater detrimental effect on DFS was observed between ctDNA positive vs negative patients for T1/2 vs. T3 vs. T4 tumors [Figure 2A], N1 vs. N2 [Figure 2B], low vs. high clinical risk groups [Figure 2C], and dMMR vs. pMMR tumors [Figure 2D]. Among ctDNA positive patients, those with dMMR vs pMMR tumors had markedly poorer 5-year OS at 34.6% vs. 52.6%, respectively (P = 0.022) [Figure 2E]. Females with ctDNA positive tumors had poorer DFS vs male patients (adjusted P_interaction = 0.039) [Figure 3, Appendix Table 1]. These findings among subgroups remained statistically significant after adjustment for multiple covariates. Significant interaction effects were observed between ctDNA detection and T stage, N stage, clinical risk group and MMR status for DFS, TTR and OS; for MMR status: P_interaction values: 0.024, 0.002, and 0.022, respectively [Appendix Table 1]. These statistically significant interaction effects (p-values: 0.0007 to 0.0236) at a nominal significance level (i.e., 0.05) were sufficient to direct further subgroup analyses. The association of ctDNA positivity with poorer DFS was statistically significant across all subgroups (all P <0.0001) [Figure 3]. With a Bonferroni adjustment to control for multiple comparisons, all of the subgroup analyses remained significant (p-values < 0.0015).

Figure 2:

5-year patient DFS by postsurgical ctDNA status stratified by (A) T stage, (B) N stage, (C) clinical risk group, and (D) MMR status. (E), OS by postsurgical ctDNA status among patients with dMMR vs pMMR stage III colon cancer treated with FOLFOX-based adjuvant chemotherapy.

Figure 3:

Forest plot displays hazard ratios (HRs) with 95% confidence intervals for DFS among patients with positive vs negative ctDNA status across multiple patient subgroups. Statistically significant associations are shown across all subgroups (all P <0.0001). With a Bonferroni adjustment to control for multiple comparisons, all of the subgroup analyses remained statistically significant (p-value < 0.0015).

Epigenetic Tumor Fraction (TF) Enhances Prognostication by ctDNA

We determined the epigenetic TF as a quantitative measure of MRD and its distribution in the 461 ctDNA positive cases where the median TF was 0.0011 (IQR, 0.0004, 0.0049). The median raw TF was higher in dMMR vs pMMR patients (0.0024 vs 0.0010, p= 0.0085) [Appendix Figure 3]. We found a nearly two-fold higher median TF in patients who recurred or died compared with those who remained recurrence-free (median ≈ 0.0016 vs 0.0008; ~2-fold increase; P = 0.0002) [Appendix Figure 4]. Univariately, the dichotomized TF was shown to significantly stratify patients by DFS, TTR and OS (all P <0.0002) [Figure 4A-C]. Among ctDNA positive patients, those with TF above vs below the median had a significantly shorter TTR [Figure 4C] and a poorer 5-year DFS of 22.1% vs. 33.4% (HR 1.56, CI 1.26-1.93, p<0.0001) [Figure 4A]. After adjustment for covariates including MMR, KRAS and BRAF^V600E status, patients with ctDNA TF above vs. below the median had a significantly poorer TTR, DFS, and OS (all p values <0.002) [Table 2].

Figure 4:

Patient DFS (A), OS (B) and TTR (C) by quantitative ctDNA tumor fraction (TF) dichotomized at the median in ctDNA-positive patients with stage III colon cancer treated with FOLFOX-based adjuvant chemotherapy.

ctDNA Status and Anatomic Site of Tumor Recurrence

Among patients who were ctDNA-positive, recurrence occurred more often in the liver compared with ctDNA-negative patients (31.9% vs. 12.9%) and was less frequent in the lung or at loco-regional sites (Appendix Figure 5). In contrast, ctDNA-negative vs positive patients who experienced relapse were more likely to have loco-regional recurrence including peritoneal involvement (34.6% vs. 20.3%) (Appendix Figure 5). Analysis of ctDNA TF by site of tumor recurrence revealed that patients with a high TF had a higher proportion of recurrences in the liver but fewer recurrences in the lung compared to patients with low TF (Appendix Figure 6).

Predictive Analysis

We assessed ctDNA status by treatment arm (FOLFOX ± cetuximab) and outcomes among KRAS wild-type tumors. Among ctDNA-negative patients, TTR, DFS, and OS were similar between arms [Appendix Figure 7A-C]. In ctDNA-positive patients, DFS and TTR did not differ by arm, but OS was significantly poorer with FOLFOX + cetuximab vs. FOLFOX alone [Appendix Figure 7C]. Similar results were seen when restricting to ctDNA-positive patients with wild-type KRAS/BRAF, pMMR, and left-sided primaries (n = 131) [Appendix Figure 7D]. Post-recurrence treatment data were unavailable.

Impact of Plasma Genotype on Patient Outcomes

Among ctDNA-positive patients, genotyping was performed using the Guardant360 assay (739 genes) to identify potential drivers of recurrence or treatment resistance. Of 434 patients with complete 3-year recurrence data (276 with recurrence, 158 without), the most frequently mutated genes—occurring at similar rates in both groups—included TP53 (52.1%), ATM (11.8%), NF1 (10.8%), PIK3CA (9.0%), and CHEK2 (8.5%) [Appendix Tables 2,3]. An Oncoprint of cfDNA alterations detected by Guardant360 revealed a higher prevalence of BRAF and DNA damage–response gene mutations (ATM, CHEK2) in recurrent cases, whereas TP53 and APC alterations were common in both groups [Appendix Figure 8]. BRAF mutations were identified in 10.8% of patients (7.8% BRAF^V600E), and were more frequent among those with recurrence (13% vs. 7%; BRAF^V600E 7.8% vs. 5.1%). Ten genes were significantly more likely to be mutated in recurrent cases (OR 1.6–11.3, p = 0.008–0.049) [Appendix Figure 9].Mutations in FLT1 (VEGFR1) and PREX2 genes best discriminated patients by recurrence. FLT1 is an RTK involved in angiogenesis, and PREX2 is involved in the Rac1/PI3K pathway. ^{22, 23} Other mutated genes that were significantly enriched in patients with recurrence include KRAS, BARD1, CDKN2A, APC, PTPRT, KDM6A, KMT2B, and BRCA2 [Appendix Figure 9]. Two genes, SRSF2 and NCR1, were more frequent in non-recurred patients (p= 0.048).

Among 420 patients with tissue MMR status, mutations that were enriched in dMMR (n = 45) vs. pMMR tumors included BRAF^V600E, POLE, POLD1, MLH1, MSH2, and MSH6 (p < 0.01). PALB2 mutations were more frequent in dMMR patients without vs. with recurrence [30.8% (4/13) vs. 3.1% (1/32); OR = 0.07, p = 0.019]. Additional genes were also associated with dMMR status, though this may reflect the higher TMB and background mutational burden in these tumors.

DISCUSSION

We report the largest study evaluating tissue-free ctDNA for MRD detection in stage III colon cancer patients. Our epigenomic ctDNA assay detects patterns of methylation that are unique to cancer and not found in healthy blood samples. While tissue-free assays have the potential for lower sensitivity in low-shedding tumors, our post-surgical detection rate of 20.4% is consistent with results (18.0% to 21%) from tissue-informed assays in stage III colon cancer patients.^7,8 Patients who were ctDNA-positive after surgery had a markedly higher risk of recurrence and death compared with ctDNA-negative patients that was independent of potential confounders. Furthermore, the integration of ctDNA with standard clinicopathological features in multivariable models was shown to offer greater precision for predicting patient outcomes. The results of our tissue-free ctDNA assay demonstrate similar prognostic performance as has been observed in studies employing tissue-informed ctDNA assays.^8,26,27 The adverse prognostic impact of ctDNA was observed across all subgroups examined. Novel findings in our cohort include a stronger detrimental effect of ctDNA positivity on prognosis in patients with lower risk tumor features e.g. T1/2 (vs. T3 or 4), N1 (vs. N2), low (vs. high) clinical risk group, and dMMR (vs. pMMR) [interaction P values 0.0002 to 0.041]. These findings underscore the ability of ctDNA analysis to identify patients at higher risk of recurrence who would have been considered lower risk using TNM staging criteria. This is clinically important since the recommended duration of adjuvant chemotherapy in stage III patients is currently based only on T and N risk classification^1,2 which is shown by ctDNA results to be insufficient for prognostication. Accordingly, we propose ctDNA testing on all resected stage III patients who are candidates for adjuvant therapy as it may support adjuvant and/or post-adjuvant therapy decision-making.

ctDNA positivity was associated with a higher frequency of liver metastases compared with loco-regional recurrence, suggesting that the latter sites may have lower levels of ctDNA shedding and/or reduced access to the systemic circulation.^18,25 Despite all patients in our cohort receiving 6 months of adjuvant FOLFOX, ~70% of post-surgical ctDNA positive patients had recurrence, suggesting that treatment was largely ineffective. Results of the AGITG Dynamic-III trial in stage III colon cancer found that treatment escalation from FOLFOX to FOLFOXIRI among ctDNA positive patients, while not randomized, failed to improve 3-year RFS,²⁸ suggesting that such patients are in need of novel therapeutic strategies.

Our finding that quantitative post-operative TF can provide further prognostic stratification is consistent with data from the prospective DYNAMIC-II and -III studies.^{28 29} It is conceivable that the TF may indicate a threshold above which adjuvant chemotherapy is unlikely to clear MRD to provide clinical benefit. In the GALAXY/CIRCULATE-Japan study, patients treated with adjuvant chemotherapy who had higher ctDNA concentrations post-surgery were less likely to have ctDNA clearance at any subsequent time point, suggesting that TF is an important factor in determining adjuvant efficacy.⁷ Patients with ctDNA-positive dMMR vs pMMR tumors had significantly worse DFS and OS, with 5-year OS <35%. Factors contributing to his observation are the significantly higher ctDNA TF in dMMR vs pMMR patients and, potentially, reduced benefit from chemotherapy. In prior studies,³⁰ stage III dMMR vs pMMR colon cancers had shorter TTR ³¹ and worse outcomes when metastatic, especially if harboring BRAF^{V600E 32,33}. An important question in dMMR tumors is whether an immune checkpoint inhibitor can achieve ctDNA clearance, and this question can be addressed in the ATOMIC trial where the addition of atezolizumab to adjuvant FOLFOX was associated with a 50% reduction in recurrence and death among stage III dMMR colon cancers.³⁴

Clinical trials evaluating ctDNA-directed adjuvant approaches have been reported.³⁵ In the AGITG DYNAMIC-III trial in stage III colon cancer patients, the de-escalation arm for adjuvant chemotherapy in ctDNA negative patients did not meet the prespecified non-inferiority boundary, although there was a suggestion of non-inferiority in the pre-planned analysis of T1-3N1 patient subgroup.²⁸.³⁵ Data from the observational GALAXY trial also suggested limited benefit from adjuvant chemotherapy in stage II-III CRC with undetectable post-operative ctDNA.⁷ Further data are anticipated from the ongoing CIRCULATE Normal America Trial in patients with high risk stage II/III colon cancer.³⁶ Adjuvant treatment decisions among ctDNA positive patients may potentially be informed by quantitative ctDNA measurements or precision oncology approaches. Genotyping of ctDNA positive cases identified several genes with increased mutation prevalence in recurred vs non-recurred patients with the strongest effect observed for FLT1 (VEGFR1), a receptor tyrosine kinase involved in angiogenesis, and PREX2 that is involved in the Rac1/PI3K signaling pathway that regulates migration and invasion.^22,23 These pathways are potentially targetable, although the mutations were relatively rare in our cohort, and more commonly observed and actionable alterations were observed in ATM, KRAS, BRAF^V600E, and PIK3CA. Validation of these findings in an external dataset is warranted.

In a predictive analysis of ctDNA status among patients with wild-type KRAS tumors, those with positive ctDNA had similar DFS but significantly worse OS when treated with FOLFOX plus cetuximab vs FOLFOX alone; this finding was also seen when restricted to left-sided tumors.²¹ Since the observation is limited to OS, the potential exists for an increase in non-cancer deaths andwe lack data for subsequent treatment in patients who developed recurrence/metastasis. Of note, adding cetuximab to FOLFOX/FOLFIRI perioperatively in patients with metastatic CRC ³⁹ led to worse OS vs. chemotherapy alone. EGFR blockade can select for resistant tumor cell clones, including RAS-mutant subclones, or upregulate alternative growth pathways (HER2, MET).⁴⁰

Our study adds significantly to the literature on MRD in CC given the large size of our cohort, the high-quality data with long-term follow-up that permitted OS assessment, and the use of a tissue-free MRD assay. Limitations include the single post-operative time point for ctDNA analysis that precludes analysis of ctDNA clearance. While the analytical and clinical specificity of our tissue-free assay is high (100% and ~98%, respectively), it is possible to have non-specific positive results and to detect second primary cancers that have differential methylation patterns that overlap with CRC.^18,41 The N0147 trial preceded results of the IDEA collaboration such that all patients received adjuvant FOLFOX for 6 months. Since Enrollment on N0147 was completed 8 years prior to the FDA approval of pembrolizumab for unresectable/metastatic dMMR tumors, patients would not be expected to have received immunotherapy.

In conclusion, ctDNA status was a robust and independent prognostic biomarker that significantly improved upon standard TNM staging in stage III CC. Our data indicate that categorization of patients into low and high clinical risk groups to guide the duration of adjuvant chemotherapy is insufficient compared to ctDNA, and that post-surgical ctDNA analysis be added to TNM staging criteria for stage III CC as it provides greater prognostic precision. Our finding that ctDNA TF provides further prognostic stratification suggests its continued evaluation for clinical implementation. Given the ability of ctDNA to refine estimates of recurrence risk and death, such data may influence the frequency of radiographic surveillance, adjuvant treatment decisions, and selection of candidates for clinical trials.

Supplementary Material

PV Appendix Figures 2-9

Appendix Figure 2. Univariate analysis of ctDNA detection rate by patient or tumor variable

Appendix Figure 3. Post-surgical ctDNA tumor fraction (TF) among MRD-positive patients according to tumor DNA mismatch repair (MMR) status. TF values are shown on a log₁₀ scale. Horizontal bars represent median and interquartile range; P value by Mann–Whitney test.

Appendix Figure 4. Post-surgical ctDNA tumor fraction (TF) distribution among MRD-positive patients according to 3-year disease-free survival (DFS) event status. TF values are shown on a log₁₀ scale. Horizontal bars represent median and interquartile range; P value by Mann–Whitney test.

Appendix Figure 5. Distribution of site of tumor recurrence based on post-surgical ctDNA status in patients with stage III colon cancer treated with FOLFOX-based adjuvant chemotherapy.

Appendix Figure 6. Distribution of recurrence sites according to post-surgical dichotomized tumor fraction (TF high vs low relative to median) in ctDNA positive patients with stage III colon cancer treated with FOLFOX-based adjuvant chemotherapy.

Appendix Figure 7A–D. Clinical outcomes by postoperative ctDNA status and treatment arm (FOLFOX vs FOLFOX plus cetuximab) in patients with stage III colon cancer and wild-type KRAS. Shown are disease-free survival (DFS; A), time to recurrence (TTR; B), and overall survival (OS; C). Panel D shows OS in patients with left-sided, stage III, KRAS/BRAF wild-type, MMR-proficient tumors according to ctDNA status and treatment arm.

Appendix Figure 8 Oncoprint depicting somatic alterations detected by Guardant360 in plasma cfDNA from patients with or without tumor recurrence at 3 years. Each column represents a patient sample (blue, non-recurrent; red, recurrent), and each row corresponds to one of the most frequently mutated genes. Colored boxes indicate mutation type (black, truncating; green, missense; red, hotspot or driver variant). Percentages on the left denote the overall frequency of each gene alteration across the cohort. Recurrent cases demonstrated a higher prevalence of BRAF and DNA-damage response gene mutations (ATM, CHEK2), whereas TP53 and APC alterations were frequent in both groups. The analysis includes 374 tumor samples, excluding 53 patients without mutations in the top 30 genes.

Appendix Figure 9. The volcano plot illustrates the association between individual gene alterations and cancer recurrence among patients with detectable postoperative ctDNA. The x-axis represents the odds of detecting each gene alteration according to recurrence status at 3 years, while the y-axis indicates the statistical significance of that association. Each dot corresponds to a single gene. Red-labeled genes met the nominal significance threshold (p < 0.05); those in the upper right quadrant (e.g., FLT1, PREX2) were more frequently altered in patients who experienced recurrence. The accompanying table (below) provides the frequency of each alteration in patients with and without recurrence, along with the corresponding odds ratios and p-values. Note that neither the figure nor the table includes adjustments for multiple hypothesis testing.

Context Summary.

Key Objective

In patients with stage III colon cancer enrolled in the NCCTG N0147 phase 3 adjuvant trial, what is the clinical utility of a tissue-free ctDNA assay for minimal residual disease detection to predict recurrence, and a genomic profiling assay to identify genomic alterations?

Knowledge generated

In the largest study evaluating a tissue-free, epigenomic ctDNA assay in resected stage III colon cancer, ctDNA positivity and quantitative tumor fraction were robust and independent predictors of recurrence and survival, supporting integration into postoperative risk assessment and adjuvant decision-making. Genotyping of ctDNA-positive samples identified mutations, including FLT1 and PREX2, that were associated with increased risk of recurrence.

Relevance (R.G. Maki)

This work represents one more step in understanding when circulating tumor DNA (ctDNA) may be useful as a prognostic biomarker. Further studies should refine the groups of patients who are best followed by ctDNA and whether therapeutic interventions can change the poorer outcomes for ctDNA-positive patients.

Relevance section written by JCO Associate Editor Robert G. Maki, MD, PhD, FACP, FASCO.

Data availability

Participant data from the N0147 trial will be shared via the NCTN Data Archive, as per NCTN policies, upon request and approval.

Code availability

All analyses were performed in SAS (SAS Institute, Cary, NC) using standard procedures; no custom code was created.