ctDNA-based molecular residual disease and survival in resectable colorectal cancer

Abstract

The interim analysis of the CIRCULATE-Japan GALAXY observational study demonstrated the association of circulating tumor DNA (ctDNA)-based molecular residual disease (MRD) detection with recurrence risk and benefit from adjuvant chemotherapy (ACT) in resectable colorectal cancer (CRC). This updated analysis with a 23-month median follow-up, including 2,240 patients with stage II-III colon cancer or stage IV CRC, reinforces the prognostic value of ctDNA positivity during the MRD window with significantly inferior disease-free survival (DFS; hazard ratio (HR): 11.99, P < 0.0001) and overall survival (OS; HR: 9.68, P < 0.0001). In patients who experienced recurrence, ctDNA positivity correlated with shorter OS (HR: 2.71, P < 0.0001). The significantly shorter DFS in MRD-positive patients was consistent across actionable biomarker subsets. Sustained ctDNA clearance in response to ACT was an indicator of favorable DFS and OS compared to transient clearance (24-month DFS: 89.0% versus 3.3%; 24-month OS: 100.0% versus 82.3%). True spontaneous clearance rate with no clinical recurrence was 1.9% (2/105). Overall, our findings provide evidence for the utility of ctDNA monitoring for post-resection recurrence and mortality risk stratification that could be used for guiding adjuvant therapy.

Fig. 1. Patient inclusion and baseline characteristics.

a, CONSORT diagram illustrating the inclusion of patients from enrollment to primary and subanalyses in this study. b, Patient characteristics of the evaluable patient population (n = 2,240). EDC, electronic data capture; RX, the presence of residual tumor cannot be assessed.

Fig. 2. ctDNA status in the MRD and surveillance windows is predictive of survival outcomes in post-surgical patients with CRC.

a,b, Kaplan–Meier estimates for DFS (a) and OS (b) stratified by ctDNA status during the MRD window (negative versus positive). P = 7.55 × 10^–162 (a); P = 1.25 × 10^–25 (b). P value for the bar plots showing the association between ctDNA status and DFS or OS events was calculated based on the two-sided chi-square test: P = 4.11 × 10⁻¹⁴⁶ (a); P = 6.97 × 10⁻²⁹ (b). c,d, Forest plot depicting the multivariate analysis (including ctDNA status during MRD window and other clinicopathological factors) for DFS (c) and OS (d) in patients with CRC. P = 1.10 × 10^–96 (c, ctDNA); P = 1.43 × 10⁻⁴ (c, MSI); P = 2.55 × 10⁻¹⁵ (d, ctDNA); P = 3.56 × 10⁻⁵ (d, BRAF V600E). e,f, Kaplan–Meier estimates for DFS (e) and OS (f) stratified by ctDNA status during the surveillance window (negative versus positive). P = 1.21 × 10⁻¹⁶³ (e); P = 1.20 × 10⁻²⁰ (f). Two-sided chi-square test for bar plots: P = 1.19 × 10⁻²¹⁷ (e); P = 1.10 × 10⁻²⁹ (f). g, Forest plot depicting the multivariate analysis (including ctDNA status during the surveillance window and other clinicopathological factors) for DFS in patients with CRC. P = 6.79 × 10⁻¹¹⁸ (g, ctDNA); P = 1.88 × 10⁻⁴ (g, sex). HRs and 95% CIs were calculated using the Cox proportional hazard model; P values were calculated using the two-sided log-rank test (a,b,e,f). Various prognostic factors and their association with DFS, as indicated by HR, were analyzed across the cohort using the two-sided Wald chi-squared test (c,d,g). The unadjusted HRs (squares) and 95% CIs (horizontal lines) are shown for each prognostic factor; the vertical dotted line represents the null hypothesis (c,d,g). The DFS and OS analyses in the MRD window were landmarked from the date of the MRD timepoint, and the analyses in the surveillance window were landmarked at 10 weeks after surgery. Median DFS/OS and percentage DFS and OS at 24 months, 30 months and 36 months were estimated from the landmark timepoint. AIC, Akaike information criterion; mDFS, median DFS; mOS, median OS; NR, not reached.

Fig. 3. ctDNA positivity during the MRD and surveillance windows was significantly associated with inferior OS of patients who had radiological recurrence.

a, CONSORT diagram depicting patients included in the OS cohorts of patients with radiological recurrence. b, Bar plot showing the association of percentage of patients with ctDNA positivity and negativity during the MRD window and any time after surgery with the site of recurrence. c, ctDNA status in the MRD window is predictive of OS across all recurrence sites in post-surgical patients with CRC. Note: Given that 226 of the 500 patients with radiological recurrence had metastatic involvement at more than one site, these individual patients were counted in more than one category. d,e, Kaplan–Meier estimates for OS stratified by ctDNA status during the MRD window (d) or surveillance window (e) of patients with CRC who had radiological recurrence during follow-up. P = 9.89 × 10⁻⁵ (d). HRs and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. The OS analysis in the MRD window was landmarked from the date of the MRD timepoint, and the analysis in the surveillance window was landmarked at 10 weeks after surgery. Median OS and percentage OS at 24 months were estimated from the landmark timepoint. ctDNA⁻, ctDNA negative; ctDNA⁺, ctDNA positive; LN, lymph node.

Fig. 4. MRD positivity was associated with worse DFS compared to MRD negativity consistently across all actionable biomarkers.

a, Prevalence of genomic biomarkers in our cohort. b,c, Kaplan–Meier estimates for DFS stratified by the presence of genomic biomarkers. P = 2.17 × 10⁻⁷ (c, TMB high); P = 3.62 × 10⁻⁷ (c, MSI high). Two-sided chi-square test for bar plot: P = 3.04 × 10⁻¹⁷ (b). d, ctDNA detection percentage rates in the MRD window by genomic biomarkers. Error bars represent 95% CI. e,f, Kaplan–Meier estimates for DFS stratified by the ctDNA status in the MRD window and genomic biomarkers. The unadjusted HRs (circles) and 95% CIs (horizontal lines) are shown for each biomarker; the vertical dashed line represents the null hypothesis (f). P = 8.75 × 10⁻¹² (e, BRAF V600E); P = 7.55 × 10⁻¹⁶² (f, all); P = 8.75 × 10⁻¹² (f, BRAF V600E); P = 3.35 × 10⁻⁵ (f, ERBB2); P = 1.01 × 10⁻⁶ (f, KRAS G12C); P = 3.37 × 10⁻⁸ (f, MSI high); P = 1.33 × 10⁻⁷ (f, TMB high); P = 4.75 × 10⁻⁷⁴ (f, RAS/BRAF WT). Two-sided chi-square test for bar plot: P = 1.01 × 10⁻¹⁵ (e). HRs and 95% CIs were calculated using the Cox proportional hazard model; P values were calculated using the two-sided log-rank test (c,e,f). Analyses in d–f were landmarked from the date of the MRD timepoint. Median DFS and percentage DFS at 24 months were estimated from the landmark timepoint. Statistical comparisons of baseline characteristics by the presence of each actionable biomarker were performed using Fisher’s exact test for categorical variables and the Mann–Whitney test for continuous variables. Multivariate logistic regression analysis was used to test the association of post-surgical MRD detection with actionable biomarkers as well as baseline characteristics.

Fig. 5. Among patients with ctDNA positivity during the MRD window, ctDNA clearance is associated with superior DFS compared to no ctDNA clearance.

a,b, Kaplan–Meier estimates for DFS (a) and OS (b) stratified by ctDNA clearance status from MRD window to 3-month timepoint in MRD-positive patients receiving ACT. This analysis was landmarked at the date of 3-month timepoint. P = 2.67 × 10⁻¹⁶ (a). Two-sided chi-square test for bar plot: P = 4.65 × 10⁻¹⁰ (a). c,d, Kaplan–Meier estimates for DFS (c) and OS (d) stratified by ctDNA clearance status from MRD window to 6-month timepoint in MRD-positive patients receiving ACT. P = 4.24 × 10⁻¹⁵ (c). Two-sided chi-square test for bar plot: P = 3.62 × 10⁻⁹ (c). HRs and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. This analysis was landmarked at the date of 6-month timepoint. Analyses were performed using R software version 4.4.0. Median DFS and percentage DFS at 24 months were estimated from the landmark timepoint.

Fig. 6. Patients turning ctDNA positive after MRD negativity or transient ctDNA clearance.

a,b, Kaplan–Meier estimates for DFS (a) and OS (b) stratified by ctDNA clearance patterns (no clearance versus transient clearance versus sustained clearance) in MRD-positive patients receiving ACT. *Based on Firth’s penalized maximum likelihood; P value from log-rank test. P = 1.78 × 10–12 (a, transient clearance); P = 3.97 × 10^–26 (a, no clearance); P = 3.72 × 10^–9 (b, no clearance). Two-sided chi-square test for bar plots: P = 6.69 × 10^–28 (a); P = 3.11 × 10^–6 (b). c, Among patients with transient clearance who recurred (n = 50), the cumulative incidence plot demonstrates a timeline of patients turning back ctDNA positive (time from surgery). d, Among MRD-negative patients who had molecular recurrence before radiological recurrence (n = 165), the cumulative incidence plot demonstrated a timeline of when the patients turned ctDNA positive. e, Kaplan–Meier estimates for OS stratified by ctDNA MRD status and molecular recurrence during the surveillance window (all-time ctDNA negative versus MRD negative with molecular recurrence versus MRD positive). P = 7.09 × 10⁻¹¹ (MRD positive). Two-sided chi-square test for bar plot: P = 2.63 × 10^–31. f, Kaplan–Meier estimates for OS stratified by timing of molecular recurrence after surgery. HRs and 95% CIs were calculated using the Cox proportional hazard model (a,e) or Cox regression with Firth’s penalized likelihood hazard model (b,f). P values were calculated using the two-sided log-rank test. These analyses were landmarked at the MRD timepoint date and were performed using R software version 4.4.0. Median DFS and OS and percentage DFS and OS at 24 months were estimated from the landmark timepoint.

Extended Data Fig. 1. ctDNA status in the MRD window is predictive of DFS across all stages in postsurgical patients with CRC.

a-f. Kaplan-Meier estimates for DFS stratified by ctDNA status during the MRD window (negative versus positive) in patients with pathologic stage I (a), stage II (b), stage III (c), stage IV (d), high-risk stage II (e), and high-risk stage III (f) CRC. P = 1.66 × 10⁻⁵ (a); P = 1.05 × 10⁻³³ (b); P = 9.55 × 10⁻⁵⁹ (c); P = 3.82 × 10⁻³⁷ (d); P = 8.30 × 10⁻³⁰ (e); P = 9.66 × 10⁻³⁶ (f). Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. This analysis was landmarked from the date of the MRD time point. Median DFS and percent DFS at 24 months were estimated from the landmark time point. Abbreviations: MRD, molecular residual disease. DFS, disease-free survival.

Extended Data Fig. 2. ctDNA status in the MRD window is predictive of DFS and OS among patients with stage I-III colon cancer and stage IV CRC.

a-b. Kaplan-Meier estimates for DFS stratified by ctDNA status in the MRD window among patients with pathological stage I-III colon cancer (a) and pathological stage IV colorectal cancer (b). c-d. Kaplan-Meier estimates for OS stratified by ctDNA status in the MRD window among patients with stage I-III colon cancer (c) and stage IV colorectal cancer (d). Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. This analysis was landmarked from the date of the MRD time point. Median DFS and percent DFS at 24 months were estimated from the landmark time point. Abbreviations: MRD, molecular residual disease. DFS, disease-free survival; OS, overall survival.

Extended Data Fig. 3. ctDNA status in the MRD and surveillance windows and subsequence radiological recurrence status.

a. Sankey plot depicting the percentage of all patients in the cohort (N = 2,240) with MRD-positivity or -negativity, the percentage of patients with subsequent ctDNA status during surveillance (after ACT or observation), and the percentage of each group with confirmed radiological recurrence. b. Among patients with ctDNA status available in both MRD and surveillance windows (N = 1,664), Sankey plot shows the percentage of patients with MRD-positivity or -negativity, the percentage of patients with subsequent ctDNA status during surveillance (after ACT or observation), and the percentage of each group with confirmed radiological recurrence. Abbreviations: MRD, molecular residual disease; NA, not available.

Extended Data Fig. 4. ctDNA status in the MRD and surveillance windows is predictive of post-recurrence survival among patients who recurred.

a-b. Kaplan-Meier estimates for PRS stratified by ctDNA status during the MRD window (a) or surveillance window (b) among patients with CRC who had radiological recurrence during follow-up. Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. All analyses were landmarked at the date of radiological recurrence. Median PRS and percent PRS at 24 months were estimated from the landmark time point. c. Bar plot demonstrating the proportion of ctDNA-negative (N = 90) and ctDNA-positive (N = 79) patients undergoing curative resection for recurrence lesions. Error bars represent 95% confidence intervals. P value was generated based on the two-sided Chi-square test. Abbreviations: MRD, molecular residual disease; PRS, post-recurrence survival.

Extended Data Fig. 5. Association of actionable biomarker status with DFS.

a-b. Kaplan-Meier estimates for DFS stratified by the presence of genomic biomarkers among patients with pathological stage I-III colon cancer (a) and pathological stage IV colorectal cancer (b). *MSI high and TMB high lines overlap with each other. Two-sided Chi-square test for bar plots: P = 5.08 × 10⁻⁷ (a) c-d. Kaplan-Meier estimates for DFS stratified by MSI status (MSI High versus MSS; c) and TMB status (TMB high versus TMB low; d) for all patients in this cohort (N = 2,040). P = 9.47 × 10⁻⁹ (c); P = 2.96 × 10⁻⁹ (d). Two-sided Chi-square test for bar plots: P = 3.27 × 10⁻¹² (c); P = 2.59 × 10⁻¹² (d). Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. This analysis was landmarked from the date of the MRD time point.

Extended Data Fig. 6. Combination of MRD status, BRAF mutation status, and MSI status is predictive of DFS.

a-b. Kaplan-Meier estimates for DFS stratified by BRAF mutation status (wild-type versus V600E) and MSI status (MSI high versus MSS) among patients who were MRD-positive (a) or MRD-negative (b). P = 4.95 × 10⁻⁷ (b, BRAF WT & MSI high). Two-sided Chi-square test for bar plots: P = 6.50 × 10⁻⁶ (b). Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. This analysis was landmarked from the date of the MRD time point. Median DFS and percent DFS at 24 months are estimated from the landmark time point. Abbreviations: WT, wild-type; MRD, molecular residual disease.

Extended Data Fig. 7. ctDNA-based MRD testing is predictive of the benefit of ACT in postsurgical patients with colon cancer.

a-f. Kaplan-Meier estimates for DFS stratified by adjuvant treatment (observation versus ACT) in MRD-positive and -negative patients with: (a-b) pathological high-risk stage II or stage III, (c-d) high-risk stage II and (e-f) stage III disease. P = 1.43 × 10⁻¹² (a); P = 5.38 × 10⁻⁵ (c); P = 8.79 × 10⁻¹⁰ (e). Two-sided Chi-square test for bar plots: P = 1.39 × 10⁻⁵ (a). *HR was adjusted by age, sex, stage (A, B), and ECOG performance status. Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. The analysis was landmarked at 2 months post-surgery. Median DFS and percent DFS at 24 months were estimated from the landmark time point. Abbreviations: ACT, adjuvant chemotherapy; MRD, molecular residual disease.

Extended Data Fig. 8. MRD-negative patients with T3N0 or T4N0 disease do not derive benefit from ACT.

a-b. Kaplan-Meier estimates for DFS stratified by adjuvant treatment (observation versus ACT) in MRD-negative patients with T30N0 (a) or T4N0 (b) colon cancer. *HR was adjusted by age, sex and ECOG performance status. Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. The analysis was landmarked at 2 months post-surgery. Median DFS and percent DFS at 24 months were estimated from the landmark time point. Abbreviations: ACT, adjuvant chemotherapy.

Extended Data Fig. 9. MRD-positive patients but not MRD-negative patients with stage IV CRC derive benefit from ACT regardless of whether they received neoadjuvant chemotherapy or not.

a-b. Kaplan-Meier estimates for DFS stratified by adjuvant treatment (observation versus ACT) in MRD-positive patients with stage IV CRC who received NAC (a) or did not receive NAC (b). P = 8.34 × 10⁻⁷ (b). c-d. Kaplan-Meier estimates for DFS stratified by adjuvant treatment (observation versus ACT) in MRD-negative patients with stage IV CRC who received NAC (c) or did not receive NAC (d). HR was adjusted by age, sex and ECOG performance status. Hazard ratios (HRs) and 95% CIs were calculated using the Cox proportional hazard model. P values were calculated using the two-sided log-rank test. The analysis was landmarked at 2 months post-surgery. Median DFS and percent DFS at 24 months were estimated from the landmark time point. Abbreviations: NAC, neoadjuvant chemotherapy; ACT, adjuvant chemotherapy.