Insights into ctDNA assessment to detect minimal residual disease (MRD) in localized colorectal cancer Authors: Ibone Labiano, David Guerrero-Setas, Ana Elsa Huerta, Elena Mata, Gorka Alkorta-Aranburu, Gonzalo R Ordoñez, Pedro Berraondo, Arturo Lecumberri, Patricia Ochoa, Irene Caseda, Rosalinda Termini, Borja Agirre-Mikelarena, Irene Amat, Natalia Castro, Hugo Arasanz, Maria Alsina, Ruth Vera Published: 2026 ABSTRACT Detection of minimal residual disease (MRD) by circulating tumor DNA (ctDNA) analysis arises as a promising strategy for localized colorectal cancer (CRC) management. Here, we present a pilot, prospective, single-center experience on ctDNA-based MRD detection with long-term follow-up and detailed clinical data. Patients (n = 47) with high-risk stage II and III colorectal adenocarcinoma were included. Tissue was sequenced with Oncomine Comprehensive Plus assay. Plasma (4 weeks post-surgery) was sequenced with Avenio ctDNA Targeted Panel V2 assays. ctDNA positivity was defined according to different criteria for their evaluation as predictive biomarker. Follow-up plasma samples were sequenced for selected patients. Predictive values varied depending on ctDNA positivity criteria. Only blood findings rendered a sensitivity and specificity of 37.5% and 84.6%, respectively. The “tissue-based” criteria increased sensitivity and specificity to 50% and 97.4%, respectively. Low concordance between tissue and plasma was observed, up to 40% of patients presenting variants only in plasma. Lung and peritoneal relapses were undetected by ctDNA. Our data support evidence on the potential of ctDNA analyses to detect MRD in high-risk stage II and III CRC. Important limitations, mainly sensitivity values need to be addressed. Further analyses are needed in order to position this technique in the routine clinical practice.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-58101-0. Introduction Colorectal cancer (CRC) represents the third most common cancer, and the second cause of cancer-related death1. Localized CRC is currently managed with surgical resection followed by the assessment of risk of recurrence based on clinical and pathological parameters, which guide adjuvant chemotherapy2,3. Based on this system, localized high-risk stage II or stage III CRC is currently managed with surgical resection followed by adjuvant chemotherapy based on fluoropyrimidines with or without oxaliplatin2,4. Nevertheless, this approach is lately under debate, a subset of patients can be cured by surgical resection without the need for adjuvant chemotherapy, and thus, could avoid derived toxicity, yet, another subset of patients will relapse despite adjuvant treatment and may be candidates for escalating therapies5–7. Clinical follow-up initiates after surgery or adjuvant chemotherapy, including CEA evaluation, CT scans, and colonoscopies. It extends for up to 5 years, with more intensive monitoring during the initial period, as approximately 80% of relapses occur within the first 3 years after surgery2,3. In this scenario, early detection of relapse is crucial to optimize clinical management of the metastatic setting and improve survival and quality of life of the patients3. In line with this, the detection of minimal residual disease (MRD), i.e., residual tumor undetectable by current clinical management, is emerging as critical to adjust and improve clinical decision making. Circulating tumor (ctDNA) is defined as the fraction of cell-free DNA (cfDNA) derived from tumor cells which carries genomic and epigenetic characteristics originating from the tumor, and fluctuates along disease evolution, mirroring tumor dynamics8. Detection of ctDNA may focus on genomic variants, with the use of next generation sequencing targeting panels of genes, or digital droplet PCR assays designed to capture specific variants. Alternatively, ctDNA detection can also target epigenetic marks such as methylation, with techniques also varying from the detection of specific regions to panels or even genome-based methylation studies9. The application of ctDNA analyses for the detection of MRD seeks to identify signals derived from residual tumor cells after surgery. Timing of plasma samples is a key aspect for successful MRD detection, and there is considerable consensus regarding a landmark time point at 4 to 12 weeks following surgery10, yet available studies reveal the convenience of serial sampling during follow-up and propose different sampling strategies, with a lack of consensus to set the appropriate sampling protocol9,11. Regarding ctDNA analysis methodologies, two main approaches can be distinguished. Blood-only or tumor agnostic approaches rely on genetic and epigenetic profiles identified directly in plasma. Tissue-informed approaches analyze surgical tumor tissue and select candidate variants to track in blood11. Different algorithms for ctDNA positivity are proposed based on specificities of the assays, including limit of detection and clinical significance of the detected variants. Accumulating evidence derived from studies following different approaches, supports ctDNA detection as a prognostic factor in localized CRC. Indeed, ctDNA-positive patients show higher risk of relapse and worse relapse-free survival (RFS), while ctDNA-negative patients show better clinical outcomes6,11,12. These findings have led to the design of clinical trials with the aim of establishing the clinical utility and feasibility of ctDNA analysis for the clinical management of localized CRC11. Despite promising results, clinical implementation of ctDNA still poses important caveats. Sensitivity values are far from optimal, particularly for lung and peritoneal relapses13. Clonal hematopoiesis of indeterminate potential (CHIP) is yet another challenge, as genetic variants derived from blood cells can be a source of false positives14. Additionally, there is lack of consensus regarding timing of plasma sampling and the specific methodology for ctDNA assessment9. Finally, limitations regarding economic costs should also be considered, most of the studies propose cost-intensive tests developed in external, highly specialized facilities requiring intense sequencing and bioinformatic resources, limiting the potential application of this approach to real-world scenarios. Taking all this data into account, although the potential for ctDNA is clear, more data is needed to leverage its implementation into the clinic. Herein, we present a pilot, prospective, single center experience of ctDNA analyses for MRD detection in high-risk stage II and stage III CRC patients. Results Patients’ characteristics and biological samples Initially, 51 patients were recruited, 4 patients were excluded due to insufficient cfDNA quantity isolated in the basal plasma samples, leading to a final cohort of n = 47 patients included in the study (Fig. 1). Patients´ characteristics are summarized in Table 1. Males represented the 63.8%, median age at diagnosis was 64 years old (range: 27–80). Most of the patients (85.1%) presented tumors in the colon, while roughly 15% were located in the rectum. Clinical stage was high-risk II (19.1%) or III (80.9%), all patients were treated with fluoropyrimidines with or without oxaliplatin for 3 or 6 months, according to current clinical guidelines. By data cut-off on 1st October 2025, 17% of the patients relapsed with a median follow-up for relapse-free survival of 44 months (range 4–52), while 12.8% patients died, with a median follow-up for overall survival of 44 months (range 9–52).Fig. 1Study design. Patients with diagnosis of localized high-risk stage II or stage III colorrectal adenocarcinoma were included in the study (n = 51). Following surgical resection, basal plasma samples were obtained at 4 weeks landmark time point, 4 patients were excluced due to insufficient basal cfDNA extraction. Thus, 47 patients were included in the study, basal and follow-up plasma (n = 17) samples were sequenced, additionally, tumor tissue DNA was also analyzed. Patient follow up was conducted according to clinical guidelines. ChT, chemotherapy; CT, computer tomography.Table 1Patients’ characteristics.Patients characteristics n = 47Sex, n (%)Male30 (63.8)Female17 (36.2)Age years, median (min–max)64 (27–80)Tumor location, n (%)Colon40 (85.10)Rectum7 (14.9)Union2High2Mid1Low2Stage, n (%)II high-risk9 (19.1)III38 (80.9)pTN, n (%)pT13 (6.4)pT26 (12.8)pT323 (48.9)pT415 (31.9)pN010 (21.3)pN125 (53.2)pN212 (25.5)MMR status, n (%)pMMR46 (7.9)dMMR1 (2.1)Adjuvant treatment, n (%)5-FU14 (29.8)5-FU + Oxaliplatin32 (68.1)Raltitrexed + Oxaliplatin1 (2.1)Relapse, n (%)No39 (83)Yes8 (17)Exitus, n (%)No41 (87.2)Yes6 (12.8)Biological samples, n (%)Blood-basal47 (100)Blood-follow up17 (36.2)Tumor tissue-resection46 (97.9)dMMR: deficient mismatch repair; MMR: Mismatch repair; pMMR: proficient mismatch repair; 5-FU: 5-fluorouracil. Study design. Patients with diagnosis of localized high-risk stage II or stage III colorrectal adenocarcinoma were included in the study (n = 51). Following surgical resection, basal plasma samples were obtained at 4 weeks landmark time point, 4 patients were excluced due to insufficient basal cfDNA extraction. Thus, 47 patients were included in the study, basal and follow-up plasma (n = 17) samples were sequenced, additionally, tumor tissue DNA was also analyzed. Patient follow up was conducted according to clinical guidelines. ChT, chemotherapy; CT, computer tomography. Table 1Patients’ characteristics.Patients characteristics n = 47Sex, n (%)Male30 (63.8)Female17 (36.2)Age years, median (min–max)64 (27–80)Tumor location, n (%)Colon40 (85.10)Rectum7 (14.9)Union2High2Mid1Low2Stage, n (%)II high-risk9 (19.1)III38 (80.9)pTN, n (%)pT13 (6.4)pT26 (12.8)pT323 (48.9)pT415 (31.9)pN010 (21.3)pN125 (53.2)pN212 (25.5)MMR status, n (%)pMMR46 (7.9)dMMR1 (2.1)Adjuvant treatment, n (%)5-FU14 (29.8)5-FU + Oxaliplatin32 (68.1)Raltitrexed + Oxaliplatin1 (2.1)Relapse, n (%)No39 (83)Yes8 (17)Exitus, n (%)No41 (87.2)Yes6 (12.8)Biological samples, n (%)Blood-basal47 (100)Blood-follow up17 (36.2)Tumor tissue-resection46 (97.9)dMMR: deficient mismatch repair; MMR: Mismatch repair; pMMR: proficient mismatch repair; 5-FU: 5-fluorouracil. Patients’ characteristics. dMMR: deficient mismatch repair; MMR: Mismatch repair; pMMR: proficient mismatch repair; 5-FU: 5-fluorouracil. Surgical tissue samples were obtained for 97.9% of the study cohort (all patients except 1). Basal blood samples were collected for all cases (n = 47, 100%), median time for blood sample collection was 29 days post-surgery (Supplementary Fig. S1). Additional blood samples were obtained from selected patients (n = 17, 36.2%) at relevant clinical time-points for each case (Table 1). MRD analyses Analyses were performed based on two criteria: i.e., only-blood (ctDNA strict) and tissue-based. Blood-only-based criteria Initially, we proposed a blood-only approach with the objective to detect MRD ctDNA in basal plasma samples. Due to the lack of germline data to distinguish true tumor-derived from CHIP-derived variants, we developed restrictive criteria to classify samples as ctDNA positive (ctDNA +) or ctDNA negative (ctDNA-). As aforementioned, we performed intensive sequencing reaching over 100 M paired-end reads for most of the samples (Supplementary Table S1), which allowed us to detect low VAFs. Based on the Avenio ctDNA Performance White Paper15, a specific limit of detection (LOD) was established for each sample depending on input cfDNA and millions of paired-end reads reached in sequencing. Samples presenting variants with VAFs above the specified LOD and classified as “Oncogenic”, “Likely Oncogenic” or “Moderate Support for Oncogenicity” according to the Cancer Genomics Consortium (CGC), and Variant Interpretation for Cancer Consortium (VICC)16 were classified as ctDNA + (Supplementary Table S1). Based on these criteria, 19.15% (n = 9) of patients were ctDNA +, featuring variants in CRC-related driver genes, such as APC, EGFR, KRAS, and TP53 (Fig. 2A, Supplementary Table S1). Among ctDNA + patients, 33.3% (n = 3) relapsed, while 66.7% did not relapse at the end of follow-up. Alternatively, 80.85% (n = 38) patients were ctDNA-, 13.16% (n = 5) relapsed, and 86.84% (n = 33) did not relapse at the end of follow-up (Fig. 2B). These results translate to a 37.5% sensitivity and a 84.62% specificity to detect micrometastatic relapse by ctDNA analysis. Regarding its prognostic value, there were no statistically significant differences in RFS between ctDNA + and ctDNA- patients while, ctDNA + patients showed shorter OS than ctDNA– patients (Hazard Ratio [HR] of 4.49, p < 0.05 (Fig. 2C,D).Fig. 2Clinical performance of the blood-only criteria. Number and percentage of (a) ctDNA + and ctDNA- patients following the blood-only. (b) Patients presenting relapse among ctDNA + and ctDNA-, accompanied by the sensitivity and specificity table (Fisher’s exact test). Kaplan Meyer curves of (c) relapse-free survival and (d) overall survival according to ctDNA blood-only criteria positivity (long-rank test). Salmon represents ctDNA + while blue represents ctDNA- according to the blood-only criteria. Clinical performance of the blood-only criteria. Number and percentage of (a) ctDNA + and ctDNA- patients following the blood-only. (b) Patients presenting relapse among ctDNA + and ctDNA-, accompanied by the sensitivity and specificity table (Fisher’s exact test). Kaplan Meyer curves of (c) relapse-free survival and (d) overall survival according to ctDNA blood-only criteria positivity (long-rank test). Salmon represents ctDNA + while blue represents ctDNA- according to the blood-only criteria. Tissue-based criteria. In an effort to improve these results, we developed a tissue-based approach. Tumor tissue derived from surgical resection was sequenced with the Oncomine Comprehensive plus assay (Thermo Scientific). Tissue and basal plasma were analyzed, focusing on the 17 genes covered by both panels, including: ALK, APC, BRAF, BRCA1, BRCA2, DPYD, EGFR, ERBB2, KIT, KRAS, MET, NRAS, PDGFRA, RET, ROS1, TP53, and UGT1A1. For this analysis, no specific LOD limit was established, all “Oncogenic”, “Likely oncogenic”, or “Moderate oncogenic” variants were taken into account. Based on this criteria, 59.6% (n = 28) of the patients presented variants in plasma samples, while 95.7% (n = 44) presented at least one variant in tissue samples (Supplementary Table S2). In plasma, TP53 was the most commonly mutated gene (found in 21% of the patients), followed by BRAF (in 17% of the patients) and KRAS (in 11% of the patients). In tissue, APC was the most commonly mutated gene (80% of the patients), followed by TP53 (74% of the patients), BRAF (46% of the patients), and KRAS (22% of the patients) (Supplementary Fig. 3 and Supplementary Table 2). Out of the 46 patients with both plasma and tissue samples analyzed, only 17.4% (n = 8) had coincident variants in both samples, while 41.3% (n = 19) of the patients presented variants in plasma which were not confirmed in tissue (Supplementary Fig. S2 and Supplementary Table S2). With no available germ-line sequencing, the most likely explanation for this discordancy is the non-tumoral and rather CHIP-related origin of the variants identified in plasma. Given that CHIP is associated with advanced age, we explored age distribution between patients with variants identified only in plasma and others, and observed no differences (data not shown). Regarding coinciding patients, 3 presented variants in DPYD both in ctDNA and tissue samples (Supplementary Table S2), this gene is essential in chemotherapy metabolism and is currently employed for chemotherapy dose adjustment. The same variant was also identified by routine clinical assessment by the Multiplex Ligation-dependent Probe Amplification (MLPA) with SALSA P103 technique. Additionally, 5 more patients coincided in their genetic variants in plasma and tissue (Supplementary Table S2). This observation was employed as the criterion for ctDNA + vs ctDNA- classification, with the rationale of focusing on those variants with confirmed tumor origin. Figure 3 shows a timeline with basal ctDNA positivity, clinical events, and follow-up for each of the patients. Following the tissue-based criteria, 10.64% (n = 5) of the patients were ctDNA +, all of them being male, while among ctDNA- patients (89.36%, n = 42) both male and female were represented, with 40.5% (n = 17) of females. No differences were observed in age distribution between ctDNA + and ctDNA- patients (Fig. 3). Regarding relapse, 80% (n = 4) of the ctDNA + patients relapsed, while among ctDNA- patients (89.36%, n = 42), only 9.5% (n = 4) relapsed and 90.5% (n = 38) did not relapse by the end of follow-up. These results represented a significant difference in the frequency of relapse between ctDNA + and ctDNA- patients, with specificity and sensitivity values of 50% and 97.4%, respectively (Fig. 4B). Accordingly, ctDNA + patients showed shorter RFS compared with ctDNA- (15 months vs not reached, HR = 14.6, p < 0.0001) and OS (36 months vs. not reached; HR = 8.8, p < 0.05) (Fig. 4C,D).Fig. 3Timeline for tissue-based criteria, clinical outcomes, and clinical characteristics. Schematic representation of a timeline for each patient representing months from diagnosis to end of follow-up, including basal plasma sample positivity according to the tissue-based criteria, adjuvant treatment period, relapse, and death or end of follow-up. Selected clinical characteristics are also included.Fig. 4Clinical performance of the tissue-based criteria. Number and percentage of (a) ctDNA + and ctDNA- patients following the tissue-based criteria. (b) Patients presenting relapse among ctDNA + and ctDNA-, accompanied by the sensitivity and specificity table (Fisher’s exact test). Kaplan Meyer curves of (c) relapse-free survival and (d) overall survival according to tissue-based criteria (log-rank test). Salmon represents ctDNA + while blue represents ctDNA- according to the tissue-based criteria. Timeline for tissue-based criteria, clinical outcomes, and clinical characteristics. Schematic representation of a timeline for each patient representing months from diagnosis to end of follow-up, including basal plasma sample positivity according to the tissue-based criteria, adjuvant treatment period, relapse, and death or end of follow-up. Selected clinical characteristics are also included. Clinical performance of the tissue-based criteria. Number and percentage of (a) ctDNA + and ctDNA- patients following the tissue-based criteria. (b) Patients presenting relapse among ctDNA + and ctDNA-, accompanied by the sensitivity and specificity table (Fisher’s exact test). Kaplan Meyer curves of (c) relapse-free survival and (d) overall survival according to tissue-based criteria (log-rank test). Salmon represents ctDNA + while blue represents ctDNA- according to the tissue-based criteria. Next, the clinical characteristics of the patients were examined in greater detail. There was only one patient with ctDNA + that did not present relapse at the end of follow-up (Pt_4; false positive), with a pT3N2 tumor (Fig. 3, Supplementary Table S3). True positive patients, defined as those ctDNA + and presenting relapse, were characterized by pT4N + (stage III) and distant metastases mainly to the liver (Fig. 3, Supplementary Table S3). Tissue-based ctDNA positivity was able to detect micrometastatic relapse before radiological confirmation, with lead times ranging from 4 to 43 months (Supplementary Table S3). False negative patients, defined as those ctDNA– but presenting relapse, accounted for 9.5% (n = 4) of the ctDNA- patients. These patients presented pT4, pN + or pN- tumors at surgical resection. One of the patients presented local relapse (Pt_28), another presented liver and bone metastases (Pt_13), while the other 2 patients (Pt_7 and Pt_45) presented peritoneal or lung metastases with low disease burden. Time from basal sample collection to relapse was similar to that observed in true positive patients, ranging from 5 to 34 months (Supplementary Table S3). Finally, ctDNA- patients not presenting relapse (true negatives) were more frequently characterized by lower pTNM stages; nevertheless, pT4N + cases were also present (n = 4, 10.5%) (Supplementary Table S3). Serial plasma sample analyses Serial plasma samples were analyzed for selected cases in order to deepen our understanding of discordant results and unravel the potential added value of a more timely analysis. There was only one false positive patient (Pt_4), which bore a variant in APC (APC p.Gln757*) at the basal ctDNA sample. Clinical history of this patient showed polyposis featuring numerous (10 to > 40) polyps histologically diagnosed as adenomas, which were resected in follow-up colonoscopies. Germline analysis in APC and MUTYH did not detect any alterations (panel TruSight Hereditary Cancer (Illumina), VariantStudio 3.0 e IGV (Integrative Genomics Viewer)) (Fig. 5A). Two additional plasma samples were analyzed. ctDNA follow-up_1 sample, obtained roughly two months after completion of adjuvant treatment, showed the same variant in APC, albeit at a slightly lower VAF. In sample ctDNA follow up_2, obtained during follow-up, the same variant was identified in APC, at a higher VAF, as well as a new variant in BRAF (V600E) (Fig. 5B). Another imaging control was performed in this patient after completion of follow-up (December 2025), with no evidence of relapse detected.Fig. 5Timeline for the tissue-based criteria, clinical outcomes, and VAF dynamics in the follow-up samples. (a) Schematic representation of patients with available follow-up samples representing months from diagnosis to end of follow-up, including basal and follow-up plasma sample positivity according to the tissue-based criteria, adjuvant treatment period, additional treatments, and clinical events, relapse, and exitus or end of follow-up. (b) VAF dynamics in the false positive patient. (c) VAF dynamics in the false-negative patient. (d) VAF dynamics in true positive patients. Salmon represents ctDNA + while blue represents ctDNA- according to the tissue-based criteria. Timeline for the tissue-based criteria, clinical outcomes, and VAF dynamics in the follow-up samples. (a) Schematic representation of patients with available follow-up samples representing months from diagnosis to end of follow-up, including basal and follow-up plasma sample positivity according to the tissue-based criteria, adjuvant treatment period, additional treatments, and clinical events, relapse, and exitus or end of follow-up. (b) VAF dynamics in the false positive patient. (c) VAF dynamics in the false-negative patient. (d) VAF dynamics in true positive patients. Salmon represents ctDNA + while blue represents ctDNA- according to the tissue-based criteria. Additional samples were obtained from 3 of the 4 false negative patients. These analyses revealed the conversion to ctDNA + of 1 of the patients. Interestingly, this patient showed liver and bone relapse and died shortly after relapse (Fig. 5A,C). Moreover, this patient showed a high proliferative index as measured by immunohistochemistry for Ki67 (15% positivity). The other two patients remained ctDNA- at follow-up, both samples were obtained after radiological relapse, one patient showing peritoneal spread and the other local relapse (Supplementary Table S3, Fig. 5A). Regarding true positive patients (n = 4), follow-up samples were obtained either during the adjuvant period or at radiological relapse. In 3 of them, the same or new variants were identified in these samples, whereas in the other, the variants identified in the basal sample were not detected in the follow-up sample. Of note, this sample was obtained 5 months before the radiological relapse (Fig. 5A,D). Finally, follow-up samples after the completion of adjuvant therapy were available for 21.43% (n = 9) of the true negative patients. These samples were confirmed as ctDNA, following the previously defined tissue-based criteria (Fig. 5A). Discussion Adjuvant chemotherapy for high-risk stage II and stage III resected CRC patients is recommended, although uncertainty still exists regarding efficacy and exposure to toxicity. Thus, there is a need for better predictive tools, in this regard, ctDNA-based MRD detection is an emerging prognostic factor that correlates with prognosis and might be used for tailoring clinical decisions on adjuvant therapy. In this study, we showed that the predictive value of ctDNA varied depending on the ctDNA positivity criteria. On the one hand, when considering only findings in plasma samples (blood-only criteria), sensitivity and specificity values were 37.5% and 84.62%, respectively, without correlation with survival outcomes. On the other hand, sensitivity and specificity significantly increased when tissue findings were incorporated. The tissue-based criteria increased sensitivity and specificity values up to 50% and 97.4%, respectively, and was predictive for RFS and OS. Nevertheless, we observed low concordance between tissue and plasma findings, with up to 40% of the patients presenting variants only in plasma, which were not confirmed in tissue. Finally, ctDNA was unable to detect lung and peritoneal relapses in most cases. Regarding concordance between tissue and plasma, it is important to note that up to 40% of the patients presented oncogenic variants in plasma samples, not confirmed in tissue. Most of the variants presented at low VAFs. This percentage is in line with previous reports sequencing paired white blood cells and plasma samples in patients with cancer, with percentages ranging from 29–52%17–19. This phenomenon is particularly challenging in the context of localized cancer, where both tumor-derived and CHIP variants may show similar VAFs20. Blood-only variants were predominantly located in TP53, BRAF, ERBB2, and KRAS genes. TP53 is a well-known CHIP gene14. BRAF, KRAS, and ERBB2, although less common, are also reported as genes involved in CHIP17,21. Moreover, genetic variants in KRAS have been associated with prior chemotherapy treatment22. Thus, CHIP arises as a critical limitation for blood-only approaches, and methods for controlling these potential false-positives are urgently needed. As previously reported, tissue-based criteria improves sensitivity and specificity values to 50% and 97.4%, in line with values reported previously in studies analyzing a single time-point post-surgery, generally established at 4 to 12 weeks after surgical intervention11. Indeed, low sensitivity is a major concern, as it reflects failure to detect relapse in a subset of patients, especially for tumors located in low-shedding sites or relapses with low disease burden, which are also associated with better clinical outcomes12. In our study, three of the four false positive patients presented local, lung, or peritoneal relapse. Moreover, when serial samples were analyzed, patients presenting local, lung, or peritoneal relapse remained negative, even in the sample obtained after relapse. By contrast, one of the false negative patients relapsed in the liver and could not be captured by ctDNA. Of note, this patient relapsed rapidly during adjuvant treatment, being ctDNA- at the basal sample but converting to ctDNA + 5 months post-surgery. These results support serial sampling, which has been reported to significantly improve sensitivity and specificity values compared to analyses focusing on a single timpe-point23–27. Indeed, ctDNA analysis limited to a single landmark time point, generally stablished at 4 to 12 weeks after surgery, fail to capture up to 10–25% of the cases in which ctDNA elevation may be a later event27. In line with this, ctDNA negativity persistence is particularly associated with good prognosis. In our study, follow-up samples were available for 9 patients, who were ctDNA- at the basal sample, all of whom remained negative. Interestingly, these findings are in line with recent suggestions that persistent ctDNA negativity after adjuvant treatment may be regarded as a surrogate marker for long-term survival outcomes12. The only false-positive patient is an interesting case with oncogenic variants detected in serial plasma samples and confirmed in tissue. This patient presented multiple polyps in the colon that were periodically resected, although germline testing ruled out an hereditary syndrome. Of note, this patient showed a new variant in the last plasma follow-up sample, although not accompanied by clinical relapse to date. Altogether, our data are in agreement with current state-of-the-art reports11 which have led to the design of interventional clinical trials proposing adjuvant chemotherapy intensification or de-escalation guided by ctDNA analyses. Clinical trials employing the tumor-guided approaches include DYNAMIC (ACTRN126150003815837,) and DYNAMIC-III trials (ACTRN1261700156632528,), the GALAXY (UMIN000039205) observational trial linked to the ALTAIR (NCT04457297) and VEGA (jRCT1031200006) trials, CIRCULATE-US (NCT05174169), CIRCULATE-SPAIN (2021–000,507-20), and SAGITARIUS (NCT06490536). Trials encompassing blood-only approaches include PEGASUS (NCT04259944), ACT-III (NCT03803553), or the PRODIGE-70-CIRCULATE trial in France (NCT04120701)11. The most advanced trials are DYNAMIC and DYNAMIC-III. DYNAMIC assessed safety of ctDNA-guided adjuvant therapy versus the standard approach for stage II colon cancer, showing no impact on RFS and OS at 5-year follow-up, together with diminished chemotherapy use7. DYNAMIC-III, investigating non-inferiority of the ctDNA-guided adjuvant therapy for stage III colon cancer, did not meet its primary end point, the ctDNA-guided arm showed slightly lower RFS28. The preliminary negative results might be related to insufficient sensitivity for ctDNA to detect MRD. Our study represents a real-world pilot experience conducted in a single center, the small sample size included in the study represents a major limitation, with 47 patients included, 5 patients classified as ctDNA + and 8 patients presenting relapse events during the follow-up period. This limitation necessarily restricts the significance of our results, which should be interpreted with caution, and hinders the performance of multivariate analyses that could confirm ctDNA as an independent factor for RFS and OS. Of note, all but one ctDNA + patients were pT4N +, while among the ctDNA- patients, pT4N +, although present, were indeed a minority. No differences were observed considering sex and age between ctDNA + and ctDNA- patients. It is important to note that our results are consistent with larger studies in the field and contribute with real-world data building on the use of ctDNA for MRD detection in CRC. By contrast, the main strengths of the study are the long-term clinical follow-up of the patients, with a median of 44 months, which largely encompasses the first 3 years after surgery, during which approximately 80% of the relapses occur2,3, thus conferring high credibility for ctDNA prediction2,4. Additionally, commercially available approaches for plasma and tissue sequencing were employed. Sample processing was performed in research facilities, thus providing real-world data on ctDNA analysis and empowering the learning process and self-controlled initiatives in the context of a public tertiary University Hospital. Future steps in the direction of leveraging the full potential of ctDNA testing for this scenario include the implementation of serial testing, which will enable real-time data on the evolution of tumor dynamics. Although currently tumor-guided approaches reach more satisfactory results, blood-only approaches hold great potential11. Our study employed different assays for tissue and plasma sequencing, a circumstance that arises from the real-world translational scenario, where tissue sequencing data is performed in routine clinical diagnosis while plasma studies have been performed with research purposes. Tissue samples were analyzed using the assay currently employed in routine clinical practice at our institution (Oncomine Comprehensive Plus Assay). As this assay is not validated for cfDNA assessment, we selected the AVENIO ctDNA Targeted Panel v2 based on the coverage of most important genes bearing driver genetic alterations in CRC, as well as the opportunity to perform deep-sequencing which enabled the detection of genetic variants at low VAFs. This scenario introduces potential discordance in the detection of genomic variants, due to the use of different software for variant calling, which may have resulted in false-negative cases in plasma samples. In line with this, ctDNA analysis for the detection of MRD needs to capture the driver genomic alteration mirroring tumor burden dynamics. The genomic landscape of CRC is characterized by the tumor suppressor gene, APC which governs colorectal carcinogenesis and presents genomic variants in 80–90% of the CRC tumors29. Nevertheless, APC remains undruggable and is not included in recommendations for NGS testing in CRC30 and thus is not the focus of commercial pan-cancer liquid biopsy assays. This scenario highlights the need for consensus in the methodologies for ctDNA analyses in technical aspects such as the specific variants to include, limits of detection and variant calling pipelines. Interestingly, assessing alternative characteristics in liquid biopsies, including epigenomics31,32, fragmentomics33, miRNAs34, extracellular vesicles35, or platelets36 may improve the performance of blood-only approaches. Among them, the combination of epigenomic and genomics analyses represents the most studied approach, with promising results outperforming genomic-only studies27,31,32. Conclusions In conclusion, our study contributes to the knowledge on the use of ctDNA analysis for MRD detection in CRC as an emerging valuable tool. Numerous clinical trials are currently ongoing, and their results, together with constantly evolving technical improvements, will shed light on the important caveats and enable the safe application of this approach in clinical practice. Methods Study design, patient recruitment, and sample collection The study was approved by the Clinical Research Ethics Committee of the HUN, in line with the ethical guidelines of the 1975 Declaration of Helsinki (sixth revision, 2008) as reflected in a priori approval by the institution’s clinical research committee (PI_2020-118). Written informed consent was obtained for all patients. Patients with histological diagnosis of localized, high-risk stage II and III CRC eligible to adjuvant chemotherapy after surgery were selected and recruited at the Oncology Department of the Hospital Universitario de Navarra (HUN) between February 2021 and December 2022. After signature of the informed consent, biological samples including tumor tissue derived from surgical resection and blood were collected. Blood samples (two 9 ml EDTA-Vacutainer® tubes) were collected at two time-points: 4 weeks after surgery (Basal sample) and at time points of interest during follow-up (Follow-up sample) for selected patients. Patient follow-up was extended until 1st October 2025 and was conducted at HUN based on current clinical guidelines. Relapse-free survival (RFS) is defined as the time from diagnosis until confirmation of relapse by imaging and clinical criteria. Overall survival (OS) is defined as the time from diagnosis until patient death. Patients not presenting relapse or death were censored at the last follow-up date (Fig. 1). Plasma ctDNA analyses: cfDNA extraction, library preparation, sequencing and variant identification Plasma samples were obtained within 1 h after blood extraction with support from the Navarrabiomed Biobank. Plasma was separated following two consecutive centrifugations (2630 × g 10 min 4ºC, 3000 × g 10 min 4ºC), 2–8 mL of plasma were obtained from each sample. Thereafter, the Avenio ctDNA Targeted Kit V2 and the Avenio ctDNA Expanded Kit V2 (Roche) assays were employed for cfDNA isolation and library preparation following manufacturer´s instructions. Briefly, cfDNA was extracted with the Avenio DNA extraction subkit based on spin-columns, quantified by Qubit HS DNA 1 × assay (ThermoFisher), and size-profile assessed by Tape Station HS D5000 kit (Agilent). Sequencing libraries were prepared by the ligation of sample adapters, and pre-enrichment PCR. Immediately, enrichment was performed by hybridization with the Targeted or Expanded panel covering 17 and 77 solid tumor-related genes, respectively, with subsequent enrichment PCR. Libraries were quantified with Qubit HS DNA 1 × assay (Thermo Fisher Scientific) and size-profile assessed by Tape Station HS D5000 kit (Agilent). Libraries were manually pooled in equimolar pools. Sequencing was performed using the 300-cycle SP or S1 flow cells on the NovaSeq 6000 (Illumina), reaching over 100 M paired-end reads for most of the samples, with the aim of obtaining extended coverage for low variant allele fraction (VAF) variants. The Avenio Oncology Analysis Software v.2.1 was employed for the secondary analysis. Data was mapped to the human genome assembly 19 GRCn37/hg19 as the standard reference genome. The clinical relevance of gene variants was annotated by referring to available public databases ClinVar, Franklin, and Varsome. All variants (unfiltered file) were manually inspected, only variants classified as “Oncogenic”, “Likely oncogenic” or “Moderate oncogenic” by the Cancer Genomics Consortium (CGC), and Variant Interpretation for Cancer Consortium (VICC)16 were taken into account. A control library prepared on the Structural Multiplex cfDNA Reference Standard (Horizon) was included in the library preparation and sequencing pool; expected single-nucleotide variants (SNV) and structural variants were properly detected. Tumor tissue analyses: DNA extraction, libraries, sequencing and variant identification DNA and RNA were extracted from formalin-fixed, paraffin-embedded (FFPE) tissue after macrodissection of tumor areas with ≥ 40% tumor content selected by a pathologist. DNA and RNA were extracted using QIAamp DNA FFPE Tissue kit (Qiagen) and High Pure FFPET RNA Isolation Kit (Roche Diagnostics), respectively. DNA and RNA concentration were determined using Qubit DNA High Sensitivity (HS) dsDNA Kit and Qubit™ RNA High Sensitivity (HS), Broad Range (BR) (Thermo Fisher Scientific). 20 ng of DNA were treated with Uracil DNA Glycosylase (Thermo Fisher Scientific) to remove deaminated cytosines that would cause non-biologically relevant C > T transitions. Complementary DNA was synthesized from 20 ng RNA using the Ion Torrent™ NGS Reverse Transcription kit (Thermo Fisher Scientific). Library preparation was performed in the Oncomine™ Comprehensive Assay Plus (OCA-plus) Chef-ready panel on the Ion Chef™ System (Thermo Fisher Scientific). This panel is designed to detect genetic variants in 517 cancer-related genes. Libraries were quantified by the Ion Library TaqMan Quantification kit (Thermo Fisher Scientific) and were diluted to a concentration of 100 pM. Automated template preparation and chip loading were also performed on the Ion 550™ chip with the Ion Chef™ System and sequenced on an Ion GeneStudio™ S5 Prime System (ThermoFisher Scientific). Sequences were analyzed with the automatic workflow Oncomine Comprehensive v3-w4.3- DNA and Fusions in Ion Reporter™ software 5.20.8.0 (Thermo Fisher Scientific, USA) to detect and annotate the somatic variants, following manufacturer’s instructions. Quality control (QCs) of sequencing output was evaluated by means of the Coverage Analysis plugin on the Torrent Suite 5.18.1 (ThermoFisher Scientific). A value < 0.5 of the Median of the Absolute value of all Pairwise Differences (MAPD) was considered as a threshold for the reliability of the CNV analysis. The data was mapped to the human genome assembly 19 GRCn37/hg19 as the standard reference genome. The clinical relevance of gene variants was annotated by referring to available public databases ClinVar, Franklin, and Varsome, only variants classified as “Oncogenic”, “Likely oncogenic” or “Moderate support for oncogenicity” by the Cancer Genomics Consortium (CGC), and Variant Interpretation for Cancer Consortium (VICC)16 were taken into account. Statistical analysis Statistical analyses and data visualization was performed with R (v 4.2.1), GraphPad (v 10.1), and SPSS 25 software. Data are described as median with minimum and maximum values for quantitative variables, and with absolute frequencies and percentages for qualitative variables. For statistical analyses, all tests were two-tailed and p values < 0.05 considered statistically significant. Contingency analyses were performed using Fisher´s exact test, and survival analyses were performed using the Kaplan–Meier method and the long-rank test. Investigators were blinded to the patients´ clinical outcomes when analyzing sequencing results. Supplementary Information Supplementary Information. Supplementary Information.