Serial Circulating Tumor DNA in Predicting and Monitoring the Effect of Neoadjuvant Chemoradiotherapy in Patients with Rectal Cancer: A Prospective Multicenter Study Free

Multidisciplinary approach of neoadjuvant chemoradiotherapy (nCRT) followed by the radical surgery is the standard of care for the locally advanced rectal cancer (LARC), which has been proven to reduce the local recurrence. However, the benefit of such approach for systemic disease control is limited. Patients' response to nCRT is highly divergent. Although the majority of patients can experience tumor response after nCRT, with a complete tumor response rate of about 10%–30% (1, 2), a substantial proportion of patients can hardly benefit from nCRT, and instead suffer from the side effects of the chemoradiation. Besides, the one-size-fits-all regimen may not be suitable for all patients with LARC who have different biological behaviors. Great efforts have been made to develop powerful methods to predict the sensitivity of tumor to nCRT, as well as approaches allowing serial monitoring of dynamic disease changes during the treatment. However, an ideal indicator still remains to be unraveled.

Serial circulating tumor DNA (ctDNA) has shown to be a state-of-the-art biomarker for monitoring the survival benefits during the multimodality treatment of patients with LARC (3–8). With advanced bioinformatic techniques, such as next-generation sequencing (NGS) and whole-genome sequencing, nowadays, processing data from circulating cell-free DNA (cfDNA) components could provide key genetic information of tumor phenotypes (9–11). Although ctDNA has proved to be a promising postsurgical predictor for long-term survival in patients with colorectal cancer, there is a lack of evidence to prove its predicting value earlier in the multidisciplinary management (4, 6, 12). Moreover, it is still under investigation whether the genomic features and the baseline ctDNA status can predict the tumor sensitivity to nCRT before commencing the treatment. Here, we report the results of a prospective multicenter clinical trial that investigated the value of tumor tissue NGS and serial ctDNA analyses in the multimodality treatment of LARC with nCRT and radical surgery. Through targeted NGS, with the currently largest panel of genes, we obtained the most comprehensive mutational information from patients' biopsies and cfDNA on status quo. In addition, via prospectively examining and monitoring the clinicopathologic parameters, our results provided the first evidence that ctDNA detection with its mutational information could present essential knowledge about the tumor response to nCRT before the surgery. Issues including the potential of using the genomic features of tumors to predict nCRT response before treatment, the value of ctDNA analyses in the dynamical surveillance of the disease, and the early prediction of tumor recurrence were also investigated. Such results could shed light on guiding the candidate selection for nCRT and refining the treatment approach. For instance, some patients will benefit from more intensified chemotherapy (e.g., induction chemotherapy before nCRT); the preoperative therapy needs to be prolonged or adjusted for patients showing poor response to nCRT; some patients with complete response to nCRT and at low risk of distant metastasis may safely avoid the radical surgery or the postoperative chemotherapy as well, etc. These possible beneficial outcomes may assist with optimized therapies for individual patients to improve the therapeutic effects and reduce side effects.

This prospective multicenter trial recruited patients with LARC treated with curative intent at eight hospitals in Beijing, China. Key eligibility criteria included: a pathologic diagnosis of rectal adenocarcinoma; pretreatment MRI (or endorectal ultrasonography if MRI was contraindicated) identified clinical T4Nany (cT4Nany) or cTanyN1b-2 disease or LARC with adverse factors, including clinical extramural vascular invasion (EMVI) positive or clinical circumferential resection margin involvement; tumor within 12 cm above the anal verge; a staging chest/abdominal/pelvic CT that demonstrated no metastatic disease; an American Society of Anesthesiologists Physical Status Classification of class I–III; and planned treatment with nCRT followed by radical surgery. Patients with a previous malignancy within the last 5 years or those of the synchronous or metachronous colorectal carcinomas were excluded.

Following the screening and baseline examinations, the enrolled patients were randomly assigned in a 1:1 ratio to receive capecitabine chemotherapy or CapeOx regimen during nCRT. All the participants received the long-term neoadjuvant radiotherapy (45–50 Gy/25 fractions/5 weeks) with three cycles of neoadjuvant chemotherapy (nCT), in which the first two cycles were concomitant with radiation. Single-agent capecitabine was given orally at a dose of 1,650–2,000 mg/m²/day divided in two doses from day 1 to 14 every 3 weeks. The CapeOx regimen consisted of a 2-hour intravenous infusion of oxaliplatin 85–100 mg/m² on day 1 and capecitabine with the same protocol as the single-agent regimen. The radical resection surgeries were performed at around 8 weeks after radiotherapy. Pathology examination of surgical specimens was performed to assess the histologic type and grade of the tumor, pathologic TNM (ypTNM) staging, and the pathologic tumor regression grade [ypTRG; following the College of American Pathologists (CAP) grading system; ref. 13], etc. Postoperatively, adjuvant chemotherapy of five to six cycles of CapeOx regimen was recommended, with other first-line regimens being optional at the discretion of the treating clinician. All patients were followed-up with routine rechecks according to the National Comprehensive Cancer Network practice guidelines version 2.2015 (14). Postoperatively, carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9) were monitored every 3 months for 2 years and then every 6 months for a total of 5 years. The chest/abdominal/pelvic CT scan was rechecked every 6 months for 1 year, and then annually for a total of 5 years. During the courses of multimodality treatment, all the clinicians were blinded to the NGS results.

Endoscopic biopsy was the main source of tumor tissues for NGS, and surgically resected tissues were also accessible for a portion of patients. Cases with unqualified tissue samples were excluded from the cohort (Fig. 1). Peripheral blood samples were collected for analyzing ctDNA, CEA, and CA 19-9 levels at four timepoints, that is, before nCRT (baseline), one cycle after the initiation of nCRT (On-nCRT), about 7 weeks after nCRT and before surgery (preop), and within 1 month after surgery (postop; Fig. 1).

The study was designed and conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Peking Union Medical College Hospital (Dong Cheng District, Beijing, China). Written informed consents were granted in sample collection, gene sequencing, and data analysis, with the obtained information authorized for publishing. This study was registered at ClinicalTrials.gov (NCT03042000).

Blood samples collected in EDTA Vacutainer Tubes (BD Diagnostics) were subjected to laboratory process within 3 hours. Peripheral blood lymphocytes (PBL) and plasma were separated with sequential centrifugation (2,500 × g, 10 minutes and 16,000 × g, 10 minutes). Genomic DNA from tissues and PBLs and cfDNA from plasma were extracted using DNeasy Blood & Tissue Kit (Qiagen) and QIAamp Circulating Nucleic Acid Kit (Qiagen), respectively. Sequencing libraries were constructed using the KAPA DNA Library Preparation Kit (Kapa Biosystems) as per the manufacturer's instruction. Barcoded libraries were hybridized to a customized panel of 1,021 genes containing whole exons, selected introns of 288 genes, and selected regions of 733 genes for both biopsy tissues and ctDNA sample. Detailed gene list was described in our previous study (15). DNA sequencing was performed with the HiSeq 3000 Sequencing System (Illumina) with 2 × 101 bp paired-end reads.

Targeted capture sequencing yielded a median depth of 753 × in tissue DNA and 1,377 × in ctDNA. Single-nucleotide variants and small insertions and deletions were called by MuTect (ref. 16; version 1.1.4). For quality control, somatic mutations in tissue samples were identified only when (i) present in <1% of the population in the 1000 Genomes Project (https://www.internationalgenome.org/), the Exome Aggregation Consortium, and the Genome Aggregation Database (https://gnomad.broadinstitute.org), (ii) not present in paired germline DNA from PBLs, and (iii) detected in at least five high quality reads containing the particular base, where high-quality reads were defined with Phred score ≥ 30, mapping quality ≥ 30, and without paired-end reads bias. ctDNA positivity was defined as, when at least one mutation in a tissue sample had also been detected in matched ctDNA. In particular, high frequency mutations in ctDNA which were persistently detected across timepoints were utilized to additionally retrieve several somatic mutations with relatively low detected frequencies. Sixteen key cancer driver genes were identified from published literature and are specially discussed in our study.

Pearson correlation test was performed to determine the linear association of mutational spectrum between current cohort and published databases. χ² test or the Fisher exact test was employed to compare difference in categorical clinicopathologic characteristics, such as pathologic complete response (ypCR)/non-ypCR etc., between ctDNA-positive versus ctDNA-negative group across time, while Wilcoxon–Mann–Whitney test was used for comparing continuous clinicopathologic parameters. Kaplan–Meier estimation was carried out to test the predictive impact of indicators, CEA/CA19-9 and ctDNA, for instance, in estimating the metastasis-free survival (MFS) preoperatively. A multivariate Cox regression analysis was carried out to examine whether the different variables were associated with MFS. Time to recurrence was calculated from the date of nCRT started until distant metastasis occurred. All data analyses were performed using R software (version 3.6.1) or GraphPad Prism software (version 8.0.2). Statistical significance was defined as P < 0.05.

Experimental approaches including patient enrolment, multidisciplinary therapy, samples collection, and data processing and analysis are shown in Fig. 1. A total of 106 patients were enrolled from August 2017 through February 2019. Two cases were excluded from analysis because of the lack of qualified biopsy tissue for NGS (Fig. 1). Patient demographics, clinicopathologic parameters, and serial ctDNA and CEA/CA19-9 levels at the four timepoints are summarized in Table 1. The median age of the 104 patients was 60 years (range, 26–74), and 64.4% were male. The baseline clinical tumor–node–metastasis (TNM) stage of patients was classified into 99 patients with (95.2%) stage cIII and five patients with (4.8%) stage cII (17). At study entry, 35% and 8.7% of patients had increased CEA and CA19-9 levels, respectively. Pathology showed poorly differentiated tumors in 17.6% (16/91) cases, with the rest being well to moderately differentiated. Except for one patient who declined the nCRT for personal reasons, all the other 103 patients received nCRT (53 cases used the single-agent capecitabine chemotherapy and 50 cases used the CapeOx regimen). Except for four patients who omitted the last cycle of nCT mainly due to the side effects, all other patients completed nCRT according to the study protocol. Six patients did not undergo the surgical treatment. Among them, one patient died from severe diarrhea and cachexia after nCRT and two patients were discovered to have disease progression during the preoperative reevaluation and were treated with additional intensified chemotherapies. The other three patients declined the surgery for personal reasons. Ninety-seven patients underwent the radical resection surgery, with no perioperative mortality. The pathologic TNM stages of patients included 30 (30.9%) ypCR, 21 (21.6%) ypI, 26 (26.8%) ypII, 16 (16.5%) ypIII, and 4 (4.1%) ypIV. The ypTRG scores included 32 cases (33.0%) of ypCAP 0, 22 (22.7%) with ypCAP 1, 40 (41.2%) with ypCAP 2, and three (3.1%) with ypCAP 3. On χ² test, no significant difference in the ypCR rate or ypTRG score was observed between the two nCT groups (P = 0.279 and P = 0.946, respectively). Among the 93 patients indicated for adjuvant chemotherapy, 75 patients (80.6%) completed the recommended five cycles, including 68% of CapeOx, 28% of singe-agent capecitabine regimen, and 4% having a halfway shift from CapeOx to capecitabine regimen. Other patients did not complete the recommended adjuvant chemotherapy regimens due to the influences of postoperative complications or patients' unwillingness. With a median follow-up of 18.8 months (range, 3.1–21.3), disease progression occurred in 13 patients (12.5%), in which nine patients were found to have distant metastasis at the preoperative reexaminations or shortly after the surgery and the other four patients were found to have distant metastasis 12–15 months after the surgery. No local recurrence was observed in this cohort. Five patients (4.8%) had died by the end of follow-up period, among which one patient died before surgery and the rest of the four patients died from disease progression.

Somatic mutations detected by targeted NGS of tumor biopsies prior to nCRT are exhibited in Fig. 2A. A total of 1,098 mutations were identified among the 104 patients, with a median of seven (range, 1–123) gene variations detected in each patient. As expected, previously reported mutations in essential driver genes, for example, APC, TP53, KRAS, FBXW7, PIK3CA, and SMAD4, were all identified with substantial frequencies in our cohort. We further compared the mutation frequencies of recurrent driver genes (>10%) in our cohort with colorectal cancer/rectal carcinoma subgroups from two public databases, Memorial Sloan Kettering Cancer Center and Catalogue of Somatic Mutations in Cancer (Fig. 2B). With the captured NGS analysis, the genetic mutation spectrum of our LARC group was highly consistent with those of the two public databases. In addition, germline mutations were detected in three patients (2.9%) including two patients with BRCA2 variations and one patient with MSH6 variation.

In this group of patients with LARC, we observed a baseline ctDNA positivity rate of 75%, which decreased rapidly to 15.6% 2–3 weeks after initiation of nCRT. It further decreased to 10.5% and 6.7% before and after surgery, respectively. Statistical analyses demonstrated that the baseline ctDNA detection rate was significantly associated with the MRI-defined EMVI (mrEMVI) status (P = 0.04), but was not associated with any other baseline clinicopathologic factors, including age, gender, tumor differentiation, and the clinical TNM stage. Neither the baseline nor the on-nCRT ctDNA detection showed significant correlation with any parameters that reflect the tumor response to nCRT, such as ypCR, ypTNM stage, ypTRG, or MRI-defined tumor regression grade (ymrTRG) (χ² test or Wilcoxon–Mann-Whitney test, P > 0.05). When analyzing with the quantitative variable, the median variant allele frequency (VAF) of baseline ctDNA showed no correlation with the tumor response parameter of ypTRG (Supplementary Fig. S1C). Notably, our results showed that none of the 29 patients experiencing ypCR presented preoperative ctDNA detection. Moreover, the preoperative ctDNA-positive rate was significantly lower in patients with ypCAP 0–1 (P < 0.001), ypCR (P = 0.02), pathologic T stage (ypT) 0–2 (P = 0.002), and postneoadjuvant MRI-defined EMVI (ymrEMVI) (-) (P < 0.002) compared with their counterparts (Table 1).

The mutation landscape of the top 20 mutated genes of the 104 patients with LARC is shown in Fig. 2A, with cases clustered by ypTRG into the well-responded (ypCAP 0–1) and the poorly responded (ypCAP 2–3) subgroups. By comparing the prevalence of mutations in the two subgroups, the mutation rates of POLD1 (P = 0.006), FAT2 (P = 0.035), and ZFHX3 (P = 0.029) were significantly higher in the ypCAP 0–1 subgroup than the ypCAP 2–3 subgroup (Fig. 2C and D). Interestingly, among the eight patients with somatic POLD1 variations, six (75.0%) experienced ypCR and the other two achieved ypCAP 1. Further analysis using the Mann–Whitney test indicated that patients harboring POLD1 mutations had a significantly higher tumor mutation burden (TMB) than patients with wild-type POLD1 genes (P < 0.001; Fig. 2E). In addition, a borderline significant trend was observed that patients in ypCAP 0–1 subgroup had a higher TMB value than their counterparts of ypCAP 2–3 subgroup (P = 0.056; Fig. 2F).

With the median follow-up of 18.8 months, 13 patients (12.5%) experienced distant metastasis. Kaplan–Meier estimation with the log-rank test showed that patients with the baseline ctDNA detection were likely to experience distant metastasis within a shorter period of time than the ctDNA undetectable patients (P < 0.05). Likewise, this correlation persisted in all the other three timepoints. The HRs revealed an increasing trend over each timepoints (baseline, undefined; on-nCRT, 6.635; preop, 19.82; and postop, 25.30; Fig. 3). When quantifying the mutation frequency of ctDNA, the correlation between the median VAF of mutations in ctDNA and the distant metastasis was also observed at all the four timepoints, with even higher statistical significance (Fig. 4B; Supplementary Fig. S2). Notably, the median VAF of baseline ctDNA >1% showed a strong predictive value for the early distant metastasis before initiation of treatment (HR, 6.549; P = 0.001; Fig. 4A and B).

As for conventional serum tumor markers, we also observed significant associations between the elevated baseline CEA (HR, 3.517; P = 0.018) or CA19-9 (HR, 5.699; P = 0.001) levels and distant metastasis (Supplementary Figs. S3 and S4). However, among the patients with normal baseline CEA/CA19-9 levels, the median VAF in baseline ctDNA >1% remained predictive of distant metastasis (HR, 16.130; P = 0.001 for CEA normal group and HR, 5.379; P = 0.005 for CA19-9 normal group, respectively; Fig. 4C and D).

According to the univariate analysis, pretreatment variables that significantly associated with MFS included median VAF of mutations in baseline ctDNA (P < 0.001), baseline CEA (P = 0.008), baseline CA19-9 (P = 0.009), tumor vertical diameter (P = 0.005), and mrEMVI (P = 0.049). No correlation between nCT regimens and MFS was observed (P = 0.622; Table 2). Post-nCRT variables that significantly associated with MFS included ymrEMVI (P = 0.006) and ypTRG (P = 0.031; Supplementary Table S1). On multivariate Cox proportional hazard analyses, only median VAF of mutations in baseline ctDNA and baseline CA19-9 remained as independent predictors of MFS (HR, 1.267; P < 0.001 and HR, 1.003; P = 0.009, respectively; Table 3).

Although most patients with LARC can benefit from nCRT, tumor response to nCRT is highly divergent. For the poor responders, nCRT brings side effects and extra medical expenses and delays the surgical treatment. Some patients even lose the opportunity for radical surgery if they develop disease progression during nCRT.

Despite extensive research, none of the traditional clinicopathologic features have been clearly defined as the predictor for patient's response to nCRT, which includes tumor size, histologic type and grade, clinical TNM stage, serum CEA level, and molecular profiling by IHC, etc. Genomic features are believed to hold great potential to predict tumor response to nCRT. However, currently related studies are few and there is a lack of concordance between cohorts owing to different methodologies and gene panels adopted. Further investigations are urgently needed.

In our study, we implemented the targeted NGS using a 1,021-gene panel to investigate the genetic alterations of tumor tissue, as well as the serial changes of ctDNA status during the multimodality treatment for patients with LARC.

To explore the potential baseline predictors for nCRT response, we analyzed parameters, including the baseline tumor size, tumor grade, clinical TNM (cTNM) staging, serum CEA and CA19-9 level, mrEMVI, MRI-circumferential resection margin (mrCRM), ctDNA detection status, as well as the mutation profile of tumor. Results showed that neither the clinicopathologic factors nor the ctDNA detection at baseline was associated with the pathologic tumor response after surgery. However, by analyzing the significantly mutated genes in this cohort, we found that harboring POLD1 mutation was significantly associated with a better response to nCRT. Previous researches have shown that POLD1, together with POLE, are essential genes for proofreading and fidelity in DNA replication. Their germline and somatic mutations can cause colorectal carcinogenesis with a hypermutated phenotype (18–20). Recently, POLE/POLD1 mutations have been proposed as one of the positive predictors for the survival benefit from immune checkpoint inhibitor (ICI) therapy (19). In addition, POLD1 mutation was reported in a patient with metastatic colon cancer who achieved an exceptional response to chemotherapy (21). In this study, tumors with POLD1 mutations appeared to have higher TMB values than those without the mutation. When further analyzing the association between TMB and tumor response to nCRT, a trend can be observed as the well-responders had higher TMB values than their counterparts, although not statistically significant (P = 0.058). We also noticed that the prevalence of POLD1 mutation in our study was a litter higher than that has been reported previously (19, 21), which may due to the deeper sequencing depth in this study, as we applied the panel sequencing instead of the whole-exome sequencing. Besides, results from this cohort, which consists of solely rectal cancer, were likely to be different from those of colorectal cancers. These results imply that POLD1 mutation and high TMB (TMB-H) may potentially be used as the predictors of good tumor response to nCRT, which needs to be proved with more evidence in future studies. Related issues are worthwhile to investigate as well, for instance, whether patients with POLD1-mutated tumors and/or TMB-H may obtain significant response to nCT alone instead of nCRT. Furthermore, currently, ICI therapy has been shown to be a promising alternative to neoadjuvant treatment for colorectal cancer in patients with microsatellite instability-high (MSI-H) and/or TMB-H (22). Given that the incidence of MSI-H is relatively lower in rectal cancer than in colon cancer, patients with LARC with POLD1 mutations, while microsatellite stable (MSS), might be the additional candidates for the neoadjuvant ICI treatment.

Consistent with results of previous studies (4, 23), we observed that ctDNA was detectable in up to 75% of patients with LARC before treatment. After one cycle of chemotherapy during the nCRT, the ctDNA-positive rate dropped sharply to 15.6%, which may reflect that most tumors responded to nCRT and shed much less ctDNA into the circulation. At the following preop and postop timepoints, ctDNA-positive rates continued to decline slightly, which may reflect the effects of further chemoradiotherapy and the radical surgery. The above results indicated that being a marker with short half-life of <2 hours, the serial ctDNA can well reflect dynamic changes of the tumor bulk in real-time, during the course of nCRT and surgery.

Similar to the results of Tie and colleagues (4) and Schou and colleagues (5), we observed no significant association of the baseline ctDNA detection with either the clinicopathologic factors at diagnosis, other than mrEMVI, or the pathologic tumor response parameters (Table 1). Unlike the previous reports (4, 5), however, our study, for the first time, revealed a significant correlation between preoperative ctDNA detection and the pathologic tumor response indicators, including ypTRG score, ypT stage, and ypCR status (Table 1). This further confirmed the value of ctDNA as an effective real-time read-out of tumor bulk, which was verified through the pathologic examination. The lack of correlation between the baseline ctDNA parameters (including the detection status and the median VAF of baseline ctDNA) and the pathologic response, however, may reflect the fact that tumors have inherently different sensitivities to nCRT.

With a median follow-up of 18.8 months, tumor progression with distant metastasis was observed in 13 of the 104 patients. Significant associations between this short- to mid-term distant metastasis and the ctDNA detection were observed at all four timepoints including baseline. Prior to the commencement of nCRT, the ctDNA detection was associated with a higher risk of distant metastasis. For the following three timepoints (On-nCRT, preoperative, and postoperative), this association became increasingly prominent. This highlights the value of ctDNA in monitoring the tumor progression. The persistent ctDNA positivity during the multimodality treatment can gradually screen out the patients at high risks of experiencing later distant metastasis, and the reliability of the prediction increases over time. Moreover, the on-nCRT timepoint seems to be the most indicative one. At this timepoint, detection of ctDNA in most patients became negative, while those with persistent positive results showed a significantly high risk of distant metastasis. In comparison with ctDNA detection, the median VAF of baseline ctDNA showed even higher sensitivity.

High VAF of ctDNA at diagnosis may be a marker for the presence of occult metastatic disease. In addition, this predictive power gradually improves by each timepoint during the multimodality treatment.

With such ability of predicting the distant treatment failure at an earlier phase, ctDNA may help in accurately classifying the stage of disease, refining the initial treatment selection, dynamically adjusting the therapeutic regimen, and planning the surgery and the postoperative treatment. Although these approaches remain to be tested in a randomized setting.

CEA and CA19-9 are conventional serum biomarkers in LARC prognosis, however, the real practice of utilizing them to predict prognosis is limited by their modest sensitivity. In this cohort of patients with LARC, with relatively advanced diseases, the abnormal rates of the baseline CEA and CA19-9 levels were merely 35% and 7.2%, respectively. For patients with normal baseline CEA and CA19-9 levels, their MFS curves can still be significantly distinguished according to the median VAF of baseline ctDNA. Comparing with CEA and CA19-9, ctDNA showed to be a much more sensitive and accurate marker in monitoring the tumor burden and predicting the prognosis. In addition to the median VAF of baseline ctDNA, our results showed that the baseline CA19-9 was also an independent predictor for MFS, suggesting that elevated level of CA19-9, although uncommon, may specifically imply the possibility of systemic disease spread in patients with LARC.

In this study, it is worth noting that mrEMVI was significantly associated with ctDNA status, which was demonstrated by the correlation between mrEMVI and ctDNA at baseline, as well as the correlation between ymrEMVI and ctDNA before surgery. mrEMVI and ymrEMVI were also shown to be significant predictors for MFS. The correlation between ctDNA and microvascular invasion has been reported previously (15). Being an indicator that can be accurately assessed by radiologists, the clinical value of mrEMVI and the potential clinical significance of its correlation with ctDNA deserve further study.

Our data validate the results of previous studies that, ctDNA has decent prognostic ability for predicting the survival of patients with cancer (4, 24). Here, we demonstrated for the first time that, the baseline status of ctDNA can predict the short- and mid-term distant metastasis before the initiation of nCRT. In previous literatures on ctDNA analyses (4, 25), however, this significant correlation was not observed until at later timepoints during the multimodality treatment. The reasons of this higher sensitivity could be the large panel size of genes we targeted in parallel and the deep sequencing depth we applied in our NGS analysis, which enabled detection of more low-frequency mutations, although such analysis requires large laboratory workload and relatively high costs at the current stage.

Limitations of this study include the modest sample size and the relatively short period of follow-up. However, the outstanding statistical power (HR, 1.267; P < 0.001) about the correlation between median VAF of baseline ctDNA and MFS in our study, together with previous results of other studies (4, 25), highlights the prognostic value of ctDNA in patients with cancer undergoing multimodality treatment. For the first time we demonstrate the correlation between the preoperative ctDNA detection and the tumor response to nCRT, as well as the prognostic value of the baseline ctDNA for the short- and mid-term distant metastasis in patients with LARC. We are waiting for the future results of this ongoing study to verify such findings and clarify the value of ctDNA for the long-term outcomes. Future studies with large sample sizes and prolonged follow-up are needed to further elucidate the related issues.

Results of this study indicate that ctDNA may be an accurate dynamic biomarker to reflect the real-time tumor bulk in patients with LARC who undergo nCRT. We demonstrated that, before the commencement of treatment, detection of ctDNA, as well as the VAF, can be used as prognostic factors for predicting distant metastasis in later stages. Besides, we observed the potential of POLD1 mutation and TMB to predict the tumor response to nCRT. In summary, we have shown that series ctDNA analysis with tumor tissue sequencing may provide accurate information to predict and monitor the tumor response to nCRT, as well as predict the prognosis of patient with LARC, which may contribute to optimizing the individualized multimodality treatment. More investigations are needed to further clarify these issues based on our findings.

J. Zhou reports grants from Beijing Science and Technology Commission (this trial was supported by the Major Grants Program of Beijing Science and Technology Commission, no. D171100002617003) during the conduct of the study. No disclosures were reported by the other authors.

J. Zhou: Conceptualization, formal analysis, writing-original draft. C. Wang: Conceptualization, formal analysis, writing-original draft. G. Lin: Supervision, funding acquisition, project administration. Y. Xiao: Investigation, methodology. W. Jia: Resources, data curation, investigation, methodology. G. Xiao: Resources, data curation, investigation. Q. Liu: Resources, data curation, investigation, methodology. B. Wu: Conceptualization, resources, data curation, investigation, methodology. A. Wu: Conceptualization, resources, data curation, investigation. H. Qiu: Conceptualization, resources, data curation, investigation. F. Zhang: Conceptualization, resources, data curation, software, investigation. K. Hu: Resources, data curation, software, investigation. H. Xue: Conceptualization, resources, data curation, software, investigation. Z. Shen: Conceptualization, resources, data curation, investigation. Z. Wang: Resources, data curation, investigation, visualization. J. Han: Conceptualization, resources, data curation, investigation, visualization. B. Niu: Conceptualization, resources, data curation, formal analysis, investigation, methodology. Y. Xu: Conceptualization, resources, software, formal analysis, visualization, methodology. Z. Yu: Software, formal analysis, visualization, methodology, writing-original draft. L. Yang: Conceptualization, supervision, visualization, methodology, writing-original draft, project administration.

The authors would like to thank all coinvestigators from the participated medical centers of this trial. We thank Dr. Yanfang Guan for expert advices and her research team. The work was supported by the Major Grants Program of Beijing Science and Technology Commission (No. D171100002617003).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.