Our pilot biomarker study investigated ctDNA during NCRT for locally-advanced rectal cancer by weekly sampling. We correlated these dynamics with initial MR imaging and MRI response (in week 2 and 5 of NCRT) as well as with outcome parameters. In most previous studies, ctDNA samples were collected once or twice during neoadjuvant treatment or solely pre- and post NRCT [17,18,19,20,21,22,23,24,25]. Thus, our study provides valuable insights into ctDNA dynamics during treatment.

Using an ultra-deep sequencing tumor-informed approach, we had high baseline detection rates of ctDNA. By weekly monitoring we could observe diverse dynamics of ctDNA during the course of neoadjuvant treatment. In the majority of patients, ctDNA levels declined during NCRT and many patients showed clearance of ctDNA after week 3 or 4. However, in some patients, after initial ctDNA decline we observed “late shedding” of ctDNA in the last weeks of treatment or even persistence of ctDNA throughout NCRT. Thus, we found various patterns of ctDNA shedding during radiochemotherapy which might reflect variable biological treatment responses.

In line with our results, a rapid decrease of ctDNA after onset of NCRT was reported in previous studies that investigated samples during neoadjuvant treatment (Zhou et al.: baseline: 75% proof of ctDNA; 2–3 weeks after NCRT initiation: 15.6% [21]; Khakoo et al.: baseline: 74% proof of ctDNA; mid NCRT: 21% [22]). This highlights the effectiveness of NCRT in rectal cancer and the potential of ctDNA as a concomitant biomarker. In future studies we suggest frequent sampling especially during the first weeks and towards the end of NCRT for further elucidation of ctDNA dynamics.

Our biomarker-study did not reveal significant correlations between dynamics of ctDNA during NCRT and pathologic response or long-term outcome of our patients. However, the small cohort size has to be considered as a limiting factor. To date, the potential of ctDNA monitoring to predict pathologic response after NCRT is still under debate and the majority of previous reports failed to proof associations [16] whilst the potential to monitor MRD after the end of treatment seems promising. A recent study by Vidal et al. evaluated ctDNA samples before and after total neoadjuvant treatment (tumor-agnostic assay) [17]. A correlation of ctDNA with pCR or ypT or ypN status could not be found. However, if ctDNA could be measured in the pre-surgery sample, a higher rate of distant recurrence and impaired overall survival was observed during follow-up [17]. Tie et al. collected liquid biopsies at baseline, 4–6 weeks after NCRT and post-surgery in 159 patients and assessed one variant per patient over time [18]. No significant association with pCR rates could be found but patients with ctDNA proof after NCRT or post-surgery had dismal recurrence-free survival. The study of Khakoo et al. included 47 patients and investigated ctDNA pre-, mid- (week 3 or 4) and post-NCRT (4–12 weeks after NCRT) as well as post surgery by monitoring up to three variants per patient [22]. An association of persistent ctDNA and the occurrence of metastases was reported. Three patients achieved pCR. In these patients, ctDNA was only detectable pre-NCRT. Apart from that observation, no significant correlations of ctDNA during NCRT or pre-surgery with pathologic response were found. In contrast, two Chinese studies report correlations of ctDNA clearance and pathologic response (pCR) [23, 25].

Challenges to compare studies and respective results imply the various approaches to detect ctDNA regarding timepoints of sampling, tumor-informed versus tumor-agnostic assays, the number of tracked variants and possible detection limits as well as heterogenous cohorts and diverse treatment regimes. Thus, the potential of ctDNA to predict pathologic response is still under investigation and especially in upcoming organ preservation strategies, further exploration of biomarkers like ctDNA with ultra-sensitive approaches is desirable. ctDNA as a biomarker for oncologic long-term outcome appears promising especially in samples after completion of treatment to monitor MRD.

Besides pathologic response in the resection specimen, we correlated the courses of ctDNA with baseline MRI and imaging response during treatment in week 2 and week 5. In this way, we investigated ctDNA as a marker to monitor treatment-response whilst neoadjuvant therapy was ongoing. Interestingly, we found a higher number of ctDNA decline over time in larger primary tumors. Furthermore, the absolute image-based tumor regression (cc) between baseline and week 2 as well as baseline and week 5 was positively correlated with ctDNA clearance during NCRT. Underlying mechanisms are unclear to date and further investigations are needed. To date, data relating ctDNA to MRI features in NCRT for rectal cancer are sparse.

Khakoo et al. correlated liquid biopsies (pre-, mid-, post-NCRT, post-surgery) with MRI response (3–6 weeks after completion of NCRT) [22]. By RECIST measurement, no association with ctDNA detection rate was seen at any time. However, the MRI tumor regression grade (mrTRG) revealed detectable ctDNA after completion of NCRT to be associated with poor mrTRG response whilst other timepoints did not correlate with mrTRG.

In a further report, the benefit of incorporating both, ctDNA features and mrTRG as complementary tools to predict pCR was suggested [25].

Zhou et al. investigated ctDNA at four times: pre- and during-NCRT as well as pre- and post-surgery [21]. Baseline detection of ctDNA was associated with baseline MRI extramural vascular invasion (EMVI) status. No correlations of ctDNA measurements pre-NCRT or 2–3 weeks after onset of NCRT and MRI response (“postneoadjuvant MRI”) were found. However, a correlation between the pre-surgical ctDNA evaluation and post-neoadjuvant MRI-defined EMVI score was reported.

Thus, the combined investigation of ctDNA and MRI features seems promising for further personalized approaches in the management of locally advanced rectal cancer.

The strength of our study is the prospective character, the mainly homogenous treatment, long period of follow-up and the tumor-informed assay based on initial tumor tissue sequencing analyzing 708 oncogenes. Therefore, in each patient multiple variants could be monitored. In contrast to others, we did not only consider the variant with the highest initial allele frequency (at baseline) [18, 25], or 1–3 variants [22] for ctDNA monitoring over time, but included all variant positions in a statistical test to determine significant residual disease. Furthermore, weekly monitoring enabled a detailed view on ctDNA dynamics and correlations with corresponding MR imaging during treatment. Weakness of our study is the relatively small cohort and a potential confounder by treatment of some patients with additional deep regional hyperthermia wherefore our results are hypothesis-generating but need to be confirmed in larger studies. Furthermore, as ultra-deep sequencing approaches are needed to detect very low tumor burden we cannot rule out detection limits with our current method despite of sequencing with a raw depth of up to 35,000x. We reported potential confounding factors like acute infections or application of Granulocyte Colony-Stimulating Factor (G-CSF) for the interpretation of ctDNA dynamics before [26]. In the recent cohort, our patients did not suffer from relevant infections or toxicities during sampling (e.g. we did not collect blood samples of the patient 112 with acute pulmonary embolism after this event any more). However, yet unknown confounders during NCRT cannot be ruled out.