Characteristics of patients
The study design and patient enrollment are presented in Fig. 1. A total of 122 patients (78 men and 44 women; median age, 52 years) were enrolled who had sought treatment for colorectal cancer (CRC) between Nov. 2012 and Apr. 2019, and had custom ctDNA fingerprints test done and monitored up to Aug. 2020. All the patients had whole-exome sequencing done on their primary tumor samples. Of them, 102 (83.6%) patients received at least one dose of chemotherapy (FOLFOX). 31 patients were excluded from subsequent analysis because they were lost of follow-up or had a follow-up time shorter than 6 months. In the remaining 71 patients, 48 patients were in advanced disease (stage IV) at the time of diagnosis and the other 23 patients were in lower or uncertain stages. 38 of the advanced stage patients had their ctDNA fingerprints monitored for 3 times or more, up to 8 times.
Genetic landscape of CRC and their predictive role for prognosis
In order to design and monitor the ctDNA fingerprints for each individual, we applied WES sequencing to eligible patients. After quality filtering, the genetic profiling including somatic mutations, DNA copy number variations (CNV) and mutational significance were assessed subsequently. Among the 122 CRC patients, 96.72% had at least one genomic alteration (Fig. 2). We identified TP53 (70%), APC (59%) and KRAS (38%) as the most frequently cancer-associated mutations (Fig. 2A), consistent with previous studies . Compared with TCGA and MSK CRC database, APC mutation frequency was significantly lower (59.0% vs 76.9% for TCGA and 59.0% vs 79.0% for MSK), similar to the result observed by Li et al (59.0% vs 65.1% for CCRC) (Fig. 2B). Furthermore, we found that KCNK15 and ABHD16B were the most common genes with CNV amplified, and the number of CNV amplified genes was greater than the number of CNV deleted genes (Figure S1).
We then evaluated the genomic alteration as predictive biomarker for CRC patients, and found that AHNAK, FRG1, MUC2 and XIRP2 mutations are associated with survival times (Fig. 2D, Fig. S2). Various of sites of AHNAK mutations had been identified and the median OS of patients with AHNAK-wt and AHNAK-mut were 30 months and 35 months respectively (HR 0.66; 95% CI 0.35–1.23; p = 0.013) (Fig. 2C). The CNV of HOXB-AS1 also has been suggested as a candidate biomarker of clinical benefits (HR 0.45; 95% CI 0.21–0.94; p = 0.025). Finally, we verified the cytoband 1q31.3, 2q24.3, 2q32.1 and 3q24.3 etc. were associated with OS (Fig. 2E), suggesting these alterations were candidate prognostic markers for CRC.
Prognostic impact of ctDNA fingerprints at baseline
The ctDNA fingerprints of each patients were designed following the method we described previously and their abundance were reported as cancer cell fraction (CCF) . In 122 enrolled CRC patients at the baseline, 81 (66.4%) had positive CCF (CCF ≥ 0.25%), and 41 (33.6%) had undetectable CCF (CCF < 0.25%) (Table S1).
We first classified the 122 patients to two groups according to their baseline CCF values, a ctDNA-high group (CCF > median) and a ctDNA-low group (CCF < median). The start time of OS is set at the time of the initiation of chemotherapy, and the cut-off date is Aug. 2020. There is a strong correlation between CCF and overall survival (OS) based on univariate analysis. The OS of the ctDNA-low group is significant longer than the ctDNA-high group (HR, 2.89; 95% CI 1.45–5.79; p = 0.0027). The median OS is 33 months in the ctDNA-low group and 28 months in the ctDNA-high group, respectively (Fig. 3A). Conversely, if we divided the patients according to their OS, the group with OS longer than the median (30 months) has a significantly higher ctDNA levels than the group with OS shorter one (Fig. 3B). Moreover, we analyzed the correlation between disease stage at diagnosis of the patients with the baseline ctDNA level. Thirty-one patients had no available stage data and were excluded from this analysis. In the remaining 91 patients, 75 were at advanced stage (stage IV, n = 75), and 16 were at lower stages (stage II, n = 5; stage III, n = 11). The advanced stage patients had a significantly higher ctDNA levels than those at the lower stages (Fig. 2C). In addition, no significant difference in the ctDNA concentration was detected to be associated with the patients’ other characteristics, such as gender and ages (Table 1) in multivariant analysis.
Table 1 Clinicopathological characteristics of patients and association with baseline ctDNA levels
ctDNA level, median (range %)
Sex (median age, years)
Primary tumor sites
0.45 0.194 —
0.789 0.524 0.244
0.842 0.133 —
ctDNA fingerprints monitoring in patients after chemotherapy
In order to evaluate the utility of ctDNA fingerprints as a therapeutic biomarker, we analyzed 102 CRC patients who had received chemotherapy (Fig. 1). As mentioned above, the start point OS of patients was corrected using the initiation of chemo-therapy instead of diagnostic time so can assess treatment response more precisely. 31 patients were therefore excluded from subsequent analysis due to loss of follow-up. Of the remaining 71 patients, 29 patients (40.8%) were deceased and 42 patients (59.2%) were alive at the cut-off date. To explore the predictive roles of baseline ctDNA in the 71 patients included, we did Kaplan-Meier analysis. The ctDNA-low group had statistically significant better prognosis than the ctDNA-high group (HR, 2.59; CI 95% 1.23–5.46; p = 0.0088) (Fig. 4A).
Based on the study by Zhang et al., the dynamics of blood ctDNA is a predictor of immunotherapy benefits in a variety of advanced cancers . We then sought to explore the predictive value of ctDNA fingerprints in CRC patients who received chemotherapy. We defined the baseline as CCF measured before the start of treatment in each patient. For patients who had multiple measurements post chemotherapy, we took the median values to represent their post chemotherapy levels. The ΔCCF of patients were defined and determined to reflect change of ctDNA fingerprints during the course of treatment (Figure S3). Of the 71 patients with OS and CCF data, 23 (32.4%) had a positive baseline (CCF ≥ 0.25%) and increased CCF (ΔCCF > 0) and had the worst outcomes; 27 (38.0%) had a positive baseline (CCF ≥ 0.25%) but decreased CCF (ΔCCF < 0) and had an intermediate outcomes; and 21 (29.6%) had undetectable CCF at both the baseline and post-treatment (CCF < 0.25%) and had the best OS (Fig. 4B; increased vs undetectable, p < 0.0001).
ctDNA fingerprints monitoring in advanced CRC patients after chemotherapy
Since baseline CCF as predictive biomarker has been confirmed in 122 CRC patients and the subset of 71 patients received standard chemotherapy CRC patients, we postulate that baseline ctDNA fingerprints can also be a benefit predictor for advanced CRC patients received chemotherapy. To explore this postulate, we focused on the 48 patients diagnosed with advanced CRC and split them into two groups based on the median of their CCF values. A significant association was found between the baseline CCF and OS. The patients in the ctDNA-high group (baseline CCF > median) had a shorter OS after chemotherapy, while the patients in the ctDNA-low group (ctDNA levels < median) had a longer OS (HR, 0.63 95% CI 0.23–1.64; p = 0.037). The median OS was 54 months for the ctDNA-low group and 51 months for the ctDNA-high group (Fig. 4C).
We further evaluated whether monitoring ctDNA change along the course of treatment have clinical value to patients. The 48 advanced CRC patients had paired ctDNA fingerprints measurements at the baseline and post chemotherapy. We classified them into three groups based on the dynamic of their ctDNA fingerprints.16 (33.3%) patients had positive baseline (CCF ≥ 0.25%) and increased CCF (ΔCCF > 0). They had a significantly shorter OS. 18 (37.5%) patients had positive baseline (CCF ≥ 0.25%) and decreased CCF (ΔCCF < 0) and 14 (29.2%) patients had undetectable CCF (CCF < 0.25%). Both the later two groups had better clinical outcomes (Fig. 4D-E; increased vs undetectable p = 0.0035). The dynamics of ctDNA fingerprints can also be characterized by the difference of their levels between the definitive test (the first done concurrent or within 10 days standard clinical evaluation) and the baseline, which we had previously defined as ΔCCF and used to predict treatment response .
From the logistic perspective in clinics, using CCF of the definite test, instead of the median value of multiple follow-up tests, is much more feasible for oncologists and will help them to make clinical decision in a timely fashion. We therefore tested ΔCCF, defined as the CCF measured at the definitive test minus the CCF measured at the baseline of individual patient, for the use of a predictive marker of early chemotherapeutic response and resistance. Of the 48 patients with advanced CRC and received chemotherapy, 39 were included in the subsequent analysis because the other 9 had undetectable ctDNA at both the baseline and post-chemotherapy measurements and were excluded. The 39 patients were divided into three groups according to their ΔCCF (Fig. 4F). 12 patients had ΔCCF > 3.0 and also had the shortest OS. 13 patients had ΔCCF < -3.0 and the best clinical outcomes. Additionally, the mortality rate between the groups are different. 11 out of the 12 patients with ΔCCF > 3.0 deceased while only 4 out of the 13 patients ΔCCF < -3.0 did so (Fig. S4). Moreover, we evaluated the degree of tumor mutation burden (TMB) and microsatellite instability (MSI) at the baseline of treatment. Only 10 TMB-high patients (TMB > 350) and 6 MSI-high patients (MSI > 3.5%) were identified and none of the group show significant association with the OS (Fig. 4F). Therefore, ΔCCF (R2 = 0.773) is a more accurate prognostic biomarker than TMB (R2 = 0.029) and MSI (R2 = 0.033).
ctDNA fingerprints as a diagnostic and prognostic marker post-chemotherapy
We tested the correlation of the changes of ctDNA fingerprints level with the standard imaging-based conventional clinical evaluation for tumor response to chemotherapy in the advanced CRC patients. In total, 34 patients with were included in the analysis because among the 38 advanced CRC patients with ctDNA fingerprints monitored longitudinally (≥ 3 times), 4 had no evaluable imaging data. Among the 34 patients, 24 had detectable ctDNA (70.6%), while the other 10 (29.4%) had undetectable ctDNA both in baseline and post treatment (Fig. S1). Of the 24 patients with positive CCF, 9 showed increased ctDNA levels (ΔCCF > 0) and the other 15 showed decreased ctDNA levels (ΔCCF < 0). The tumor response imaging clinical data of those 34 patients were provided in the Supplementary Table S2. Based on the trends of ΔCCF and clinical imaging performance of the 34 patients, they were divided into three groups (Fig. 5). A "consistent" group (18 patient, 53.0%), which includes patients with disease progression by imaging standard also had increased ΔCCF or patients with disease remission also had decreased/undetectable ΔCCF. An “inconsistent” group (5 patients, 14.7%) which includes patients with opposite trends of imaging diagnosis as that predicted from ΔCCF; and a third “uncertain” group (11 patients, 32.4%) which includes patients with temporal progression or remission disease by imaging diagnosis although their ΔCCF stayed in one direction. These results demonstrated that at least over half of patients can benefit from ctDNA fingerprints test to portend disease progression and support its use as a diagnostic biomarker in combination with imaging-based diagnosis.