Early change in circulating tumor DNA as a potential predictor of response to chemotherapy in patients with metastatic colorectal cancer

The impact of ctDNA changes after chemotherapy on the clinical outcomes of patients with metastatic colorectal cancer (mCRC) remains unclear. The present study evaluated the clinical implications of the early change in ctDNA levels as a predictor of objective response and clinical outcome in mCRC patients who received chemotherapy. We investigated the effects of after/before ratio of ctDNA levels 2 and 8 weeks after initiation of second-line chemotherapy, on objective response rate (ORR), progression-free survival (PFS), and overall survival (OS). ctDNA was detected using amplicon-based deep sequencing with a molecular barcode encompassing >240 hotspot mutations in 14 colon cancer-related genes. In multivariate analysis, as compared to baseline, patients with lower ctDNA level (≤50%) 8 weeks after initiation of chemotherapy showed significantly longer PFS and OS than the patients with higher (>50%) ctDNA level. In patients achieving a partial response or stable disease, the after/before ratio of ctDNA level 8 weeks after initiation of chemotherapy was significantly lower than those in patients with progressive disease. The present study suggests that an early change in the ctDNA level might serve as a biomarker to predict the chemotherapeutic efficacy and clinical outcomes in patients with mCRC.

Moreover, because the response rate of second-line chemotherapy is likely to be lower than those of first-line chemotherapy, the development of a new surrogate marker for clinical response (survival) after second-line chemotherapy other than tumor shrinkage is important to provide mCRC patients with effective second-line chemotherapy. In the present study, we aimed to investigate the correlation between early response of ctDNA and clinical response after chemotherapy in mCRC patients using a deep-sequencing system with NGS and evaluated the ctDNA response which might ease the clinical decision-making process.

Detection of somatic mutations in plasma.
Association between early change in ctDnA levels and clinical outcomes after second-line chemotherapy. To assess the clinical significance of the early change in ctDNA levels in the patients with mCRC after chemotherapy, we investigated the association of ctDNA levels at 2 and 8 weeks after initiation of the second-line chemotherapy with progression-free survival (PFS) and overall survival (OS). ctDNA levels at 2 weeks (median ctDNA level, 6.8%; range, 0% to 65.1%) and 8 weeks (median ctDNA level, 3.8%; range, 0% to 72.7%) after initiation of the chemotherapy were likely to be lower than in the baseline (median ctDNA level, 17.8%; range, 0.17% to 78.1%) as shown in Fig. 2A (2 weeks vs baseline; P = 0.09, 8 weeks vs baseline; P = 0.20).
The changes in ctDNA levels in patients with progressive disease (PD) and partial response (PR) or stable disease (SD) before and after the chemotherapy are shown in Fig. 2B. Kaplan-Meier estimates indicated that the patients who showed ≤50% after/before ratio in their ctDNA levels 2 weeks after initiation of the chemotherapy had significantly longer PFS than those with >50% (median PFS: 5.8 vs 3.3 months; HR, 0.33; 95% CI, 0.10-1.04; P = 0.04, and median OS: NA vs 7.8 months; HR, 0.30; 95% CI, 0.07-1.29; P = 0.08, Fig. 3A,B). Similarly, patients who showed ≤50% after/before ratio in their ctDNA levels 8 weeks after initiation of the chemotherapy had significantly longer PFS and OS than those with >50% (median PFS: 5.6 vs 2.1 months; HR, 0.16; 95% CI, 0.06-0.46; P = 0.0001, Fig. 3C, and median OS: 14.1 vs 8.1 months; HR, 0.10; 95% CI, 0.02-0.52; P = 0.001; Fig. 3D). To evaluate the clinical validity of changes in absolute ctDNA counts during the second-line chemotherapy for mCRC, we further estimated Kaplan-Meier curves of PFS and OS using absolute ctDNA counts. Kaplan-Meier estimates indicated that there were no significant differences of PFS and OS between the patients who showed ≤50% after/before ratio of absolute ctDNA counts 2 weeks after initiation of the chemotherapy and those with >50% (Supplemental Fig. 1A,B). On the other hand, patients who showed ≤50% after/before ratio of absolute ctDNA counts 8 weeks after initiation of the chemotherapy had significantly longer PFS and OS than those with >50% (median PFS: 5.8 vs 2.1 months; HR, 0.18; 95% CI, 0.07-0.49; P = 0.0002, Supplemental Fig. 1C, and median OS: NA vs 9.6 months; HR, 0.17; 95% CI, 0.03-0.81; P = 0.01; Supplemental Fig. 1D). Moreover, we evaluated Kaplan-Meier curves of PFS and OS using optimal cut off value of after/before ratio of ctDNA levels calculated by the receiver operating characteristic curves. A cutoff value of 84.1% for after/before ratio in their ctDNA levels allowed the best stratification at 2 weeks, with a sensitivity of 42.9% and a specificity of 91.7% (AUC, 0.67; 95%CI, 0.41-0.93) and 39.4% at 8 weeks with a sensitivity of 90.0% and a specificity of 75.0% (AUC, 0.83; 95%CI, 0.67-0.99). Kaplan-Meier estimates indicated that the patients who showed ≤84.1% after/before ratio in their ctDNA levels 2 weeks after initiation of the chemotherapy had significantly longer PFS than those with >84.1% (median PFS: 5.6 vs 2.1 months; HR, 0.28; 95% CI, 0.09-0.95;
In the univariate Cox proportional hazard analysis, primary tumor location, lung metastasis, changes in ctDNA level (after/before ratio of ctDNA level), CEA and CA19-9 levels (after/before ratio of CEA and CA19-9) 8 weeks after initiation of the chemotherapy were predictive factors for PFS (Table 2). Similarly, change in ctDNA levels and CA19-9 levels 8 weeks after initiation of the chemotherapy were predictive factors for OS (Table 2). In multivariate analysis, change in the ctDNA level 8 weeks after initiation of the chemotherapy was a predictive factor for both PFS (HR, 0.17; 95% CI, 0.06-0.47; P = 0.0006; Table 2) and OS (HR, 0.10; 95% CI, 0.02-0.52; P = 0.006; Table 2).
Association between early change in ctDnA after chemotherapy and therapy response. We next analyzed the association between objective response and change in ctDNA levels 2 weeks and 8 weeks after initiation of second-line chemotherapy in mCRC patients. Objective response rate (ORR) and disease control rate (DCR) were 6.9% and 55.2% (2 PR and 14 SD), respectively. Although there was no significant difference of change in the ctDNA levels 2 weeks after initiation of the chemotherapy between PD and SD or PR groups, patients achieving SD or PR showed significantly lower ctDNA level 8 weeks after initiation of chemotherapy compared to those with PD (2 weeks: P = 0.25, 8 weeks: P = 0.006, Fig. 4A). Furthermore, patients with ≤50% after/before ratio of ctDNA levels 8 weeks after initiation of the chemotherapy showed better objective response compared to those with >50% (P = 0.003, Table 3). Changes in ctDNA levels significantly correlated with tumor shrinkage 8 weeks after initiation of second-line chemotherapy (r = 0.52, P = 0.006, Fig. 4B). These results suggest that ctDNA could act as a prognostic biomarker to predict the outcomes of second-line chemotherapy in mCRC patients.

Discussion
The present study demonstrates a significant association between early changes in ctDNA levels and the ORR, PFS, and OS in mCRC patients treated with second-line chemotherapy. The mCRC patients whose ctDNA levels decreased up to ≤50% 2 weeks and 8 weeks after initiation of chemotherapy showed a better response to the chemotherapy. Previous reports showed that early tumor shrinkage (ETS) is pivotal early predictors of treatment efficacy in mCRC patients, especially in those treated with molecular-targeted drugs [22][23][24] . Here, the change in ctDNA levels at 2 weeks after initiation of chemotherapy is shown as a possible predictor of PFS (HR, 0.33; 95%CI, 0.10-1.04; P = 0.059, in univariate analysis; Table 2). Furthermore, the change in ctDNA level at 8 weeks after initiation of chemotherapy was an independent predictor of PFS (HR, 0.17; 95%CI, 0.06-0.47; P = 0.0006, in multivariate analysis; Table 2) and OS (HR, 0.10, 95%CI, 0.02-0.52; P = 0.006, in multivariate analysis; Table 2). On the other hand, ETS, which is defined as the relative change in the sum of the longest diameter of the tumor at 8 weeks compared to the baseline (≥20% tumor shrinkage) in this study, was not a significant predictor of PFS (HR, 0.17; 95%CI, 0.02-1.33; P = 0.09, in univariate analysis; Table 2) or OS (HR, 0.48; 95%CI, 0.15-1.54; P = 0.22, in univariate analysis; Table 2) in our study. Our data suggest that after/before ratio of the ctDNA level after second-line chemotherapy could be a better predictor of chemotherapeutic efficacy than ETS in mCRC patients 20,25 .
Tumor markers such as CEA and CA19-9 are widely used to monitor the tumor burden and progression of mCRC during chemotherapy 26,27 . Previous reports indicated that CEA levels and survival are inversely correlated in patients receiving a combination of chemotherapy and anti-EGFR inhibitor 28,29 . In our study, univariate analysis and log-rank test showed that the mCRC patients showing reduced after/before ratio in CEA and CA19-9 levels 8 weeks after initiation of the chemotherapy was inversely correlated with PFS (CEA: HR, 0.37; 95%CI, 0.14-0.96; P univariate = 0.036, P log-rank = 0.03, CA19-9: HR, 0.31; 95%CI, 0.10-0.99; P univariate = 0.049, P log-rank = 0.04; Table 2 and Supplementary Fig. 3A,C). In addition, the mCRC patients exhibiting lower after/before ratio of CA19-9 levels 8 weeks after initiation of the chemotherapy was inversely correlated with OS (HR, 0.07; 95%CI, 0.01-0.49; P univariate = 0.006, P log-rank = 0.0005; Table 2 and Supplementary Fig. 3D). However, the change (after/ before ratio) in ctDNA level at 8 weeks remained as an independent indicator of both PFS (P = 0.0006; Table 2) and OS (P = 0.006; Table 2) in multivariate analysis. Reportedly, the change in ctDNA after completion of cycle 1 of chemotherapy could be a successful predictor for response to chemotherapy while CEA failed as a predictive marker 20 . Collectively, these findings suggest that ctDNA might serve as a reliable predictive biomarker for early therapeutic response. In-depth research is warranted to further establish the critical role of ctDNA as a predictive marker for the response to chemotherapy in patients with mCRC.
The sample size in the current investigation was small. Moreover, the frequencies of mutated genes in ctDNA of patients with CRC in our study were inconsistent with those in the tissue DNA reported in the mutation database including The Cancer Genome Atlas (TCGA) 30,31 . The mutation frequency of the APC gene in CRC tissue has been reported to be ~80% 30,32 . However, we observed only 20.7% in plasma, which was a significantly lower frequency than those reported in tissues previously 33 . This inconsistency might be partially due to the insufficient coverage of mutations in APC gene that the gene panel used in this study could guarantee. Further technical improvement in the mutation detection system to detect additional mutations and gene rearrangements could increase the sensitivity for mutation detection in the plasma of mCRC patients.
In conclusion, we unraveled that in mCRC patients receiving chemotherapy, early changes in ctDNA levels represent highly sensitive early predictor of treatment response 20 . Future prospective clinical trials with large sample size should be conducted to validate the clinical impact of change in ctDNA and support its application as a novel early predictive biomarker for the response to second-line chemotherapy. (2019) 9:17358 | https://doi.org/10.1038/s41598-019-53711-3 www.nature.com/scientificreports www.nature.com/scientificreports/

Materials and Methods
patients. This study aimed to study was to study the correlation between early ctDNA response and PFS, OS, ORR in mCRC patients treated with second-line chemotherapy (Table 1). Twenty-nine mCRC patients, who were treated with second-line chemotherapies at Cancer Institute Hospital, Japanese Foundation for Cancer Research, were prospectively enrolled in this study from February 2017 to March 2018. TNM Classification of Malignant Tumors (7th edition) was used to determine the tumor and nodal status. This present study was approved by the Institutional Review Boards of the Japanese Foundation for Cancer Research (Tokyo, Japan, registry number 2017-1009). Written informed consent was obtained from all the patients for the use of their plasma and tissue samples. All methods were performed in accordance with the Declaration of Helsinki.
Blood samples, ctDnA isolation, and sequencing. Blood samples were collected in EDTA tubes as per the manufacturer's instructions. Time points for collecting the blood samples were just before the initiation of second-line chemotherapy and 2 weeks and 8 weeks after initiation of the second-line chemotherapy. Plasma from the blood was obtained by centrifugation at 1600 g for 10 min at 4 °C, followed by another spin at 16 www.nature.com/scientificreports www.nature.com/scientificreports/ for 10 min at 4 °C to remove the cell debris. cfDNA was extracted from 2 mL plasma using a MagMAX cfDNA Isolation Kit (Thermo Fisher Scientific, USA) following the manufacturer's instructions. Preparation and quality control of the libraries, template preparation, and sequencing were performed as previously described 34 . Fourteen genes with >240 hotspots (Single nucleotide variants and short indels), including AKT1, BRAF, CTNNB1, EGFR, ERBB2, FBXW7, GNAS, KRAS, MAP2K1, NRAS, PIK3CA, SMAD4, TP53, and APC were covered in this assay 34,35 . Clean reads were mapped to the human reference genome (hg19) sequence. The Torrent Variant Caller was used to filter and call the mutations in targeted regions of each gene 34,35 . The limit of detection for each variant was 0.15% in this study. ctDNA level in plasma was defined as the highest allele frequency of the detected mutant alleles at each time point in each patient when two or more mutations were detected. tumor tissue DnA sequencing. Genomic DNA was extracted from fixed paraffin-embedded tissues obtained from biopsies or surgical resections as previously described 36,37 . For tissue KRAS and NRAS test, RASKET KIT (MBL, Japan), which applies the polymerase chain reaction-reverse sequence-specific oligonucleotide method (PCR-rSSO), was used following the manufacturer's protocol. We examined twelve mutations in RAS exon 2, eight in RAS exon 3, and four in RAS exon 4 using Luminex 100/200 (Luminex, Japan) and UniMAG (MBL, Japan) system as described previously 38,39 . Statistical analyses and tumor assessment. Tumor response was assessed by CT imaging using RECIST guideline, version 1.1. PFS was defined as the time from the 1st day of the second line treatment to either the first objective evidence of disease progression or death from any cause. OS was defined as the time from the 1st day of the second line treatment until the time of death. PFS and OS were estimated using the Kaplan-Meier method and the statistical significance of the correlation between the clinical outcome and clinical parameters (ctDNA, CEA, and CA19-9 levels) was assessed using the log-rank test. ORR denotes the proportion of patients who have a complete response or PR to the second-line chemotherapy, and DCR indicates the proportion of patients who have a complete response or PR or SD to the therapy 40 . ETS is the relative change in the sum of the longest diameter of the tumor at week 8 compared to the baseline (≥20% tumor shrinkage) 24   www.nature.com/scientificreports www.nature.com/scientificreports/ provided two-sided P values, and P < 0.05 was considered significant. In Cox proportional hazard analysis, factors with P < 0.05 in the univariate analysis were included in the multivariate analysis (Backward stepwise methods). Statistical analyses were carried out using the statistical software, "EZR" (Easy R), which is based on R and R commander 42 .

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.