STAR★Methods

Key resources table

REAGENT or RESOURCE | SOURCE | IDENTIFIER
Critical commercial assays

Guardant360® (74-gene panel assay) | Guardant Health, Inc. | https://www.guardanthealth.com/; RRID:SCR_023455
RASKET-B KIT | Medical & Biological Laboratories (MBL) Co., Ltd. | Cat# 4505; https://www.mblbio.com/

Deposited data

De-identified clinical and ctDNA alteration profiles | This study | Available from the lead contact upon reasonable request

Software and algorithms

R version 4.2.1 | R Project for Statistical Computing | https://www.r-project.org/; RRID:SCR_001905
GraphPad Prism 7 | GraphPad Software | https://www.graphpad.com/; RRID:SCR_002798
EZR (Easy R) version 1.61 | Saitama Medical Center, Jichi Medical University | https://www.jichi.ac.jp/saitama-sct/SaitamaHP.files/statmedEN.html

Other

UMIN Clinical Trials Registry (UMIN-CTR) | UMIN-CTR | Trial registration number: UMIN000029315; https://center6.umin.ac.jp/cgi-open-bin/ctr_e/ctr_view.cgi?recptno=R000033509

Experimental model and study participant details

Patients

This retrospective, multi-institutional study evaluated a cohort of patients with metastatic colorectal cancer (mCRC) enrolled in the nationwide plasma genomic profiling platform, SCRUM-Japan GOZILA, across 31 core cancer institutions in Japan. All data collection and clinical protocols were approved by the Institutional Review Board of the Cancer Institute Hospital of the Japanese Foundation for Cancer Research (approval number: 2021-GB-009). The study was conducted in strict compliance with the ethical principles mandated by the Declaration of Helsinki. The protocol was described on the hospital’s website, and participants were given the opportunity to opt out of the study. Data from next-generation sequencing (NGS; Guardant360®, Guardant Health, Inc., Redwood City, CA, USA) were used to confirm the subsequent ctDNA genotypes (RAS, BRAF, and others) following chemotherapy. Patients with mCRC enrolled in the GOZILA study between March 2018 and February 2022, with confirmed results from tissue RAS and BRAF testing before chemotherapy initiation, were eligible, regardless of the treatment line, as shown in Figure S1. The GOZILA study is a nationwide plasma genomic profiling study involving 31 core cancer institutions in Japan. Patients with metastatic gastrointestinal cancer were eligible for enrollment.

A total of 1,391 patients who underwent baseline tissue sequencing and subsequent post-treatment ctDNA next-generation sequencing were eligible for final survival analysis. All participating subjects in this cohort were Japanese ethnicity. The median age of the entire cohort at the time of plasma sampling was 61.0 years (range, 25.0–85.0 years). In terms of biological sex distribution, the cohort comprised 800 male patients (57.5%) and 591 female patients (42.5%), with sex data missing or unknown for 2 patients (0.1%).

To rigorously assess the potential confounding influence of baseline demographic characteristics on clinical outcomes, both biological sex and age were integrated into the univariable Cox proportional hazard models for OS. Univariable analysis demonstrated that neither biological sex (Male vs. Female: HR, 0.90; 95%CI, 0.78–1.05; p = 0.21) nor age group (<65 vs. ≥65 years: HR, 0.95; 95% CI, 0.61–1.50; p = 0.84) exerted a statistically significant impact or interaction on survival outcomes. Consequently, these demographic variables were not identified as independent prognostic indicators and were excluded from the final multivariable Cox regression models.

Method details

Blood samples, ctDNA isolation and sequencing

NGS of ctDNA was performed using the Guardant360® system (Guardant Health, Inc., Redwood City, CA, USA), a liquid biopsy platform leveraging a 74-gene panel. This assay was configured to detect single-nucleotide variants, insertions/deletions, fusions, and copy number alterations across the complete coding regions of target oncogenes, including KRAS, NRAS, and BRAF, alongside the analytical determination of microsatellite instability status. For sample acquisition in the SCRUM-Japan GOZILA study, 2 × 10 mL of whole blood was collected from each participant into cell-free DNA BCT® tubes (Streck, Inc., La Vista, NE, USA) containing proprietary cellular stabilization reagents to prevent the lysis of nucleated cells and subsequent genomic DNA contamination.23,24 The average sequencing depth was 15,000×, and the ctDNA fraction was determined from the maximum variant allelic fraction. Blood specimens were maintained at ambient temperature and shipped to a centralized laboratory at Guardant Health for processing within standard pre-analytical stability windows. Cell-free DNA (cfDNA) was isolated from plasma via automated magnetic bead-based extraction protocols. Following extraction, cfDNA underwent oligonucleotide hybridization capture and targeted library preparation utilizing unique molecular identifiers (UMIs) to convert individual double-stranded DNA fragments into digital sequences, thereby suppressing polymerase chain reaction (PCR) errors and sequencing artifacts. Sequencing was executed on high-throughput NGS platforms (Illumina, Inc., San Diego, CA, USA) with a target average sequencing depth of coverage per base. The definitive ctDNA fraction for each clinical specimen was quantitatively determined from the maximum variant allele frequency (VAF) observed among the detected somatic alterations.

Tumor tissue DNA sequencing

All enrolled patients were histopathologically diagnosed with metastatic colorectal cancer (mCRC) through microscopic evaluation of tissue specimens obtained via diagnostic biopsy or surgical resection. Prior to the initiation of first-line treatment for advanced disease, routine clinical RAS (KRAS and NRAS) and BRAF mutational profiling of the primary or metastatic tumor tissue was mandated using validated standard-of-care assays. As the primary representative measurement method utilized across cohorts, the RASKET-B KIT (Medical & Biological Laboratories [MBL] Co., Ltd., Nagoya, Japan) was applied using genomic DNA extracted from formalin-fixed paraffin-embedded (FFPE) tumor tissue sections in absolute accordance with the manufacturer's operational protocol. The assay applies a multiplex PCR-based reverse sequence-specific oligonucleotide (PCR-rSSO) methodology combined with xMAP® technology on the Luminex platform, allowing for high-throughput, multiplexed fluorometric detection of target alleles within a single reaction well. Through specific biotinylated primer amplification and hybridization to sequence-specific oligonucleotide probes conjugated to distinct fluorescent microbeads, we rigorously evaluated 12 discrete types of RAS exon 2 mutations (G12S, G12C, G12R, G12D, G12V, G12A, G13S, G13C, G13R, G13D, G13V, and G13A), 8 types of RAS exon 3 mutations (A59T, A59G, Q61K, Q61E, Q61L, Q61P, Q61R, and Q61H), 4 types of RAS exon 4 mutations (K117N, A146T, A146P, and A146V), and the BRAF exon 15 V600E hotspot mutation. Fluorescent signals were swept and analyzed on a Luminex analyzer to determine positive mutational status based on predefined institutional cut-off bead intensities.

Endpoints and data collection

The primary endpoint was to compare OS among cohorts of patients with mCRC according to changes in their tumor genomic profiles from baseline tissue to subsequent ctDNA testing. Patients were divided into six groups: persistent RAS WT, persistent RAS MT, change from RAS MT to RAS WT (Neo_RAS_ WT), change from RAS WT to RAS MT (acquired RAS MT), persistent BRAF V600E (BRAF MT), and change from RAS MT to RAS WT (Neo_BRAF_ WT). BRAF MT other than V600E was not considered. Furthermore, patients with Neo_RAS_ WT were categorized into two groups based on genetic alterations: Group A, tissue-confirmed pretreatment RAS MT and no post-treatment RAS MT detected by ctDNA analysis; however, Group A included both patients with true biological clearance of RAS mutations and undetectable ctDNA levels as patients with insufficient tumor-derived DNA shedding would fulfill the Group A criteria irrespective of their underlying RAS mutational status. To address this limitation, Group B was defined more stringently to include patients with tissue-confirmed pretreatment RAS MT and undetectable post-treatment RAS MT, but with other post-treatment ctDNA alterations remaining after excluding those potentially associated with clonal hematopoiesis (CH), thereby confirming the presence of ctDNA and providing greater confidence that RAS mutation absence reflects genuine clonal elimination rather than assay insensitivity. Mutations in TP53, HRAS, PDGFRA, and GNAS genes were considered potential CH alterations.42,43 Secondary endpoints included the differences in clinicopathological factors among the five groups. Data on the following features were collected: patient age and sex, primary tumor location, metastatic site, treatment lines at the time of sampling, and history of anti-EGFR or anti-VEGF antibodies. OS was measured from the time of first-line treatment initiation to the date of death.

Quantification and statistical analysis

Clinicopathological factors and distribution frequencies were compared between the Persisitent_RAS_ WT cohort (reference group) and other molecular subgroups. Continuous variables were evaluated using the Mann–Whitney U test, and categorical variables were analyzed using the Chi-square test or Fisher's exact test as appropriate. Overall survival (OS) was calculated from the date of first-line therapy initiation to the date of death. Survival curves were estimated using the Kaplan–Meier method, and statistical differences between the genomic dynamics cohorts were evaluated using the two-sided log-rank test. The exact number of individual patients for each cohort is detailed in Table 1 and the respective figure legends. To identify independent prognostic factors, variables achieving a p-value < 0.05 in the univariable analysis were subsequently entered into the multivariable Cox proportional hazard regression models. Key statistical values, including hazard ratios (HRs), 95% confidence intervals (CIs), and exact values, are explicitly reported in the text, figures, and tables. A two-sided value p < 0.05 was considered statistically significant for all analyses. All statistical calculations were performed using the “EZR” software (Saitama Medical Center, Jichi Medical University, Saitama, Japan), which is a graphical user interface for R (The R Foundation for Statistical Computing, Vienna, Austria, version 4.2.1) and R Commander.44