Study design and participants
Participants were recruited from Samsung Medical Center and Seoul St Mary’s Hospital (Seoul, Korea) between 2012 and 2023 through prospective and retrospective cohorts. Retrospective cases were selected based on the availability of archived primary tumour and matched normal blood samples, along with sufficient clinical information, and were enroled regardless of disease stage or survival status at accrual. All retrospective samples were obtained shortly after diagnosis, typically at curative-intent surgery. Detailed study design, inclusion and exclusion criteria, and the CONSORT diagram are provided in Supplementary Methods and Supplementary Fig. 1. The study was approved by the Institutional Review Boards of both institutions (Samsung Medical Center: SMC 2022-05-050 and SMC 2013-04-005; Seoul St Mary’s Hospital: KC21TISI0007 and KC22TISI0292) and conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. Written informed consent, including consent to publish de-identified clinical and genomic data, was obtained from all prospective participants. For retrospective cases, the use and publication of de-identified data were approved by the relevant institutional review boards, with a waiver of informed consent where applicable. A subset of participants was enroled as part of substudy components within the clinical trials NCT03131089 and NCT06334471, which informed specific aspects of the study design.
Sample preparation for sequencing
We performed WGS using CancerVision assay as previously reported58. In brief, WGS was performed on tumour samples obtained as part of routine clinical care either via surgery or biopsy and stored as fresh frozen tissue. For biopsy sample cores were retrieved first for routine pathology, followed by at least one additional core for cancer WGS. Biopsy sample cores were retrieved first for routine pathology, followed by at least one additional core for cancer WGS. For the matched normal samples peripheral blood was used. DNA extraction and library preparation was performed at the Inocras in a Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory. We used the Allprep DNA/RNA Mini Kit (Qiagen) for DNA extraction, and TruSeq DNA PCR-Free (Illumina) for library preparation. Sequencing was performed on the Illumina NovaSeq6000 platform (Illumina) with an average depth of coverage of 40x for tumour and 20x for blood. Quality assessment of the WGS data is available in Supplementary Table 10.
For whole-transcriptome sequencing, total RNA was extracted using the AllPrep DNA/RNA Mini Kit (Qiagen) in accordance with the manufacturer’s protocol. Total RNA was quantified using a Qubit Fluorometer (Invitrogen) and the purity and integrity were assessed by the TapeStation RNA ScreenTape (Agilent Technologies). Total RNA-sequencing analysis enabled detection of both coding and noncoding RNA, along with other long intergenic noncoding RNA (lincRNA), small nuclear RNA (snRNA) and small nucleolar RNA (snoRNA). RNA-sequencing libraries were constructed using the KAPA RNA HyperPrep Kit, with RiboErase (Roche Molecular Systems, following the manufacturer’s protocol. We quantified and assessed the libraries using KAPA Library Quantification Kits for Illumina Sequencing platforms according to the qPCR Quantification Protocol Guide (KAPA Biosystems, KK4854) and the TapeStation D1000 ScreenTape (5067–5582, Agilent Technologies) recommendations. The generated libraries were sequenced by using a paired-end 150-bp read protocol on a NovaSeq 6000 platform (Illumina) with the S4 reagent kit (Illumina).
Genomic analysis and interpretation
Comprehensive genomic analysis and interpretation were conducted using the CancerVision platform (Inocras). WGS data were aligned to the GRCh38 human reference genome using bwa-mem (v.0.7.17-r1188)59. Preprocessing included duplicate marking and generation of compressed reference-oriented alignment map (CRAM) files. Somatic SNVs and short indels were called using Mutect2 (GATK v.4.0) and Strelka2 (v.2.9.10)60,61. High-confidence somatic variants were selected based on the following criteria: ≥2 variant reads in tumour, ≤1 variant read in matched normal, mapping quality ≥15, variant allele frequency (VAF) ≤ 5% in tumour, and population allele frequency ≤1% in the panel of normals.
Tumour purity, ploidy, and segmented CNV profiles were estimated with Sequenza (v.3.0.0)62 and somatic structural variations were identified using Delly (v.0.7.6)63. High-quality structural variations were defined as those with ≥2 variant reads in tumour, ≤1 in normal, mapping quality ≥15, and population allele frequency ≤5% in the panel of normals. GISTIC (v.2.0.23) was applied to identify recurrently amplified or deleted genomic regions64. AmpliconArchitect (v.1.2) was used to detect and characterize ecDNA amplicons42. Variant annotation was performed using the Ensembl Variant Effect Predictor (VEP, release 112)65. Variant call format (VCF) files were processed with bcftools (v.1.9). All variants, both germline and somatic, were subjected to rigorous manual review and curation within Inocras’s proprietary genome browser.
Identification of protein-coding driver genes
Protein-coding driver genes were identified using the IntOGen pipeline (v.2023), which integrates seven independent driver gene identification methods: dNdSCV, OncodriveFML, OncodriveCLUSTL, cBaSE, MutPanning, HotMaps3D and smRegions17. These methods collectively assess the selective advantage of somatic mutations in cancer genomes.
... continue reading