Background: Molecular analysis of solid tumor has traditionally relied on tissue biopsies;however, these samples can present several issues, particularly when the analysis is performed by Next Generation Sequencing (NGS). Some tumors may be difficult to access, small, diffuse or even difficult to locate. In addition, tumor heterogeneity may be an issue while inconvenience and cost of sampling, patient discomfort and even potential harm to the patient should be considered. Liquid biopsy is an alternative approach that overcomes many of these limitations while adding potential benefits of providing a sample that may be a more up-to-date, dynamic and representative reflection of tumour activity or status.

Here, we present the results of a multi-centre validation of a cost-effective and efficient capture-based ctDNA assay, together with integrated variant analysis and reporting software. The assay targets single nucleotide variants (SNVs), indels and copy number variants (CNVs) from 74 cancer-associated genes. Testing was performed at four independent laboratories using 54 reference and clinical samples (including multiple replicates and repeated reference samples per run across all sites). Clinical ctDNA samples were obtained from patients with lung cancer (23), breast cancer (11), colorectal cancer (8), cholangiocarcinoma (4), melanoma (2), pancreatic cancer (2), bone cancer (1) and brain cancer (1). Sequencing was performed on Illumina sequencing platforms. The limit of blank was established at 0.11% for SNV variant fraction and 0.154% for Indel variant fraction, while limit of detection was set at 0.25%. Sensitivity at the limit of detection and above for reference samples from the site with the most experience of the protocol was determined to be 99.0% for SNVs and 95.6% for indels with a specificity of >99%, while overall sensitivity for reference samples from all sites was found to be 97.1% for SNVs and 91.6% for indels. The difference was likely due to unfamiliarity with the protocol and sequencing performance of different flowcells. Sensitivity in clinical samples was 97.1% (with two false negative variants also being detected above the limit of blank but below the limit of detection). For CNVs, a reference sample containing 3 known CNVs and 4 previously characterised clinical samples with CNVs (repeated across 15 independent tests) were analysed with all CNVs being correctly detected (above our set threshold of 6 copies), giving a clinical sensitivity of 100%. CNV clinical specificity (at this threshold) was also 100%. The OncoSELECT panel