Defining the analytical and clinical sensitivity of the ARTIC method for the detection of SARS-CoV-2 (preprint)

ABSTRACT

The SARS-CoV-2 ARTIC amplicon protocol is the most widely used genome sequencing method for SARS-CoV-2, accounting for over 43% of publicly-available genome sequences. The protocol utilises 98 primers to amplify ~400bp fragments of the SARS-CoV-2 genome covering all 30,000 bases. Understanding the analytical performance metrics of this protocol will improve how the data is used and interpreted. Different concentrations of SARS-CoV-2 control material were used to establish the limit of detection (LoD) of the ARTIC protocol. Results demonstrated the LoD was a minimum of 25-50 virus particles per mL. The sensitivity of ARTIC was comparable to the published sensitivities of commercial diagnostics assays and could therefore be used to confirm diagnostic testing results. A set of over 3,600 clinical samples from three UK regions were then evaluated to compare the protocols performance to clinical diagnostic assays (Roche Lightcycler 480 II, AusDiagnostics, Roche Cobas, Hologic Panther, Corman RdRp, Roche Flow, ABI QuantStudio 5, Seegene Nimbus, Qiagen Rotorgene, Abbott M2000, Thermo TaqPath, Xpert). We developed a Python tool, RonaLDO, to perform this validation (available under the GNU GPL3 open-source licence from https//github.com/quadram-institute-bioscience/ronaldo). Positives detected by diagnostic platforms were generally supported by sequencing data; platforms that used RT-qPCR were the best predictors of whether the sample would subsequently sequence successfully. To maximise success of sample sequencing for phylogenetic analysis, samples with Ct <31 should be chosen. For diagnostic tests that do not provide a quantifiable Ct value, adding a quantification step is recommended. The ARTIC SARS-CoV-2 sequencing protocol is highly sensitive, capable of detecting SARS-CoV-2 in samples with Cts in the high 30s. However, to routinely obtain whole genome coverage, samples with Ct <31 are recommended. Comparing different virus detection methods close to their LoD was challenging and significant discordance was observed.