ABSTRACT
Background. The quantitative level of pathogens present in a host is a major driver of infectious disease (ID) state and outcome. However, the majority of ID diagnostics are qualitative. Next-generation sequencing (NGS) is an emerging ID diagnostics and research tool to provide insights, including tracking transmission, evolution, and identifying novel strains. Methods. We built a novel likelihood-based computational method to leverage pathogen-specific genome-wide NGS data to detect SARS-CoV-2, profile genetic variants, and furthermore quantify levels of these pathogens. We used de-identified clinical specimens tested for SARS-CoV-2 using RT-PCR, SARS-CoV-2 NGS Assay (hybrid capture, Twist Bioscience), or ARTIC (amplicon-based) platform, and COVID-DX software. A training (n=87) and validation (n=22) set was selected to establish the strength of our quantification model. We fit non-uniform probabilistic error profiles to a deterministic sigmoidal equation that more realistically represents observed data and used likelihood maximized over several different read depths to improve accuracy over a wide range of values of viral load. Given the proportion of the genome covered at varying depths for a single sample as input data, our model estimated the Ct of that sample as the value that produces the maximum likelihood of generating the observed genome coverage data. Results. The model fit on 87 SARS-CoV-2 NGS Assay training samples produced a good fit to the 22 validation samples, with a coefficient of correlation (r2) of ~0.8. The accuracy of the model was high (mean absolute % error of ~10%, meaning our model is able to predict the Ct value of each sample within a margin of ±10% on average). Because of the nature of the commonly used ARTIC protocol, we found that all quantitative signals in this data were lost during PCR amplification and the model is not applicable for quantification of samples captured this way. The ability to model quantification is a major advantage of the SARS-CoV-2 NGS assay protocol. Left. Observed genome coverage (y-axis) plotted against Ct value (x-axis). The best-fitting logistic curve is demonstrated with a red line with shaded areas above and below representing the fitted error profile. RIGHT: Model-estimated Ct values (y-axis) compared to laboratory Ct values (x-axis) with grey bars representing estimated confidence intervals. The 1:1 diagonal is shown as a dotted line. Conclusion. To our knowledge, this is the first model to incorporate sequence data mapped across the genome of a pathogen to quantify the level of that pathogen in a clinical specimen. This has implications in ID diagnostics, research, and metagenomics.
ABSTRACT
Background. As the SARS-CoV-2 (SCV-2) virus evolves, diagnostics and vaccines against novel strains rely on viral genome sequencing. Researchers have gravitated towards the cost-effective and highly sensitive amplicon-based (e.g. ARTIC) and hybrid capture sequencing (e.g. SARS-CoV-2 NGS Assay) to selectively target the SCV-2 genome. We provide an in silico model to compare these 2 technologies and present data on the high scalability of the Research Use Only (RUO) workflow of the SARS-CoV-2 NGS Assay. Methods. In silico work included alignments of 383,656 high-quality genome sequences belonging to variant of concern (VOC) or variant of interest (VOI) isolates (GISAID). We profiled mismatches and sequencing dropouts using the ARTIC V3 primers, SARS-CoV-2 NGS Assay probes (Twist Bioscience) and 11 synthesized viral sequences containing mutations and compared the performance of these assays using clinical samples. Further, the miniaturized hybrid capture workflow was optimized and evaluated to support high-throughput (384-plex). The sequencing data was processed by COVID-DX software. Results. We detected 101,432 viruses (27%) with > = 1 mismatch in the last 6 base pairs of the 3' end of ARTIC primers;of these, 413 had > = 2 mismatches in one primer. In contrast, only 38 viruses (0.01%) had enough mutations ( > = 10) in a hybrid capture probe to have a similar effect on coverage. We observed that mutations in ARTIC primers led to complete dropout of the amplicon for 4/11 isolates and diminished coverage in additional 4. Twist probes showed uniform coverage throughout with little to no dropouts. Both assays detected a wide range of variants (~99.9% coverage at 5X depth) in clinical samples (CT value < 30) collected in NY (Spring 2020-Spring 2021). The distribution of the number of reads and on target rates were more uniform among specimens within amplicon-based sequencing. However, uneven genome coverage and primer dropouts, some in the spike protein, were observed on VOC/VOI and other isolates highlighting limitations of an amplicon-based approach. Conclusion. The RUO workflow of the SARS-CoV-2 NGS Assay is a comprehensive and scalable sequencing tool for variant profiling, yields more consistent coverage and smaller dropout rate compared to ARTIC (0.05% vs. 7.7%).