k-mer-Based Metagenomics Tools Provide a Fast and Sensitive Approach for the Detection of Viral Contaminants in Biopharmaceutical and Vaccine Manufacturing Applications Using Next-Generation Sequencing

Adventitious agent detection during the production of vaccines and biotechnology-based medicines is of critical importance to ensure the final product is free from any possible viral contamination. Increasing the speed and accuracy of viral detection is beneficial as a means to accelerate development timelines and to ensure patient safety. Here, several rapid viral metagenomics approaches were tested on simulated next-generation sequencing (NGS) data sets and existing data sets from virus spike-in studies done in CHO-K1 and HeLa cell lines. It was observed that these rapid methods had comparable sensitivity to full-read alignment methods used for NGS viral detection for these data sets, but their specificity could be improved. A method that first filters host reads using KrakenUniq and then selects the virus classification tool based on the number of remaining reads is suggested as the preferred approach among those tested to detect nonlatent and nonendogenous viruses. Such an approach shows reasonable sensitivity and specificity for the data sets examined and requires less time and memory as full-read alignment methods. IMPORTANCE Next-generation sequencing (NGS) has been proposed as a complementary method to detect adventitious viruses in the production of biotherapeutics and vaccines to current in vivo and in vitro methods. Before NGS can be established in industry as a main viral detection technology, further investigation into the various aspects of bioinformatics analyses required to identify and classify viral NGS reads is needed. In this study, the ability of rapid metagenomics tools to detect viruses in biopharmaceutical relevant samples is tested and compared to recommend an efficient approach. The results showed that KrakenUniq can quickly and accurately filter host sequences and classify viral reads and had comparable sensitivity and specificity to slower full read alignment approaches, such as BLASTn, for the data sets examined.