Lower respiratory tract infections (LRTIs) have high morbidity and mortality rates. However, traditional etiological detection methods have not been able to meet the needs for the clinical diagnosis and prognosis of LRTIs. The rapid development of metagenomic next-generation sequencing (mNGS) provides new insights for the diagnosis and treatment of LRTIs; however, little is known about how to interpret the application of mNGS results in LRTIs. In this study, lower respiratory tract specimens from 46 patients with suspected LRTIs were tested simultaneously using conventional microbiological detection methods and mNGS. Receiver operating characteristic (ROC) curves were used to evaluate the performance of the logarithm of reads per kilobase per million mapped reads [lg(RPKM)], genomic coverage, and relative abundance of the organism in predicting the true-positive pathogenic bacteria. True-positive viruses were identified according to the lg(RPKM) threshold of bacteria. We also evaluated the ability to predict drug resistance genes using mNGS. Compared to that using conventional detection methods, the false-positive detection rate of pathogenic bacteria was significantly higher using mNGS. It was concluded from the ROC curves that the lg(RPKM) and genomic coverage contributed to the identification of pathogenic bacteria, with the performance of lg(RPKM) being the best (area under the curve [AUC] = 0.99). The corresponding lg(RPKM) threshold for identifying the pathogenic bacteria was -1.35. Thirty-five strains of true-positive virus were identified based on the lg(RPKM) threshold of bacteria, with the detection of human gammaherpesvirus 4 being the highest and prone to coinfection with Pseudomonas aeruginosa, Acinetobacter baumannii, and Stenotrophomonas maltophilia. Antimicrobial susceptibility tests (AST) revealed the resistance of bacteria containing drug resistance genes (detected by mNGS). However, the drug resistance genes of some multidrug-resistant bacteria were not detected. As an emerging technology, mNGS has shown many advantages for the unbiased etiological detection and the prediction of antibiotic resistance. However, a correct understanding of mNGS results is a prerequisite for its clinical application, especially for LRTIs. IMPORTANCE LRTIs are caused by hundreds of pathogens, and they have become a great threat to human health due to the limitations of traditional etiological detection methods. As an unbiased approach to detect pathogens, mNGS overcomes such etiological diagnostic challenges. However, there is no unified standard on how to use mNGS indicators (the sequencing reads, genomic coverage, and relative abundance of each organism) to distinguish between pathogens and colonizing microorganisms or contaminant microorganisms. Here, we selected the mNGS indicator with the best identification performance and established a cutoff value for the identification of pathogens in LRTIs using ROC curves. In addition, we also evaluated the accuracy of antibiotic resistance prediction using mNGS.
Keywords: genomic coverage; lg(RPKM); lower respiratory tract infections; metagenomic next-generation sequencing; receiver operating characteristic curve; relative abundance.