SUMMARYBackgroundCurrent diagnostic techniques are inadequate for rapid microbial diagnosis and optimal management of patients with suspected sepsis. We assessed the impact of metagenomic sequencing and host response profiling individually and in combination on microbiological diagnosis in these patients.MethodsIn this cohort study of 200 consecutive patients with suspected sepsis we evaluated three molecular diagnostic methods with blood specimens: 1) direct bacterial DNA detection and

more »

... rization with metagenomic shotgun next generation sequencing and contaminant sequence removal using Bayesian inference; 2) direct viral DNA and RNA enrichment and detection with viral capture sequencing; and 3) transcript-based host response profiling with a previously-defined 18-gene qRT-PCR assay. We then evaluated changes in diagnostic decision-making among three expert physicians in a chart review by unblinding our three molecular test results in a staged fashion.FindingsMetagenomic shotgun sequencing confirmed positive blood culture results in 14 of 26 patients. In 17 of 200 patients, metagenomic sequencing and viral capture sequencing revealed organisms that were 1) not detected by conventional hospital tests within 5 days after presentation, and 2) classified as of probable clinical relevance by physician consensus. Host response profiling led at least two of three physicians to change their diagnostic decisions in 46 of 100 patients. Finally, we report on potential bacterial DNA translocation in 8 patients who were originally classified by physicians as noninfected and show how host response profiling can guide interpretation of metagenomic shotgun sequencing results.InterpretationThe integration of host response profiling, metagenomic shotgun sequencing, and viral capture sequencing synergistically enhances the utility of each of these approaches, and may improve the diagnosis of infections in patients with suspected sepsis.FundingNational Institutes of Health, Chan-Zuckerberg Biohub Microbiome Initiative, and the Thomas C. and Joan M. Merigan Endowment at Stanford University.RESEARCH IN CONTEXTEvidence before this studyOur PubMed search for articles matching the terms ("metagenomic" OR "cell-free DNA") and "infect*" in the title/abstract and using the "Human" species filter from inception to September 30, 2019 yielded 463 articles. Many proof-of-concept and validation studies illustrating how metagenomic sequencing can diagnose infections have been previously reviewed. Our search identified only nine studies which applied metagenomic shotgun sequencing to blood specimens, likely because there is a relatively low signal-to-noise ratio with this specimen type in this setting. In a study of 358 febrile sepsis patients, plasma cell-free DNA sequencing detected causative agents missed by standard-of-care testing in 15% of patients, but also detected bacterial organisms adjudicated as commensals in 10% of patients. Recently, a proof-of-concept study used machine learning to integrate metagenomic sequencing and transcriptional host response profiling to differentiate pathogens from commensal organisms in respiratory specimens, albeit with only a small derivation cohort to train host response signatures.Added value of this studyOur 200-patient study assessed the clinical utility of combining both metagenomic sequencing and a previously-defined host response assay to diagnose sepsis. We developed a rigorous chart review approach to measure whether our assays' results could change a physician's diagnostic decision-making, without having to commit the assays into patient care. Metagenomic sequencing revealed previously-undetected and clinically relevant organisms in 17 of 200 patients, and host response profiling led at least two of three physician chart reviewers to change their diagnostic classifications in 46 of 100 patients. We also report on potential bacterial DNA bloodstream translocation in 8 of 40 patients who were originally classified by physicians as noninfected and show how host response profiling can guide interpretation of metagenomic shotgun sequencing results. Finally, we present a statistical algorithm for contaminant removal from metagenomic sequencing data using Bayesian inference.Implications of all the available evidenceCurrent diagnostic techniques are inadequate for rapid microbial diagnosis and optimal management of patients with suspected sepsis. Metagenomic sequencing, which offers the promise of hypothesis-free testing to discover new organisms that would have otherwise been missed, is already being introduced into clinical practice. However, interpretation of results from this powerful approach can be difficult, given that a large fraction of positive results represents reactivated viruses, chronic infections, commensal organisms, and contamination. Host response profiling can serve as an objective adjunct in interpreting ambiguous metagenomic sequencing results. As host response assays are introduced into clinical practice, we suggest that all patients undergoing metagenomic sequencing be simultaneously tested with one of these assays. For now, we urge clinicians to carefully interpret metagenomic sequencing results with the utmost regard for patient safety and antimicrobial stewardship.