Optimizing Detection of Clinically Relevant Microbes from Whole Blood of Febrile Neutropenia Patients

Abstract

Febrile Neutropenia is a potentially fatal side effect of cancer chemotherapy. Weakened immune systems make FN patients vulnerable to infection by a wide range of opportunistic pathogens. Culture-positive rates in these patients is notoriously low. Treatment consists of broad-spectrum antibiotics that are often ineffective in treating the patient and may be harmful in some cases. The advancement of sequencing technology and the decrease in associated costs suggests an imminent shift from culture to molecular based diagnostics. Described here is a streamlined bioinformatics workflow for analyzing whole genome shotgun reads from blood in 6 Febrile Neutropenia and 5 bone marrow transplant patients. To ensure that taxonomic assignments are accurate, binary mixtures of bacteria (with varying taxonomic distance across a log-scale of abundance) and a mock bacterial community in known staggered abundance were sequenced and subjected to the bioinformatics workflow. Reported are the effects of using various bioinformatics steps, strategies, and parameters to establish standard protocols for the expanded use of NGS for analyzing clinical samples. Quality control using modest parameters resulted in significant reductions in total number of reads ranging from 38% to 57% and showed bias in removing viral reads. Host screening against the human genome removed 42% to 96% of reads from NF samples, including Human Parvovirus B19 an endogenous microbe in the human genome representing a potentially causative organism. Time series analysis of bone marrow transplant samples revealed the dynamic nature of microbial communities and identified growing systemic infection of Escherichia coli. Consistent presence of contaminating organisms including Human Endogenous Virus and Cutibacterium acnes was observed in samples with few total microbial sequences, compared to samples with potentially causative organisms with a higher abundance of microbial reads. Taken together, this work lays the foundation for bioinformatics pipelines that use open-source software, require minimal computational resource, and provide rapid and accurate identification of known microorganisms.