Assessment of methods for serum extracellular vesicle small RNA sequencing to support biomarker development

Extracellular vesicles (EVs) have great potential as a source for clinically relevant biomarkers since they can be readily isolated from biofluids and carry microRNA (miRNA), mRNA, and proteins that can reflect disease status. However, the biological and technical variability of EV content is unknown making comparisons between healthy subjects and patients difficult to interpret. In this study, we sought to establish a laboratory and bioinformatics analysis pipeline to analyse the small RNA content within EVs from patient serum that could serve as biomarkers and to assess the biological and technical variability of EV RNA content in healthy individuals. We sequenced EV small RNA from multiple individuals (biological replicates) and sequenced multiple replicates per individual (technical replicates) using the Illumina Truseq protocol. We observed that the replicates of samples clustered by subject indicating that the biological variability (~95%) was greater than the technical variability (~0.50%). We observed that ~30% of the sequencing reads were miRNAs. We evaluated the technical parameters of sequencing by spiking the EV RNA preparation with a mix of synthetic small RNA and demonstrated a disconnect between input concentration of the spike-in RNA and sequencing read frequencies indicating that bias was introduced during library preparation. To determine whether there are differences between library preparation platforms, we compared the Truseq with the Nextflex protocol that had been designed to reduce bias in library preparation. While both methods were technically robust, the Nextflex protocol reduced the bias and exhibited a linear range across input concentrations of the synthetic spike-ins. Altogether, our results indicate that technical variability is much smaller than biological variability supporting the use of EV small RNAs as potential biomarkers. Our findings also indicate that the choice of library preparation method leads to artificial differences in the datasets generated invalidating the comparability of sequencing data across library preparation platforms.