Haemolysis Detection in MicroRNA-Seq from Clinical Plasma Samples

Abstract

The abundance of cell-free microRNA (miRNA) has been measured in blood plasma and proposed as a source of novel, minimally invasive biomarkers for several diseases. Despite improvements in quantification methods, there is no consensus regarding how haemolysis affects plasma miRNA content. We propose a method for haemolysis detection in miRNA high-throughput sequencing (HTS) data from libraries prepared using human plasma. To establish a miRNA haemolysis signature we tested differential miRNA abundance between plasma samples with known haemolysis status. Using these miRNAs with statistically significant higher abundance in our haemolysed group, we further refined the set to reveal high-confidence haemolysis association. Given our specific context, i.e., women of reproductive age, we also tested for significant differences between pregnant and non-pregnant groups. We report a novel 20-miRNA signature used to identify the presence of haemolysis in silico in HTS miRNA-sequencing data. Further, we validated the signature set using firstly an all-male cohort (prostate cancer) and secondly a mixed male and female cohort (radiographic knee osteoarthritis). Conclusion: Given the potential for haemolysis contamination, we recommend that assays for haemolysis detection become standard pre-analytical practice and provide here a simple method for haemolysis detection.

Keywords: bioinformatics; biomarker; haemolysis; microRNA; plasma; prediction.

Figure 1

The number of mature miRNA species identified in an individual sample increases with read depth for both haemolysed (dark red) and non-haemolysed (blue) samples. However, the number of mature miRNA species identified for a given read depth is significantly lower (ANOVA, p-value = 1.68 × 10-9) in samples affected by haemolysis when compared to a non-haemolysed sample of equal read depth (n = 121).

Figure 2

(a) A comparison of the derived haemolysis metric and the ΔCq measure of haemolysis shows a clear correlation. We identified 13 samples (named) that we suggest should be discarded or used with caution in further analysis. (b) Histogram of haemolysis metric values from the 121 samples in our experiment, coloured according to their ΔCq (miR-23a-3p-miR-451a) classification, indicate a minimum haemolysis metric of ≥ 1.9 for samples previously identified as haemolysed.