Evaluation of variability in cell-free DNA extraction efficiency from plasma and urine and spike-in normalization

Abstract

Cell-free DNA (cfDNA) is a promising new biomarker for clinical use in e.g. oncology and transplantation medicine. Urinary cfDNA is also gaining interest as a non-invasive biomarker. However, cfDNA extraction is not standardised, leading to a variety of different methods being used with varying efficiencies and size-specificities. In this study, we aimed to assess the variability in cfDNA extraction efficiency for multiple cfDNA extraction methods (QIAamp Circulating Nucleic Acid Kit, Zymo Quick-DNA Urine Kit, Q Sepharose protocol (Qseph)) using the artificial spike-in CEREBIS and compare the contribution of variable extraction efficiency to overall variability in cfDNA quantities determined using droplet digital PCR (ddPCR). We found reproducible extraction efficiencies specific for each method with 84.1% (± 8.17) in plasma, 58.7% (± 11.1) for Zymo, as well as 30.2% (± 13.2) for Qseph based on the 180 bp CEREBIS (Construct to Evaluate the Recovery Efficiency of cfDNA extraction and BISulphite modification) spike-in. Additionally, while the largest proportion of the technical variability was observed between extractions, it was almost negligible compared to the biological variability. Normalization of urinary cfDNA using creatinine in urine reduced the variability, whereas when normalizing for CEREBIS-based extraction efficiency this was not consistently observed. With overall relatively consistent extraction efficiencies within each cfDNA extraction method, normalization for extraction efficiency using CEREBIS thus did not show a clear benefit but might be considered for comparisons between extraction methods.

Keywords: Cell-Free DNA; Extraction efficiency; Liquid biopsy; Pre-analytics; Spike-in.

Fig. 1

Fragment size distribution for one sample each extracted from plasma using QIA and urine samples using Zymo and Qseph. The same sample is shown for both urinary cfDNA extractions. The expected size range for Zymo is 100 bp – 23 kb and 25–23 kb for Qseph. Fluorescence intensities at a given size are not comparable between methods, due to variable input used. This graph was capped at fluorescence intensity of 40 to increase the details and the full figure is shown in the Supplementary Fig. S6.

Fig. 2

Variance component estimates of ddPCR-based cfDNA quantities in plasma samples with and without extraction efficiency adjustment. The stacked bar plots represent the total variance component estimates from the nested ANOVA for each assay in the technical plasma setup (a) and the biological plasma setup (c) with the colours indicating the nested factors. Only the variance component estimate of the inter-extraction variability is shown in (b) for the technical plasma setup and (d) for the biological plasma setup. The results using the cp/mL as well as the CER180bp extraction efficiency adjusted cp/mL (cpml/EE180) are shown.

Fig. 3

Nested ANOVA results from urine samples with adjustment for extraction efficiency and/or urinary creatinine. The variance component estimates are shown for the complete model of the urine long setup (a) with the nested factors distinguished by colour. Inter-individual (b) as well as inter-extraction (c) variance component estimates are shown for the urine long setup. The graphs are separated by extraction method and the analyses for cp/mL unadjusted, and adjusted for creatinine (cpml/UCr), extraction efficiency (cpml/EE) or both (cpml/Ucr/EE) are depicted. Total variance component estimates for the urine combination setup are shown in (d) and the estimates for the inter-individual (e) as well as inter-extraction (f) variability are shown separately. Only samples extracted with Qseph are shown for the urine combination setup with the respective extraction efficiency (EE) and/or urinary creatinine (UCr) adjustments.