Human plasma and serum extracellular small RNA reference profiles and their clinical utility

Abstract

Circulating extracellular RNAs (exRNAs) have the potential to serve as biomarkers for a wide range of medical conditions. However, limitations in existing exRNA isolation methods and a lack of knowledge on parameters affecting exRNA variability in human samples may hinder their successful discovery and clinical implementation. Using combinations of denaturants, reducing agents, proteolysis, and revised organic extraction, we developed an automated, high-throughput approach for recovery of exRNAs and exDNA from the same biofluid sample. We applied this method to characterize exRNAs from 312 plasma and serum samples collected from 13 healthy volunteers at 12 time points over a 2-month period. Small RNA cDNA library sequencing identified nearly twofold increased epithelial-, muscle-, and neuroendocrine-cell-specific miRNAs in females, while fasting and hormonal cycle showed little effect. External standardization helped to detect quantitative differences in erythrocyte and platelet-specific miRNA contributions and in miRNA concentrations between biofluids. It also helped to identify a study participant with a unique exRNA phenotype featuring a miRNA signature of up to 20-fold elevated endocrine-cell-specific miRNAs and twofold elevated total miRNA concentrations stable for over 1 year. Collectively, these results demonstrate an efficient and quantitative method to discern exRNA phenotypes and suggest that plasma and serum RNA profiles are stable over months and can be routinely monitored in long-term clinical studies.

Keywords: biofluid DNA isolation; biofluid RNA isolation; exRNA biomarker; exRNA reference profiling; extracellular nucleic acids.

Fig. 1.

exRNA isolation from serum and plasma. (A) Schematic overview of extracellular RNA and DNA isolation procedure detailing addition of calibrator or radiolabeled DNA and RNA size markers. (B) Examples of RNA and DNA isolations from plasma and serum samples visualized by phosphorimaging of ³²P-labeled spike-in DNA and RNA tracers. Less than 0.1 pmol of labeled tracer nucleic acids (input lane) were added with the denaturant to individual 450-μL biofluid samples at the first step of the procedure. (C) Average composite read length distributions of 5′P/3′OH-containing exRNAs in platelet-depleted EDTA plasma and serum samples. (D) Individual miRNA concentration ranges in EDTA plasma and serum. miRNA concentrations were calculated using read abundance data of Set1 calibrators added with the denaturant at the beginning of the RNA isolation. Boxplots show first to third interquartile ranges and their medians for plasma (black diamonds) and serum samples (red squares) of all 13 study participants.

Fig. 2.

Unsupervised clustering of miRNAs from human biofluid samples reveal distinct miRNA signatures for serum and plasma and segregate subject P12 from otherwise indistinguishable study participants. For each sample, individual calibrator and miRNA read frequencies from Dataset S2 are log₂-transformed and color-coded. Study subject metadata and library preparation details are color-coded and indicated as annotation. This figure features samples with ≥2,000,000 hg19-mapped reads; an equivalent panel of samples with a lower cutoff ≥100,000 reads is included as Fig. S3. (A) The calibrator heatmap reports unsupervised clustering of two sets of 10 synthetic 22-nt 5′P/3′OH-containing oligoribonucleotides spiked into samples at the beginning of RNA isolation (Set1) and before small RNA cDNA library preparation (Set2). (B) The miRNA heatmap is based on the union of the top 90% of miRNAs present in any sample. Sample order (columns) and miRNA arrangement (rows) were determined by hierarchical clustering. All biofluid samples from study subject P12 (green) clustered together yet were different from the rest of samples.

Fig. 3.

Differences in miRNA abundance based on type of biofluid, individual P12, gender, female hormonal cycle, and prandial state. MA plots are based on merged miRNA read counts from Dataset S3 and tabulated values describing lfc values, and their significance can be found in Dataset S4. Significant differences in abundance are indicated by red dots for P_adj < 0.05. Blood-based miRNAs are labeled in red (RBCs) and green (platelets/PBMCs). Select examples of additional miRNAs discussed in the text are labeled black when their abundance variance is greater than fourfold (lfcs ≤ −2 or ≥2), while smaller magnitudes are shown in gray. A count of 10 normalized reads, which has been chosen as a lower threshold for data analyses to eliminate noise, is indicated by a dashed vertical line. Count normalization in this analysis considers all miRNA sample reads, unless stated otherwise. (A and B) Serum vs. plasma, excluding individual P12, considering all miRNA reads (A) or utilizing miR-451 and 144 as standards (B) for SFE/miRNA read count normalization. (C) Unsupervised clustering separating miRNA signatures of biofluid representatives and those of RBC, PBMC, and platelets; only a few abundant miRNAs show enrichment in some of these major cellular or cell-derived blood constituents (red, green: higher relative abundance in RBC and platelets/PBMCs, respectively). (D–F) Study subject P12 vs. other male study participants, plasma (D and F) and serum (E). In F, we used Set1 calibrator reads for SFE/read count normalization. (G) Effect of gender, comparing female vs. male (plasma). For the corresponding serum comparison, see Fig. S4A. (H) Effect of menstrual cycle, comparing follicular vs. luteal state, in plasma. For the corresponding serum comparisons, see Fig. S4E. (I) Effect of fasting: preprandial vs. 4-h postprandial state, in plasma. For the 1-h preprandial vs. postprandial state comparison, see Fig. S4F.

Fig. 4.

miRNA abundance variability in individual P12 compared with other individuals. (A) Boxplots featuring Set1-calibrator normalized read counts of select miRNAs, which are either prevalent in plasma (pink) and serum (sage) and show no significant difference (miR-451, 144, 223); or which are tissue specifically expressed and are significantly more abundant in individual P12 (miR-1, 122, 200a, 375, 320). miRNA abundance was computed using Set1-normalized count data. (B) Receiver operating characteristic curves based on miRNA 375 and 320 read counts.