Plasma microRNA Interindividual Variability in Healthy Individuals, Pregnant Women, and an Individual with a Stably Altered Neuroendocrine Phenotype.

BACKGROUND
Extracellular RNAs (exRNAs) in biofluids are amenable to quantitative analysis and proposed as noninvasive biomarkers for monitoring organ function. Cell-lineage-specific microRNAs (miRNAs) are present in plasma as soluble ribonucleoproteins or enclosed in exRNA carriers and transported through the vasculature. However, more extensive studies of healthy individuals are needed to gain insights into the variability of plasma miRNA abundance and composition.


METHODS
The exRNA composition of platelet-depleted plasma collected twice from 236 healthy individuals was characterized by small RNA sequencing. Plasma of pregnant women featuring dramatically increased placental miRNAs and samples from subject P12 with noticeably increased epithelial- and neuroendocrine-origin miRNAs were included for comparison. The miRNA content of 10 000g and 100 000g pellet fractions of plasma generated by ultracentrifugation was also determined. Data analysis methods included Pearson correlation, differential gene expression, and unsupervised clustering.


RESULTS
The abundance changes for more variable miRNAs in plasma of normal individuals correlated between coexpressed cell-lineage-specific miRNAs of the liver, neuroendocrine organs, epithelial cells, and muscle. ExRNA of pellet fractions contained <2% of total plasma miRNA with modest enrichment of lineage-specific and variable miRNAs compared to supernatant. The abundance fold changes of miRNAs observed in pregnancy and P12 compared to normal exceeded interquartile variability of healthy individuals. The neuroendocrine miRNA signature of P12 persisted for more than 4 years and was absent in other individuals.


CONCLUSIONS
This study defines the framework and effect size for screening of extensive plasma collections for miRNA phenotypes and biomarker discovery.

While the earliest observations of exNA in biofluids date back 5 decades (11,12), their systematic study became increasingly prominent with the introduction of deep-sequencing technologies. Fragmented exNA persists in nuclease-rich biofluids in ribonucleoproteins (RNPs) (13) or nucleosomes and may be further shielded in larger assemblies or carriers, including extracellular vesicles (EV) of different sizes and lipoproteins (12,14,15). miRNAs in circulation are stably bound to members of the Argonaute protein family (16), including soluble Argonaute miRNPs (13). A fraction of plasma miRNAs are also encapsulated in larger exRNA carriers and have been implicated in intercellular molecular communication (14). However, since plasma miRNA concentrations are approximately 100-fold below those of hormones (5), their role may be limited to innate immune signaling rather than guiding sequence-specific mRNA destabilization (17).
Progress in the exRNA field has been hampered by an inability to efficiently and reproducibly recover their small quantities present in biofluids rich in secreted nucleases (18,19). Various methods for isolation of exRNA carriers have been proposed, including precipitation, affinity interactions, gel filtration, and ultrafiltration (15,20). Differential centrifugation is still considered an effective method to separate larger exRNA carriers, including EVs and lipid-enclosed fractions from smaller soluble RNA-binding protein complexes in plasma (21).
Here we applied a protocol developed for automatable exNA isolation starting with 0.5 mL of plateletdepleted plasma yielding low nanogram quantities of exRNA and similar amounts of exDNA when using an additional fractionation step. We characterized the plasma miRNA composition of 236 nonpregnant and 20 pregnant individuals and included additional samples of subject P12, featuring a stable neuroendocrine phenotype (22) and his family, reporting results from >500 samples processed by a uniform procedure.

COLLECTION AND FRACTIONATION OF EDTA PLASMA
All clinical investigation was performed according to Declaration of Helsinki principles, following review and approval by the institutional review boards at Columbia University (IRB no. AAAS2157) and Rockefeller University (IRB no. TTU0707, KMA950, and KMA983). Enrollment, informed consent, metadata, blood collection, and low-volume exRNA isolation were performed as previously described (22). Collected samples and metadata are summarized in Supplemental Table 1 and individually reported in Supplemental Dataset 1 in the Online Supplement. For plasma fractionation, up to 108 mL of blood was collected into multiple 9-mL K 3 EDTA Vacutainer tubes (Greiner BioOne 455036). Platelet-depleted plasma was obtained after 2 sequential centrifugations at 2500g for 15 min at room temperature, each followed by the transfer of the supernatant into a fresh tube (22), if required, 50-mL tubes (Falcon no. 352098) were used to accommodate the larger collection volumes. Pelleted exRNA carrier fractions and corresponding carrier-depleted supernatant fractions were obtained by differential centrifugation of 11 mL of platelet-depleted plasma. Subfractions included pelleted 10 000g sediment and an aliquot of the 10 000g supernatant obtained after a 30 min centrifugation at 4 C; pelleted 100 000g sediment obtained from the remaining 10 000g supernatant, and an aliquot of the resulting 100 000g supernatant, obtained after a 90 min centrifugation at 4 C. Further details are provided in the Online Supplement, Materials and Methods.
RNA ISOLATION, SMALL RNA COMPLEMENTARY DNA LIBRARY PREPARATION, AND DEEP SEQUENCING Automated RNA isolation was conducted as previously described (22) with modifications detailed in the Online Supplement, Materials and Methods. Briefly, samples were randomized and organized in batches of 24 samples of 450 mL plasma or its subfractions. An aliquot of Calibrator Set1, composed of 10 21-nt 5 0 -phosphorylated RNA oligoribonucleotides, was added with the denaturant to each sample. An aliquot of isolated sample RNA along with an aliquot of spiked-in calibrator Set2 was subjected to complementary DNA (cDNA) library preparation, including size selection to enrich for 19 nt up to 45 nt RNA using 15% polyacrylamide gel electrophoresis gels and 32 P-labeled reference oligonucleotides as size markers. cDNA libraries were generated by Next-Generation Illumina sequencing using single reads (large cohort, n ¼ 496 samples; P12 family, n ¼ 12 samples; 4 plasma subfractions, n ¼ 17 samples each).

DATA PROCESSING AND BIOINFORMATICS ANALYSES
Sequencing data was processed by our in-house RNAsequencing data analysis platform using extraction lengths of 16 nt to 45 nt, a minimum 3 0 overlap of adapter with 0/1 mismatches of 4/5 nt, respectively, for sequence extraction and removal of low complexity sequences. For mapping demultiplexed reads, profile annoDB_jul2014c was used, which included reference sequences for the definitions of human and virus RNAs based on miRBASE 18 with modifications and HG19 (23). Annotated reads and their corresponding RNA classes are reported in RNA summary tables (listed for each cohort in Supplemental Table 3 and every sample in Supplemental Dataset 2), including miRNA read count and read frequency (RF) tables (Supplemental Datasets 3 and 4). Sequencing data was deposited in National Center for Biotechnology Information Gene Expression Omnibus (24) under accession number GSE181810.
Reads annotated as calibrators, expression system (plasmid and E. coli), size markers, and ligation adapters were considered as "technical reads"; the remainder was regarded as "sample reads." miRNA concentrations were calculated from the sum of Set1 calibrator read counts, considering their added molar amounts and volumes of biofluid samples.

UNSUPERVISED CLUSTERING
RFs of cluster-mirs and calibrators were used to generate log 2 -transformed heatmaps as previously described (22). Briefly, we considered the most abundant cluster-mirs for every chosen dataset until a cumulative !85% (top 85%) RF threshold was reached. We used the union of all individual cluster-mir selections to generate a combined top 85% cluster-mir RF table for unsupervised clustering using pheatmap (Raivo Kolde, 2015; pheatmap: Pretty Heatmaps; R package, v.1.0.8). Other thresholds were used where indicated.

DIFFERENTIAL GENE EXPRESSION (DGE)
Tabulated shared raw reads of cluster-mirs reported by RNA-sequencing data analysis platform were used to perform differential expression analyses using DESeq2 (25), to generate DGE plots and read abundance boxplots as previously described (22). Comparisons included genders, pregnancy, and the presence of a neuroendocrine phenotype (P12). For DGE comparisons of pellet and supernatant fractions, an invariant mean RNA composition prerequisite is not met, resulting in exaggerated scaling of DESeq2 normalized counts. We instead calculated relative cluster-mir frequencies based on cluster-mir counts per total cluster-mir counts, scaled to reads per million, and reported changes in relative abundance as fold-change median read per million ratios. log 2 -transformed abundance ratios were plotted as scatterplots against log 2 -transformed median abundance values. RF comparisons showed no normal distributions; therefore, the significance of an abundance change was reported as a Wilcoxon signed-rank test p value, adjusted for multiple measurements using the Benjamin-Hochberg correction (p adj ). Differences were considered significant for p adj < 0.05. All DGE results were reported in Supplemental Dataset 5.

Results
We collected platelet-depleted plasma from a large cohort of 173 female and 63 male healthy volunteers of diverse races and ages at 2 visits spaced 2 weeks apart. We included 4 formerly collected aliquots from study subject P12 (22), who displayed a stable neuroendocrine miRNA signature in plasma, and 20 plasma samples of pregnant women spanning all trimesters T1 to T3 of pregnancy, with an expected placenta miRNA contribution (5) (large cohort: n ¼ 496 total samples, Supplemental Table 1). As a separate cohort, we collected platelet-depleted plasma of 3 unstudied P12 family members along with a sample from P12, 4 years after the initial discovery of the phenotype, along with samples of 2 healthy volunteers, which were all represented as 2 replicates in the cDNA library preparation (P12 family cohort: n ¼ 2 Â 6 samples, Supplemental Table 1).
To determine extracellular miRNA abundance, samples were processed in 25 batches of up to 24 3 0adapter-multiplexed samples, each spiked with external calibrator oligonucleotide sets during processing (22,26). Calibrators were added into the denaturant at the beginning of the exRNA isolation (Set1) and before cDNA library preparation (Set2). Based on median values representing this cohort, 2.1 million sequence reads per sample were obtained, of which a median of 82% mapped to miRNAs (Supplemental Table 3). Small cytoplasmic RNAs, including transfer RNAs and Y RNAs, accounted for 7% of all reads and 4.9% mapped to the human HG19 in regions without specific noncoding RNA annotation. Calibrator sets were present at 38 000 reads per sample, plasmid and bacterial sequence carried over from the production of recombinant RNA ligase used for library preparation showed 34 400 reads per sample.
To reduce redundant miRNA abundance information, we summed up the reads of miRNAs originating from cistronically expressed miRNAs (cluster-mirs) (27). Cluster-mir read frequencies were normalized over total cluster-mir read counts, although some analytical software packages required as input the actual clustermir read counts relying on their internal normalization processes.

DISTINCT CELLULAR AND TISSUE SOURCES INDEPENDENTLY CONTRIBUTE TO MIRNA VARIABILITY IN PLASMA
We assessed cluster-mir variability and covariation across samples of the large cohort of healthy controls omitting P12 and pregnant women samples and keeping male and female samples in separate groups. To determine the variability of abundance (normalized read counts) for any detectable cluster-mir across all individuals, cluster-mir QCD values were calculated from normalized cluster-mir read counts and plotted against their midhinge normalized read count values (Fig. 1, A). Furthermore, fitted dispersion values (QCD fit ) were calculated as a function of read abundance by applying exponential fitting and considering all detectable cluster-mirs. Cluster-mirs with QCD values that exceeded a threshold of 1.66Â QCD fit were considered as highly variable (Fig. 1, A, Supplemental Fig. 1, A). Actual QCD values never exceeded 2.5Â QCD fit , defining an upper threshold for variability encountered across normal healthy individuals (Fig. 1, A).
The highly variable cluster-mirs of each gender group were subjected to Pearson correlation analysis to identify cluster-mirs of common cellular origin (Fig. 1, B). The abundances of liver-specific cluster-mir-122 and clustermir-885 (28) were correlated (R ¼ 0.61) (Fig. 1, B, Supplemental Fig. 1, B). Cluster-mir-33b, located in an intron of a gene predominantly transcribed in the adrenal gland and the liver (29), was moderately correlated with ductal-epithelial-cell-enriched cluster-mir-205 (30) (R ! 0.52). Cluster-mir-134(43) was highly correlated with cluster-mir-127(8) (R ¼ 0.75), both enriched in the brain and in neuroendocrine tissues (3,31), and with normalized abundance data of highly variable cluster-mirs (bold) to all mirs were compared using Spearman correlation. To facilitate comparisons, the 2 gender-specific cluster-mir correlation tables were arranged in one array and outlined in blue (maledatasets) and pink (female datasets). Cluster-mirs correlated at R ! 0.5 were grouped and are separated by lines. For clustermirs whose lineage-specificity could solely be inferred by their genomic location/coexpression with tissue-specific genes, the suggested sources are set in italics.

MIRNA SIGNATURES COMPARED TO NORMAL CONTROLS
We conducted unsupervised hierarchical clustering of representative samples. From each participant of the large cohort and an earlier pilot study (22), we selected the sample dataset with the highest miRNA read depth, and included all datasets from P12. Based on their combined top 85% of cluster-mir RFs, the clustering algorithm grouped all samples of P12 to one end of the distance tree, while samples from pregnant individuals appeared on the other side (Fig. 2). A similar structuring was observed when all samples were considered (Supplemental Fig. 2).
DGE of cluster-mir abundance comparing female (n ¼ 344) to male (n ¼ 126), excluding samples of pregnant women and subject P12, revealed 33 significant (p adj 0.05) and 20 additional highly significant (p adj 0.001) differences, all within 1.5-fold and mostly below the computed high-variability threshold of 1.66Â QCD fit (Supplemental Fig. 3, A). Only clustermir-124-1(3), which was twofold enriched in women, marginally exceeded our threshold of 2.5Â QCD fit maximum variability.
The !2-fold gender-specific differences noted in our earlier DGE study (Supplemental Fig. 3, B), which did not identify cluster-mir-124-1(3) members, fell within our maximum variability limit and should therefore be reinterpreted as interindividual rather than gender-associated differences, as it was based on only 6 male and 6 female participants (22). DGE analysis of pregnant (n ¼ 20) versus agematched nonpregnant women (n ¼ 260) samples revealed a 50-fold increased abundance of placentaspecific (5) cluster-mir-498(46) and was highest (!250fold) for the trimester T3 subgroup (Fig. 3, A and B). We also noted additional highly significant changes, albeit at lower magnitude fold, including cluster-mir-199b, 199a-1(3), 148a, 188 (8), differentiating T1 from later pregnancy stages and nonpregnant samples (Fig. 3,  A). These differences also drove unsupervised clustering distinguishing early from late pregnancy stages (Fig. 3,  C). Most pregnancy-associated differences were well outside the limits defined by our 2.5Â QCD fit maximum variability thresholds of healthy control samples.

THE NEUROENDOCRINE SIGNATURE OF P12 IS ABSENT IN FAMILY MEMBERS AND OTHER INDIVIDUALS
The unique and stable neuroendocrine miRNA plasma signature of P12 and its absence in nearly 250 healthy individuals prompted us to investigate a genetic contribution to this phenotype. We obtained plasma samples from both parents and a brother of P12 (P12 family cohort), which were processed together with a newly collected sample of P12 and two normal male reference samples. Unsupervised clustering revealed that the neuroendocrine signature of P12 now persisted for at least 4 years following our initial discovery, while other P12 family members were indistinguishable and clustered together with normal controls (Fig. 4, A).
DGE of P12 within the P12 family cohort as well as across 69 male control study participants from the large and pilot cohort showed a >25-fold increase in neuroendocrine cluster-mir-375 and a >10-fold increase in ductal-epithelial-cell-specific cluster-mir-205, and epithelial-cell-specific 200a(3) and 141 (2) in P12 compared to control (Fig. 4, B and C). Cluster-mir-320, which is more broadly expressed including neuroendocrine cells (3), was increased fivefold in P12 compared to controls. All miRNA content differences in plasma of P12, including many lower-fold cluster-mir changes, surpassed the limits defined by the 2.5Â QCD fit maximum variability threshold in healthy control plasma samples.

PLASMA
We processed samples from P12 and his family, 2 healthy male volunteers, 2 nonpregnant, and 9 pregnant women. We applied a large-volume sequential differential ultracentrifugation strategy to obtain sufficient RNA from pellets to generate high-quality cDNA libraries. As individual subfractions, we characterized 10 000g and 100 000g pellets and their corresponding supernatants (n ¼ 17 samples each, Supplemental Table 1). Based on Set1 calibrator spike-ins, total miRNA median concentrations represented by pelletable exRNA carriers in plasma were 120 fM, and 15 fM identified in the 10 000g and 100 000g pellet fractions, respectively, and 4.6 pM and 5.8 pM for soluble miRNAs in the corresponding depleted supernatants, respectively (Fig. 5, A). This comparative analysis excluded the sample of P12, which had increased supernatant miRNAs consistent with overall increased plasma miRNAs (22), while their miRNA concentrations represented by pelletable exRNA carriers stayed within range defined by the control group.
We conducted unsupervised hierarchical clustering of fractionated samples for normal control and P12 family using their combined top 75% of cluster-mir RFs across all samples considered in this analysis (Fig. 5, B, Supplemental Fig. 4, A). The clustering algorithm separated the supernatant fractions from pellet fractions and 10 000g from 100 000g pellet fractions. Plasma supernatants of P12 clustered separately from all other samples.
DGE of pellet versus supernatant fractions excluding P12 revealed 22-fold and eightfold enrichment of myeloidlineage-and platelet-specific cluster-mir-223 in 10 000g and 100 000g pellets relative to supernatants, respectively, suggestive of collection of platelet and myeloid cell debris Red dots indicate statistically significant differences in abundance. log 2 -fold-change values based on 2.50Â multiples of calculated QCD fit values, previously used as thresholds for maximum variabilities observed in our datasets of normal healthy individuals (Fig 1, A), were indicated as dashed lines. (B) Boxplots comparing normalized read counts of selected cluster-mirs of controls and pregnancy plasma. (C) Unsupervised clustering of pregnancy samples based on select cluster-mirs, which help to discern pregnancy states.
1682 Clinical Chemistry 67:12 (2021) Fig. 4. The neuroendocrine cluster-mir signature of P12 is still present in plasma collected 3 years after the first observation. (A) Unsupervised clustering of cluster-mirs of plasma samples of P12, his family, and 2 healthy volunteers. The cluster-mir heatmap is based on the combined top-85% of cluster-mir RFs across all samples considered in this analysis. Sample order (columns) and cluster-mir arrangement (rows) were determined by hierarchical clustering. The calibrator heatmap reports unsupervised clustering for calibrator sets. (B) DGE plots based on cluster-mir read counts and shown as log 2 -fold-change values. Red dots indicate significant differences in abundance. log 2 -fold-change values based on 2.5Â multiples of calculated QCD fit values, previously used as thresholds for maximum variabilities observed in our datasets of normal healthy individuals (Fig 1, A), were indicated as dashed lines. (C) Boxplots for normalized read counts of selected cluster-mirs of subject P12 and controls. Select sample subgroups are individually listed: female (F) and male (M) samples of the large cohort and pilot study representatives (LC and PS), including subject P12 samples collected at the study onset (start); a sample of P12 taken 2 years (2 y.) after the discovery of this subject's phenotype; samples of P12 family members, which include samples of P12, collected 4 years (4 y.) after the discovery of this subject's phenotype, and samples of 2 additional healthy controls. and/or exRNA carriers residual to platelet-depleted plasma (Fig. 5, C, Supplemental Fig. 5, A). Red-blood-celldominating cluster-mir-144(2) was sixfold depleted in the 10 000g fraction and changed insignificantly in the 100 000g fractions relative to supernatants. Hepatocytespecific cluster-mir-122 was !1 000-fold depleted in the 10 000g pellet and changed insignificantly in 100 000g pellet fractions relative to supernatants. The cluster-mir-885 followed the cluster-mir-122 pattern, indicating that hepatocytes contributed exRNA carriers present in the 100 000g pellet and predominantly miRNPs to circulation. Abundance-correlated cluster-mir-134(43) and mir-127 (8) show !5and !8-fold enrichment in 10 000g and 100 000g pellet fractions, respectively, compared to the corresponding supernatants. Neuroendocrine-origin cluster-mir-375 detectable in supernatants of all normal controls was strongly depleted from pellet fractions, indicative of soluble miRNP of cluster-mir-375 in plasma (Supplemental Fig. 5, A). In P12, with its plasma abundance increased !25-fold (Fig. 4, B), cluster-mir-375 distributed similarly and was strongly depleted in pellet fractions (Fig. 5, C, Supplemental Fig. 5, A).  and 205 of normal controls were unaltered in abundance in pellet versus supernatant, while 141(2) appeared marginally enriched in 10 000g pellets. In P12, with plasma abundances increased !10-fold (Fig. 4, B), cluster-mir-200a (3), 141 (2), and 205 were depleted in pellets, albeit the absolute content of epithelial cluster-mirs in pellets was comparable to those of normal controls (Supplemental Fig. 5, A). The persistence of the P12-specific epithelial cluster-mirs in supernatant indicates a unique soluble miRNP contribution paralleling the partitioning of cluster-mir-375. DGE of 10 000g versus 10 000g supernatant did not reveal any statistically significant differences (Fig. 5, C).
Clustering of cluster-mir profiles from pregnant and nonpregnant controls also separated supernatant and pellet fractions (Fig. 6, A, Supplemental Fig. 4, B). Samples were further subgrouped by pregnancy according to the presence of placenta-derived cluster-mir-498(46). Cluster-mir-498(46) was about threefold enriched in 10 000g and 100 000g pellet fractions compared to their corresponding depleted supernatants, slightly above the significance threshold (p adj ¼ 0.07) (Fig. 6, B and C). Enrichment or depletion in pellets for other cluster-mirs was comparable to those noted above, indicating that only a small fraction of placenta-specific cluster-mir-498(46) was present in pelletable exRNA carriers with most remaining as soluble miRNP in supernatant following ultracentrifugation.

Discussion
In this study, we expanded our earlier efforts of defining baseline plasma miRNA profiles (22) by increasing the numbers of study participants more than 20-fold. To better understand the natural variation across normal individuals, we coprocessed plasma samples of individuals with well-established extracellular miRNA signatures. We believe ours to be the largest study of biofluid miRNAs following a uniform protocol (32).
Using correlation analysis of abundance variation encountered in normal individuals, we identified several groups of cluster-mirs characteristic of contributions from myeloid-lineage cells and platelets, liver, epithelial cells, muscle, and neuroendocrine organs. The interquartile range of these variations typically covered less than 1 order of magnitude across normal healthy individuals (Supplemental Fig. 1). In plasma of patients with heart failure, however, changes were >18-fold for muscle-specific cluster-mir-1-1(4) and mir-133b (2) and up to 143-fold for 208a, 208b, and 499 (8). In liver transplant patients experiencing rejection disease, cluster-mir-122, and 885 increased 7.6-fold (33). The abundance in cluster-mir-375 was increased about 25fold in P12, while cluster-mir-498(46) increased from 50-to !250-fold from trimester T1 to T3. In summary, cell-type specific miRNAs in plasma show limited variability in abundance in normal individuals. Therefore, miRNA abundance changes falling outside the baseline variability defined as 2.5-fold QCD fit , have biomarker value as exemplified by P12 and pregnancy. Potential applications of extracellular miRNAs as biomarkers have been discussed previously (1,(4)(5)(6)(8)(9)(10).
EVs and other exRNA carriers have been extensively discussed as sources for extracellular miRNAs in biofluids and as potential biomarkers in disease states (4,34). The low magnitude of cell-lineage-specific, nonhematopoietic-origin miRNA enrichment in pelleted exRNA carriers over supernatants questions the value of fractionation by ultracentrifugation for enhancing the biomarker value of plasma miRNAs. Neither of the celltype-specific and highly variable miRNAs reached a >10-fold enrichment in pellet over supernatant. In P12 plasma, where neuroendocrine mir-375 and epithelial cluster-mirs were !25-fold and !10-fold increased compared to male controls, respectively, the increases were due to soluble miRNP contributions. This may suggest a P12-specific release mechanism related to cellular turnover or pyroptosis (35). Our results do not exclude the possibility of physiological or disease conditions resulting in increased formation of EVs or other exRNA carriers featuring lineage-specific miRNA signatures, where fractionation approaches may be beneficial to differentiate disease states. However, further separation of pelleted exRNA carriers and their characterization by small RNA sequencing in the cohorts studied here limited by volume will likely fail technically as 2% of total miRNAs are contributed by all exRNA carriers collected by centrifugation as pellets. Assuming a mean of !10 8 exRNA carriers per ml of plasma (36,37), including EV and lipoproteins, the resulting number of miRNA molecules per carrier would be less than 1, in agreement with other observations (37).
Additional studies will be required to determine how other lineage-specific RNA classes, and in particular mRNAs, distribute between exRNA carriers versus soluble RNA-binding proteins fractions, requiring methods not selective for 5 0 P/3 0 OH RNAs (38,39) and more efficient than ligation-based methods (40). Given that most exRNAs of !100 nt size were targeted by RNases and, consequently, lack distinctive size or sequence features enabling enrichment, and given that nonlineage-specific ribosomal RNA fragments predominate over other RNA classes (38,39), the characterization of informative exRNAs such as ex-mRNA fragments by sequencing will likely require sequence-specific enrichment or depletion methods. For clinical testing, RT-qPCR-based techniques-limited in the number of discernable targets per reaction but highly sensitive and not requiring RNA enrichment-may be preferred over RNA-sequencing approaches once suitable RNA markers have been established.
In summary, we have documented high reproducibility and interpersonal variability of extracellular miRNAs across hundreds of normal control plasma samples and our ability to readily identify biologically relevant miRNA phenotypes outside of baseline variability. Additional fractionation of plasma by ultracentrifugation in an attempt to accumulate nonhematopoietic, organ-or cell-type-specific miRNAs in pelleted carrier fractions versus soluble miRNPs in the carrier-depleted supernatants was not beneficial for the cohorts studied here and merely increased the burden for collecting, storing, and processing large volumes of biofluids.

Supplemental Material
Supplemental material is available at Clinical Chemistry online.
Human Genes: To reduce redundant miRNA abundance information in the paper, we summed up the reads of miRNAs originating from cistronically expressed miRNAs (cluster-mirs), as explained in citation (27). Cluster-mirs were named according to the lowest numbered mir and stating the number of coexpressed mirs (i.e., stem-loop precursors) in round brackets unless representing a single-member cistron. To comply with the prerequisite to report human genes, we list all clustermirs discussed in the paper in the order of appearance, along with the human miRNA genes they contain. For each gene, we report the gene symbol, followed by the accepted gene name.