Identification of extracellular miRNA in human cerebrospinal fluid by next-generation sequencing

In this Methods paper, the authors identify optimal ways for RNA extraction and detection of miRNAs from body fluids. A comparison of ten commercially available RNA isolation kits and an optimized RNA-isolation method is used to profile small RNAs present in human cerebrospinal fluid and plasma using next-generation sequencing. This can be used for the discovery of novel RNAs in patient biofluids and could be of importance to provide biomarkers for disease.

Abstract

There has been a growing interest in using next-generation sequencing (NGS) to profile extracellular small RNAs from the blood and cerebrospinal fluid (CSF) of patients with neurological diseases, CNS tumors, or traumatic brain injury for biomarker discovery. Small sample volumes and samples with low RNA abundance create challenges for downstream small RNA sequencing assays. Plasma, serum, and CSF contain low amounts of total RNA, of which small RNAs make up a fraction. The purpose of this study was to maximize RNA isolation from RNA-limited samples and apply these methods to profile the miRNA in human CSF by small RNA deep sequencing. We systematically tested RNA isolation efficiency using ten commercially available kits and compared their performance on human plasma samples. We used RiboGreen to quantify total RNA yield and custom TaqMan assays to determine the efficiency of small RNA isolation for each of the kits. We significantly increased the recovery of small RNA by repeating the aqueous extraction during the phenol-chloroform purification in the top performing kits. We subsequently used the methods with the highest small RNA yield to purify RNA from CSF and serum samples from the same individual. We then prepared small RNA sequencing libraries using Illumina’s TruSeq sample preparation kit and sequenced the samples on the HiSeq 2000. Not surprisingly, we found that the miRNA expression profile of CSF is substantially different from that of serum. To our knowledge, this is the first time that the small RNA fraction from CSF has been profiled using next-generation sequencing.

INTRODUCTION

miRNAs are regulators of mRNA transcription and translation, they are stable, and often specifically enriched in a particular tissue or during essential cellular processes (Chen et al. 2008; Eacker et al. 2009; Rosa et al. 2009; Schratt 2009; Casalini et al. 2010; Li et al. 2012). There is a growing interest in examining extracellular miRNAs from biofluids to identify biomarkers for cancer, cardiovascular disease, diabetes, and many other disorders. In the current study, we developed a way to use small volumes of biofluids, 1 mL of plasma, serum, and CSF, for next-generation sequencing (NGS) using the Illumina platform. Sequencing provides us with the opportunity to examine all miRNAs at one time, including variations in expressed sequences (isomiRs), and the potential to identify novel miRNAs associated with disease.

Recent advances in small RNA library preparation have made sequencing less expensive, improved sample throughput, and allowed the output to be more quantitative (Aldridge and Hadfield 2012; Pritchard et al. 2012). The TruSeq Small RNA Sample Preparation Kit introduces the indexing barcode during PCR amplification of the library after adapter ligation. This method significantly reduces sample bias over previous indexing/barcoding approaches where a barcode was ligated directly to the miRNA (Van Nieuwerburgh et al. 2011). Researchers can barcode up to 48 samples and load as many as desired per lane of a sequencing flow cell, depending upon the amount of small RNA coverage required by the experiment.

Blood is among one of the easiest biofluids to obtain, and miRNAs assayed from blood may be able to provide us with important markers for kidney, liver, and cardiovascular disease, and many types of cancer. However, it is possible that the miRNA changes related to neurological insult found in the blood would be masked by unrelated or indirectly related systemic miRNA changes. Cerebrospinal fluid (CSF), because it bathes the central nervous system and comes into contact with the injured tissue, might be a better source of miRNA expression changes related to central nervous system injury. Recent publications by Cogswell et al. (2008) and Haghikia et al. (2012) found altered miRNA expression profiles in CSF associated with Alzheimer’s disease and multiple sclerosis, respectively. Baraniskin et al. (2011) identified differentially expressed miRNAs associated with primary central nervous system lymphoma in CSF. Because CSF has such a low quantity of RNA, in order to examine miRNAs in CSF, these researchers used qRT-PCR and were required to either pool samples or limit the study to a small number of miRNAs.

Advances in RNA sequencing, primarily in the amplification and use of whole transcriptome RNA or mRNA to examine differential gene expression, has significantly increased the number and type of experiments that have been completed in the last few years (Oshlack et al. 2010; Fang et al. 2012; Robles et al. 2012). When using the whole transcriptome (including large noncoding RNAs) or mRNA as starting material, it is possible to begin Illumina TruSeq RNA sample preparation with 0.1–4 μg of total RNA. In some cases, as little as 500 pg of total RNA can be used to begin cDNA amplification from total RNA, thereby creating enough starting material to go into the Illumina platform (NuGEN Ovation RNA-Seq System V2).

NGS platforms for miRNA, such as Illumina’s TruSeq Small RNA Sample Prep Kit, require at least 1 μg of total RNA as starting input, an amount not attainable from a biofluid sample of 1 mL of plasma, serum, or CSF. Therefore, small RNA sequencing has remained underutilized for biomarker discovery due to the fact that many clinically relevant research projects do not have the sample volume required to provide enough RNA for small RNA library production.

Our primary purpose for this study was to perform small RNA deep sequencing in samples that had both small volumes and low RNA content (1 mL of serum, plasma, or CSF). It was critical to begin the process by identifying the RNA purification method with the highest recovery of extracellular small RNA. We tested commercially available RNA isolation kits on human plasma and ways to maximize small RNA recovery. Although the extracellular RNA content in serum and plasma is low, RNA levels are consistently detectable and quantifiable when isolated from volumes such as 200 μL. In contrast, when we measure the RNA isolated from 1–3 mL of CSF, RNA quantitation is sometimes below our limit of detection. Therefore, we used plasma for our initial RNA isolation tests and then tested the best kits on our CSF samples.

After isolation, the total RNA yield of each sample was assessed by Quant-iT RiboGreen (RiboGreen; Invitrogen). This method is the most sensitive for nucleic acid detection and least affected by possible contaminants. The small RNA yield is not effectively measured when we assay the total RNA; therefore, we used miRNA-specific TaqMan qRT-PCR assays against endogenous miRNAs and exogenous synthetic miRNAs to measure the recovery of the small RNA species in each of the samples. We used the four isolation kits that performed the best using the plasma samples to test RNA isolation from CSF samples.

Because the total RNA amounts that go into the TruSeq Small RNA Sample Prep Kit from biofluids are so low that they are often not accurately quantifiable, we tested a range of RNA isolated from different CSF sample volumes (from 0.5 to 1.5 mL). We prepared small RNA libraries from the total RNA isolated from each of those CSF volumes, sequenced them, analyzed the data, and found that the data from each volume correlated very well with the other samples. To our knowledge, this is the first report of using NGS to profile the extracellular miRNA from CSF.

RESULTS

Samples and RNA isolation kits

To allow us to make systematic comparisons among different RNA purification methods, we prepared a large number of uniform plasma samples. We used plasma; though it has low amounts of total RNA, the RNA is sufficient to be consistently quantified by RiboGreen and Taq. Plasma samples (∼1 mL each) collected from 20 subjects were thawed on ice, pooled, separated into 200-μL aliquots, flash-frozen in liquid nitrogen, and stored at −80°C (Fig. 1). When we tested the commercially available RNA isolation kits, individual 200-μL sample aliquots were thawed in the presence of the initial denaturant for the respective kit and tested in triplicate. The amount of denaturant that was added to each 200-μL aliquot is listed in Table 1.

FIGURE 1.

Work flow. In order to make a uniform pool of samples for use in all of the RNA isolation kit comparisons, we thawed 20 samples of ∼1 mL each on ice and mixed the samples to create a single pool. The pool was divided into 200-μL aliquots, flash-frozen, and stored at −80°C until they were used.

TABLE 1.

Summary of the RNA isolation methods

Known quantities of synthetic exogenous small RNAs (referred to as spike-ins) were added as a positive control for quantification after RNases were inactivated by the kit’s respective protein denaturant. The spike-ins were composed of a mix of three synthetically generated miRNAs that lack sequence homology to any known human small RNAs (cel-miR-39, cel-miR-54, and cel-miR-238) (Mitchell et al. 2008).

Ten commercially available RNA-isolation kits were compared (Table 1): MaxRecovery BiooPure RNA Isolation Reagent (BiooPure; BiooScientific), mirVana miRNA Isolation Kit (mirVana; Ambion), mirVana PARIS (PARIS; Ambion), TRI Reagent RT (MRC RT; Molecular Research Center), TRI Reagent RT-Blood (MRC RT-B; Molecular Research Center), TRI Reagent RT-Liquid Samples (MRC RT-LS; Molecular Research Center), RNAzol (Molecular Research Center), miRNeasy (Qiagen), PureLink microRNA Isolation Kit (PureLink; Invitrogen), and mirPremier (Sigma). The mirPremier kit from Sigma was found to be better suited for isolation of small RNA from solid tissue and was unsuitable for biofluids in our hands and was consequently dropped from the experiments. With the exception of PureLink, each of the remaining nine kits utilizes proprietary mixes of guanidinium-thiocyanate and phenol-chloroform in order to denature proteins and isolate RNA in the aqueous phase. BiooPure, TRI Reagent RT, TRI Reagent RT-Blood, TRI Reagent RT-Liquid Samples, and RNAzol all rely upon ethanol or isopropanol precipitation and centrifuge sedimentation (pelleting) of the RNA. In contrast, mirVana, mirVana PARIS, and miRNeasy employ solid-phase extraction techniques involving the addition of ethanol to the aqueous phase containing the RNA and subsequent binding of the RNA to a glass-fiber-based column by centrifugation. PureLink also utilizes guanidinium-thiocyanate as a protein denaturant; however, RNA larger than 200 nt is captured on one silica-based column in the presence of a low concentration of ethanol. A higher concentration of ethanol is added to the flow-through, allowing for small RNA to be immobilized when the solution is passed through a second silica-based column.

Total RNA yield

Quantification of the total RNA yield was determined by Quant-iT RiboGreen RNA reagent (Invitrogen). The performance of the RNA isolation protocol was assessed in triplicate, and total RNA yields were calculated from RiboGreen measurements and reported as the average (Fig. 2). Using the manufacturer-provided protocols from the nine RNA extraction kits yielded considerably different total RNA amounts, ranging from 0 to 16 ng (Fig. 2). In addition, the yield varied among the technical replicates for each kit, even within any one type of purification method. Based on our results, the MRC and PureLink kits were the most variable. Furthermore, there were notable differences in yield when utilizing ranges for centrifugation speed and temperature suggested by MRC (Fig. 2).

FIGURE 2.

Average recovery of total RNA from kits using manufacturer-provided protocols. Total RNA yield was assessed by RiboGreen Quant-it assay (ng) from 200 μL of plasma. Isolations were performed in triplicate and the average displayed. There is variability from isolation to isolation as depicted by the error bars (standard error of the mean). For the MRC and RNazol kits, different temperature or centrifugations were tested according to the manufacturer’s recommendations.

Small RNA yield

In order to compare the recovery of small RNA species isolated from the different kits, we used TaqMan qRT-PCR assays for both endogenous and synthetic miRNAs. We measured the exogenously added Caenorhabditis elegans synthetic small RNAs recovered by each kit, along with two endogenous human miRNAs: hsa-26A and hsa-222. We calculated the crossing point (Cp) value for the highest possible recovery of the C. elegans miRNAs by diluting them in RNase-free water to a final concentration equal to what was added to the samples. This value represents maximal efficiency of recovered miRNAs: cel-miR-39 Cp 17.30, cel-miR-54 Cp 18.74, and cel-miR-238 Cp 21.81. The average Cp values calculated for the recovery of miRNAs from plasma samples using each isolation kit are displayed in Figure 3 (isolations for each kit were performed in triplicate).

FIGURE 3.

miRNA recovery of C. elegans spike-ins and two endogenous human miRNAs after each isolation kit using Taq qRT-PCR. miRNA yield was measured by TaqMan qRT-PCR. Crossing points (Cp) were compared across (A) three different synthetic C. elegans miRNA cel-238, cel-54, and cel-39 (spike-ins) and (B) two endogenous human miRNA hsa-222 and hsa-26A. The lowest Cp values indicate the highest amount of miRNA recovered and the best kit performance. The black line drawn across the figure represents the lowest Cp values and the highest recovery.

Based on the TaqMan results for miRNA yield and the RiboGreen results for total RNA isolation, the four kits that recovered the most RNA were mirVana, mirVana PARIS, BiooPure, and Qiagen miRNeasy (Figs. 2, 3). Overall, we found that the mirVana PARIS kit was best suited to our experiments in three ways: (1) It gave the highest yield in total RNA and high yield in small RNA; (2) there is no size exclusion on the recovery of small RNAs; Qiagen, on the other hand, states that the miRNeasy kit does not recover small RNAs <18 nt and the manufacturers of the PARIS kit state that the kit can recover down to 10-mers; and (3) the volume of aqueous solution containing the RNA + ethanol that has to be passed through the column is significantly less using the mirVana PARIS kit than when using the mirVana kit. When starting with 1 mL of biofluid, the final volume that must be passed through the mirVana column, 700 µL at a time, is ∼24 mL. The mirVana protocol states that you can use a vacuum to draw the liquid through the column. In our hands, however, the recovery of miRNA was less when we used a vacuum. This could be due to loss of glass fibers from the column due to the suction. We found that even with centrifugation of the mirVana or PARIS columns, there was a slight loss of glass fibers—presumably with RNA still attached to them.

Maximization of RNA recovery by repeated extraction

Organic phase separation for nucleic acid purification requires that the upper aqueous phase containing the RNA be carefully removed from the interphase and the lower organic phase. In an effort to isolate the aqueous layer with the least amount of contamination from the interphase material, some residual RNA-containing aqueous solution is ultimately left behind. To maximize RNA recovery, we rehydrated the interphase and the organic phase left behind and re-extracted the phenol-chloroform solution with water (Fig. 4). We hoped this simple procedure would increase both total RNA and the small RNA yield. While this method is not sophisticated, none of the kits suggest adding liquid back to the remaining interphase and organic layers after the first aqueous phase has been removed and performing a second phenol-chloroform extraction. Several of the kits do suggest a second phenol-chloroform extraction of the first aqueous layer that is removed in order to further clean up the RNA and remove contaminants.

FIGURE 4.

Work flow for first and second extractions. (A) The RNA and denaturing solution are mixed with phenol-chloroform and centrifuged. (B) The aqueous phase is removed and placed in a fresh tube. (C) RNase-free water equal to the volume of the aqueous phase that was removed is added back to the residual interphase and organic layers. (D) Solution is mixed and centrifuged. (E) The aqueous layer is removed and placed into a clean tube as Extraction 2.

After addition of phenol-chloroform and centrifugation, the aqueous layer of the extraction was carefully removed, measured, and set aside (Extraction 1). Instead of discarding the residual interphase and organic layer from the extraction, we added another volume of RNAse-free water (equal to the volume removed in Extraction 1) to the organic layer and repeated the extraction. We mixed the sample once again in the manner specified by each kit, separated the phases again by centrifugation, and carefully removed the aqueous phase again (Extraction 2) (Fig. 4). We continued to process these two extractions in parallel according to the downstream instructions called for by the respective kit.

While we expected some increase in the recovered RNA, we were surprised to find that the total and small RNA yield was substantially improved by the second extraction with water. To illustrate the increase in RNA recovery using two separate phenol-chloroform extractions in our top kit choices, we acquired 800 µL of fresh-frozen plasma aliquots from two different subjects. We separated the plasma into 200-µL aliquots to be tested in each of the four kits and added a known quantity of spike-in C. elegans miRNAs. We also acquired 8 mL of CSF from two different subjects, separated them into 2-mL aliquots, added C. elegans miRNAs, and tested 2 mL in each of the four kits (Ambion mirVana, Ambion PARIS, BiooPure, and Qiagen miRNeasy).

We quantified the RNA yield in Extraction 1 and Extraction 2 separately by RiboGreen assay (Fig. 5A,B). Quantification of RNA in Extraction 2 from plasma indicates that there is still a large amount of RNA that can be recovered by repeating the extraction. In some cases, such as with the PARIS kit, we were able to more than double our total RNA yield by repeating the extraction. For example, plasma total RNA for subject 1 using the PARIS kit was 48.7 ng by combining 23.35 ng from Extraction 1 with 25.35 ng from Extraction 2. CSF total RNA for subject 1 was 15.8 ng by adding 9.2 ng from Extraction 1 to 6.6 ng from Extraction 2, using the PARIS kit.

FIGURE 5.

Repeated extraction of the organic phase results in higher RNA yield. (A) Fresh-frozen plasma from two subjects (subject 1 and subject 2) was used for RNA isolation using the top four kits: mirVana and mirVana PARIS (Ambion), miRNeasy (Qiagen), and BiooPure (BiooScientific). Total RNA was recovered and quantified from repeated extractions (black = Extraction 1 and gray = Extraction 2). PARIS kit yielded the highest amount of RNA from both subjects. The yield was more than doubled by the second extraction. (B) Fresh-frozen CSF samples from two subjects were used to compare the efficiency of the top four RNA isolation kits. The RNA recovered in Extraction 1 and Extraction 2 is displayed.

We really wanted to know if the isolation of small RNA was increased by this method. We compared the yield of small RNA recovered after isolation from plasma using qRT-PCR for the spiked-in C. elegans miRNAs as well as two endogenous human miRNAs in Extraction 1 (Fig. 6A) and Extraction 2 (Fig. 6B). Recovery of small RNA was markedly increased, and in some cases doubled, by the repeated extraction. We tested extractions on the same sample for a third and fourth time, but the recovery of RNA was very low (data not shown). We also tested the recovery of small RNA from CSF using the four best kits. After quantitation of the CSF with RiboGreen in triplicate, there was so little RNA remaining from the CSF samples that we were able to examine the recovery of only one cel miRNA (cel-238) in Extraction 1 (Fig. 6C) and Extraction 2 (Fig. 6D). Again, in the CSF samples, the recovered miRNAs were greatly increased by performing the second extraction.

FIGURE 6.

miRNA yields calculated from plasma and CSF with repeated extractions using qRT-PCR. (A) miRNA recovered in Extraction 1 was measured by TaqMan qRT-PCR in fresh-frozen plasma samples from two subjects (subject 1 and subject 2). Crossing point values (Cp) were compared across three different synthetic C. elegans miRNA cel-238, cel-54, and cel-39 (spike-ins) and two endogenous human miRNA hsa-222 and hsa-26A. The lowest Cp values indicate the highest amount of RNA present and best performance, highlighted by the black line. (B) Extraction 2 recovery of miRNA is displayed for each kit. (C) The Cp values for two different subject CSF samples for Extraction 1. There was only enough RNA remaining after RiboGreen for cel-238. (D) Cp values for cel-238 recovered from two CSF samples in Extraction 2.

miRNA from CSF sequenced

In order to determine whether we can use the small amounts of RNA that can be recovered from the volumes of CSF typically given to us by clinical collaborators, we isolated RNA from a range of starting volumes using a pool of CSF. We chose to use CSF because the total RNA and miRNA fraction has not yet been profiled by NGS. While the TruSeq small RNA kit recommends 1 µg of total RNA to start, 1 mL of CSF only yields ∼15–30 ng of total RNA (Fig. 5B).

We thawed ten 1 mL-samples in the presence of 2× denaturing solution from mirVana PARIS, thoroughly mixed the samples together in a pool, isolated the RNA, and aliquoted the CSF in 0.5, 0.75, 1.0, 1.25, and 1.5 mL volumes in duplicate. To maximize yield, we repeated the extraction of the organic layer as before and combined the RNA from the first and second extractions. Since the total and small RNA are almost immeasurable at these starting volumes of CSF, we isolated RNA from each volume and used the entire amount of isolated RNA for sequencing. We followed sample preparation according to the Illumina TruSeq small RNA kit with one alteration. In order to avoid extensive adaptor dimers forming in the library preparation, we reduced the reagents from the Illumina TruSeq small RNA kit by half. This increased our library preparation success rate and decreased the number of adaptor-only contaminating sequences.

The number of reads (raw counts) that mapped to known mature miRNAs in miRBase was more than 1 million for each sample tested and ranged from 1,003,030 to 4,849,671 mapped reads (Supplemental Table 2). We calculated Spearman rank correlations by comparing the 0.5- to 1.25-mL starting volumes with the 1.5-mL volume. The correlations were >0.95 for miRNAs with more than five counts (data shown in Supplemental Table 1). We repeated this experiment using RNA isolated with the BiooPure RNA isolation kit, which also performed very well, and attained nearly identical sequencing results for 0.5- to 1.5-mL starting volumes (data not shown). These data indicate that we can obtain reproducible results from as little as 0.5 mL of human CSF.

The top 50 most abundant miRNA from the pooled CSF samples are presented in Table 2. The complete list of known miRNA identified is provided in Supplemental Table 2. One of the advantages of sequencing the miRNA is the potential to assay all the miRNA present, including novel miRNA. Using miRDeep2 prediction software, we identified potential new miRNAs from the CSF samples. These sequences are presented in Supplemental Table 3.

TABLE 2.

Top 50 most abundant miRNAs identified in human CSF

Comparison of miRNA sequences identified in serum and CSF

Next, we wanted to compare the miRNA identified in serum to miRNA present in CSF from the same subjects. We examined the miRNA profiles for five different subjects from whom both CSF and serum were collected (Supplemental Table 4). We found that, while there are a large number of miRNA that are unique to either CSF or serum, most miRNAs are present in both the CSF and serum of each subject (Fig. 7A). To determine to what extent the miRNA expression profiles of CSF and serum differ between subjects and within a subject, we correlated the normalized read counts of miRNA in CSF and serum between subjects as well as within subjects. We found that the expression profile of miRNA in CSF is highly correlative across the five subjects (median Spearman correlation = 0.87208229). The profiles of miRNA in serum are also highly correlative across the five subjects (median Spearman correlation = 0.8497603). Not surprisingly, when we compared the miRNA profile of CSF with serum within single subjects, we found less correlation (median Spearman correlation across five subjects = 0.6047526). Figure 7B shows a scatter plot of miRNA from CSF vs. serum combined from five subjects. These data indicate that the miRNA profile of a single biofluid is more similar among different subjects than the miRNA profiles of serum and CSF within any one subject. See Supplemental Table 4 for the complete list of miRNA and the read counts in CSF and serum for the five subjects studied.

FIGURE 7.

Comparison of miRNA profiles from human CSF and plasma. (A) We counted the number of miRNAs that had more than two reads across five subjects and averaged the numbers. On average, serum and CSF from subjects had 532 and 486 miRNAs, respectively. Of these, 353 were present in both biofluids, while 179 and 133 were unique to serum and CSF, respectively. The small RNA expression profiles for CSF and serum across different subjects were highly correlative (Spearman correlation: CSF = 0.87208229 and serum = 0.8497603) but different between CSF and plasma within a subject (median Spearman correlations across five subjects = 0.6047526). (B) Scatter plot of CSF and serum miRNA from five subjects combined.

DISCUSSION

The purpose for identifying an efficient RNA isolation kit and introducing slight protocol modifications was to be able to maximize the recovered small RNA from low-RNA content samples to go forward into NGS small RNA library preparation. A significant challenge to using NGS for miRNA profiling from biofluids has been the low amount of RNA. Blood and CSF samples given to researchers by clinicians or obtained from tissue banks are often no more than 0.2 to 1 mL in volume. Biofluid samples are even more limited when derived from animal models of disease, such as transgenic mice. The top four isolation methods, based on small RNA yield, were (1) mirVana PARIS (Ambion), (2) mirVana (Ambion), (3) miRNeasy (Qiagen), and (4) BiooPure. The kit chosen to purify RNA for sequencing was based upon several factors: highest small RNA yield, no recovery restriction on small RNA size, ease of use, and practicality of the protocol. The main disadvantage of using Qiagen miRNeasy for miRNA discovery is that it does not allow for the purification of miRNA species smaller than 18 nt. An analysis of reported human miRNA indicates that there are a total of ∼68 human miRNAs that are smaller than 18 nt, making up ∼3% of all mature human miRNA identified to date (miRBase; as of October 2012). Our sequencing data collected from plasma samples isolated using the mirVana PARIS kit detected nine of the 68 known miRNAs that are <18 nt in length. While BiooPure yielded a high amount of small RNA, it recovered the least amount of total RNA in our hands. This is a consideration if one plans to do downstream experiments to identify the full RNA profile (whole transcriptome and small RNA).

We discovered that by repeating the phenol-chloroform extraction with RNase-free water, we could increase our detection of miRNA by almost double. It seems reasonable that we might increase our small RNA yield even more by doing a third or fourth extraction. When we tried this, however, we found that the additional extractions resulted in only a modest increase in yield and did not warrant the additional steps and required processing time (data not shown). We found that the combination of the first and second extractions were sufficient for acquiring enough small RNA for downstream sequencing assays.

It is possible to use these sequencing protocols with small but clinically relevant biofluid sample sizes. Using the RNA isolation protocol described here, we were successfully able to use CSF in downstream sequencing assays. It is possible to sequence miRNA from as little as 0.5 mL of CSF using the methods outlined in the current study. To our knowledge, this is the first time the small RNA fraction of CSF has been sequenced. We surveyed our sequencing results from five subjects’ CSF alongside the miRNA counts from normal human brain tissue sequenced by Hua et al. (2012) and Skalsky and Cullen (2011). There are many miRNAs that reflect expression levels similar to those observed in brain tissue, but there are also some miRNAs that are more abundant in either the CSF or the brain.

We next applied our methods to sequence small RNA from CSF and serum derived from the same patients. Comparison of small RNA present in serum and CSF from five different subjects revealed that profiles of small RNA are more similar from different subjects for either CSF or serum and are less similar to one another, even within the same subject. This is not surprising given that miRNAs in CSF are likely derived from neural cells whereas serum miRNAs are likely collected from all tissues in the body. A list of the miRNA identified in CSF alongside a list of miRNAs identified in normal brain tissue can be found in Supplemental Table 5.

For the first time, we present an approach to sequence extracellular miRNA from human CSF. The methods described here can be used to identify extracellular small RNA in small, clinically obtainable volumes of biofluids and plasma from patient samples and even transgenic mouse models of disease. These methods can be applied to identify novel biomarkers or mechanisms of pathology, or to monitor drug efficacy for a variety of diseases including cancer, neurological diseases, and traumatic brain and spinal cord injury. The results of the sequencing experiments demonstrate that sequencing small RNAs from small starting volumes can provide us with robust, reproducible data.

MATERIALS AND METHODS

Clinical samples

All clinical samples included in the current study were obtained from subjects who had given informed consent, and studies were performed under the guidelines of Institutional Review Board (IRB)-approved protocols at St. Joseph’s Hospital and the Translational Genomics Research Institute (TGen).

Patient plasma, serum, and CSF samples were obtained from Dr. Roberto Bomprezzi (Barrow Neurological Institute at St. Joseph’s Hospital and Medical Center). Blood draws were performed from the antecubital veins directly into Vacutainer potassium EDTA tubes (BD Vacutainer) as a routine part of the neurological workup. Within 2 h of the blood draw, samples were processed for plasma or serum isolation. CSF was obtained by lumbar puncture, and samples were spun down to pellet cells, and the supernatant removed and flash-frozen in liquid nitrogen for subsequent RNA isolation.

As a preface to this study, to ensure systematic comparison between different RNA purification methods, the plasma samples were thawed on ice, pooled, separated into 200-µL aliquots, flash-frozen in liquid nitrogen, and stored at −80°C until the initial denaturant for the respective kit was added. Each RNA extraction method was tested in triplicate for each kit and/or variation using these 200-µL plasma samples as starting material.

RNA extractions

Ten commercially available kits were compared in the current study for the purification of biofluids: BiooPure (BiooScientific), mirVana (Ambion), mirVana PARIS (Ambion), TRI Reagent RT (MRC), TRI Reagent RT-Blood (MRC), TRI Reagent RT-Liquid Samples (MRC), RNAzol (MRC), miRNeasy (Qiagen), and PureLink microRNA (Invitrogen). One of the kits, mirPremier (Sigma), was not found suitable for purifying biofluids as the initial lysate was unable to pass through the column.

For all extractions, we first followed the manufacturer-provided protocol with minor modifications. RNA purifications were performed on virtually identical samples (see clinical samples above) in triplicate for each kit and were rehydrated as called for by the commercially available protocol.

All purifications were performed at room temperature unless a protocol specified a different temperature. For all nine kits, we followed the protocol for total RNA isolation that included recovery of small RNA. In the case of the MRC kits, the protocol allowed for a range of temperatures and centrifugation speeds; the upper and lower limits of those parameters were tested. RNA purifications were performed and quantified side-by-side in triplicate for each kit.

Where applicable, reserved for procedures involving phenol-chloroform phase separation, we rehydrated the interphase and organic layer and subsequently re-extracted to maximize recovery of nucleic acids. This procedure was utilized for the following RNA purification methods that relied upon phase separation: BiooPure, mirVana, mirVana PARIS, TRI Reagent RT, TRI Reagent RT-Blood, TRI Reagent RT-Liquid Samples, and miRNeasy. The phenol was extracted a second time with an equal volume of nuclease-free water to obtain residual aqueous material left at the interface. The two extractions were kept separate throughout and assayed independently for total RNA and miRNA content but were combined for downstream sequencing experiments. After column washes, the RNA was rehydrated on the column and centrifugation allowed the RNA eluate to be collected. The protocol for the MRC kits allowed for incubation temperatures ranging from 4°C to 25°C and centrifugation speeds between 4000g and 12,000g; the upper and lower limits of those parameters were also tested. All RNA was precipitated and recovered by either centrifugation (pellet) or elution (column) in molecular biology grade, nuclease-free water (Life Technologies) in the volume and temperature recommended by the kit.

Determination of RNA yield

Quantification of total RNA yield was determined by Quant-iT RiboGreen RNA reagent (Invitrogen) utilizing the low-range assay in a 200-µL total volume in the 96-well format (Costar). This protocol allows for quantification of 1–50 pg/µL, the linearity of which is maintained in the presence of common post-purification contaminants such as salts, ethanol, chloroform, detergents, proteins, and agarose (Jones et al. 1998). Individual samples were assayed in triplicate, and the means were calculated. The three replicates from the same treatment were averaged. We used the low-range assay (1–50 pg/µL) in a 200-µL total volume of working reagent in a 96-well format and read on a plate reader (BioteK Synergy HT).

In order to simplify the quantification of samples processed with different kits and having varying final volumes, we removed half of the eluent from each sample and adjusted the volume to a final volume of 60 µL for every sample. For example, if kit A recommends to elute in 50 µL and kit B recommends elution in 100 µL, 25 µL and 50 µL, respectively, were removed, and each volume was adjusted to a final volume of 60 µL. The concentration in that 60 µL represents half of the recovered RNA and made downstream assays (i.e., loading 1 µL of each sample into the RiboGreen assay) much easier to process and interpret.

Real-time RT-PCR

Input RNA was reverse transcribed using a small-scale reaction with the TaqMan miRNA Reverse Transcription Kit using miRNA-specific primers, and real-time RT-PCR (qPCR) was performed using TaqMan miRNA-specific stem–loop primers as described previously (Mitchell et al. 2008).

In order for the recovery of RNA across all samples isolated with different kits to be directly comparable, irrespective of the volume in which the RNA was rehydrated, the RNA input into the reverse transcription (RT) was 50% of the total elution volume scaled up to a set volume of 60 µL across all samples. 1.67 µL was added to the reverse transcription mix. The cycle number at which the fluorescence passes a fixed threshold (Cp) is reported. Probe sequences were (Mitchell et al. 2008): cel-miR-39: UCACCGGGUGUAAAUCAGCUUG, cel-miR-54: UACCCGUAAUCUUCAUAAUCCGAG, cel-miR-238: UUUGUACUCCGAUGCCAUUCAGA, hsa-miR- 26A: UUCAAGUAAUCCAGGAUAGGCU, hsa-miR-222: CUCAGUAGCCAGUGUAGAUCCU.

Synthetically generated C. elegans miRNAs, which lack sequence homology to the current human miRNA database (miRBase V. 16), were utilized in the current study to correlate absolute cycle threshold data generated by qRT-PCR to the number of molecules of that species present, as previously described (Mitchell et al. 2008). Briefly, the synthetic oligonucleotides, generated with 5′ phosphate and 3′ hydroxyl groups to match the molecular structure of RISC complex-processed mature miRNAs (Mitchell et al. 2008), have sequence homology to C. elegans miRNAs cel-miR-39, cel-miR-54, and cel-miR-238 (miRBase 16; ordered as custom RNA oligonucleotides from IDT). A mix of these miRNAs at 25 fmol each was prepared and flash-frozen in 10-µL aliquots. A volume of 1.5 µL of the mix was added to each sample after RNase inactivation. For determining the maximal C. elegans recovery, we diluted the 1.5-µL spike-in mix equivalent to the final amount tested in the samples. Because half of the isolated RNA content of the samples is diluted in 60 µL, we put half of the spike-in mix in 60 µL (0.75 µL in 60 µL of RNase-free water). In order to make this even more similar to the samples, half was removed and brought up to 60 µL. 1.67 µL was then used in the reverse transcription reaction (5-µL reaction). 28.9 µL of water was added to the cDNA, and 2.25 µL was used in the Taq reaction (as in Mitchell et al. 2008). We used Cp values of up to 35 accurately to score RNA yield as previously reported (L Chen et al. 2009; Y Chen et al. 2009). CSF samples, because they have so little RNA, were processed for RT and qPCR slightly differently. Half of the eluted volume was put in 60 µL, as in the plasma samples above. The 60-µL sample was then dried down to 6 µL, 1.67 µL went into the RT reaction, and we added 28.9 µL water. We took 2.25 µL of the RT reaction forward into Taq. When we calculate the return of spike-ins for this experiment using just spike-in mix and water, the Cp values are cel-miR-39 (Cp 15.33), cel-miR-54 (Cp 16.65), and cel-miR-238 (Cp 17.79).

Small RNA sequencing

Total RNA was purified from a pool of CSF created from six subject samples using the mirVana PARIS kit and the modified protocol as described. The pooled sample was then separated into aliquots of 500, 750, 1000, 1250, and 1500 µL. After elution of RNA in 100 µL of nuclease-free water, the total RNA was precipitated as described by mixing eluate with ammonium acetate to a final concentration of 2 M, adding four volumes of ethanol, chilling overnight at −20°C, then centrifuging at 16,000g for 30 min, followed by two 80% ethanol washes. RNA was resuspended in 6 µL of water, the entire volume of which was introduced into half of the TruSeq Small RNA Sample reagents, followed by 15 cycles of PCR to amplify the library.

We clustered a single read v3 flow cell and performed small RNA deep sequencing on the HiSeq 2000 using the RNA isolated from the 0.5- to 1.5-mL aliquots of CSF.

Sequencing data analysis

Raw fastq sequences were generated and de-multiplexed using the Illumina CASAVA v1.8 pipeline. The FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and FASTX toolkit (http://hannonlab.cshl.edu/fastx_toolkit) were used for Quality Check [ensured that fastq reads are in entirely normal (green tick: ≥Q28) range in the QC report] and to preprocess the reads prior to mapping, respectively. The fastx clipper tool was employed to remove the Illumina 3 prime adaptor (TGGAATTCTCGGGTGCCAAGG) sequences. Post-clipped reads were then run through mirDeep2 analysis Pipeline (Friedlander et al. 2012). Sequences were aligned using mapper.pl to Human genome (hg18) and miRBase_v16 and further processed using miRDeep2.pl scripts. The .csv files for miRNA expression from the mirDeep2 outputs were used for the analysis. Reads per million were calculated as follows: Number of sequenced reads/total reads × 1,000,000.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.