Important considerations for microRNA extraction methods from whole blood and peripheral blood mononuclear cells

MicroRNAs are non-coding RNAs that negatively regulate mRNA expression and play significant roles in both health and disease. Differential microRNA expression has been used to aid diagnosis and discriminate disease stages. The accuracy and reliability of microRNA expression measurement is of utmost importance. For diagnostic investigations, microRNA expression in human peripheral blood is commonly detected using total RNA extracted using different methods. To date, no convincing data have been available showing whether microRNA expression levels are comparable when total RNA has been extracted from whole blood or peripheral blood mononuclear cells (PBMCs). The present study examined miR-146a-5p and miR-155-5p expression using total RNA extracted in parallel from whole blood and PBMCs of 14 healthy volunteers. MicroRNA expression levels were significantly different between whole blood and PBMCs. No significant difference was observed in microRNA expression between fresh and cryopreserved PBMCs (p=0.125 for both). Further observations revealed that gender differences did not influence miR-146a-5p or miR-155-5p expression regardless of using whole blood (p = 0.797 and 1.00 respectively) or PBMC (p = 0.190 and 0.898 respectively). The design of this and several are not justified by the PCR there may


Introduction
MicroRNAs (miRNAs) are cell-and therefore tissue-specific, and their expression levels impact protein translation (Sood et al., 2006). Nearly, 2000 microRNA (miRNA) sequences have been identified in humans (Kozomara & Griffiths-Jones, 2014). Numerous studies have reported specific miRNA expression levels in peripheral blood (PB) as markers of disease (Mookherjee & El-Gabalawy, 2013;Patnaik et al., 2012). Although miRNA expression levels could be inevitably influenced by the way that total RNA is extracted, often studies reporting differential miRNA expression levels fail to emphasise the impact of RNA extraction methods. This could at least partially lead to significant controversies and inconsistencies in the literature related to miRNA research.
PAXgene Blood RNA System (PAXgene Blood RNA Tube and Paxgene Blood miRNA Kit [PAXM]) has been the gold standard for PB collection as the stabilising reagent present in the tube prevents RNA degradation and inhibits changes in gene expression due to the collection procedure (Guenther & McCluskey, 2008). However, it should be emphasised that the miRNA expression detected in whole blood is the overall outcome from total hematocytes rather than the lymphocyte fraction only. Recent studies have shown that erythrocytes also contain a high proportion of miRNAs (Hamilton, 2010). Bayatti et al. have shown that the presence of globin (globular proteins such as haemoglobin) in PB can impact RNA expression and that globin depletion can decrease total RNA quality and yield in particular when extraction is performed using the PAXgene Blood RNA kit (Bayatti et al., 2014). However, it is necessary to evaluate the miRNA expression levels in an independent cohort, as miRNAs are in general highly stable, and therefore their expression might not be affected by globin treatment. Likewise, Mookherjee et al. have only shown that miRNA expression levels correlate between whole blood and PBMCs using PAXgene Blood RNA System and mirVana miRNA kit (MM), respectively (Mookherjee & El-Gabalawy, 2013). They have not shown that the expression levels are comparable or that they agree. This has raised some confusion as the PAXgene Blood RNA system extracts from whole blood while MM from PBMCs. In addition, researchers have only tested for the relationship between one method and another rather than for agreements as recommended by Bland et al., (Bland & Altman, 1986). Performing a comprehensive statistical analysis is required when two methods are compared (Burd, 2010).
In this investigation we have measured the expression of miR-146a-5p and miR-155-5p which have extensively been investigated in whole blood and PBMCs due to their critical functions in the innate and adaptive immune system (Curtale et al., 2010;Schulte et al., 2013). We sought to compare miRNA expression levels in PB collected from 14 healthy volunteers using both extraction methods (PAXM and mirVana PARIS [MP]). Overall, we sought to address (i) whether miRNA expression detected in whole blood is comparable to that of isolated PBMCs and (ii) whether the degree of haemolysis impacts on the detected levels of miRNA expression in whole blood and PBMCs. We also investigated whether there is a difference in miR-146a-5p and miR-155-5p expression between males and females as well as fresh and cryopreserved PBMCs. The experimental design for this investigation is depicted in Figure 1.

Volunteer cohort
Whole blood was collected from 14 healthy volunteers following approval from the Newcastle and North Tyneside 2 Research Ethics Committee (STEMDIAGNOSTICS: REC-07/H0906/131) and informed consent was obtained from every volunteer for both blood collection and miRNA testing. Samples from 14 healthy volunteers (5 males and 9 females) were used to quantify miRNA expression in whole blood and PBMC.
Sample collection and storage PB (2.5 ml) was collected in PAXgene Blood RNA Tubes (PreAnalytiX GmbH, Switzerland [Catalog No: 762165]) containing an RNA stabilising agent that lysed the blood cells and stabilised the intracellular RNA. The tubes were stored at -20°C until processed. PB was also collected in sodium (Na)-heparin (Sigma, UK) containing tubes for peripheral blood mononuclear cell isolation using graduated centrifugation over Lymphoprep™ (STEMCELL Technologies, Manchester, UK

Data analysis
The qPCR results were analysed using SDS v2.4 software and normalised using SNORD48 as the reference control which was selected by testing a panel of four controls for stable expression within the whole blood and PBMCs. The comparative ∆∆C q method was used to calculate fold-changes (ΔC q = C q microRNA of interest -C q reference control , Relative Quantification [RQ] = 2 -ΔΔCq and LOG transformed = LOG 2 RQ). Fold-change was logarithm-transformed as qPCR data are non-linear (exponential), and is transformed to decrease the heterogeneity of variance (McDonald, 2009) and also to identify the outliers present in the data (Rieu & Powers, 2009). Standard curves for three samples from both PAXM and MP were generated and a 95% confidence interval slope of the line was used to calculate the PCR efficiency (E) using the standard formula; E=10 -1/Slope and % efficiency = (E-1) × 100. Mean efficiency was then calculated. Results were analysed and plotted using GraphPad PRISM v5.0 software (GraphPad Software, Inc, USA). Mann-Whitney U t-test was used to assess difference between two groups and Kruskal-Wallis one-way analysis of variance (ANOVA) for multiple groups. Spearman's test was used to determine correlation. Bland-Altman was performed to test whether two methods agreed and if one could be interchanged with another. Significance was set at p<0.05.

Quality of RNA extracts
Total RNA was extracted using three different extraction methods; PAXM (PAXgene Blood miRNA), MP (mirVana PARIS) and MM (miRVana miRNA) from whole blood and PBMCs. RNA purity was assessed by detecting the absorbance ratios at 260 nm/280 nm and at 260 nm/230 nm. The ratio (260 nm/280 nm) for whole blood was 1.95 -2.35 and for isolated PBMCs was 2.00 -2.27 (both MP and MM respectively). The ratios were ≥ 1.8 -2.0; thus extraction was free from protein contamination which is usually absorbed at 280 nm. Absorbance ratios detected at 260 nm/230 nm showed ratios below the accepted contamination-free range of 1.5 -2.0 (PBMCs: 0.18 -1.83 and whole blood: 0.15 -1.49). Therefore, the samples may have been affected by contaminants absorbed at 230 nm such as guanidine isothiocyanate present in all the three extraction kits. For RNA purity, the peak of each total RNA plot was also analysed as it could indicate contamination by phenol and/guanidine isothiocyanate. PBMCs had plot peaks at 260 nm thus confirming contaminant-free samples. However, peaks at 260 nm were absent for total RNA extracted using the PAXM ( Figure 2). This further suggests that guanidine salts may have been the cause of contamination in whole blood samples extracted using PAXM. Quantitative PCR quality controls NEC and NTC controls were used to ensure that the RT and qPCR reactions were contaminant-free. Each control displayed no amplification (Cq > 36). This was particularly important for total RNA extracted from PBMCs, as they were non-DNase treated. Thus, any amplification in the controls may have suggested either nonspecific binding of primers or presence of contamination such as genomic DNA. In addition we calculated the average efficiency of our real-time qPCR reactions (E=97.8%) which confirmed absence of reaction inhibitors. Determining qPCR efficiency is important as inhibitory compounds can affect miRNA expression and result in false positives.

Reference control for microRNA expression in peripheral blood
We determined the most appropriate reference gene for this investigation by testing a panel of four controls that have been known for stable expression ( Figure 3A) and MP ( Figure 3B) method. Therefore, SNORD48 was used as the reference control to normalise miR-146a-5p and miR-155-5p expression in each sample.
Whole blood extracted using PAXM exhibits higher degree of haemolysis Erythrocyte haemolysis has been reported to alter miRNA measurements in whole blood, plasma, serum and tissues (McDonald et al., 2011;Pritchard et al., 2012). The total RNA extracted using the three different methods was examined for degree of haemolysis by quantifying miR-451-5p and miR-23a-3p (Dataset b). Normalised ∆Cq values (miR-23a-3p -miR-451-5p) greater than seven were considered as an indicator of haemolysis. Our results showed that there was a significantly high degree of haemolysis in the total RNA extracted using the PAXM method with the ∆Cq values in the range of 9 -11. Haemolysis was low in total RNA extracted using either of MP or MM, ∆Cq>3 (Figure 4).

MicroRNA expression in whole blood and PBMCs
With inclusion of stringent quality controls, we assessed miR-146a-5p and miR-155-5p expression in total RNA extracted in parallel from whole blood using PAXM (n=14) and PBMCs using MP (n=14) (Dataset c). Our results showed that there was no correlation ( Figure 5, miR-146a-5p: r=-0.352, p=0.217 and miR-155-5p: r=0.380, p=0.180) between PAXM and MP in the expression of both miR-146a-5p and miR-155-5p. In a PCR reaction, it is assumed that the target expression doubles at every reaction cycle. Bland-Altman analysis also showed that the two methods did not agree as the bias was greater than 1 which equated to more than one qPCR cycle difference between the two methods ( Figure 5A and 5B). Mookherjee et al. had used MM to extract total RNA from PBMCs, then correlated miRNA expression between the PAXM and MM method (Mookherjee & El-Gabalawy, 2013). To eliminate the possibility that using MP was the reason for the non-correlation and disagreement, we tested the three different extraction methods (PAXM, MP and MM) for both miRNAs in a randomly selected cohort of five healthy volunteers ( Figure 6) (Dataset d). The results demonstrated that PAXM and MP as well as PAXM and MM do not correlate nor agree. However, MP and MM methods agree with one another and can be interchanged as the bias between the two methods for both miR-146a-5p and miR-155-5p was only 0.769 (SD=0.307) and 0.892 (SD=0.802), respectively. Interestingly, normalised miRNA expression was significantly different only between PAXM and MM methods (miR-146a-5p and miR-155-5p: p<0.01). There was a higher miRNA expression in PBMCs than in whole blood for both miRNAs (Figure 7).

Impact of cryopreservation on microRNA expression
Many samples from patients participating in clinical studies are cryopreserved and stored as PBMCs for later use. Therefore, it is important to know the specific effects of cryopreservation on miRNA expression in PBMCs. We evaluated the impact of cryopreservation by extracting total RNA using MP kit from paired fresh and cryopreserved PBMCs (n=5) and performing RT-qPCR respectively (Dataset e). There was no correlation in miRNA expression between fresh and cryopreserved PBMCs, but the bias for both miR-146a-5p and miR-155-5p was less than 1 ( Figure 8A and B). In addition, there was no statistically significant difference between fresh and cryopreserved PBMCs for both miR-146a-5p and miR-155-5p expression (p=0.125) ( Figure 8C and D, respectively). Thus, expressions in fresh and cryopreserved samples do agree with one another but more samples would need to be tested as the standard deviation of bias was high for both miRNAs (miR-146a-5p: SD=2.284 and miR-155-5p: SD=1.342), which indicates a large sample variability.

MicroRNA expression in males and females
In order to determine whether miR-146a-5p and miR-155-5p expression differed between males (n=5) and females (n=9), further analysis was performed (Figure 9) (Dataset c). No significant difference was observed between males and females for miR

Discussion
Globally, miRNA expression levels are used to classify diseases and also to distinguish the diseased from the healthy population. However, non-uniform technical procedures have led to the many controversies and inconsistencies in miRNA expression results observed in various studies. In addition, failure to test the degree of haemolysis in studies using PB has led to only few reproducible experimental results. Thus, in an attempt to elucidate the importance of protocol standardization and show the impact of technical variations in miRNA expression investigations, we examined several parameters when measuring the expression of immune related miRNAs (miR-146a-5p and miR-155-5p).
First, we performed RT-qPCR to test whether the expression of miR-146a-5p and miR-155-5p is the same in whole blood and PBMC samples using three different commercially available extraction methods (PAXM, MP and MM). Our results showed that there was no agreement between PAXM and both MP and MM for miR-146a-5p and miR-155-5p expression. PBMCs constitute only a fraction of the cells present in PB and therefore lack granulocytes, platelets and erythrocytes (Min et al., 2010). Due to the unique miRNA expression pattern in each cell-type the relative proportions of cells in blood may have an effect on the overall miRNA expression profile and the expression of their protein targets (Min et al., 2010). Some studies have shown that mature miRNA expression signature in erythrocytes is similar to that in whole blood while different when compared to PBMCs (Chen et al., 2008). We have clearly demonstrated a high degree of haemolysis in whole blood samples processed using PAXM by measuring the difference between miR-451-5p which is a marker of erythrocytes (Rasmussen et al., 2010) and miR-23a-3p that has been shown to be unaffected by haemolysis (Blondal et al., 2013). We have shown that there is a lower miR-146a-5p and miR-155-5p expression in whole blood compared to PBMCs. Several studies have shown that total RNA yield from whole blood decreases after the use of DNase step in the PAXM protocol (Asare et al., 2008;Bayatti et al., 2014;Debey et al., 2004). However, if PAXM total RNA is not DNase-treated, there may be a possibility of DNA contamination. Contrary to our results, Mookherjee et al. found a linear correlation between miR146a-5p and miR-155-5p expression in whole blood and isolated PBMCs collected from a healthy population (Mookherjee & El-Gabalawy, 2013). In their work, they did not measure the degree of haemolysis in the samples, which may partly explain the discrepancy between the two studies. Furthermore, our study compared both the strength (correlation) and level of agreement between the two methods whilst Mookherjee et al. examined only the correlation (Bland & Altman, 1986). This highlights the importance of performing the correct statistics when two methods are compared with regards to their equivalence and interchangeability (Burd, 2010).
Our results suggest that isolation of total RNA from PBMCs is more reliable with higher RNA purity than isolation from whole blood via PAXgene Blood RNA Tubes. However, in clinical practice, it is easier to collect PB in PAXgene Blood RNA tubes as they have a shelf-life of two to five years without any RNA degradation.
of both miRNAs. This disagreement could be due to technical differences between the studies as Gaarz et al., (2010) who used TRIZOL reagent for extraction, while we used the MP kit. Bland-Altman plots showed that both fresh and cryopreserved methods agreed with each other and cryopreserved samples had lesser variability, which could be explained by the fact that any erythrocyte carried-over from the gradient centrifugation step is removed on freeze-thawing.
In conclusion, our study shows differences in miR-146a-5p and miR-155-5p expression in isolated PBMCs and whole blood. We suggest that PBMCs are not the ideal source to study and correlate miRNA protein targets where the miRNA expression had been measured in whole blood as the miRNA expression pattern in whole blood is not comparable to that in PBMCs. We also highlight the importance of having a stringent set of technical controls and performing the correct statistics to increase the reliability and reproducibility of miRNA expression studies.
Immediate stabilisation is vital as storage of blood cells induce changes in the miRNA composition (Gaarz et al., 2010). Thus, extraction methods from whole blood must be optimised to either eliminate erythrocyte contamination or consider the expression as cumulative and design downstream experiments for miRNA protein target studies accordingly.
We evaluated whether miR-146a-5p and miR-155-5p expression was different between males and females as previous studies have shown that miRNA expression can be gender-dependent (Ji et al., 2009). We found no gender-dependent differences in the expression of miR-146a-5p and miR-155-5p.
Finally, we assessed the impact of cryopreservation on the expression of both miRNAs. Gaarz et al. had shown that miRNA expression profiles of fresh and cryopreserved isolated PBMCs were not comparable (Gaarz et al., 2010). However, in our investigation we have shown that there is no significant difference in the expression

Kenneth Whitaker Witwer
Molecular and Comparative Pathobiology, Johns Hopkins University, Baltimore, MD, USA The design of this study is flawed, and several conclusions are not justified by the results. Since the PCR technique seems sound (and the authors are commended for including information such as PCR efficiency), there may be several subcomponents of the study that could be separated out and published as a smaller paper (e.g., the results of the gender and freezing experiments). I would encourage the authors to remove immediately the baseless claims about putative problems with the PAXgene reagents or protocol. A major (and wrong) conclusion of the paper is that the PAXgene method results in putatively excessive hemolysis, or, as the Figure 4 title incorrectly states, "Erythrocyte contamination". The putative difference from the other methods is described as "significantly" high, and although no statistical analysis is reported, one could hardly quibble with a ten-order-of-magnitude difference. However, this difference has nothing to do with hemolysis as a processing variable and in fact reflects well, not poorly, on the PAXgene method. The PAXgene tube is designed to lyse blood cells and stabilize RNA. This study is thus not an examination of methods, hemolysis or RBC contamination! It is simply a confirmation that lysed RBCs contain RBC RNA, whereas PBMCs do not contain RBC RNA. 1.
There appears to be a fundamental confusion about what constitutes hemolysis vs. red blood cell contamination. Where is the evidence that "haemolysis has been reported to alter miRNA measurements in whole blood, plasma, serum, and tissues"? The two cited references examine the effects of hemolysis in plasma and serum, where lysis of cells can indeed have a great effect. It is clear that the very low abundance extracellular RNA fraction of blood is influenced by an influx of cellular RNA. It is unclear how degree of hemolysis could affect detection of very high abundance cellular miRNAs in whole blood (where erythrocytes dominate anyway) or in cells like PBMC that are purified away from other components.

2.
The title does not describe the study. The study is presented incorrectly as a comparison of RNA extraction methods and of RNA expression in whole blood and PBMCs. Unfortunately, too many variables are changed at once in the study to allow any conclusions to be drawn.

3.
The title of the manuscript has also been changed to clarify the message of the "I would encourage the authors to remove immediately the baseless claims about putative problems with the PAXgene reagents or protocol. A major (and wrong) conclusion of the paper is that the PAXgene method results in putatively excessive hemolysis, or, as the Figure 4 title incorrectly states, "Erythrocyte contamination". The putative difference from the other methods is described as "significantly" high, and although no statistical analysis is reported, one could hardly quibble with a ten-order-of-magnitude difference. This article has no intention at all to claim any problems with the PAXgene protocol or to examine haemolysis or RBC contamination. The key points we intended to raise in this manuscript are 1) the use of total RNA extracted from whole blood or isolated PBMC could lead to differential results of miRNA quantification as miRNAs expressed in erythrocytes are present in total RNAs extracted from whole blood but not isolated PBMCs; 2) This inevitable fact is partly accountable for many conflicting reports in the existing literatures; 3) This is an important factor that researchers should consider when conducting miRNA research so as to extract total RNAs from most appropriate starting material (whole blood or isolated PBMC). Particularly, total RNAs extracted from the same starting material should be used for analysis of specific miRNAs and their downstream protein targets. These points have been clarified in the revised manuscript.
Furthermore, Figure 4 was intended to demonstrate the fact that erythrocyte related miRNA is present and detectable when using total RNAs extracted from whole blood (not to show red blood cell contamination). The title of Figure 4 and the related contents in the text have been revised to clarify this point that it is not a contamination per se but a factor that needs to be considered when analysing, reporting results and investigating miRNA protein targets. We believe this is a confusion caused by the inappropriate use of terminology rather than a flaw in study design. The revised manuscript has clarified these points. The discussion point of the manuscript is neither about haemolysis nor red blood cell 2.
contamination but simply about the fact that the presence of red blood cells in whole blood method may give rise to different miRNA quantification results compared to those where isolated PBMCs were used to extract total RNAs.
"The title does not describe the study. The study is presented incorrectly as a comparison of RNA extraction methods and of RNA expression in whole blood and PBMCs. Unfortunately, too many variables are changed at once in the study to allow any conclusions to be drawn. For example (see Figure 1): substrate type (whole blood vs PBMC); initial treatment (PAXgene tube for whole blood, heparin for PBMC); cell isolation (none for whole blood, centrifugation over a gradient for PBMC); and RNA isolation method (PAXgene for whole blood, two nearly identical Ambion kits for PBMC). The only variable that can be isolated is the use of the mirVana PARIS vs miRNA kit." This study was not a comparison of RNA extraction methods. The aim of this study was to find out whether miR-146a-5p and miR-155-5p expression detected in PBMCs mirrored that quantified in total RNA extracted from whole blood. For this reason two main extraction methods had to be used one for extraction from whole blood and the other from PBMCs. Our results clarified the misconceptions reported in the literature suggesting that the levels of miRNAs detected in whole blood and PBMCs correlate with each other. In fact miR-146a-5p and miR-155-5p are both miRNAs that are detectable in acellular sources such as serum and plasma which again reiterates that the miRNA expression quantified in whole blood would be varied from that of PBMCs.

3.
"Heparin, as a PCR inhibitor, may not be the best anticoagulant for use in qPCR-based studies. No special steps are described for removal of heparin." The use of Heparin in this study was driven by the logic that Heparin is one of the most commonly used anti-coagulants for collecting PBMCs from clinical samples. We reported over 90% efficiency of the PCR reactions demonstrating the fact that the data presented in this study had not been compromised by the use of Heparin. The quality of RNA was also high as the 260/280 ratios were above 1.8 as stated in the manuscript.

4.
Minor points "There are no error bars to indicate biological or technical variability in Figure 3 or 4. " Error bars have been added to illustrate technical variability in Figure 3 and Figure 4, respectively. As the variability is very small the error bars are hardly visible, hence inserted the sentence "The error bars represent technical variability" in the figure legends. 1.
"The workflow of Figure 1 does not indicate that the right-hand-side pathway involved only PBMC, not whole blood." The workflow of Figure 1 has been modified with an added step to indicate that the right-hand-side pathway involved only PBMCs and not whole blood.

2.
"The PAXgene Blood RNA Tube and the PAXgene Blood miRNA Kit are presented as the 3.
"gold standard" in the field, yet the citation is a publication by the company, and it does not mention the miRNA kit. This should be replaced by a more objective assessment." The company reference has been substituted with an original research article by Viprey et al (2012) which highlights the reliability and suitability of the whole blood collection method using the PAXgene Blood RNA tubes. It is worth noting that the focus of this report is to evaluate the effect of using total RNAs collected via the PAXgene Blood RNA protocol on miRNA quantification rather than on mRNA expressions.
"More clarity is needed in the use of terms. For example, there may or may not be RNA "expression" differences between genders or cell types. In contrast, differences in RNA quantitation observed upon freezing, or addition of PCR inhibitors, or after purification of the same sample with different kits, are not due to "expression" differences. The underlying expression in the cell remains the same; the ability to characterize it changes." Great attention has been given to using terms with clarity and accuracy in the revised manuscript.

4.
"Citations of the literature are somewhat thin." In contrast to mRNA research where large pools of literatures have investigated numerous factors that could impact on the outcome of mRNA quantification, the miRNA research is still at its infancy and very few studies have investigated the impact of various total RNA extraction methods on the outcome of miRNA quantification although many articles have been published reporting associations of differential miRNA expression with diagnostic or prognostic status of diseases. This manuscript is particularly aimed to inform the miRNA community that different sources and methods used for extracting total RNAs could have significant impact on the outcome of miRNA quantification.

5.
Competing Interests: No competing interests to declare.
The benefits of publishing with F1000Research: Your article is published within days, with no editorial bias • You can publish traditional articles, null/negative results, case reports, data notes and more • The peer review process is transparent and collaborative • Your article is indexed in PubMed after passing peer review • Dedicated customer support at every stage • For pre-submission enquiries, contact research@f1000.com