The affiliation of authors Rebecca Sanders, Carole Foy and Jim Huggett with the commercial company LGC does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials.
Conceived and designed the experiments: RS DJM CAF JFH. Performed the experiments: RS. Analyzed the data: RS JFH. Wrote the paper: RS JFH.
Gene expression measurements detailing mRNA quantities are widely employed in molecular biology and are increasingly important in diagnostic fields. Reverse transcription (RT), necessary for generating complementary DNA, can be both inefficient and imprecise, but remains a quintessential RNA analysis tool using qPCR. This study developed a Transcriptomic Calibration Material and assessed the RT reaction using digital (d)PCR for RNA measurement. While many studies characterise dPCR capabilities for DNA quantification, less work has been performed investigating similar parameters using RT-dPCR for RNA analysis. RT-dPCR measurement using three, one-step RT-qPCR kits was evaluated using single and multiplex formats when measuring endogenous and synthetic RNAs. The best performing kit was compared to UV quantification and sensitivity and technical reproducibility investigated. Our results demonstrate assay and kit dependent RT-dPCR measurements differed significantly compared to UV quantification. Different values were reported by different kits for each target, despite evaluation of identical samples using the same instrument. RT-dPCR did not display the strong inter-assay agreement previously described when analysing DNA. This study demonstrates that, as with DNA measurement, RT-dPCR is capable of accurate quantification of low copy RNA targets, but the results are both kit and target dependent supporting the need for calibration controls.
Measuring RNA by reverse transcription real-time quantitative PCR (RT-qPCR) is an established approach for investigating gene expression and viral diagnostics. It is well known that the RT step, required to convert RNA to complementary DNA (cDNA), is imprecise and that different reverse transcriptase enzymes (RT
Digital (d)PCR is continuing to gain recognition in the field as an extremely precise and reproducible methodology offering the potential for accurate, robust and highly sensitive measurement without the need for a standard curve
While this may be true, many studies have demonstrated that the variability inherent in the RT component of the process far outweighs that observed from the PCR step when performing qPCR
In this study we investigated how this characteristic of the RT might affect cDNA production and ultimately influence the dPCR measurement by performing RNA analysis by RT-dPCR and assessing the repeatability, linearity and sensitivity of dPCR measurement. We prepared a Transcriptomic Calibration Material (TCM) and measured both synthetic and endogenous targets, comparing RT-dPCR analysis to UV, and evaluated how different assays and commercially available one-step RT-qPCR kits perform using both endogenous targets and synthetic process controls.
LoBind® tubes were employed throughout this study (Eppendorf, Cambridge, UK). Primer and probe sequences for dPCR were designed in-house using Primer Express, software version 3 (Life Technologies, Paisley, UK) and obtained from Sigma (Dorset, UK). Primers/assays were positioned across different RNA secondary structure motifs (predicted using MFOLD
Six synthetic (ERCC developed targets; External RNA Control Consortium) RNA transcripts (ERCC-00013, −00025, −00042, −00099, −00113, and −00171) were selected for investigation (supplied in plasmid DNA format, courtesy of Dr Marc Salit, NIST, USA). For brevity, the ERCCs shall be subsequently identified without the preceding zeros. Concentrations of plasmid were assigned by the supplier using UV spectrophotometry and converted to copy number using published methods
ERCC RNA was prepared from the corresponding plasmid DNA, as described previously
Three human cell lines were employed for production of complex background material for endogenous target selection; Hep-G2 (organ: liver, disease: hepatocellular carcinoma), SaOS-2 (organ: bone, disease: osteosarcoma) and Hs 683 (organ: brain, disease: glioma), (all cell lines from ATCC, Teddington, UK). Culturing details given in
Based on confluency and cell size, eight to fourteen flasks were prepared for each cell type, as outlined in
Total RNA was extracted from cell lysates by following the standard TRIzol protocol (Invitrogen), details given in
Following total RNA extraction, DN
Pooled cell line RNA stocks were diluted in RSS to 250 ng/µL (Hep-G2 and Hs683) or 100 ng/µL (SaOS-2), and the complex background material prepared by mixing different proportions of each cell line RNA to a final concentration of 50 ng/µL (Proportions: 0.755 Hep-G2, 0.205 Hs 683, 0.04 SaOS-2). A mix containing all six ERCC transcripts was spiked into the mixed ratio cell line solution, at approximately 1.0E+06 copies/µL (final concentration), to produce the Transcriptomic Calibration Material (TCM) for analysis. The TCM solution was aliquoted (150 µL) to generate 245 replicate units prior to storage at −80°C.
dPCR experiments were performed using the Fluidigm Biomark platform. Both 12.765 and 48.770 chip formats were utilised. Assays were first optimised using the Prism 7900 HT real-time PCR system (Life Technologies) before transfer to the Biomark. One-step RT-dPCR utilised AgPath-ID one-step RT-PCR reagents (Ambion). Master reactions comprised RT-PCR buffer/master mix (1×), RT enzyme (1×), GE sample loading reagent (1×, Fluidigm), sequence-specific gene assay (
Experiment | Assay | RNA target copies per panel |
Replicates |
One-Step RT-qPCR Kit Comparison by dPCR | ERCC-25 and ERCC-99 | ∼1896 | 1 panel/assay, 3 replicate chips |
Comparison between dPCR and UV Measurement | All six ERCCs | ∼200–400 | 3 panels/assay |
RT-dPCR Quantification Sensitivity | ERCC-25 and ERCC-99 | ∼500, 250 100, 50, 25, 10 or 5 | 6 panels/dilution/assay, 2 replicate chips |
Evaluation of Reverse Transcriptases | ERCC-25, ERCC-99, UBC and MMP1 | ∼1886 | 3 panels/assay duplex, 2 replicate chips |
Dilutions are quoted based on RNA copies per dPCR panel. RNA concentrations were estimated by UV and converted to copy number using published methods
A count of partitions showing positive amplification can be made and an absolute target concentration elucidated. “Estimated copies” or “Copies per panel” refer to the number of targets on the panel following a Poisson correction, to account for the fact that some positive partitions will contain more than one molecule. As the number of positive partitions increases, so does the probability that some partitions will contain more than one target molecule. See
For one-step kit comparison, two further commercial kits were evaluated; Quantitect Probe one-step RT-PCR Kit (Qiagen) and Superscript III Platinum one-step RT-qPCR system w/ROX (Invitrogen). Both the Ambion (Multiscribe) and Invitrogen (Superscript III) RT
Initially, quantification was assessed for two external (ERCC-25 and ERCC-99) targets in both uniplex and duplex formats, between the three commercial one-step RT-qPCR kits: AgPath ID (Ambion), Quantitect (Qiagen) and Superscript III (Invitrogen). RT-dPCR was performed using Fluidigm Biomark 12.765 dPCR chips, n = 1 panel, plus three replicate experiments. Sample was diluted to approximately 1896 copies per panel (or 2062 copies/µL added to master mix), based on UV estimates. Following this, ERCC-25 and ERCC-99, plus two endogenous (UBC and MMP1) targets were compared between the kits. These assays were analysed in duplex: ERCC-25 with ERCC-99 (duplex A), UBC with MMP1 (duplex B), and ERCC-25 with UBC (duplex C). Sample was diluted to approximately 1886 copies per panel (or 1640 copies/µL added to master mix, for ERCC targets), based on UV estimates. RT-dPCR was performed using Fluidigm Biomark 12.765 dPCR chips, n = 3 replicate panels, plus two replicate experiments.
Measurement variability of six ERCC targets was tested using RT-dPCR evaluated as above (AgPath ID kit, Ambion). ERCC targets were spiked into cell line-derived total RNA at approximately 1.0E+06 copies/µL (estimated by UV), enabling evaluation of potential assay bias. Sample was diluted to approximately 200–400 copies per panel. RT-dPCR was performed using Fluidigm Biomark 48.770 dPCR chips, n = 3 replicate experiments. Assays were analysed in uniplex.
An evaluation of RT-dPCR quantification sensitivity was performed using ERCC-25 and ERCC-99 assays. Based on UV estimated values, sample was diluted in 0.5% Tween 20 (Sigma) to approximately 500, 250, 100, 50, 25, 10 and 5 copies per panel (equivalent to 3077, 1538, 615, 308, 154, 62 and 18 copies/µL, respectively). Volumetric dilutions were performed independently for each dilution, rather than sequentially, to avoid volumetric error propagation during dilution steps. RT-dPCR was performed using Fluidigm Biomark 48.770 dPCR chips, n = 6 panels per dilution, plus three replicate experiments. Assays were analysed in duplex.
All statistical analyses were performed using MS Excel 2007 and the R statistical programming environment (
The square of the copy numbers was needed in order to stabilise the difference in variance between groups. Standard uncertainties have 2 degrees of freedom, converted to expanded uncertainty with coverage factor (k) = 4.3.
Weighted regression was used to stabilise the different variance between groups. Standard uncertainty estimates were made to 3 significant figures and have 2 degrees of freedom, with k = 4.3 to convert to expanded uncertainties. Only dispersion due to plate-to-plate variation was included. dPCR plate-to-plate variability was estimated by pooling the data for all six ERCCs. The relative standard deviation was approximately 7.59% (46,000 copies) with 12 degrees of freedom (18 data points minus the six estimated group means).
A linear mixed model fit was used with experiment as random effect. Additionally, an ANOVA was applied removing experiment from the model and applying a classical fixed effect linear model fit (with only assay and dilution as factors).
The analysis was split into four groups, one per assay. The square root of the copy numbers was needed in order to stabilise the difference in variance between groups. The Qiagen kit always resulted in zero positive partitions for MMP1, which was therefore removed from the data set.
Three commercially available kits were compared for quantitative performance by RT-dPCR. The three kits were initially assessed using both uniplex and duplex formats for quantification of two synthetic RNA targets: ERCC-25 and ERCC-99 (
Three different one-step RT-qPCR kits were compared in both uniplex and duplex formats, by dPCR. Two external targets, ERCC-25 and ERCC-99 were analysed. Error bars: 95% Confidence intervals. n = 3 replicate panels. Equivalent UV estimates: ERCC-25 1185 copies/panel, 95% CI 17.34. ERCC-99 1185 copies/panel, 95% CI 26.19.
Consistent ratios for ERCC-25:ERCC-99 between uniplex and duplex measurements were not maintained between kits suggesting an assay-dependent as well as a kit associated difference (
Method | Format | ERCC- | Positive Partitions | Copies per panel |
Ratio |
Standard Uncertainty |
Ambion | Duplex | 25 | 627 | 1316 | 1.37 | 0.051 |
99 | 546 | 959 | ||||
Uniplex | 25 | 639 | 1383 | 1.47 | 0.076 | |
99 | 541 | 944 | ||||
Invitrogen | Duplex | 25 | 295 | 373 | 3.31 | 0.223 |
99 | 104 | 113 | ||||
Uniplex | 25 | 335 | 442 | 5.18 | 0.262 | |
99 | 81 | 85 | ||||
Qiagen | Duplex | 25 | 68 | 71 | 4.22 | 0.588 |
99 | 17 | 17 | ||||
Uniplex | 25 | 89 | 95 | 5.57 | 0.906 | |
99 | 17 | 17 |
Copies per panel calculated from the number of positive partitions using the Poisson correction.
Ratio of ERCC-25/ERCC-99 dPCR values with standard uncertainties. Ratios calculated using copies per panel. Standard uncertainty calculated by dividing the standard deviation by the square root of n (number of replicate measurements).
To investigate this disparity further, RT-dPCR measurements using the Ambion kit were compared when measuring a further four ERCC targets (all six present within the TCM) (
Six external targets (ERCC-13, −25, −42, −99, −113 and −171) were assessed by both one-step dPCR, utilising the Ambion one-step RT-qPCR kit, and UV measurement. Error bars: 95% Confidence intervals. n = 3 replicate dPCR experiments or UV measurements.
An additional aim was to identify RT-dPCR sensitivity and linearity of measurement for low copy targets. This was performed utilising the Ambion kit alone, due to its superior capabilities throughout our initial analyses. A dilution series of two synthetic RNA targets, ERCC-25 and ERCC-99, were analysed in duplex (
Assessment of RT-dPCR quantification sensitivity, using independent dilutions and quantifying ERCC-25 and ERCC-99 external targets in a duplex format. n = 6 panels per dilution, plus two replicate experiments. UV data based on initial UV quantification of stock and predicted target levels following volumetric dilutions. (A) & (B) dPCR sensitivity. (B) Focus on lowest level target dilutions. Error bars: 95% Confidence intervals. (C) Precision of dPCR quantification compared to UV.
There was a significant difference identified between the two targets agreement with UV values, p<0.0001 (
In order to investigate the applicability of our findings to real samples, the same three, one-step RT-qPCR kits were tested to compare measurement of endogenous targets alongside external controls in various duplex combinations in the TCM (
Three different one-step RT-qPCR kits were compared in different duplex formats, by dPCR. Quantification for external (ERCC-25 and ERCC-99) and endogenous (MMP1 and UBC) targets was evaluated. ERCC-25 with ERCC-99 (duplex A), UBC with MMP1 (duplex B), and ERCC-25 with UBC (duplex C). In the key/tabulated values, the assay in brackets is the duplex partner for the assay whose positive partition values are being displayed. Error bars: 95% Confidence intervals. n = 3 replicate panels, plus two replicate experiments.
To establish whether different plex pairings influenced RT-dPCR results, duplex reactions were performed pairing different targets (Duplex A: ERCC-25+ ERCC-99. Duplex B: MMP1+ UBC. Duplex C: ERCC-25+ UBC). As observed above, there was a significant difference between the kits, but no significant difference observed in dPCR values between ERCC-25 or UBC when assessed in different duplex reactions using the Ambion reagents (ABC), p = 0.061 and 0.92, respectively. Therefore, for these targets, assays did not influence the quantification result of their duplex partners.
In this study we used a Transcriptomic Calibration Material (TCM) containing synthetic RNA transcripts in a complex background made of mixtures of human cell line total RNA. This was used to both evaluate dPCR measurement and demonstrate the applicability of the TCM for supporting accurate RNA enumeration by RT-dPCR.
The findings from the one-step kit comparison by dPCR (
The analysis method was shown to significantly affect the RNA quantification result. There may be a number of reasons explaining the significant difference observed between dPCR and UV methodologies. While dPCR makes an absolute count of specific amplified cDNA target molecules, UV cannot discriminate between nucleic acid species, non-target RNA and fragmented/degraded/non-amplifiable targets
Our linearity and sensitivity data clearly show a pattern of increased variability with the increase of dilution factor below 50–100 estimated copies. We have previously demonstrated that when analysing DNA targets, dPCR is highly precise down to 16 copies/panel
The magnitude of the quantification difference between kits was not consistent between different targets, both synthetic and endogenous, suggesting an additional assay specific and kit associated bias. There was a greater difference between kits when measuring endogenous targets than for synthetic targets. Furthermore, both Invitrogen (1 positive partition) and Qiagen (0 positive partitions) kits were unable to provide satisfactory quantification values for MMP1 despite being measured with six replicates totalling some 4590 reactions. However, as the Ambion kit only measured on average 112 MMP1 positive partitions, it would suggest that this transcript was below the limit of detection for the two former kits. Measurement, or specifically enzyme, efficiency/sensitivity is an important consideration when measuring low abundance RNA targets, in order to avoid false negative results and our data suggests that choice of kit is crucial for ensuring the most sensitive result when performing RT-dPCR. It should also be noted that while MMP1 target was present at low abundance in the dilutions tested, evaluation of a more concentrated sample may circumvent the sensitivity issues associated with the two kits. Therefore, this must also be considered when validating protocols, and where possible, low copy measurements should be avoided.
One of the most striking findings of this study is the large inter assay and inter kit difference in the estimated copies for a given target. There are a number of potential causes for these observations. It is clear from our data that some, if not all, of the kits analysed during this study were not measuring all the RNA molecules that were present. There may be a number of different reasons for this. The assumption that DNA measurement by dPCR can be precise, reproducible, and absolute cannot be readily extrapolated to the measurement of RNA
In our recent study, we documented a dPCR phenomenon termed molecular dropout
Template secondary structure and position of the assay is known to impact on the RT-qPCR reaction
The recommendation from the MIQE guidelines
There may be other factors contributing to RT yields. For example, the samples used were sourced from cell line lysates. Co-extracted inhibitors may affect different reverse transcriptases to different degrees. Furthermore, components of total RNA, such as rRNA and tRNA may additionally inhibit RT
As may be seen from this comparison, despite the accuracy conferred by dPCR, analysis of RNA using RT-dPCR needs to be approached with caution. While for RNA measurement the precision of the RT-dPCR technique is high, it nonetheless introduces increased variability into the measurement value than dPCR alone
For accurate RNA analysis by RT-dPCR it is possible that unknown measurements should be properly correlated to an appropriate measurement standard, with a well-defined value and uncertainty
This study has shown that dPCR is capable of making precise measurements of synthetic and endogenous RNA molecules in a complex RNA background. RT-dPCR quantification of RNA targets was significantly lower than that derived from UV values suggesting a possible underestimation bias. Furthermore, absolute measurements differed between the three one-step kits assessed, with bias in detection sensitivity. Linearity and precision were sustained for duplex dPCR measurement of synthetic RNA using the Ambion kit, while sensitivities differed between RNA targets. dPCR is unencumbered by the restraints of calibration curve measurements, however, the employment of dPCR-specific calibrant materials (reference samples) would facilitate greater accuracy for absolute quantification. Furthermore, use of the TCM shows the applicability of RT-dPCR for the target-dependent selection of suitable RT enzymes. This study is novel in demonstrating application of RT-dPCR for absolute quantification of RNA endogenous and synthetic targets. Our findings give strong weight to the applicability of RT-dPCR to measurement fields including RNA diagnostics and RNA viral measurement.
RNA Secondary Structure Predictions from mFold. (A) MMP1, (B) UBC, (C) ERCC-13, (D) ERCC-25, (E) ERCC-42, (F) ERCC-99, (G) ERCC-113 and (H) ERCC-171. Green highlighted regions indicate amplicon. Folding predictions were performed at 45°C (temperature of RT step).
(DOCX)
Integrity assessment of Synthetic RNA Transcripts. 2100 Bioanalyzer quantification for all six synthetic targets was comparable to nanodrop concentration estimates (p = 0.660, with an average fold change between the two measurements of 1.02).
(DOCX)
Typical dPCR output data from this study. Both amplification plots and heatmaps are shown. Amplification plots display ΔRN versus cycle number. Heatmaps are the corresponding schematic representations of positive partitions as detected by the Biomark instrument. Black = no amplification. Red = FAM amplification. Blue = HEX amplification. Threshold was adjusted to eliminate cross talk between the filters (FAM versus HEX). (A) One-Step RT-qPCR Kit Comparison by dPCR. (B) Endogenous versus Synthetic Targets.
(DOCX)
Primer and probe sequences.
(DOCX)
ERCC RNA concentration and copy number estimates.
(DOCX)
Assay Positions.
(DOCX)
MIQE checklist for authors, reviewers and editors.
(DOCX)
Materials and Methods.
(DOCX)
dPCR Calculations Explained.
(DOCX)
We are grateful to Dr Alexandra Whale for critical review of the manuscript. We are also grateful to colleagues Dr Gary Morley (LGC) for preparation of cell lysates, Dr Jesus Minguez and Dr Simon Cowen for Statistical input and Dr Marc Salit for kind provision of ERCC plasmid DNA reference standards (NIST, USA).