This study reports the results of an evaluation of the performance of three commonly used commercially available miRNA quantification platforms. The focus was particular on performance in relation to minute levels RNA input, i.e. in the range of what can be extracted from 250 μl of human plasma. Consistent with previous reports we found qRT-PCR based platforms to have higher sensitivity than microarray based platforms . In fact in our hands the GeneChip miRNA 2.0 platform was not sensitive enough to reliably produce signals with the plasma RNA input levels studied. We nevertheless continued assessing the two remaining qRT-PCR platforms.
Using samples with known miRNA contents we were able to show that at abundant miRNA levels the technical reproducibility and sensitivity of these platforms was good and comparable. However, at low miRNA levels, particularly at 50/200 copies or below, the sensitivity of the miRCURY platform was significantly higher than for the TaqMan platform. Consistent with this we found that the number of miRCURY assays with a high degree of linearity (r2 ≥ 0.9) across four log scales of miRNA copies was significantly higher than for the TaqMan platform. Our evaluation of the two platforms' ability to recover four-fold differences revealed that at the investigated miRNA concentrations both platforms have an acceptable recovery. However, based on the better sensitivity and linearity of the miRCURY platform it is likely that at lower miRNA concentrations the results of a similar recovery analysis would have been in favor of the miRCURY platform. We did not address recovery of fold-changes less than four-fold, but expect that the performance of both platforms will decline with decreasing fold-changes.
Considering the inclusion of the reported sensitivity improving pre-amplification step in the TaqMan protocol it was surprising to find that at low miRNA levels the sensitivity and linearity of the miRCURY platform was better than the TaqMan platform. This could indicate that the sensitivity boosting effect of the pre-amplification step is less prominent than reported . However, it should be noted that other differences between the platforms also exist, and alternatively, these could also play a role for the better sensitivity of the miRCURY platform. For example, for cDNA synthesis the miRCURY platform uses a universal approach with poly(A) end-tailing and oligo(dT) primed reverse transcription, while the TaqMan approach megaplexes > 300 miRNA specific stem-loop primers for initiation of reverse transcription. Potentially the universal approach may be more robust and sensitive than the megaplexed approach. Another possibility is the inclusion of LNAs in the miRCURY primer designs. LNAs make assay design nearly independent of miRNA GC content and compensates for many of the compromises one otherwise would have to make with a short miRNA target sequence of just ~22 nucleotides. Noticeably, It has previously been reported that inclusion of LNAs improves both PCR specificity and sensitivity .
We also assessed the specificity of the platforms and strikingly noticed that while the TaqMan platform generally showed no false positives in the no-RT control nearly 10% of the miRCURY assays were positive (even though detected at late Ct's, > 38). This could indicate that the miRCURY platform is less specific than the TaqMan platform. However, our analyses of the synthetic samples revealed that within the operative range, i.e. from the first detection to the detection threshold, the two platforms produce nearly identical numbers and rates of false positives (Figure 1). Hence, for practical purposes the specificity of the two platforms appears equal. Importantly, the number of false positives produced by both platforms increased exponentially with increasing detection thresholds. Hence, for both platforms the detection threshold should be chosen with great care to obtain acceptable false discovery rates when profiling biological samples. Sequence similarity between miRNAs has been reported to be potential cause of false positive detections [11, 26]. Consistent with this our analyses indicated that a significant fraction of the false positives, for both platforms, in particular at low Ct's was related to sequence homology (Figure 2). The latter is critical as this is also the detection range of the true positives making it practically impossible to distinguish the false from the true positives. As expected we found an inverse relationship between the number of nucleotide mismatches and the likelihood of a false positive detection (Table 1). While this was observed for both platforms the relationship was more pronounced for the miRCURY than the TaqMan platform. Along the same line the fraction of non-homology related false positives was larger for the TaqMan than the miRCURY platform (Figure 2). We do not understand the basis of the non-homology related false positives, but we perceive non-homology related amplification as more unspecific than homology related. In summary, our specificity evaluations indicate that the LNA based miRCURY platform compared to the stem-loop based TaqMan platform performs poorer in no-RT controls and slightly better, in terms of sequence specificity, in template containing samples. With both platforms it seems prudent to keep homology in mind when interpreting data from clinical samples and it may be advisable to consider carefully the likelihood of assay cross-reaction before taking a particular miRNA further, e.g. to in vitro studies.
We do not find it likely that the specificity issues of qRT-PCR will ever be completely eliminated; however, considering the ongoing technological shift to small RNA sequencing the issue may also soon be alleviated. The advantage of sequencing is that it is not hindered by variability in melting temperatures, coexpression of nearly identical miRNA family members, or post-transcriptional modifications. Moreover, it enables identification of novel miRNAs and unlike qRT-PCR the validity of sequencing results are not sensitive to changes in the miRNA sequences registered in miRBase. Currently the minimal input requirements for most small RNA liberary preparations are limiting for the use of the technology for profiling low abundance miRNA samples, such as human plasma. However, multiple approaches are being pursued to bring input requirements down, and the results are promising . Other limitations include the RNA ligation and the PCR amplification steps in the library construction protocols both of which bear inherent biases . Furthermore the tools for computational analysis of next generation sequencing data are still in their infancy. However, the technology has the potential to replace qRT-PCR as the preferred tool for profiling low abundance miRNA samples.
For now though, our data indicate that of the three tested platforms, the miRCURY platform with its better sensitivity and linearity in the low miRNA concentration range should be the platform of choice for analysis of low abundance miRNA samples. For analysis of samples with high miRNA abundance both qRT-PCR-platforms can be used, and probably also the GeneChip platform. However, when choosing platform one should keep in mind the poorly amplifying well phenomena we observed for some samples with the TaqMan platform. It caused artificial outlier measurements in the affected wells and while these outliers relatively easy can be identified by replication this approach may be prohibited by the availability of sample material and/or funding.