Gene expression profiling on several hundred cells isolated by cell-sorting technologies, or by LCM prompts the development of new amplification procedures. Over recent years, academic, private sector and commercial researchers have been developing tools and methods to generate expression profiles from small amounts of biological material. Improvements have been made in all steps of the process of expression profiling experiments such as sample collection, RNA purification, or acquisition and analysis of expression data. Over the last few years, improvements in RNA amplification methods now allow the synthesis of sufficient targets to perform microarray hybridisation on as little as picogram amounts of input RNA. In this study, we evaluated and compared four commercial RNA amplification protocols using picogram amounts of input RNA. This comparative study will be useful to researchers when planning new experiments involving samples derived from a few cells.
This comparative study was based on several assessments of amplification products. First of all, electrophoresis was performed and examined in combination with cRNA or cDNA yields. This analysis is performed routinely prior to hybridisation, as it provides an estimate of the overall amplification performance. Unusual traces or insufficient yield correspond to either abnormal input RNA, or amplification failure. The poor yield obtained with the protocol proposed by Epicentre is due to amplification failures since the quality of the commercial RNA has been checked. Experimental errors might be the cause of these failures. However, as amplifications were carried out in duplicate by two experienced operators in two laboratories, the quality of the batch and its reproducibility may be responsible for these errors.
The overall size of aRNA, estimated by the electrophoretic profile, has been previously described as a good metric when looking at within-sample fidelity [26
]. As expected, a high level of consistency in amplified product sizes and electrophoretic profiles was observed within each technology, since all amplifications were performed using the same input RNA. The repeated observation of products of very large size (8,000–12,000 nt) obtained with the Arcturus RiboAmp™ kit are surprising and might be due to the transcription of non-specific ligation of cDNA templates. As for other microarray technologies, probes designed for Affymetrix gene expression assay are biased towards the 3' end of the transcripts: the length of the target is therefore not a critical metric provided the targets cover the first 700 bases from the 3' end of the transcripts. However some probes (10% at most) represent sequences over 600 bases from the 3' end, and short amplification products may therefore not address these probes. The impact of this issue can be easily estimated by the 3'/5' ratios of housekeeping genes. Affymetrix HG U133 plus 2.0 GeneChips include probe sets selected in the 5' region of housekeeping genes in addition to conventional probes within a maximum of 600 nt from the 3' end. The signal intensity ratio of the 3' probe over the 5' probe (3'/5' ratio) is a good metric to evaluate the qualitative performance of first strand cDNA synthesis and aRNA transcription (or cDNA replication for Nugen WT-Amplification™). Abnormally high ratios were obtained when using T7-IVT-based amplification chemistries suggesting that the 3' biased expression issue should be taken into consideration when analysing these data. The lower 3'/5' ratios observed in the Nugen WT-Amplification™ approach could be explained by the fact that i) in addition to the conventional poly-T priming for the initial reverse transcription, the system proposed by Nugen also includes random priming, and ii) amplification is performed in one cycle thus avoiding shortening of targets at the 5' end by performing a random primed reverse transcription of first-round amplified aRNA.
The percentage of present call has been previously described as a good metric when evaluating the sensitivity of a method [30
]. Gene expression analysis performed on small amounts of input RNA, can be logically expected to give a reduced sensitivity. However, only Arcturus RiboAmp™ chemistry showed significantly reduced sensitivity compared to the standard procedure. Poor sensitivity was also observed when conducting experiments using Nugen WT-Amplification™ system using the lowest amounts of input RNA (50 pg and 100 pg), suggesting that the minimum input limits of the protocol had been reached.
With respect to hybridisation specificities, the Nugen technology shows a fundamental difference as it generates a single-stranded cDNA target whereas others protocols yields cRNA targets. As RNA/DNA interactions are stronger than DNA/DNA interactions, cDNA hybridisations should theoretically be more specific but also less sensitive. However, the protocol proposed by Nugen was not associated with decreased sensitivity and the lower average background value of the chips hybridised with cDNA targets (Nugen) suggests reduced non-specific target/background interactions but additional studies should be conducted to evaluate the potential increased specificity of the probe/target interaction and its biological impact on expression profiling.
Practical criteria should be considered to choose an amplification approach to carry out expression profiling experiments. These include completion time, handling difficulties and labour intensiveness. Major differences were observed between the various protocols evaluated in terms of these criteria. IVT-based amplifications (Ambion MessageAmp™, Arcturus RiboAmp™, and Epicentre TargetAmp™) involve numerous steps and are therefore time-consuming and labour-intensive. However, the high quality of technical support and optimisation of chemistries (most reagents are conveniently pre-dispensed and pre-mixed) simplify the handling of the Ambion and Arcturus systems. The Nugen technology is fundamentally different from the IVT-based amplifications. Fewer steps are needed to achieve amplification, and the process does not include delicate RNA handling during the amplification process. The Nugen WT-Amplification™ protocol is therefore shorter and easier to complete, and consequently less error-prone. Ultimately, investigators will need to compare experimental results obtained by different laboratories. Experimental variables (such as operator) on expression data could be limited if fewer steps are needed to achieve amplification.
When expression profiling experiments are performed on just a few cells, it is currently impossible to strictly assess the quality and quantity of the purified RNA. Even the most modern and accurate spectrophotometers or microfluidic-based electrophoresis chips require at least 100 pg of RNA to characterise nucleic acids extracted from such minute samples. The results presented here show that a two-fold variation of the amount of input RNA had almost no impact on the transcriptomes analysed when starting with picogram amounts of RNA. Moreover, high reproducibility within a given protocol (correlation coefficient of about 0.95) was observed for amplifications that were carried out in two different laboratories by two different operators. This shows that trained laboratories could conduct amplifications with the technologies proposed by Ambion, Arcturus, or Nugen and produce data that could effectively be compared across laboratories using the same amplification protocols. However, the results obtained in this study indicate large variations between different protocols. This suggests, as previously shown, that i) when conducting expression profiling experiments the same amplification protocol should be used in order to maximize the comparability of the results [26
], and ii) that RNA amplification affects the expression measurements [35
]. However, researchers who already have a large volume of Affymetrix two-round amplification data, and who want to conduct a project including comparison of these data, might decide to choose the Ambion MessageAmp™ amplification protocol despite its very high bias towards the 3' to 5' ends, as expression profiles estimated according to this protocol exhibited the best correlation with the results of Affymetrix amplification (correlation coefficient of 0.89).
Amplifications were performed by two operators in two different laboratories in order to mimic "real experimental conditions" of expression profiling. In our hands and under these experimental conditions, Nugen WT-Amplification™ pico protocol appeared the most suitable. Additional experiments were performed in order to further assess this system. Amplifications were performed on amounts of input RNA ranging from 50 pg to 1 ng to test the system at the upper and lower limits of input amounts corresponding to a few cells isolated by FACS or LCM. The results obtained show that the overall quality, comparability and reproducibility of expression measurements were very good for RNA input amounts ranging from 250 pg to 1000 pg. The system was less efficient with 100 pg and 50 pg of RNA inputs with respect to sensitivity and reproducibility. It is important to analyse this result in the light of manufacturer's specifications indicating that the minimal RNA input of the system is 500 pg.
The results of a differential gene expression experiment using Nugen WT-Amplification™ chemistry or Affymetrix one-round IVT amplification were also compared. If we strictly count the number of probesets with the same status (Decrease, Increase, Marginal Increase, Marginal Decrease or No Change) in both chimistries, discrepancies were observed only for 25% of probesets. However the direction of differential expression diverged in few measurements. These results show that most genes have similar differentially expressed patterns and that one amplification protocol does not create an artificial variability of the measurements compared to the other. The observed discrepancies do not appear to be due to the decreased sensitivity of one of the two protocols because the same proportion of genes was modulated when using the Affymetrix and Nugen protocols. Although the effects of RNA amplification on differential gene expression measurements have been previously reported [26
], a large proportion of these discrepancies might be due to fundamental differences between the two approaches, such as the influence of the molecular nature of the targets (RNA or DNA) on expression measurement, or the use of random priming to synthesise double-stranded cDNA templates.
The preservation of the relative abundance levels of gene transcripts is an important issue when performing RNA amplification prior to genome-wide expression measurements. This issue has been widely studied in comparing expression measurements i) from the amplification of different amounts of RNA [11
], ii) from amplified and unamplified materials [14
], iii) using different amplification procedures [11
], iv) to a gold standard amplification procedure[26
], v) using different expression evaluation methods such as quantitative-reverse-transcription-Ploymerase-Chain-Reaction (qRT-PCR) [21
]. Most of these studies were performed using IVT-based amplification methods. Those methods were different from the ones evaluated in this report, and were not optimised for picogram amounts of input RNA. Only two studies were conducted using Nugen amplification technology but not the WT-Amplification™ pico system evaluated here [21
]. The general conclusion that could be drawn from these articles is that either IVT-based or Nugen-like RNA amplifications were globally able to maintain relative transcripts abundance with only slight differences. However, it has been pointed out that, in most of these reports, the statistical analyses or the comparison with qRT-PCR data were restricted only to subsets of genes such as outliers or highly expressed genes [31
]. Furthermore, several studies have seriously questioned the preservation of the relative transcript abundance during RNA amplification. Sequence dependent biases [35
], and a drop of fidelity for low expressed transcripts [10
] have been demonstrated. In addition, it has been shown that low expressed genes were subject to stochastic fluctuations that increase as the sample size decreases [31
]. However, we obtained high correlations between technical replicates using either Ambion, Arcturus or Nugen methods (Figure ). It shows that no or negligible stochastic fluctuations occur at these levels of input RNA (500 pg and 250 pg). On the other hand, stochastic event could explain the lower reproducibility observed for smaller amount of input RNA (50 pg and 100 pg) using Nugen method.
Therefore, a) we do not recommend to perform amplification from less than 250 pg of input RNA, and b) when it is possible, researchers should try to increase the RNA input quantity to at least 500 pg to be more confident in the biological interpretation of the results, particularly concerning low express transcripts.