Comparative evaluation of linear and exponential amplification techniques for expression profiling at the single-cell level
© Subkhankulova and Livesey; licensee BioMed Central Ltd. 2006
Received: 20 December 2005
Accepted: 8 February 2006
Published: 7 March 2006
Single-cell microarray expression profiling requires 108-109-fold amplification of the picogram amounts of total RNA typically found in eukaryotic cells. Several methods for RNA amplification are in general use, but little consideration has been given to the comparative analysis of those methods in terms of the overall validity of the data generated when amplifying from single-cell amounts of RNA, rather than their empirical performance in single studies.
We tested the performance of three methods for amplifying single-cell amounts of RNA under ideal conditions: T7-based in vitro transcription; switching mechanism at 5' end of RNA template (SMART) PCR amplification; and global PCR amplification. All methods introduced amplification-dependent noise when mRNA was amplified 108-fold, compared with data from unamplified cDNA. PCR-amplified cDNA demonstrated the smallest number of differences between two parallel replicate samples and the best correlation between independent amplifications from the same cell type, with SMART outperforming global PCR amplification. SMART had the highest true-positive rate and the lowest false-positive rate when comparing expression between two different cell types, but had the lowest absolute discovery rate of all three methods. Direct comparison of the performance of SMART and global PCR amplification on single-cell amounts of total RNA and on single neural stem cells confirmed these findings.
Under the conditions tested, PCR amplification was more reliable than linear amplification for detecting true expression differences between samples. SMART amplification had a higher true-positive rate than global amplification, but at the expense of a considerably lower absolute discovery rate and a systematic compression of observed expression ratios.
Whole-genome expression profiling has many applications in areas of research where acquisition of small, highly specific tissue or cell samples is required for accurate expression analysis, such as oncology, neuroscience and development biology [1, 2]. The application of array technology or sequencing-based expression profiling technologies, such as SAGE , to single cells, however, requires either a considerable increase in the sensitivity of those assays or the amplification of the input RNA . Amplification of the starting RNA population is in common use to generate labeled microarray targets from limiting amounts of RNA. Commonly used amplification techniques are based on two different approaches: linear isothermal amplification by in vitro transcription (IVT) of the cDNA population into labeled complementary RNA (cRNA), typically using T7 RNA polymerase [5, 6], and PCR amplification of the entire population of cDNA following reverse transcription [7–9].
The most commonly used mechanism for linear isothermal RNA amplification is based on T7 RNA polymerase-mediated IVT . Several protocols based on this technique have been developed and used for microarray analysis [10–12]. Linear isothermal RNA amplification can increase the starting amounts of mRNA up to 1,000-fold in one round, while second or possibly third rounds of amplification are possible [6, 13]. Amplified RNA (aRNA) samples have been shown to generate reproducible microarray data when compared with non-amplified mRNA and closely approximate original samples [13–15]. It has been found, however, that the resulting microarray data can vary depending on details of the amplification protocol, including the amount of starting material , whether antisense or sense RNA is produced , and the number of rounds of amplification performed. In addition, time-dependent RNA degradation during IVT can introduce noise to the resulting microarray data .
Several PCR-based methods of RNA amplification have been developed as an alternative to linear IVT-based techniques. These include global PCR amplification following polyadenylation, which we shall call global amplification (GA) , 3' end amplification (TPEA)  and strand-switching-mediated reverse transcription amplification, commonly known as switching mechanism at 5' end of RNA template (SMART) [19, 20]. PCR has a number of potential advantages over linear isothermal amplification: it is faster, more cost effective, with an almost unlimited degree of amplification [18, 21, 22]. The disadvantage of relatively simple PCR-based exponential amplification is a general assumption that it introduces unacceptable biases to microarray data.
A key development in single-cell PCR amplification was the introduction of a strategy to normalize the size distribution of the resulting cDNA fragments such that the range of cDNA lengths falls between several hundred and a thousand bases [7, 8]. This was achieved by restricting the initial reverse transcription to the most 3' sequence by limiting deoxyribonucleotide concentrations and the time of reaction . Global cDNA amplification with this method enables amplification of picograms of mRNA with preservation of relative abundance of cDNAs through amplification as high as 1011-fold under ideal conditions .
Previously we found that SMART-amplified cDNA results in a systematic underestimation of the magnitude of gene-expression differences between samples when amplifying microgram amounts of total RNA . Subsequent work has also found that SMART cDNA generates reproducible data while introducing systematic changes in gene-expression ratios compared to those observed from unamplified material [21, 22, 24]. The performance of SMART in amplifying single-cell amounts of RNA has not been investigated, however, nor has it been compared to other methods for single-cell RNA amplification.
To date, little consideration has been given to direct comparison of existing methods of mRNA amplification in side-by-side experiments, particularly within the same range of amplification. In this study we investigated whether linear and exponential techniques for amplifying single-cell equivalents of total RNA introduce biases to microarray data, the nature of those biases and the levels of noise for each particular amplification method. Our goal was to define which technique is more acceptable for picogram-level expression profiling.
To do so, we analyzed the reproducibility of each method, in terms of the errors each method introduced into the amplified cDNA population, and how each method performed in identifying truly differentially expressed genes while minimizing the rate of false positives in the resulting datasets. To estimate this, we compared data generated using each method with data generated from unamplified RNA from the same sources. The three methods we studied were T7-based in vitro amplification (IVTIII), and two PCR-based methods, SMART and GA.
Overall, we found that under the conditions tested of amplifying picogram quantities of total RNA, PCR amplification outperformed IVT in several key areas. The two PCR-based methods were found to have complementary advantages: SMART had a high true-discovery rate but a low absolute number of differentially expressed genes, whereas GA identified the largest number of true positives at the expense of a considerably higher false-positive rate. An analysis of the performance of the two PCR-based methods in generating data from single-cell equivalents of total RNA and from single mammalian neural stem cells confirmed those findings.
Experimental design and yields of amplified DNA/cRNA
The objective of this study was to compare amplification techniques and choose the most reliable method for single-cell expression profiling. Therefore, all methods were tested using single-cell amounts of total RNA. It has been estimated that a single cell contains approximately 0.1 pg of mRNA or 10 pg of total RNA. This amount would need to be amplified 108-109-fold to generate enough DNA/RNA targets for hybridization on two-color microarrays. Dilution of a complex population of total RNA down to low concentrations can, however, cause a sampling effect, resulting in random representation of different species of mRNA in each aliquot.
To minimize this source of error we took a relatively high amount, 10 ng of total RNA, for the initial reverse transcription reaction for each of the PCR amplification methods assessed: GA [7, 8] standard SMART  and modified SMART (SM37) One-fifth of the reversed transcribed cDNA was used for the initial ten cycles of PCR amplification for both SMART and GA, after which 1/200 of that PCR product was used for a further 28 cycles of PCR amplification.
Indirect labeling of cDNA or cRNA with Cy3 or Cy5 did not introduce a bias to microarray data
To estimate the contribution of labeling to microarray data noise, we performed a set of self-self hybridizations (Figure 1): both unamplified or amplified aminoallyl-labeled DNA/cRNA were divided into two parts and each was coupled with either Cy3 or Cy5 NHS-esters followed by co-hybridization on the same slide.
In microarray analysis, data are frequently represented by the MA plot . This data representation plots gene-expression log ratios, log2(R/G) or M values, against the log mean intensities, log2 or A values, where R and G represent the Cy3 and Cy5 intensities for a given spot. If we assume that d1 = log2R and d2 = log2G, then
M = log2 = d1 - d2
A = log2 = 1/2log2 (R·G) = 1/2(d1 + d2)
Thus, an MA plot for two technical replicates is the same as the Bland-Altman plot for two measurements : the x axis shows the mean of the results of the two measurements ([d1 + d2]/2), whereas the y axis represents the absolute difference between the measurements ([d1 - d2]). The Bland-Altman plot may also be used to assess the repeatability of a method by comparing repeated measurements using one single method on a series of subjects. The coefficient of repeatability (CR) can be calculated as 1.96 (or 2) times the standard deviations of the differences between the two measurements (d2 and d1) .
Average coefficients of repeatability for two targets hybridized on the same array
0.29 ± 0.03
1.00 ± 0.11
0.265 ± 0.004
0.64 ± 0.014
1.24 ± 0.12
0.80 ± 0.065
1.03 ± 0.21
0.42 ± 0.07
0.60 ± 0.13
0.87 ± 0.16
0.94 ± 0.11
All methods of RNA amplification introduce errors to microarray data
Amplification of RNA by each method resulted in higher CR values compared with unamplified targets. Nevertheless, SMART-generated replicates demonstrated very low variability between each other, with a CR value (0.27) similar to unamplified DNA (0.16) (Table 1). For GA- and IVTIII-amplified pairs of replicates, CR values were 0.49 and 0.59, respectively, and particularly high variability between replicates was found for SM37-amplified targets (CR = 0.71).
Amplification errors that generate differences between two technical replicates are also reflected in the number of genes calculated as differentially expressed in two hybridizations. Only three outliers were selected for unamplified replicates by a Bayesian method at a P value of 0.01 with threshold LOD greater than 0. Surprisingly, SMART-generated replicates did not possess any outliers at all, whereas the other methods of amplification resulted in 75-200 false-positive differentially expressed genes (see below for further discussion of this effect).
Expression ratios from PCR-amplified cDNA correlate best with those from unamplified cDNA
From the analysis of the errors introduced by amplification reported above, the most reliable RNA amplification technique was found to be SMART PCR-based exponential amplification. To test this further and to analyze the ability of each method to correctly identify differences between RNA samples, we performed a model experiment in which gene expression was compared between two different cell lines.
To do so, single-cell equivalents of total RNA from each cell type were amplified using each method by the strategy outlined above. Gene expression was compared between the two cell types using each method independently. We repeated microarray hybridizations four times for each type of amplification as well as for unamplified cDNA targets, including two independent replicates for each cell line and two dye-swap hybridizations.
Differential expression discovery rates of each method identify strengths and weaknesses of each approach
Summary of expression data obtained by different methods of amplification
Number of outliers
True positive, percentage of selected
Reflecting the overall lower absolute number of genes identified as differentially expressed by SMART, the numbers of the false negatives were slightly lower for all of the other methods (false negative outliers were calculated as DNA outliers minus true positives). Thus all of the methods identify more differentially expressed genes than SMART (compared with data generated from unamplified cDNA), but at the expense of a considerably higher false-positive rate (ranging from 7-12-fold more false-positive genes for those methods).
SMART PCR-based amplification linearly decreases the expression ratios in amplified samples
Analysis of these data generated an interesting observation on the statistical behavior of microarray data for amplified targets. As expected, the distribution of averaged expression values across hybridizations (M values) when comparing gene expression between different cell lines was wider than that observed between replicates and considerably higher than for self-self hybridizations (see Figure 3). Similarly, CR values for virtual arrays (see Table 1) were higher in hybridizations comparing between cell types than in replicate or self-self hybridizations. Thus, whereas the CR value for unamplified cDNA was approximately 1.0, linear T7 amplification resulted in a wider ratio distribution (CR = 1.30), possibly as a result of introducing random noise. GA cDNA demonstrated CR values very close to unamplified samples, but SMART-based amplifications, particularly the standard SMART method, generally decreased variability between two hybridized targets, both in replicates or difference hybridizations (see Table 2).
The number and percentage of true-positive outliers of the outliers with the highest LOD score for tested amplification techniques
Cutoff level (top number of outliers)
SMART and GA amplification performance under real-world conditions
To remove the possible effects of sampling in diluting total RNA to single-cell equivalents and to isolate the effects of amplification on introducing errors into array data, all of the above experiments were carried out under ideal conditions in which the starting material for reverse transcription was 10 ng, the equivalent of 100-1,000 cells. To test whether the findings on the performance of the two PCR-based methods under ideal conditions is predictive of their performance under real-world conditions, we subsequently used each method to amplify total RNA from the two cell lines at a range of concentrations from 1 ng to 10 pg, covering the range of single-cell equivalent amounts of total RNA. As in the experiments described above, cDNA amplified by each method was used to address two questions: the reproducibility of each method (and noise introduced by each method) and the ability of each method to preserve representation of the original starting material, as reflected in their ability to identify true expression differences between the two cell lines.
Summary of expression data obtained by SMART and GA amplification techniques applied to 10 pg of total RNA from NIH 3T3 fibroblast cells
Number of outliers
True positive, percentage of selected
Number and percentage of true-positive outliers with the highest LOD score for SMART and GA amplification techniques applied to 10 pg of total RNA from NIH 3T3 fibroblast cells
Cutoff level (top number of outliers)
Exponential amplification methods generate reliable data from picogram amounts of RNA
Despite the fact that both linear and exponential RNA amplifications are commonly used methods for expression profiling, little consideration has been given to side-by-side examination of different amplification techniques, particularly for the purpose of single-cell RNA expression profiling. The goal of this study was to test the performance of the most widely used amplification techniques in generating expression data from single-cell amounts of RNA. In addition, we estimated the levels of error for each type of amplification.
Analysis of both technical replicates and test/reference samples hybridized on oligonucleotide microrrays revealed that PCR based-amplifications, and particularly SMART technology, are competitive with, and may outperform, T7-based linear amplification for amplifying picogram amounts of total RNA. We present several findings that demonstrate that PCR amplification generates reproducible microarray data, but with some key shortcomings.
First, PCR-amplified targets possess the least difference between technical replicates among the amplification techniques: CR values for SMART were close to those of unamplified samples (0.27 vs 0.16; see Figure 2), followed by GA and then linear amplification. Secondly, the correlation between unamplified and amplified targets is highest for SMART amplification (r2 = 0.75), closely followed by GA (r2 = 0.6), with (r2 = 0.51; see Figure 4). In addition, the rate of true positives is highest for SMART-amplified cDNA (80%) compared with 53% for GA and 39% for linear amplification. However, a critical difference between SMART and the other methods is the lower overall absolute number of truly differentially expressed genes identified (336 by SMART, as opposed to 492 by GA and 633 by linear amplification).
Extending this approach to analyzing PCR-amplified single cell equivalent amounts of total RNA and also RNA from single neural stem cells found that SMART again outperformed GA in the key areas of CR and true positive rates (see Figure 8, see Tables 4 and 5). SMART also had the lowest overall absolute number of differentially expressed genes, however. We conclude that GA is inherently more noisy than SMART amplification, as reflected in its higher false-discovery rate, but has a higher absolute discovery rate. These results are consistent with previously published findings that exponential amplification methods may yield reproducible results from the picogram range of total RNA [16, 18] and be more precise than linear RNA amplification [18, 22].
SMART PCR-based amplification results in compression of microarray expression ratios
It is noteworthy that the distribution of log ratios for SMART-amplified samples is considerably narrower (for both technical replicates and test/reference hybridizations) than for any other method (see Figure 3). We previously observed this compression effect of SMART amplification when amplifying microgram amounts of total RNA , finding that it results in a systematic reduction in the magnitude of expression differences between two samples. Therefore, it is likely that the SMART-based technical replicate data appear less noisy than data generated by other methods because of the compression effect on log ratio distribution.
Consistent with our previous findings , we found that the decrease in log ratios was also linear when amplifying single-cell amounts of total RNA. In this case, the estimated coefficient of linearity is 2.5. Such a relationship means that the real expression differences between tested samples should be 22.5, or 5.6, times higher than that calculated from the microarray data. No such compression was observed with the other PCR-based amplification method (GA) or with linear isothermal amplification. Global amplification of picogram amounts of total RNA 1011-fold has previously been found to substantially increase expression ratios , whereas the 108-109-fold GA amplification reported here resulted in a minor change in expression ratios. It is possible that the alteration in ratios seen here with GA amplification could change to the extent of that observed in the work of Iscove and colleagues  with the additional 103-fold amplification used in that study.
Although all techniques tested here successfully amplified the starting population of total RNA, we present strong evidence that they also introduce errors to microarray data. Some of the variation is systematic and could be possibly negotiated if reference and test targets are synthesized by the same method. Others are random and could be decreased by replicate hybridizations. In the present investigation we observed that averaging microarray data decreases the values of CR in replicate hybridizations (see Table 1) and increases the correlation between unamplified and amplified targets in test/reference hybridizations by reducing the random component of noise (see Figure 4). The reduction in the contribution of random noise to the false-discovery rate by increasing the number of biological replicate hybridizations could make GA amplification an attractive option for single-cell expression profiling, given the overall higher absolute discovery rate of this method, compared to SMART.
Overall, PCR-amplified samples demonstrate a higher correlation between each other then with T7-amplified targets (see Figure 4), indicating a systematic bias intrinsic to technically similar amplification methods. These data are in good agreement with previous observations of the systematic bias related to the type of hybridization technique which has been demonstrated for both linear and exponential amplifications [28, 29].
Noise in microarray data depend on the rate of RNA amplification
The variability of amplified targets may depend on many factors, including the technical basis of amplification, details of the amplification method and the degree of amplification required. As single cell profiling requires 107-109-fold amplification of the original mRNA population, the number of PCR cycles or number of rounds of linear amplification can become a critical source of errors. Consistent with this, Petalidis and colleagues  previously demonstrated a reduction in the discovery rate of differentially expressed genes with numbers of PCR cycles in microarray analysis of SMART-amplified targets when amplifying microgram amounts of total RNA.
For T7-based amplification we also have shown that Pearson's correlation coefficients decreased from r2 = 0.90-0.95 for the first round of amplification to r2 = 0.7-0.8 for the second round, and finally to r2 = 0.5-0.6 for the third round, when amplified targets were correlated with unamplified cDNA (T.S. and F.J.L., unpublished data). One of the sources of variability in T7-amplified samples may be a time-dependent degradation of amplified cRNA that results in shortening of cRNA species . Thus, if each round of linear amplification increases slightly the levels of error, the cumulative effect of three rounds may result in a relatively poor approximation of original mRNA sample.
The decision as to which amplification technique to use for expression profiling of limiting biological samples depends on several parameters, among them the quality and quantity of RNA, laboratory facilities and the experimental goal. If a goal is to obtain the largest possible number of the differentially expressed genes, GA would be the technique of choice, particularly if the resources are in place to analyze enough cells to reduce the noise in this system. Nanogram amounts of total RNA make it possible to restrict linear amplification to two rounds of T7-based amplification with sufficient yields of labeled targets and high-quality data. Finally, if the rate of true positives is required to be as high as possible, the relatively low false-positive rate of the SMART amplification technique is a useful approach. The decision to use this method should, however, take into account the overall lower number of differentially expressed genes that this approach is likely to identify and also that the real difference in gene expression levels between any two tested samples is likely to be systematically higher than observed using this approach. Further improvements in PCR-based amplification techniques, such as reducing the losses associated with RNA extraction, improved strand switching in the case of SMART, and careful choice of buffers and PCR conditions, may yield even more reproducible results from the picogram range of total RNA.
Materials and methods
For all experiments, total RNA was isolated from mouse fibroblast (3T3) or mouse ovarian surface epithelium (OV) cell lines using TRI reagent (Amersham Biosciences, Little Chalfont, UK). The ovarian cells were a kind gift of Cristian Brocchieri (University of Cambridge, Department of Oncology & Hutchison/MRC Research Centre). For generating fluorescently labeled cDNA from unamplified RNA, 100 μg of total RNA from 3T3 or OV cell lines was labeled with amino-allyl dUTP during reverse transcription followed by coupling with Cy3 or Cy5 NHS esters (; Cy3 and Cy5 Mono-Reactive Dye Packs, Amersham Biosciences).
The sequences of the oligonucleotides used for the different amplification technologies were as below:
SMART cDNA amplification
cDNA synthesis for SMART was performed essentially as described . Total RNA (10 ng) was mixed with 10 pmol of SM1 primer and 10 pmol template-switching SM2 primer in volume of 5 μl. The reaction mixture was incubated at 70°C for 2 minutes and then placed on ice for 2 minutes. The following reagents were then added, 2 μl 5× first strand buffer (Gibco, Carlsbad, USA), 1 μl 20 mM DTT, 1 μl 10 mM dNTPs and 1 μl PowerScript RT (BD Clontech, Mountain View, USA), and the reaction was incubated at 42°C for 1 hour. A 2 μl aliquot of the first strand cDNA was then used for PCR amplification. A 2 μl aliquot of the first-strand cDNA was then used for PCR amplification. The following reagents were added, 80 μl dH2O, 10 μl 10× Advantage 2 PCR Buffer (BD Clontech), 2 μl 10 mM dNTPs, 4 μl SMPCR primer and 2 μl 50× Advantage 2 polymerase mix and the reaction mixture was subjected to the cycling program: 95°C for 1 minute and then a variable number of cycles (10 or 28) of 95°C for 15 seconds, 65°C for 30 seconds and 68°C for 6 minutes. cDNA synthesis in a slightly modified SMART technique (SM37) was performed at 37°C rather then at 42°C, and SM1 primer was replaced with primer SM37, followed by PCR amplification with both SMPCR and SM3 primers.
Global polyadenylated PCR amplification (GA)
10 ng of total RNA (1 μl) was mixed with 3.5 μl of ice-cold stock buffer (25.14 μl DEPC water, 1 μl anchored primer (10 ng/μl), 5 μl 10× reaction buffer (PCR buffer, Roche), 2.5 μl 100 mM DTT, 1 μl 2.5 mM dNTPs, 0.25 μl NP-40, 1 μl RNase inhibitors mix (1:1 mixture rRNasin (Promega, Madison, USA) and Prime (Brinkman/Eppendorf, Hamburg, Germany)) and incubated for 1 minute at 65°C followed by placing on ice for 2 minutes. Then 0.5 μl of RT mix (3 ml PowerScript RT, 0.5 μl of RNase inhibitor mix) was added to RNA and the reaction was incubated at 37°C for 90 minutes. The reaction was stopped by heating to 65°C for 10 minutes and cooled to 4°C. To perform poly(A) tailing of synthesized cDNA 5 μl of TdT mix (0.15 μl 100 mM dATP, 0.5 μl 10× reaction buffer, 0.3 μl 25 mM MgCl2, 3.55 dH2O, 0.25 μl TdT (Roche, Lewes, UK), 0.25 μl RNaseH (Roche)) was added to the reaction mixture and the reaction was incubated for 20 minutes at 37°C followed by inactivation at 65°C for 10 minutes. A 2 μl aliquot of the first-strand polyadenylated cDNA was then used for PCR amplification. The following reagents were added: 67 μl dH2O, 10 μl 10× Taq PCR Buffer (Takara Bio, Shiga, Japan), 10 μl MgCl2, 2 μl 2.5 mM dNTPs, 2 μl anchored primer (1 μg/μl) and 1 μl LA Taq (Takara Bio) and the reaction mixture was subjected to the cycling program: 95°C for 1 minute, 37°C for 5 minutes, 72°C 20 minutes (once) and then a variable number of cycles (10 or 28) of 95°C for 30 seconds, 67°C for 1 minute and 72°C for 6 minutes. To avoid sampling effects (see Results for further details), 10 ng of total RNA was always taken for cDNA synthesis in all amplifications. This amount is approximately 1,000 times higher then the amount of total RNA in a single cell (around 10 pg). To adjust the amount of RNA to single-cell content, one-fifth of the resulting cDNA was used for first-round PCR. After ten cycles of exponential amplification, 1/200 of the amplified product was taken for a further 28 PCR cycles. When the starting amounts of RNA were l ng, 100 pg, or 10 pg, all of the reverse-transcribed cDNA was used for initial PCR amplification. After ten cycles of PCR, each amplified cDNA was diluted to single-cell equivalents (1 ng starting material was diluted 1/100, 100 pg diluted 1/10), and a second amplification round of 28 PCR cycles was carried out. PCR products were purified with the CyScribe GFX Purification kit (Amersham Biosciences) and indirectly labeled with aminoallyl dNTP using Klenow DNA polymerase (BD Biosciences, Franklin Lakes, USA) followed by coupling with Cy3 or Cy5 NHS esters .
Neuronal stem cells were obtained from dissections of developing mouse neocortex at embryonic day 11.5 (E11.5). Dissected neocortices were dissociated with a papain dissociation system (Worthington Biochemical Corporation, Lakewood, USA). Single cells were picked using glass capillary tubes, washed in PBS and placed in PCR tubes. A group of 12 cells were pooled, mixed with 100 ng of polyinositol as a carrier and total RNA was isolated using TRI-reagent (Amersham Biosciences). RNA was dissolved in water and divided into four parts for amplification by SMART or the GA technique (two replicates for both methods).
T7 based amplification
Linear amplification was performed using the MessengerAmp II aRNA kit (Ambion, Austin, USA). Ten nanograms of total RNA was used for the first-round amplification; 1/10 of the resulting cRNA (cRNAI) was used for the second round of amplification, and then 1/100 of second-round cRNA (cRNAII) was used for the third round of amplification. cRNA was indirectly labeled with amino-allyl-UTP during in vitro transcription and coupled with Cy3 or Cy5 NHS-esters as described .
Expression microarrays containing 23,232 65-mer oligonucleotides (Sigma-Genosys, UK) were printed on CodeLink slides (Amersham Biosciences). Hybridized arrays were scanned in an Axon microarray scanner at a resolution of 10 μm at maximum laser power and photomultiplier tube voltage of 60-80%. Image analysis and feature analysis were performed with GenePix Pro 4.0 (Axon Instruments, Foster City, USA).
All statistical analysis was conducted using the R environment  and the R package Statistics for Microarray Analysis . Log intensity ratios for each spot were obtained with background subtraction. Data normalization was performed using print lowess normalization using the Limma package . Differential genes were identified using an empirical Bayesian method with threshold at LOD score of zero or higher (if specified) . The Pearson correlation coefficient and CR were calculated as described .
We thank Cristian Brocchieri for providing the ovarian epithelial cell line and James Smith for technical support. This work was supported by the EU FP6 programme and the Wellcome Trust.
- Player A, Barrett JC, Kawasaki ES: Laser capture microdissection, microarrays and the precise definition of a cancer cell. Expert Rev Mol Diagn. 2004, 4: 831-840. 10.1586/1473722.214.171.1241.PubMedView ArticleGoogle Scholar
- Kamme F, Salunga R, Yu J, Tran DT, Zhu J, Luo L, Bittner A, Guo HQ, Miller N, Wan J, Erlander M: Single-cell microarray analysis in hippocampus CA1: demonstration and validation of cellular heterogeneity. J Neurosci. 2003, 23: 3607-3615.PubMedGoogle Scholar
- Velculescu VE, Zhang L, Vogelstein B, Kinzler KW: Serial analysis of gene expression. Science. 1995, 270: 484-487.PubMedView ArticleGoogle Scholar
- Livesey FJ: Strategies for microarray analysis of limiting amounts of RNA. Brief Funct Genomic Proteomic. 2003, 2: 31-36. 10.1093/bfgp/2.1.31.PubMedView ArticleGoogle Scholar
- Eberwine J, Yeh H, Miyashiro K, Cao Y, Nair S, Finnell R, Zettel M, Coleman P: Analysis of gene expression in single live neurons. Proc Natl Acad Sci USA. 1992, 89: 3010-3014.PubMedPubMed CentralView ArticleGoogle Scholar
- Van Gelder RN, von Zastrow ME, Yool A, Dement WC, Barchas JD, Eberwine JH: Amplified RNA synthesized from limited quantities of heterogeneous cDNA. Proc Natl Acad Sci USA. 1990, 87: 1663-1667.PubMedPubMed CentralView ArticleGoogle Scholar
- Brady G, Iscove NN: Construction of cDNA libraries from single cells. Methods Enzymol. 1993, 225: 611-623.PubMedView ArticleGoogle Scholar
- Brady G, Billa F, Knox J, Hoang T, Kirsh IR, Voura EB, Hawley R, Cumming R, Buchwald M, Siminovitch K: Analysis of gene expression in a complex differentiation hierarchy by global amplification of cDNA from single cells. Curr Biol. 1995, 5: 909-922. 10.1016/S0960-9822(95)00181-3.PubMedView ArticleGoogle Scholar
- Dixon AK, Richardson PJ, Lee K, Carter NP, Freeman TC: Expression profiling of single cells using 3 prime end amplification (TPEA) PCR. Nucleic Acids Res. 1998, 26: 4426-4431. 10.1093/nar/26.19.4426.PubMedPubMed CentralView ArticleGoogle Scholar
- Liu CL, Schreiber SL, Bernstein BE: Development and validation of a T7 based linear amplification for genomic DNA. BMC Genomics. 2003, 4: 19-10.1186/1471-2164-4-19.PubMedPubMed CentralView ArticleGoogle Scholar
- Patel OV, Suchyta SP, Sipkovsky SS, Yao J, Ireland JJ, Coussens PM, Smith GW: Validation and application of a high fidelity mRNA linear amplification procedure for profiling gene expression. Vet Immunol Immunopathol. 2005, 105: 331-342. 10.1016/j.vetimm.2005.02.018.PubMedView ArticleGoogle Scholar
- Yang IV, Chen E, Hasseman JP, Liang W, Frank BC, Wang S, Sharov V, Saeed AI, White J, Li J, et al: Within the fold: assessing differential expression measures and reproducibility in microarray assays. Genome Biol. 2002, 3: research0062-10.1186/gb-2002-3-11-research0062.PubMedPubMed CentralGoogle Scholar
- Zhao H, Hastie T, Whitfield ML, Borresen-Dale AL, Jeffrey SS: Optimization and evaluation of T7 based RNA linear amplification protocols for cDNA microarray analysis. BMC Genomics. 2002, 3: 31-10.1186/1471-2164-3-31.PubMedPubMed CentralView ArticleGoogle Scholar
- Jenson SD, Robetorye RS, Bohling SD, Schumacher JA, Morgan JW, Lim MS, Elenitoba-Johnson KS: Validation of cDNA microarray gene expression data obtained from linearly amplified RNA. Mol Pathol. 2003, 56: 307-312. 10.1136/mp.56.6.307.PubMedPubMed CentralView ArticleGoogle Scholar
- Wang E, Miller LD, Ohnmacht GA, Liu ET, Marincola FM: High-fidelity mRNA amplification for gene profiling. Nat Biotechnol. 2000, 18: 457-459. 10.1038/74546.PubMedView ArticleGoogle Scholar
- Goff LA, Bowers J, Schwalm J, Howerton K, Getts RC, Hart RP: Evaluation of sense-strand mRNA amplification by comparative quantitative PCR. BMC Genomics. 2004, 5: 76-10.1186/1471-2164-5-76.PubMedPubMed CentralView ArticleGoogle Scholar
- Spiess AN, Mueller N, Ivell R: Amplified RNA degradation in T7-amplification methods results in biased microarray hybridizations. BMC Genomics. 2003, 4: 44-10.1186/1471-2164-4-44.PubMedPubMed CentralView ArticleGoogle Scholar
- Iscove NN, Barbara M, Gu M, Gibson M, Modi C, Winegarden N: Representation is faithfully preserved in global cDNA amplified exponentially from sub-picogram quantities of mRNA. Nat Biotechnol. 2002, 20: 940-943. 10.1038/nbt729.PubMedView ArticleGoogle Scholar
- Matz M, Shagin D, Bogdanova E, Britanova O, Lukyanov S, Diatchenko L, Chenchik A: Amplification of cDNA ends based on template-switching effect and step-out PCR. Nucleic Acids Res. 1999, 27: 1558-1560. 10.1093/nar/27.6.1558.PubMedPubMed CentralView ArticleGoogle Scholar
- Zhu YY, Machleder EM, Chenchik A, Li R, Siebert PD: Reverse transcriptase template switching: a SMART approach for full-length cDNA library construction. Biotechniques. 2001, 30: 892-897.PubMedGoogle Scholar
- Vernon SD, Unger ER, Rajeevan M, Dimulescu IM, Nisenbaum R, Campbell CE: Reproducibility of alternative probe synthesis approaches for gene expression profiling with arrays. J Mol Diagn. 2000, 2: 124-127.PubMedPubMed CentralView ArticleGoogle Scholar
- Petalidis L, Bhattacharyya S, Morris GA, Collins VP, Freeman TC, Lyons PA: Global amplification of mRNA by template-switching PCR: linearity and application to microarray analysis. Nucleic Acids Res. 2003, 31: e142-10.1093/nar/gng142.PubMedPubMed CentralView ArticleGoogle Scholar
- Livesey FJ, Furukawa T, Steffen MA, Church GM, Cepko CL: Microarray analysis of the transcriptional network controlled by the photoreceptor homeobox gene Crx. Curr Biol. 2000, 10: 301-310. 10.1016/S0960-9822(00)00379-1.PubMedView ArticleGoogle Scholar
- Puskas LG, Zvara A, Hackler L Jr, Van Hummelen P: RNA amplification results in reproducible microarray data with slight ratio bias. Biotechniques. 2002, 32: 1330-1334, 1336, 1338, 1340.PubMedGoogle Scholar
- Chenchik A, Zhu YY, Diatchenko L, Li R, Hill J, Siebert PD: Generation and use of high-quality cDNA form small amounts of total RNA by SMART PCR. Gene Cloning and Analysis by RT-PCR. Edited by: Siebert P, Larrick J, Natick MA. 1998, USA: Biotechniques Books, 305-319.Google Scholar
- Dudoit S, Yang YH, Callow MJ, Speed TP: Statistical methods for identifying differentially expressed genes in replicated cDNA microarray experiments. Stat Sinica. 2002, 12: 111-139.Google Scholar
- Bland JM, Altman DG: Measuring agreement in method comparison studies. Stat Methods Med Res. 1999, 8: 135-160. 10.1191/096228099673819272.PubMedView ArticleGoogle Scholar
- Wilson CL, Pepper SD, Hey Y, Miller CJ: Amplification protocols introduce systematic but reproducible errors into gene expression studies. Biotechniques. 2004, 36: 498-506.PubMedGoogle Scholar
- Nygaard V, Loland A, Holden M, Langaas M, Rue H, Liu F, Myklebost O, Fodstad O, Hovig E, Smith-Sorensen B: Effects of mRNA amplification on gene expression ratios in cDNA experiments estimated by analysis of variance. BMC Genomics. 2003, 4: 11-10.1186/1471-2164-4-11.PubMedPubMed CentralView ArticleGoogle Scholar
- Richter A, Schwager C, Hentze S, Ansorge W, Hentze MW, Muckenthaler M: Comparison of fluorescent tag DNA labeling methods used for expression analysis by DNA microarrays. Biotechniques. 2002, 33: 620-628, 630.PubMedGoogle Scholar
- t Hoen PA, de Kort F, van Ommen GJ, den Dunnen JT: Fluorescent labelling of cRNA for microarray applications. Nucleic Acids Res. 2003, 31: e20-10.1093/nar/gng020.View ArticleGoogle Scholar
- Ihaka R, Gentleman R: R: A language for data analysis and graphics. J Comput Graph Stat. 1996, 5: 299-314.Google Scholar
- Smyth GK: Limma: linear models for microarray data. Bioinformatics and Computational Biology Solutions using R and Bioconductor. Edited by: Gentleman R, Carey V, Dudoit S, Irizarry R, Huber W. Springer, New York, 397-420.
- Lönnstedt I, Speed T: Replicated microarray data. Stat Sinica. 2002, 12: 31-46.Google Scholar
- Jenssen TK, Langaas M, Kuo WP, Smith-Sorensen B, Myklebost O, Hovig E: Analysis of repeatability in spotted cDNA microarrays. Nucleic Acids Res. 2002, 30: 3235-3244. 10.1093/nar/gkf441.PubMedPubMed CentralView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.