Our purpose was to analyse moderate amplifications on tissues of similar but also different molecular complexities to analyse the relevance and biases in gene profiling studies following RNA amplification. To this aim, we selected two protocols (IVT one round and global PCR) and five tissues: three embryonic tissues of equivalent molecular complexities as revealed by SAGE data on ovoid, tubular and filamentous stages in pig 28 and two adult tissues of different complexities (brain and ovary) as described by human SAGE and EST data 29.

To focus the design on technological variability, we reduced the biological variability as much as we could (Fig. 1). For example, the tissues (embryos, brain and ovaries) were collected on a limited number of pregnant cows. Total RNA from brain and ovary was extracted from different pieces of tissue and pooled thereafter to get a single RNA pool for each tissue. Similarly, RNA from individual embryos was extracted and pooled per stage since biological pooling was acceptable there due to our methodological focus 30. Global PCR or IVT one round was then applied to 1 μg of total RNA from each tissue. Concomitantly, poly A+ RNA from brain and ovary was purified from the corresponding pools of total RNA and used in slot-blots to assess the biological quality of the amplified material. This has not been done on embryos due to a limited amount of material. We chose slot-blots instead of real-time PCR to assess the quality of the amplified material since a validation of a global PCR by a PCR did not seem reasonable to us due to similar enzymatic drawbacks.

Amplified material from each tissue (brain, ovary, ovoid embryos, tubular embryos and filamentous embryos) was indirectly labelled using "random" hexamers. As advised 31 replicates were emphasized and deliberately focused on technical points. Three or two independent targets for each tissue and each protocol (target replicates) were thus generated and hybridised to 4 replicates of the same array (array replicates), so that 48 measurements per probe were generated for somatic (3 targets × 2 tissues × 2 protocols × 4 arrays) and embryonic samples (2 targets × 3 tissues × 2 protocols × 4 arrays). To find out gene expression differences between protocols or embryonic stages, appropriate statistical analyses have been applied on each set of data.

Since drawbacks were reported for both IVT and PCR based protocols which could originate from a too long IVT (degradation effect reported by Spiess 32) or too many cycles of PCR (saturation effect reported by Cha or Nagy 3334), we first challenged our protocols on our tissues to define optimal amplification conditions. The protocols were tested using increasing in vitro transcription time or PCR cycle number, with a special look at 5 transcripts: 3 endogenous and 1 exogenous transcripts in brain and ovary, 1 endogenous transcript in developing embryos. Transcripts encoding EF1α, L23a and Cytochrome oxidase III were selected as somatic controls because of a differential expression between brain and ovary, an easy detection on slot blots with poly A+ RNA and a different length: 1.7, 0.9 and 0.7 kb, respectively (preliminary data, not shown). As a result, an in vitro transcription of 10 h and 2 rounds of 12 PCR cycles on 1/10 of the reverse transcription looked optimal since (i) the size of the "spiking" transcripts was conserved (ii) no shortage or degradation of the amplified material was observed and (iii) the amount of spiking transcripts had increased linearly with the time of the IVT or the number of cycles in the PCR [see Additional file 1].

Comparing the 3 amplified targets generated on the somatic tissues, it appeared that the anti-sense RNA (or aRNA) obtained after IVT corresponded to molecules of 0.1 to 4 kb with a mean size of 600 bp while the cDNA fragments generated by global PCR were reduced in size: 0.1 to 1 kb, with a mean size of 150 bp [see Additional file 2A–B]. Interestingly cDNA populations were similar for brain and ovary (panel B) whereas aRNA populations displayed slightly different patterns (panel A). However, the 3 aRNA targets generated from brain or ovary were similar [see Additional file 2C]. These results underlined a good reproducibility in the production of target replicates, a slightly different distribution of RNA species between tissues with IVT and a more homogeneous pattern between tissues with PCR.

To further assess the quality of the amplified material generated on each tissue by IVT or PCR, we took advantage of the endogenous and exogenous transcripts which were used to calibrate the protocols and studied their expression in amplified versus un-amplified material (somatic samples: Fig. 2A, embryonic samples, Fig. 2B). The L23a mRNA, slightly more expressed in ovary than in brain according to poly A+ RNA, kept the same profile after amplification (IVT or PCR). Conversely, the stronger expression of EF1α in the ovary was either increased (IVT) or attenuated (PCR) whereas the stronger expression in brain of the Cytochrome Oxidase III mRNA was weakened after IVT and abolished after PCR. Interestingly enough the exogenous mRNA, which was equally added in the RNA from brain and ovary (see Methods), appeared more expressed in the ovary after IVT but not after PCR. Using the Interferon-tau transcript as endogenous control for bovine developing embryos (Fig. 2B) we showed that the amount of this sequence increased from the ovoid to the filamentous stage after amplification (IVT or PCR) as it does in vivo (reviewed in 35). On this very small set of transcripts it was obvious that under and over representations occurred during amplification, but to which extent and with which impact on gene expression differences?

As previously advised by N'Guyen 36, we first determined the amount of labelled target to be used for each hybridisation so that no additional signal appeared but the intensity of the positive signals increased when the amount of target did (50, 125, 250 and 500 ng for aRNA or cDNA labelled targets; data not shown). On this basis, 125 ng of each target has been hybridised to each array with no particular focus on the relative amplification rates and the subsequent equivalence between these targets.

Considering the somatic hybridisations, PCR amplified targets gave a double amount of valid signals as compared to IVT amplified ones (Fig. 3A), when valid meant observed on 2 thirds of the arrays. In this case, common signals represented 90% of the IVT signals but only 45% of the PCR ones. This however was not true when valid meant detected on all the arrays with intensities 2 times over the background. Indeed, such a stringent calculation gave similar numbers of signals with IVT or PCR amplifications: 112 versus 96 with brain and 167 versus 146 with ovary, respectively. Conversely, embryonic targets (Fig. 3B) displayed similar numbers of signals at each stage but 2 to 3 times more signals than somatic ones, as expected from an array enriched in embryonic probes. Moreover, common signals between embryonic targets represented 70 to 80% of the signals generated by IVT and PCR amplifications.

When analysed per protocol, the technical replicates (target replicates: 3 for the somatic tissues; 2 for the embryonic stages; array replicates: 4 per target) proved to be nicely correlated as evidenced by the corresponding scatter plots (Fig. 4A–B). Briefly, the coefficients of correlation were between 0.85–0.97 for the hybridisations after PCR or IVT amplification: 0.85 to 0.96 for the somatic signals and 0.95 to 0.97 for the embryonic ones (Fig. 4D). However, the correlation between PCR and IVT amplified products was much lower (0.39 to 0.67 for somatic hybridisations; 0.58 to 0.67 for embryonic ones; Fig. 4C–D). The scatter plots revealed additionally a large number of signals with a very low coefficient of correlation which corresponded to signals of high intensity with PCR amplifications but low intensity with IVT and vice versa (Fig. 4C). These signals generated a crab-like figure of high interest with respect to amplification distortions.

We thus confirmed a high correlation within methods, an intermediate correlation between methods and evidenced a divergence between methods for at least a subset of the array. We thus aimed at its characterisation.

Gene expression differences between amplification methods (global PCR and IVT one round) were identified with the TMEV 3.0 software. Analysing the whole array (1920 EST), 341 gene expression differences were identified whatever the tissues. Interestingly most of them were localised into the "crab claws" previously identified in the scatter plots between PCR and IVT amplified samples (Fig. 5A). Before any other analysis this result suggested that these differences were amplification discrepancies. Repeating the analysis on the core array (987 EST, see Methods), 154 gene expression differences appeared significant between PCR and IVT amplified samples. These ones localised differently (Fig. 5B), showing that the crab claws were mostly due to the mitochondrial sequences (among which 12SrRNA) which were largely redundant (33%) within the whole array. Though biologically not challenging, this drawback (due to the fact that the embryonic library was neither normalised nor sequenced when arrayed) highlighted a specific PCR bias towards 12SrRNA sequences which could not be visualised on arrays containing highly selected cDNAs. Reverse transcription being mostly achieved on total RNA, this however was of methodological interest since those sequences incorporate a part of the isotope during the labelling.

Applying a clustering analysis to the relevant differences identified between amplified samples from embryonic and somatic tissues (n = 154), 109 appeared at first glance attributable to IVT and 45 to PCR (Fig. 6A). However, in the absence of unamplified standards such as mRNA targets or total RNA targets, one cannot distinguish higher IVT expressions due to IVT induced over-expressions or PCR induced under-expressions, and vice versa. We thus named these groups of genes Panel 1 and Panel 2 instead of IVT and PCR. Most of these deviating expressions corresponded to genes involved in similar processes: 75% and 71% in physiological and cellular processes, respectively (Fig. 6B). As expected from the clustering results, the deviating gene differences from Panel 1 fell into the highest intensities of the IVT data and lowest intensities of the PCR data whereas those from Panel 2 showed the opposite distribution (Fig. 6C). Obviously, the density of these deviating genes over unamplified data would have been of high interest to sort out the part of IVT and/or PCR defaults in these deviating expression differences. 64 and 32 genes referenced in the Unigene Bos taurus index were respectively identified within Panel 1 and Panel 2 [see Additional files 3 and 4]. In Panel 1, the endogenous controls encoding RPL23a and EF1α were recognised thus confirming that some of the deviating expressions we observed on Fig. 6 were due to IVT induced over-expressions (see Fig. 2). To further know whether molecular features such as transcript size, GC content or presence of hairpins could partly explain such deviations, we explored these features on the amplification affected sequences from both Panels.

We first found (Fig. 7A–E) that the sequences from Panel 2 (i) displayed a reduced size as compared to those from Panel 1: 200 pb against 350 pb (ii) corresponded to significantly smaller sized cDNAs: 850–950 pb against 900–1800 pb (iii) were more frequently located in the 3' end of the cDNAs: 25 to 50% of the Bt. length and (iv) displayed a lower GC content. Nevertheless, this last difference stopped being significant when the full length cDNAs were compared (Fig. 7D–E) likely due to a "buffering" effect of the coding regions where the GC contents are often closer to 45%. Considering hairpins as potential pausing sites, dA stretches as internal oligo-dT priming sites and promoter-like sequences as alternative RNA polymerase initiating sites, we then observed that sequences from Panel 2 contained also more hairpins (60% versus 37%) and A stretches (10% versus 5%) than those from Panel 1. They displayed however similar contents of promoter-like sequences (Table 1).

Conclusively, it appeared that systematic biases arose during both amplification procedures independently of (i) the sample used: brain, ovary or embryos, (ii) the enzymatic properties initially inferred (exponential or linear) and (iii) the preliminary optimisation of the protocols. These distortions affected 16% of the core array (154/987) and involved different subsets of genes (Panels 1 and 2) which harboured different molecular properties.

Knowing from above that systematic biases arose during amplification (global PCR and IVT one round) and affected 16% of the core array (987 EST), we wondered whether gene expression differences identified between embryonic stages with amplified samples could be both reliable and relevant.

49 gene expression differences were identified between stages (ovoid, tubular and filamentous) with PCR amplified samples and 28 with IVT amplified ones. Among these, 14 were IVT specific, 35 PCR specific and 14 were commonly identified (Fig. 8A). The common ones (n = 14 EST) encoded 4 genes referenced in the Unigene Bos taurus index and corresponded to transcripts identified in another study using IVT amplified samples only 26. We showed therein that c12, c93, c88 and TKDP1 transcripts were differentially expressed among these stages (c12, c93: Northern blots; TKDP1 37). The IVT specific differences (n = 14) encoded 8 genes referenced in the Bos taurus index, 4 of which were known as reliable differential expressions: IFN-tau (our endogenous control for embryos, Fig. 2B), Cox2 38, c12 and PAG11 26. Similarly, the PCR specific differences (n = 35) encoded 15 genes referenced in the Bos taurus index, 5 of which were also known as reliable differences: c12, c93, TKDP1, PAG11 and IFN-tau. Surprisingly, they were not identified as common differences between PCR and IVT amplified samples. Looking in more details at the corresponding EST it appeared clearly that, although located in the same Bt., they did not overlap. Extending this analysis to the list of specific differences (n = 49; 35 +14) we found that the EST from the PCR group were frequently located at the 3'end of the referenced cDNAs (or Bt.), as compared to those from the IVT group (Fig. 8B), and displayed reduced sizes (Fig. 8C). Last but not least, a few differences identified between embryonic stages with PCR amplified samples (2 Bt./15) matched with those identified in Panel 2 (2 Bt./32) whereas no intercept was detected with Panel 1.

Since these differential patterns were detected with amplified embryonic targets, we compared their relative expression ratios between amplified and unamplified RNA (Table 2). We thus quantified c12, c93 and PAG11 expression levels between stages as revealed by former Northern blots 26 and performed Real-Time PCR on 2 new transcripts: Cox2 (identified only by IVT targets) and IFN-tau (identified by IVT and PCR targets through non overlapping EST). From these results, one clearly sees that the differential ratios between stages were biologically similar, though not identical. Indeed, the differential ratios for Cox2 looked smaller at some stages with IVT targets whereas the differential ratios of IFN-tau, c12 and PAG11 appeared smaller with PCR targets. Nevertheless, only one inverted ratio appeared between IVT and PCR: the c12 ratio between tubular and filamentous stages.

Estrus synchronized heifers of the Charolais breed were inseminated (day 0) and day 12 to day 17 blastocysts were collected by non surgical flushing in warm PBS. Ovoid blastocysts (1–12 mm) came from collects at 12 dpi (day post insemination) whereas tubular and early filamentous stages (50–60 mm and 140–160 mm) were obtained at 14 to 15 and 16 to 17 dpi, respectively. Brain and ovaries were collected on Day-50 pregnant cows. To take adult somatic tissues, animals were humanly put down in the accredited experimental slaughterhouse of INRA under the supervision of veterinary services.

Total RNA from ovoid (n = 4), tubular (n = 4) and filamentous (n = 4) embryos was extracted with RNA-Plus™ (QBioGene). RNA quality was first verified by intact ribosomal bands on a 1% agarose gel (28S and 18S) and A260/280 absorbance ratios. Total RNA from brain and ovary was isolated in the same way. RNA quality was also verified by intact ribosomal bands on a 1% agarose gel (28S and 18S) and A260/280 absorbance ratios. A spiking mRNA was then added to brain and ovary as 1% of the estimated polyA+ amount to test whether highly expressed genes can be biased through amplification. This CG03 mRNA from A. thaliana was in vitro synthesized (with a T7 Megascript kit, Ambion) from the c554 containing plasmid, given to us by H. Hofte (LBC, INRA Versailles, France). Brain and ovary polyA+ RNA were further extracted using a Dynabeads mRNA purification kit (Dynal).

Amplified RNA from each sample was synthesized with the MessageAmp™ aRNA Kit (Ambion) according to the manufacturer instructions. Briefly, 1 μg of total RNA was incubated with 500 ng of an anchored T7-(dT) primer in 12 μl (water) at 70°C for 10 min. The 1rst cDNA strand was synthesized by the addition of 2 μl first-strand buffer, 1 μl RNAse inhibitor, 4 μl dNTP mix and 1 μl reverse transcriptase mix and incubation at 42°C for 2 h. Second-strand synthesis was performed by the addition of 63 μl DEPC-treated water, 10 μl second-strand buffer, 4 μl dNTP mix, 2 μl DNA polymerase, 1 μl RNAse H- and incubation at 16°C for 2 h. DNA was extracted with phenol:chloroform:isoamyl alcohol and precipitated in ethanol with 20 μg glycogen (Ambion). In vitro transcription was carried out at 37°C for 10 h in a 20 μl reaction volume. 1 μl DNAse was added and incubated at 37°C for 30 min. RNA was purified on Mini Quick Spin RNA columns (Roche Diagnostic) and its quality verified on RNA 6000 lab-chips (BioAnalyser 2100; Agilent Technologies).

aRNA was retro-transcribed and directly labelled with [α-³³P]dATP as described for polyA+RNA 42. 500 ng of aRNA was mixed with 500 ng of random hexamers in a volume of 25 μl. The mixture was incubated at 70°C for 10 min and chilled on ice. cDNA was synthesised by the addition of 5 μl 10× PCR buffer, 5 μl 25 mM MgCl₂, 5 μl 0,1 mM DTT, 2,5 μl 10 mM mix dGTP, dCTP and dTTP, 2,5 μl water, 50 μCi [α-³³P]dATP and 200 U Superscript II (Invitrogen) at 42°C for 50 min. The RNA template was removed by the addition of 1 μl RNAse H- and incubation at 37°C for 20 min.

Amplified cDNA was prepared as described 24 with few modifications. Briefly, 1 μg total RNA was incubated with 1 μl 10 μM oligo(dT), 1 μl 10 mM dNTPs, 1 μl 10% NP40, 1 μl 20 mM DTT, 2 μl first-strand buffer 5×, 1 μl RNAse inhibitor (Ambion) at 65°C for 2 min, at room temperature for 3 min and cooled on ice. cDNA was synthesised by the addition of 200 U Superscript II (Invitrogen) and 2 U AMV (Gibco BRL) and incubation at 42°C for 30 min. First-strand cDNA were poly(dG)-tailed by incubation with 1 μl 20 mM dGTP, 4 μl TdT buffer 5×, 2,5 μl water, 2,5 μl TdT enzyme (Promega) at 37°C for 1 h. The first PCR was performed in a volume of 50 μl using 1/10 of the RT and the second PCR was performed on 1/4 of the first PCR. Samples were incubated at 94°C for 10 min before the two rounds of PCR cycles (12 cycles each; 94°C for 2 min, 63°C for 50 sec and 72°C for 6 min). PCR products were then purified using Qiaquick PCR purification (Qiagen) and their quality verified on DNA 7500 lab-chips (BioAnalyser 2100; Agilent Technologies).

PCR-amplified cDNA was labelled with [α-³³P]dATP using random hexamers and Klenow included in Atlas SMART Probe Amplification kit (Clontech). 500 ng of amplified cDNA was mixed with 500 ng random hexamers in a volume of 34 μl. The mixture was incubated at 98°C for 8 min and at 50°C for 3 min. After addition of 5 μl 10× buffer, 5 μl dNTPs for ATP label, 5 μl [α-³³P]dATP and 1 μl Klenow, the reaction mixture was incubated at 50°C for 30 min and stopped with 2 μl 0,5 M EDTA. Labelled targets were then purified on Sephadex columns (G-50).

Real-time PCR was carried out in a final volume of 30 μl with 1 μl of diluted reverse transcriptions (1/100; 1/1000) in a 1× SYBR green Master Mix (Applied Biosystems) with 0.3 μM of gene-specific primers. Reactions were run on ABI Prism 7000 HT (Applied Biosystems). The presence of a specific and unique PCR product was checked by ABI Prism melting curves. The relative quantification of the initial amount of target was extrapolated from the appropriate standard curve, which was generated simultaneously while using serial dilutions of the corresponding PCR product. IFN-tau and Gapdh primers were as published 4344 but Cox2 primers were a kind gift from G. Charpigny. Their sequences (unpublished so far) will be available upon request gilles.charpigny@jouy.inra.fr.

125 ng of polyA+RNA, aRNA or cDNA were spotted and cross-linked to HybondN+ membranes (Amersham) at 80°C for 2 h. DNA probes encoding IFN-tau, CG03, EF1α, RPL23a or Cytochrome oxidase III were [α-³²P]dCTP-labelled using the Ready-Prime kit (Amersham). Apart from CG03, those DNA probes originate from the array. Hybridisations were conducted at 65°C for 16 h and washes performed once in 2 × SSC, 0,1% SDS at 65°C for 30 min and twice in 0,1 × SSC, 0,1% SDS at 65°C for 10 min. Slot blots were then exposed to phosphor-imaging for 24 hours and signal intensities quantified with the ImageQuant 3.3 software (Molecular Dynamics).

The bovine embryonic array used here originates from a bovine cDNA library established at the ovoid stage, starting from 1.6 μg of RNA and using the Cap Finder cDNA kit from Clontech as described in Degrelle et al. 26. Briefly, cDNA inserts from the arrayed library were amplified by PCR using the flanking primers from the Cap Finder kit and selected for spotting after a short run on a 2.5% agarose gel. 1855 probes were then spotted and fixed (UV light, 1 min, 1200 J, twice) onto nylon N+ membranes (8 cm × 12 cm, Amersham Biosciences) with a 5 × 5 pattern (BioRobotics). This was achieved with the kind help of C. Matingou and G. Piétu at the Genexpress Laboratory headed by C. Auffray (CNRS FRE 2571, Villejuif, France). The library has been called "bcai" and indexed in TGI and NCBI database as "#FJB" and "15979", respectively 4546 and the array published as "INRA-BDR Bovine D14 Embryo 1K" (GPL6284) in NCBI Gene Expression Omnibus database 47. Bacterial clones are available upon request at the CRB GADIE (INRA Jouy en Josas, France 48)

Each target was hybridized to 4 array replicates using ExpressHyb™ Hybridization Solution (Clontech) at 68°C overnight. Arrays were washed four times in 2 × SSC, 1% SDS and once in 0.1 × SSC, 0.5% SDS at 68°C for 30 min each. They were then exposed to phosphor-screens for 7 days. The hybridization signals were quantified with the Imagene 3.1 software from BioDiscovery (Proteigene) on the PICT plateform (INRA Jouy en Josas, France). These raw datasets are accessible in NCBI Gene Expression Omnibus database (experimental series "GSE9929" 47). Internal controls within the array corresponded to 65 probes and either positive or negative controls were as expected in all the hybridizations. A signal was considered "valid" when the Imagene software did not flag it (flag = 0) and when the same signal was observed on 2 thirds of the arrays, namely: 8 out of 12 for the somatic targets and 5 out of 8 for the embryonic ones.

All the plots (scatter plots, histograms) were performed on R environment 49.

Biological processes were analysed through Gene Ontology annotations 52 considering the Indentation 1. EST size, GC content, EST position according to the referenced mRNA of the Bos taurus gene index 45 were performed using Perl scripts and box plot function from the R environment 49. Presence of hairpins, dA stretches and sequences similar to RNA polymerase promoters was evaluated using the palindrome and fuzznuc programs of the Emboss package 53.