TY - JOUR
AU - Robert, Claude
AB - Abstract High throughput methods deliver large amount of data serving to describe the physiological treatment that is being studied. In the case of microarrays, there would be a clear benefit to integrate the published data sets. However, the numerous methodological discrepancies between microarray platforms make this comparison impossible. This incompatibility is magnified when considering the peculiar context of transcript management in early embryogenesis. The total RNA content is known to profoundly fluctuate during development. In addition, the mRNA population is subjected to poly(A) tail shortening and elongating events, a characteristic of stored and recruited messengers. These intrinsic factors need to be considered when interpreting any transcript abundance profiles during early development. As a consequence, many methodological details affect microarray platform performances and prevent compatibility. In an effort to maximize our microarray platform performance, we determined the various sources of variation for every one of the main steps leading to the production of microarray data. The five main steps involved in sample preparation were evaluated, as well as conditions for post-hybridization validation by qRT–PCR. These determinations were essential for the implementation of standardized procedures for our Research Network but they can also provide insight into the compatibility issues that the microarray community is now facing. bovine, embryo, microarray, mRNA, oocyte Introduction As it is the case for many other fields, the study of early mammalian development has been using microarray platforms for the past decade as a mean to describe the molecular mechanisms underlying observed physiology. Microarrays have proven to be a valuable high throughput approach that can efficiently describe the abundance of transcripts of a sample and provide, by comparative assessment, gene lists of candidates, whose expression is statistically different between treatments. Although studies reporting gene lists are abundant, it has been very difficult to fully integrate the information provided by these reports and gain the knowledge that was anticipated from this tremendous increase in information. In fact, the full power of microarray technology remains unexploited, in most part because of the numerous different types of microarray platforms, as well as important discrepancies in sample preparation. These sources of technological variations impair the comparison of data sets produced by different teams. For instance, the RNA extraction strategy, either aiming at the total RNA fraction or at isolating the poly(A)-bearing mRNA, will result in different transcript abundance measurements. Such initial discrimination of the messenger population based on poly(A) tail length is an inherent consideration that is of prime importance, especially in view of the peculiar nature of oocytes and early embryos, in which mRNAs are stored in a deadenylated form and recruited for translation in a time-dependant manner through lengthening of the poly(A) tail (Bachvarova et al., 1985; Wickens 1990; Bachvarova 1992; Paynton et al., 1994). There are many microarray platforms available, either custom made or commercial (Brown et al., 1999; Lipshutz et al., 1999; Hughes et al., 2001; Nuwaysir et al., 2002; Albert et al., 2003; Barczak et al., 2003). Aside from the Affymetrix platform, which is more restrictive in terms of sample preparation and data collection, all of the ‘spotted or printed’ microarray platforms can be adapted according to user needs and preferences and as such, greatly differ structurally and methodologically (Pennie 2000; Carter et al., 2003; Vallee et al., 2006; Lee et al., 2007). Therefore, because the basic principle of microarrays relies on relative abundance comparison, the choice between the diverse available options, from sample preparation to hybridization, has a tremendous impact on our ability to compare data sets across platforms. The generation of data using a microarray platform involves numerous steps for which performance and quality are essential for the creation of valuable gene lists. When minute amounts of starting material, such as mammalian oocytes and pre-hatching embryos are involved, five main steps are involved, in addition to the data validation step: (i) RNA extraction either total or poly-adenylated; (ii) RNA amplification; (iii) hybridization sample labeling; (iv) sample hybridization; (v) microarray wash and data acquisition; the data are then processed through an analysis pipeline (the topic of data analysis will not be addressed herein) and lead to the final step upon result analysis; (vi) data set validation using quantitative PCR. The scope of the present report is to share and provide optimization details from a microarray platform used to study the molecular mechanisms involved in pre-hatching development in bovine. We currently aim to standardize the microarray data generation procedures amongst our Research Network dedicated to the study of early embryo development. Furthermore, these standard operative procedures could maximize data set compatibility and offer a novel opportunity to compare treatments across data sets and consequently greatly enhance the potential for discovery of microarray platforms. Materials and Methods Note that all chemicals were obtained from Sigma-Aldrich (St. Louis, MO) unless otherwise stated. Biological sample collection To test the different methods, bovine immature oocytes at the germinal vesicle (GV) stage were used. Typically, pools of 10 oocytes/early embryos are utilized for microarray probe preparation. The material was collected as previously described (Vigneault et al., 2007). Total RNA extraction All extractions were conducted as recommended by the manufacturers, aside from some tests performed to reduce the elution volumes and avoid downstream concentration steps. Essentially, two types of approach were compared, e.g. the guanidium isothiocyanate-phenol cell disruption for RNA extraction (Trizol, Invitrogen, Berlington, Canada) and silica-based ion exchange columns (PicoPure RNA Isolation Kit, Molecular Devices, Sunnyvale, CA; Absolutely RNA Microprep Kit, Stratagene, La Jolla, CA; RNeasy Mini Kit, Qiagen, Mississauga, ON, Canada). For Trizol RNA extraction, the RNA pellet was dissolved in 10 µl of RNAse-free water and a DNAse I treatment (DNAse I amplification grade, Invitrogen) was performed according to the manufacturer's recommendation. In all column-based extractions, cells were disrupted using the supplied extraction buffer. Nucleic acids were bound to the membrane, washed and genomic DNA contamination was removed using an on-column DNAse treatment. Conditions for the DNAse treatment was as detailed in the user manual. In order to obtain the adequate volume (e.g. 10 µl) for the reverse transcription or amplification step, three different approaches were tested for the Qiagen and Stratagene kits: (i) elution of the column with 12 µl of elution buffer; (ii) two elutions to obtain a volume of 100 µl and concentration of RNA by isopropanol precipitation; (iii) two elutions followed by ethanol precipitation. In both cases, RNA was precipitated by adding 0.1 volume of sodium acetate and 15 µg/ml of linear acrylamide (Applied Biosystems, Streetsville, Canada) as a co-precipitant. 1.5 volume (for isopropanol precipitation) or 2.5 volumes (for ethanol precipitation) were added to the mixture, samples were left at −80°C for half an hour and centrifuged at 16 000 g, 20 min at 4°C. The supernatant was carefully removed and washed with 70% ethanol. The RNA pellets were resuspended in 10 µl of RNAse-free water. For Picopure RNA extractions, RNA was eluted in 11 µl as recommended by the manufacturer. Total RNA yields were determined using a 2100 Bioanalyzer apparatus (Agilent Technologies, Mississauga, Canada) and samples were run on RNA Pico Chips (Agilent Technologies). For extraction efficiency evaluations and repeatability assessment of the chosen method, the total RNA of three pools of 10 denuded oocytes was extracted using the PicoPure columns (Molecular Devices). The DNAse treatment (Qiagen) was performed directly on the column. Double concentrations of DNAse I (relatively to the amount recommended in the user manual) were used to ensure complete digestion of contaminating genomic DNA and the polydI-dC carrier contained in the extraction buffer. A synthetic RNA coding for green fluorescent protein (GFP) (Vigneault et al., 2004) bearing a poly(A) tail was spiked in the sample before or after column extraction. Eluates were reverse transcribed using the SensiScript reverse transcriptase (Qiagen) and quantification of the GFP content was performed by quantitative PCR as described below. Messenger RNA isolation For mRNA isolation using a poly(A) capture or selection method, magnetic beads either containing oligo-dT primers (DynaBeads, Invitrogen) or a GripNA Lock Nucleic Acids (mTrap) (Active Motif, Carlsbad, CA) were tested. An opposite strategy rather aiming at the removal of the major rRNA components and leaving mRNAs in the sample was also tested. The RiboMinus (Invitrogen) protocol was followed as recommended by the manufacturer but RNeasy MinElute Cleanup (Qiagen) was used to concentrate the resulting sample. Briefly, magnetic beads covered with DNA sequences complementary to the 18S and 28S rRNAs were used to remove these rRNAs from samples. For all of these methods, the recommended protocol was not suitable for the low input materials obtained when working with oocytes or early embryos. Tests were conducted to reduce volumes and adapt conditions to this context. For all these tests, a total RNA sample isolated from a bovine ovary was utilized. This total RNA sample was extracted by PicoPure RNA Isolation Kit (Molecular Devices) following manufacturer's instructions. The performance of the RNA capture/fractionation methods was assessed on a 2100 BioAnalyzer (Agilent Technologies). Transcriptome amplification To globally amplify RNA samples, four methodological approaches were tested. Two of them (RiboAmp HS (Molecular Devices); RampUP (Genisphere, Hatfield, PA) rely on in vitro transcription to copy the templates. Sample preparation demands the introduction of an RNA polymerase promoter sequence on every template. The RiboAmp HS introduces the promoter sequence on the 3′ end during the reverse transcription reaction through the use of a chimerical poly(T) primer that bears the T7 RNA polymerase promoter in 5′. RampUP is designed to introduce two RNA polymerase promoter sequences in tandem at the 5′ end of all templates. As a consequence of the RNA polymerase positioning, RiboAmp produces antisense RNA while RampUP generates sense RNA. The third method was the Ovation RNA Amplification System V2 (referred to as RiboSPIA) (Nugen, San Carlos, CA), which is based on the strand displacement activity of a DNA polymerase. The templates are copied by the constant binding and degradation events of a DNA:rNA chimerical primer that recruits the polymerase and initiate strand elongation, which is in turn followed by the specific digestion of the RNA portion of the bound primer by RNAse H; therefore, creating a binding site for another intact chimerical primer. This amplification results in the production of single-stranded cDNAs. All of these methods are isothermal and are advertized as being linear. The last one, TransPlex Whole Transcriptome Amplification (Sigma-Aldrich), is a PCR-based amplification and utilizes random primers to ensure coverage of total RNA without the 3′ bias. The end product is therefore double-stranded DNA. To evaluate the different amplification methods, total RNA of a single pool of 240 denuded GV stage oocytes was extracted using a PicoPure column (Molecular Devices). A DNAse treatment was performed directly on the column. Sample integrity was determined by size profiling using a 2100 BioAnalyzer (Agilent). Aliquots containing 10 oocyte equivalents were used to test amplification strategies. A least three technical replications were performed for each methods. The amplification reactions' kinetics were determined for the RiboAmp HS, the RampUP and the TransPlex WT. The first two methods were followed using a broken beacon designed to specifically target the bovine actin, beta (ACTB) sequence (either sense or antisense depending on the reaction's output) (Gilbert et al., 2009a). For the third method, the unspecific intercalating SybrGreen dye was utilized to detect the production of DNA molecules during Transplex WT amplifications. All kinetic assessments were performed in a LightCycler apparatus (Roche, Laval, Canada) using plastic capillaries. To avoid the potential impact of the detection compound on the reactions' efficacy, other samples were run in parallel under recommended conditions in a standard thermal cycler (PT-200 from MJ Research; Bio-Rad, Mississauga, ON). The amplification yields were measured using a NanoDrop 1000 micro-spectrophotometer (NanoDrop, Wilmington, DE). Size distribution of the amplification output was determined by microfluidic using the appropriate DNA or RNA LabChip run on a 2100 BioAnalyzer system (Agilent Technologies). Hybridization samples labeling Three different sample labeling approaches were tested. The first involves the classical direct incorporation of dye-bound nucleotides performed by reverse transcribing antisense RNA (aRNA) samples, the second incorporates, during the amplification step, modified nucleotides containing an amino-allyl group that binds the dyes during a subsequent step, while the third approach involves the addition of dyes directly on amplified samples through unmodified guanine targeting. For direct incorporation, Cy-dyes-coupled dCTP were incorporated using the CyScribe First-Strand cDNA Labeling Kit with CyScribe GFX Purification Kit according to the manufacturer's instructions (GE Healthcare, Baie d'Urfe, Canada). The incorporation of amino-allyl coupled UTP (Applied Biosystems) was conducted during the second aRNA amplification round (RiboAmp HS kit, Molecular Devices). Two micromilligrams of amino-allyl containing aRNA samples was used to chemically bind Alexa Fluor 555 or 647 (Invitrogen) according to the manufacturer's protocol. The third type of amplification strategy involved the use of the Universal Linkage System (ULS) to chemically bind the fluorophores to guanine residues. Four different kits were tested: uLS aRNA Labeling Kit (Kreatech, Amsterdam, The Netherlands); Ulysis (Invitrogen); Turbo Labeling (Molecular Devices) and ULS Fluorescent Labeling Kit for Agilent arrays (Kreatech). The first two kits use Alexa dyes while the others use Cy-dyes. Labeling reactions were performed as recommended by the manufacturers with one modification: the supplied unincorporated dye removal columns were used for Turbo Labeling (Molecular Devices) only; for all other kits, sample clean up was performed with a Picopure RNA purification kit (Molecular Devices) and samples were eluted in a final volume of 11 µl. Quantification for dye incorporation was performed using a NanoDrop 1000 (NanoDrop). Microarray hybridization To determine the optimal hybridization conditions, total RNA from ovarian tissue was isolated using Trizol reagent (Invitrogen) and reverse transcribed (SuperScript II reverse transcriptase, Invitrogen) in presence of amino-allyl dUTPs (Applied Biosystems). Dye addition (Alexa Fluor 647, Invitrogen) was carried out according to Invitrogen's protocol. Only AlexaFluor 647 was used in order to eliminate the bias caused by differential fluorescence intensity between dyes. Two micrograms of labeled-cDNA was added to each hybridization buffers tested (Table I) and solutions were deposited on our custom cDNA microarray under a Lifterslip (Thermo Fisher Scientific, Waltham, MA). This microarray contains a collection of bovine cDNAs specifically targeting expressed sequence tags found in the oocyte and early embryos (Vallee et al., 2009). Slides were hybridized for 18 h in a temperature-controlled hybridization station (SlideBooster (Advalytix, San Francisco, CA)). As a reference condition, an hybridization using the DIG buffer (Roche) was performed in a plastic chamber (Corning, Corning, NY) submerged in a water bath overnight. Details regarding hybridization temperatures and washing conditions are found in Table I. Following washes, the slides were dried off by centrifugation at room temperature at 1200 g for 5 min. Table I Hybridization buffer composition and washing conditions. Buffer  Recommended hybridization temperature  Washing conditions  SlideHyb buffers*  50°C  2 × 15 min at 55°C, 2× SSC–0.5% SDS      2 × 15 min at 55°C, 0.5× SSC–0.5% SDS      1–3 times at RT°C with 1× SSC      1 time with water  DIG buffer  42°C  1 × 10 min at RT°C1× SSC–0.2% SDS      1 × 10 min at 50°C1× SSC–0.2% SDS      1 × 5 min at RT°C1× SSC–0.1% SDS      1–3 times at RT°C with 1× SSC      1 time with water  ArrayBooster buffer  37°C  1 × 10 min at 30°C, 2× SSC–0.1% SDS      1 × 1 min at 30°C, 1× SSC      1 × 9 min at 30°C, 1× SSC      1 × 10 min at 30°C, 0.2× SSC  Buffer  Recommended hybridization temperature  Washing conditions  SlideHyb buffers*  50°C  2 × 15 min at 55°C, 2× SSC–0.5% SDS      2 × 15 min at 55°C, 0.5× SSC–0.5% SDS      1–3 times at RT°C with 1× SSC      1 time with water  DIG buffer  42°C  1 × 10 min at RT°C1× SSC–0.2% SDS      1 × 10 min at 50°C1× SSC–0.2% SDS      1 × 5 min at RT°C1× SSC–0.1% SDS      1–3 times at RT°C with 1× SSC      1 time with water  ArrayBooster buffer  37°C  1 × 10 min at 30°C, 2× SSC–0.1% SDS      1 × 1 min at 30°C, 1× SSC      1 × 9 min at 30°C, 1× SSC      1 × 10 min at 30°C, 0.2× SSC  *Washing conditions were identical for all SlideHyb buffers. View Large For investigations on global amplification methods and dye incorporation (results shown in Figs 7 and 8), labeled samples were hybridized using SlideHyb buffer #1 (Applied Biosystems) at 50°C for 18 h in the SlideBooster hybridization station (Advalytix). Microarray data collection and processing Microarrays were scanned using the VersArray ChipReader System (Bio-Rad, Mississauga, ON, Canada). Microarray image processing was performed with the ArrayPro Analyzer software (Media Cybernetics, San Diego, CA). Background correction was performed by subtracting the locally determined background value for each spot. Data sets were normalized using Loess and the intensities for each channel were plotted against each other in order to calculate correlation values. Testing the reverse transcriptases Four reverse transcriptase enzymes were tested on a unique pool of 50 GV oocytes. For every enzyme, three 1 GV equivalent aliquots were reverse transcribed using an oligo-dT(18) (ReadyMade Primers, IDT, Coralville, IA). The four enzymes tested are the Sensiscript Reverse Transcriptase (Qiagen), the Superscript III Reverse Transcriptase (Invitrogen), the Transcriptor Reverse Transcriptase (Roche) and the qScript cDNA SuperMix (Quanta Biosciences, Gaithersburg, MD). For all protocols, manufacturer recommendations for oligo-dT concentration, incubation time and temperature were followed. RNA abundance measurements by quantitative real-time PCR Primers for each candidate were designed using the Primer3 web interface (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi). Sequences, size of amplified product, GeneBank accession numbers, as well as annealing temperatures are presented in Table II. The qPCR measurements were conducted using the LightCycler FastStart DNA Master Sybr Green I (Roche) in a LightCycler apparatus (Roche) following manufacturer instructions. A detailed description of our Real-time PCR amplification procedure is described in (Gilbert et al., 2009a). Table II Characteristics of PCR primers for real-time abundance measurements. Name  Sequence  Annealing temperature (°C)  Fluorescence acquisition temperature (°C)  Product size (bp)  Accession number  H2A.1  Up 5′-GTCGTGGCAAGCAAGGAG-3′  57  88  182  XM_583411    Low 5′-GATCTCGGCCGTTAGGTACTC-3′          HMGA1  Up 5′-GCCTCCAAGCAGGAAAAGG-3′  58  89  125  BT006774    Low 5′-TTGCTTCCCTTTGGTCGG-3′          HMGB1  Up 5′-ATTTGAAGACATGGCAAAGG-3′  57  82  325  NM_176612    Low 5′-TTTCCCTTTAGCTCGGTATG-3′          TEAD2  Up 5-′GGCAGATCTACGACAAATTCC-3′  57  88  270  NM_001103347    Low 5′-GTTCCGTCTCCACCTTCTC-3′          GFP  Up 5-′GCA GAAGAACGGCATCAAGGTGAA-3′  58  86  144  *    Low 5′-TGGGTGCTCAGGTAGTGGTTGT-3′          GAPDH  Up 5′-CCAACGTGTCTGTTGTGGATCTGA-3′  57  84  194  NM_001034034    Low 5′-GAGCTTGACAAAGTGGTCGTTGAG-3′          ACTB  Up 5′-ATCGTCCACCGCAAATGCTTCT-3′  57  80  80  NM_173979    Low 5′-GCCATGCCAATCTCATCTCGTT-3′          ACTB-Broken Beacon sens  Fluo: 5′-FAM-CCACGTACGAGTCCTTCTGGCCCA-3′  42  NA  NA  NM_173979  Quencher: 5′-GGACTCGTACGTGG-DABCYL-3′          ACTB-Broken Beacon anti-sens  Fluo: 5′- FAM-TGGGCCAGAAGGACTCGTACGTGG-3′  42  NA  NA  NM_173979  Quencher: 5′-GTCCTTCTGCCCA-DABCYL-3          Name  Sequence  Annealing temperature (°C)  Fluorescence acquisition temperature (°C)  Product size (bp)  Accession number  H2A.1  Up 5′-GTCGTGGCAAGCAAGGAG-3′  57  88  182  XM_583411    Low 5′-GATCTCGGCCGTTAGGTACTC-3′          HMGA1  Up 5′-GCCTCCAAGCAGGAAAAGG-3′  58  89  125  BT006774    Low 5′-TTGCTTCCCTTTGGTCGG-3′          HMGB1  Up 5′-ATTTGAAGACATGGCAAAGG-3′  57  82  325  NM_176612    Low 5′-TTTCCCTTTAGCTCGGTATG-3′          TEAD2  Up 5-′GGCAGATCTACGACAAATTCC-3′  57  88  270  NM_001103347    Low 5′-GTTCCGTCTCCACCTTCTC-3′          GFP  Up 5-′GCA GAAGAACGGCATCAAGGTGAA-3′  58  86  144  *    Low 5′-TGGGTGCTCAGGTAGTGGTTGT-3′          GAPDH  Up 5′-CCAACGTGTCTGTTGTGGATCTGA-3′  57  84  194  NM_001034034    Low 5′-GAGCTTGACAAAGTGGTCGTTGAG-3′          ACTB  Up 5′-ATCGTCCACCGCAAATGCTTCT-3′  57  80  80  NM_173979    Low 5′-GCCATGCCAATCTCATCTCGTT-3′          ACTB-Broken Beacon sens  Fluo: 5′-FAM-CCACGTACGAGTCCTTCTGGCCCA-3′  42  NA  NA  NM_173979  Quencher: 5′-GGACTCGTACGTGG-DABCYL-3′          ACTB-Broken Beacon anti-sens  Fluo: 5′- FAM-TGGGCCAGAAGGACTCGTACGTGG-3′  42  NA  NA  NM_173979  Quencher: 5′-GTCCTTCTGCCCA-DABCYL-3          H2A.1, Histone H2A.l; HMGA1, high-mobility group AT-hook 1; HMGB1, high-mobility group protein B1; TEAD2, TEA domain family member 2. GFP, green fluorescent protein; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; ACTB, actin, beta. *Amplified from pCMS-eGFP (PT3268-5). NA, not applicable. View Large Statistical analyses For evaluation of RNA extraction method performances (Results shown in Fig. 1) and associated RNA abundance measurements of candidate genes (Results shown in Fig. 2), significant differences were calculated using the SAS software (SAS-Institute inc., Cary, NC). One-way ANOVA with Fisher's LSD tests were conducted. When ANOVA criteria were not met (normality and homogeneity of variance), data were transformed to log. For RNA extraction column efficiency assessment (Results shown in Fig. 3), data were analyzed through an unpaired t-test performed on Prism 5.03 software (GraphPad inc., La Jolla, CA). Correlation values of microarray data sets were calculated by the Array-Pro Analyzer 3.5 microarray data analysis software (MediaCybernetics). The significance of difference analyses to determine the impact of the source of reverse transcriptase on RNA abundance measurements (Results shown in Fig. 10) was performed by a one-way ANOVA with Tuckey's multiple comparison test on the Prism 5.03 software. For every test, differences were considered statistically significant at the 95% confidence level (P < 0.05). Results RNA extraction Total RNA extraction Three parameters were considered to determine the most appropriate total RNA extraction method for bovine oocytes and early embryos. 1) Efficiency of extraction, as measured by the yield of the methods. 2) Reproducibility of the procedure. This parameter was estimated by calculating the variance between replicates. 3) Elution volume, which is of importance for downstream applications as concentration steps usually result in loss of sample. Figure 1 presents the first two parameters. The lowest yields were obtained with the RNeasy columns (Qiagen). The first essential step during RNA extraction is cell solubilization. We noticed that the extraction buffer supplied with the columns was not harsh enough to completely dissolve the bovine oocyte structure, as observed under stereo microscope (data not shown). By contrast, the Trizol reagent provided very good yields, surpassing almost all of the column-based approaches. The repeatability of Trizol extraction suffered mostly from the partial loss of sample during precipitation and pellet resolubilization. For most column-based RNA extraction procedures, the elution volume is problematic as it is too high to proceed with downstream sample preparation, therefore requiring a concentration step. As such, we compared reducing the elution volume to an acceptable range (12 µl), precipitation using a standard method (ethanol) or precipitation using isopropanol. The alcohol used for precipitation had an impact on final absorbance measurements. We could not figure out whether it was caused by higher levels of non-RNA contaminants or simply by a better nucleic acids recovery. To answer this question, quantitative PCR were performed to measure the abundance of transcripts in samples isolated using two of the most promising methods, e.g. PicoPure RNA extraction kit and Absolutely RNA microprep-isopropanol precipitation, concomitantly with the traditional Trizol. For all four gene candidates, the RNA abundance were similar across extraction methods (Fig. 2) indicating that the divergence in yields determined by 260 nm absorbance are not caused by different total RNA contents. Based on our initial aforementioned criteria, we chose the PicoPure RNA extraction kit as our total RNA extraction method of choice when working with oocytes and early embryos. To more precisely determine the efficiency and repeatability of the columns, tests were conducted by spiking samples prior or after total RNA extraction (as full recovery controls). Quantification of spikes was performed by quantitative PCR during which all samples were submitted to the same reverse transcription process and all contained similar RNA contents. The mean recovery was determined to be 84.30% with a mean variation coefficient of 20% (Fig. 3). Figure 1 View largeDownload slide Total RNA yields from different extraction and concentration methods. Q: rNeasy Mini Kit, (Qiagen); MD: PicoPure total RNA columns (Molecular Devices); S: absolutely RNA MicroPrep columns (Stratagene); Trizol: trizol reagent (Invitrogen). Concentration strategies: 12 µl: elution with 12 µl of elution buffer; IPA, precipitation with isopropanol. EtOH, ethanol precipitation. Different letters are representative of significant statistical difference (P < 0.05). Figure 1 View largeDownload slide Total RNA yields from different extraction and concentration methods. Q: rNeasy Mini Kit, (Qiagen); MD: PicoPure total RNA columns (Molecular Devices); S: absolutely RNA MicroPrep columns (Stratagene); Trizol: trizol reagent (Invitrogen). Concentration strategies: 12 µl: elution with 12 µl of elution buffer; IPA, precipitation with isopropanol. EtOH, ethanol precipitation. Different letters are representative of significant statistical difference (P < 0.05). Figure 2 View largeDownload slide Relative quantification of transcript abundance for selected candidates in total RNA samples extracted using different methods. H2A.1, h2A histone family; member A.1; HMGB1, high-mobility group box 1; HMGA1, high-mobility group AT-hook 1; TEAD2, tEA domain family member 2. Trizol, Trizol reagent (Invitrogen); MD, PicoPure total RNA columns (Molecular Devices); S-IPA, absolutely RNA MicroPrep columns (Stratagene) with isopropanol precipitation. Different letters are representative of significant statistical difference (P < 0.05). Figure 2 View largeDownload slide Relative quantification of transcript abundance for selected candidates in total RNA samples extracted using different methods. H2A.1, h2A histone family; member A.1; HMGB1, high-mobility group box 1; HMGA1, high-mobility group AT-hook 1; TEAD2, tEA domain family member 2. Trizol, Trizol reagent (Invitrogen); MD, PicoPure total RNA columns (Molecular Devices); S-IPA, absolutely RNA MicroPrep columns (Stratagene) with isopropanol precipitation. Different letters are representative of significant statistical difference (P < 0.05). Figure 3 View largeDownload slide Recovery efficiency of PicoPure columns (Molecular Devices) for total RNA extraction. Green fluorescent protein (GFP) RNA bearing a polyA tail was exogenously added to the samples prior or following total RNA extraction. Both sample types were submitted to reverse transcription and GFP was quantified using qPCR. The difference was not statistically different (P = 0.90). Figure 3 View largeDownload slide Recovery efficiency of PicoPure columns (Molecular Devices) for total RNA extraction. Green fluorescent protein (GFP) RNA bearing a polyA tail was exogenously added to the samples prior or following total RNA extraction. Both sample types were submitted to reverse transcription and GFP was quantified using qPCR. The difference was not statistically different (P = 0.90). Messenger RNA isolation Instead of isolating total RNA from oocytes and early embryos, it is also possible to isolate the mRNA subfraction. Considering that the main benefit from this approach is to get rid of non-mRNA contamination, we tested three different approaches to isolate and fractionate total RNA samples and thus remove major ribosomal RNAs. The first two methods target the isolation of poly(A) bearing RNAs by pull down using either an oligo-dT or a locked nucleic acids (GripNA) (mTrap) both recognizing poly(A) stretches. For the method using an oligo-dT (Dynal), size profiling clearly indicates the presence of remaining large ribosomal subunits (Fig. 4C). It is unclear why these RNAs that do not bear poly(A) tails were still present in the samples. Also, in mTrap profiles (Fig. 4D) a peculiar peak in the pre-18S region is observed. This peak is of unknown nature but could correspond to the GripNA molecules. It is noteworthy to mention that due to the atypical mRNA and rRNA contents in oocytes and early embryos (Gilbert et al., 2009b), these tests were conducted with total RNA samples from somatic tissues, therefore containing the recommended content of starting material. When we reduced the starting material to the projected levels found in mammalian oocytes or early embryos (80 ng of total RNA), the isolates were free of contaminating ribosomal RNA but also contained very little RNA (not shown). Both of these results indicate that at higher RNA concentrations, the prevalent non-poly(A) bearing rRNAs were most probably also isolated by being carried by the excess beads in the solution, whereas at lower concentrations, the eluate was clean of contaminating rRNAs but poor yields, indicative of partial isolation of polyadenylated RNA or the selection of a subfraction of polyadenylated RNAs, were achieved. Figure 4 View largeDownload slide Electrophoresis profile of the different mRNA isolation techniques. (A) Total RNA control profile before subfraction (B) after 18S and 28S rRNA depletion with the RiboMinus (Invitrogen) (C) after mRNA enrichment with an oligo dT (DynaBeads, Dynal); (D) after mRNA enrichment with an oligo-dT gripNA (mTrap, Active Motif). Figure 4 View largeDownload slide Electrophoresis profile of the different mRNA isolation techniques. (A) Total RNA control profile before subfraction (B) after 18S and 28S rRNA depletion with the RiboMinus (Invitrogen) (C) after mRNA enrichment with an oligo dT (DynaBeads, Dynal); (D) after mRNA enrichment with an oligo-dT gripNA (mTrap, Active Motif). A third method was tested, which relies on the opposite strategy, its goal being to remove contaminating 18S and 28S rRNAs and leaving behind all other types of RNA, including the poly(A) RNAs (RiboMinus, Invitrogen). This method reduced by 80% the abundance of large rRNAs. Size profiling of the remaining fraction showed a similar dynamic range and profile for non-ribosomal RNAs than the initial sample, indicating that most mRNAs remained in the sample (Fig. 4B). Sample amplification Sample amplification can be achieved by different means. We tested four different strategies; two of them utilize the intrinsic properties of RNA polymerases, whereas the others are based on the action of DNA polymerases. The yields and size distribution of the output are found in Table III and Fig. 5. The method that generated the highest yield was the RiboAmp HS (mean of 52.5 µg) about ten times higher than the output of the PCR-based Transplex WT amplification (mean of 5.7 µg). By comparison, the RiboSPIA and RampUP mean outputs were 6.5 µg and 12.5 µg respectively. Although both strategies based on in vitro transcription (RiboAmp HS and RamUP) are targeting a priming site on opposite end of the mRNAs, both approaches generated similar size profiling with an average fragment size of 400 nt (Fig. 5). The RiboSPIA produced considerably longer fragments with an average of 1000 nt. The presence of fragments larger than 4 kb is clearly evident from size profiling (Fig. 5). The randomly primed PCR based approach produced fragments of mainly 100 to 300 bases in length with a peak of unknown nature at the 17 kb size mark. Considering that in eukaryotic, the average protein contains 400 amino acids and that the average length of the untranslated regions (UTR) in mammals are 200 and 800 nucleotides for the 3′ and 5′ UTR respectively, the average mRNA would therefore be 2.2 kb in length (Mignone et al., 2002), the template used for amplification of this extremely large fragment is in all logic not transcripts. Considering that RNA samples were treated with DNAse, it is improbable that remaining genomic DNA was utilized as template and can account for the large fragment. Not knowing the composition of the various buffers, we cannot rule out non-nucleic acid contaminants as the cause of the large fragment. Table III Final yields following transcriptome amplification using different methods. Samples  Ovation RNA Amplification System V2 RiboSPIA (NuGEN) final yield (µg)  TransPlex Whole Transcriptome Amplification (Sigma-Aldrich) final yield (µg)  RiboAmp HS (Molecular Devices) final yield (µg)  RampUP (Genisphere) final yield (µg)  Oocyte_pool 1  7.1  5.8  51.2  9.6  Oocyte_pool 2  5.8  5.8  53.9  9.6  Oocyte_pool 3  6.7  5.6  52.3  18.4  Samples  Ovation RNA Amplification System V2 RiboSPIA (NuGEN) final yield (µg)  TransPlex Whole Transcriptome Amplification (Sigma-Aldrich) final yield (µg)  RiboAmp HS (Molecular Devices) final yield (µg)  RampUP (Genisphere) final yield (µg)  Oocyte_pool 1  7.1  5.8  51.2  9.6  Oocyte_pool 2  5.8  5.8  53.9  9.6  Oocyte_pool 3  6.7  5.6  52.3  18.4  View Large Figure 5 View largeDownload slide Size distribution of amplified products from different transcriptome amplification methods. (A) RiboSpia (NuGen); (B) RiboAmp HS (Molecular Devices); (C) TransPlex Whole Transcriptome (Sigma) and (D) RampUP (Genisphere). Size profiling was performed by using the 2100 BioAnalyzer apparatus (Agilent). For methods B and D, a two round amplification is required. Size distributions following the second round are presented. Figure 5 View largeDownload slide Size distribution of amplified products from different transcriptome amplification methods. (A) RiboSpia (NuGen); (B) RiboAmp HS (Molecular Devices); (C) TransPlex Whole Transcriptome (Sigma) and (D) RampUP (Genisphere). Size profiling was performed by using the 2100 BioAnalyzer apparatus (Agilent). For methods B and D, a two round amplification is required. Size distributions following the second round are presented. The kinetic of three of these amplification strategies were determined by following product generation in real-time. Due to the specific nature of the RiboSPIA output (single-stranded DNA), the kinetic could not be assessed. The two in vitro transcription driven methods (RiboAmp HS and RampUP) are advertized as being linear since the amplification is primed on only one end rather than on both, as it is the case with PCR. Real-time measurements of fragment generation provided evidences that these methods are not linear and reach a plateau phase fairly quickly (Fig. 6). The Transplex WT PCR conditions provided by the manufacturer are efficient in keeping the reaction from reaching the known plateau phase. At completion (17 cycles), the kinetics only started to slow down, indicating that the linear phase was maximized, as predicted by the manufacturer. Figure 6 View largeDownload slide Kinetics of transcriptome amplifications. For methods requiring two amplification rounds, only the final round is presented. (A) RiboAmp HS (Molecular Devices). The kinetic was followed in real time using a labeled broken-beacon targeting the bovine actin, beta (ACTB) antisense strand. (B) RampUP (Genispheres). The kinetic was followed using a broken-beacon targeting the bovine ACTB sense strand. (C) TransPlex WT (Sigma). The kinetic was followed in real-time using SybrGreen. All reactions were followed from start to completion following the manufacturer's instructions. Figure 6 View largeDownload slide Kinetics of transcriptome amplifications. For methods requiring two amplification rounds, only the final round is presented. (A) RiboAmp HS (Molecular Devices). The kinetic was followed in real time using a labeled broken-beacon targeting the bovine actin, beta (ACTB) antisense strand. (B) RampUP (Genispheres). The kinetic was followed using a broken-beacon targeting the bovine ACTB sense strand. (C) TransPlex WT (Sigma). The kinetic was followed in real-time using SybrGreen. All reactions were followed from start to completion following the manufacturer's instructions. The amplified products were labeled and hybridized on microarrays (Fig. 7). The overall intensity signals were compared between every amplification methods. Surprisingly, the highest correlation (R = 0.69) was obtained between both DNA polymerase-based methods (TransPlex WT and RiboSPIA). When both RNA polymerase methods were compared, the correlation value only reached 0.29. In fact, aside for the above mentioned mildly acceptable correlation value, all tested amplification strategies generated samples that share very weak similarities. Figure 7 View largeDownload slide Scatter plots of spot intensities resulting from microarray samples prepared using different global amplification methods. Correlation values are indicated in the upper right corner. Labeling was performed according to the nature of the sample (RNA, single-stranded or double-stranded DNA). Hybridization conditions were identical for all samples. Figure 7 View largeDownload slide Scatter plots of spot intensities resulting from microarray samples prepared using different global amplification methods. Correlation values are indicated in the upper right corner. Labeling was performed according to the nature of the sample (RNA, single-stranded or double-stranded DNA). Hybridization conditions were identical for all samples. Hybridization sample labeling To label the amplified material to be hybridized on the microarray, different options are available. These options can be categorized as direct incorporation, where the fluorophore is directly coupled to one of the nucleotide during amplification or duplication of the material, or indirect incorporation, where the fluorophore is added chemically to the output of the amplification. We tested one direct and two indirect labeling methods (Fig. 8). The direct incorporation of nucleotides was performed during the reverse transcription of the amplified RNA samples. As observed and well documented (Dobbin et al., 2005; Lu et al., 2008), direct dye incorporation resulted in unbalanced labeling due to the size and charge differences between the two dyes (data not shown). Due to extensive sample handling and presence of an important dye bias, this labeling strategy was not further tested and alternative methods were considered. Figure 8 View largeDownload slide Self-to-self hybridization to determine the impact of dye incorporation. (A) Indirect labeling by incorporation of amino-allyl groups containing nucleotides during the global amplification followed by chemical coupling of the dyes (Alexa 555 and 647). (B to C) Direct coupling of dyes (Cy3 and Cy5) using the Universal Linkage System (ULS) following the global amplification. (B) TURBO labeling kit (Molecular Devices); (C) ULS aRNA labeling kit for Agilent (Kreatech Diagnostics). Figure 8 View largeDownload slide Self-to-self hybridization to determine the impact of dye incorporation. (A) Indirect labeling by incorporation of amino-allyl groups containing nucleotides during the global amplification followed by chemical coupling of the dyes (Alexa 555 and 647). (B to C) Direct coupling of dyes (Cy3 and Cy5) using the Universal Linkage System (ULS) following the global amplification. (B) TURBO labeling kit (Molecular Devices); (C) ULS aRNA labeling kit for Agilent (Kreatech Diagnostics). There are two main types of indirect labeling. The first incorporates an amino-allyl modified nucleotide that serves as an anchor for the fluorophore during a downstream chemical addition, while the second, which does not require the introduction of modified nucleotides during the sample amplification step, utilizes the nitrous atom in position 7 of the guanine base to chemically bind a fluorophore coupled to a reactive group (Universal Linkage System (ULS)), which is a derivative of the anticancer agent cisplatin. The incorporation of amino-allyl modified nucleotides resulted in considerable reduction of the amplification output (about 30 versus 50 µg). However, the method proved to be an efficient means of labeling (Table IV and Fig. 8). When amplifying samples using RiboAmp, the reduction in yield during sample amplification is a minor negative impact considering the requirement of 1–2 µg per microarray hybridization. It may underperform with other means of sample amplification that result in much lower yields. Table IV Dye incorporation efficiency from different methods aiming to label amplified nucleic acid samples. Method  Dye  Number of trials  Average incorporation yield  SEM  Direct incorporation  Cy3  3  27.8  4.3  Cy5  3  27.5  9.0  Indirect amino-allyl  Alexa-555  8  31.9  3.0  Alexa-647  8  28.1  1.9  Indirect ULS (Turbo)  Cy3  3  21.3  1.8  Cy5  3  13.5  1.2  Indirect ULS (Ulysis)  Alexa-555  3  15.9  2.8  Alexa-647  3  19.5  0.4  Indirect ULS (Kreatech)  Alexa-555  10  44.5  3.4  Alexa-647  10  29.1  1.9  Indirect ULS (Kreatech)  Cy3  3  28.9  5.2  Cy5  3  25.6  2.1  Method  Dye  Number of trials  Average incorporation yield  SEM  Direct incorporation  Cy3  3  27.8  4.3  Cy5  3  27.5  9.0  Indirect amino-allyl  Alexa-555  8  31.9  3.0  Alexa-647  8  28.1  1.9  Indirect ULS (Turbo)  Cy3  3  21.3  1.8  Cy5  3  13.5  1.2  Indirect ULS (Ulysis)  Alexa-555  3  15.9  2.8  Alexa-647  3  19.5  0.4  Indirect ULS (Kreatech)  Alexa-555  10  44.5  3.4  Alexa-647  10  29.1  1.9  Indirect ULS (Kreatech)  Cy3  3  28.9  5.2  Cy5  3  25.6  2.1  View Large Two different manufacturers of ULS-based labeling kits were tested. Overall, this labeling strategy performed extremely well. The main considerations to select the appropriate labeling method were labeling efficiency, cost per sample, ease of use of the protocol and quality of hybridization as determined by the level of unspecific binding outside the spotted areas of the microarray, as well as sresults from the dye bias test. To test this last criterion, self-hybridizations were conducted for all of the indirect labeling strategies. The correlation coefficients indicate that the incorporation of amino-allyl UTP during the amplification step, followed by coupling of the Alexa fluors, provides a reliable method (r = 0.99) (Fig. 8). The labeling of aRNA using an ULS system coupled to Cye dyes resulted in lower but still very acceptable correlation values (0.94 and 0.97) (Fig. 8). The samples labeled with the Turbo labeling system (Molecular Devices) generated higher background (data not shown) which we believe is caused by the sample clean up step that is designed to remove the unincorporated dyes. Since all the other kits also use size-exclusion columns to perform this task, we modified the sample clean up step by replacing the columns with standard ion-exchange columns (PicoPure, Molecular Devices). The background problem was not observed following this modification. Sample hybridization and washing The quality of sample hybridization depends on two main aspects, e.g. the ability of the sample's labeled molecules to find their immobilized complementary targets (specificity) and the lack of unspecific binding, which is visualized by the presence of fluorescence observed in the DNA-free areas of the microarray. Binding specificity can be optimized by modifying the hybridization temperature, changing buffer salt contents and varying the stringency of the post-hybridization washes. We chose to optimize the quality of hybridization by comparing the performance of different buffers. Aliquots of a labeled sample were mixed in six different hybridization buffers and allowed to hybridize on our microarray under the recommended conditions. Buffer composition and washing conditions are detailed in Table I. The microarray buffer test kit offered by Applied Biosystems (SlideHyb buffer 1, 2, 3 and 4) was utilized to highlight the impact of buffer composition on results. The SlideHyb 1 was first tested at different hybridization temperatures and the one providing the best signal-to-noise results was set to be the buffer testing temperature for other SlideHyb buffers. From Fig. 9, it is possible to appreciate the diversity in background signals generated from the same hybridization sample. Amongst the tested buffer, SlideHyb buffer 1 gave the best target signal intensities combined with a low overall background. The reference condition (panel G) performed in a static chamber resulted in lower overall signals (observable by the higher background) indicating movement during hybridization provides better results. Figure 9 View largeDownload slide Impact of hybridization buffer on the prevalence of unspecific background signal. All hybridizations were conducted in a microarray hybridization station aside from panel (G) which is a control hybridization performed in a Corning chamber submerged in a water bath. (A–D) SlideHyb 1, 2, 3 and 4, respectively (Applied Biosystems); (E) ArrayBooster buffer (Advalytix); (F) DIG buffer (Roche); (E) DIG control in hybridization chamber. Temperature conditions can be found in Table I. Figure 9 View largeDownload slide Impact of hybridization buffer on the prevalence of unspecific background signal. All hybridizations were conducted in a microarray hybridization station aside from panel (G) which is a control hybridization performed in a Corning chamber submerged in a water bath. (A–D) SlideHyb 1, 2, 3 and 4, respectively (Applied Biosystems); (E) ArrayBooster buffer (Advalytix); (F) DIG buffer (Roche); (E) DIG control in hybridization chamber. Temperature conditions can be found in Table I. Impact of reverse transcriptase source on RNA abundance measurements Following hybridization, microarray data are pre-processed for normalization and analyzed through different statistical means to generate lists of differentially abundant candidates between the biological contrasts of interest. Estimation of the validity of gene lists is generally performed by measuring the abundance of selected candidates by quantitative RT–PCR. In this method, as it is the case for nearly all procedures quantifying or amplifying RNA samples, reverse transcription is a pivotal step that can profoundly influence downstream results. The main issue arises from the fact that reverse transcriptases lack robustness and easily terminate their progression before reaching the 5′ end of the transcript. Manufacturers provide different sources of reverse transcriptase supplied with optimized conditions. Interestingly, some enzymes are designed to handle low input material, which is of importance when working with oocytes and early embryos. We tested the impact of the source of the reverse transcriptase on the RNA abundance assessment of two housekeeping genes (Fig. 10). Technical replication was performed from a common RNA pool extracted from 50 GV oocytes where identical aliquots (1 GV oocyte equivalent) served as the source of templates. This ensured that the results were solely representative of the efficiency of the reverse transcriptase in handling minute samples. We observed that the source of the reverse transcriptase has a significant impact on the following RNA abundance measurements. Under typical conditions for early development studies, the qScript performed with the highest efficiency, closely matched by the Sensiscript. The two other tested enzymes resulted in lower quantification. Figure 10 View largeDownload slide Impact of the source of the reverse transcriptase on transcript quantification during data validation by real-time PCR. A single GV stage oocyte equivalent was used for all replicates. Two gene candidates (actin, beta (ACTB) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were used to avoid sequence-specific preference. Different letters are representative of significant statistical difference (P < 0.05). Figure 10 View largeDownload slide Impact of the source of the reverse transcriptase on transcript quantification during data validation by real-time PCR. A single GV stage oocyte equivalent was used for all replicates. Two gene candidates (actin, beta (ACTB) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) were used to avoid sequence-specific preference. Different letters are representative of significant statistical difference (P < 0.05). Discussion The study of embryonic development or of the impact of the in vitro culture environment on RNA abundance levels using microarray platforms is common practice and has been reported in the literature for several years (Hamatani et al., 2004; Hamatani et al., 2006; Vallee et al., 2009; Zhang et al., 2009). Currently, the integration of the data produced by different microarray platforms has proven to be extremely challenging and the numerous important discrepancies in gene lists suggest that methods and sample handling have a profound impact on results. Technical imprecision leads to data variation, which in turn reduces the potential to find statistically significant differences between the biological contrasts being studied, which could explain the result discrepancies between published reports. However, the observed divergences are so profound that we believe that the different techniques being used to prepare samples also have a deep impact on these differences. In order to determine to what extent these sources of variations impact our own microarray platform and to provide users with standardized and optimized procedures that are best adapted to the peculiar context of oocytes and early embryos, we undertook a series of tests to compare different methodological options for microarray hybridization sample preparation as well as for the following quantitative RT–PCR validation step. Method selection was based on various criteria that are of prime importance in the context of a multi-users platform. We sought to provide a simple handling pipeline with a minimal number of steps as each can become a source of sample loss and thus imprecision. The efficiency and repeatability of each method were considered and to a lesser extent, the cost per sample and time requirements. We observed that every step of the sample handling pipeline is important. As such, even buffer composition for oocytes/embryos collection can impact downstream RNA abundance measurements. Indeed, the addition of polyvinyl alcohol (PVA) to the oocytes/embryos wash medium during sample collection negatively impacts the yield of the subsequent total RNA extraction (data not shown). However, we also observed that some steps can have a more profound impact on microarray results than others. We classified the impact of these technical sources of variation in two main categories: (i) loss in quality and repeatability, (ii) loss of physiological relevance of the microarray results. The first category is directly linked with protocol optimization, whereas the second is tied to the peculiar RNA management context that is pre-hatching development. This developmental window is characterized by dramatic fluctuation in the nature of the biological samples; while cell number increases, their size diminishes, total RNA content fluctuates, so does its proportions in the diverse type of RNAs (mRNA, rRNAs, tRNAs, miRNAs, etc.) (Gilbert et al., 2009b). Furthermore, this period includes the maternal-embryonic transition (MET) (Braude et al., 1988; Schultz 1993; Memili et al., 1998; Wang et al., 2000), which consequently implies two distinct states of transcriptional potential for these cells. Prior to the MET, de novo protein synthesis is sustained by the translation of maternal RNA stocks stored in the gamete during its growth (Pratt et al., 1983; Fair et al., 1997; Paynton 1998). It is well accepted that in these transcriptionally incompetent blastomeres, the fate of the maternal mRNAs is managed through shortening and lengthening of the poly(A) tail (Bachvarova et al., 1985; Wickens 1990; Bachvarova 1992; Paynton et al., 1994). This phenomenon represents an important aspect to consider when selecting the RNA extraction procedure, as well as the priming strategy for the reverse transcription step. For instance, an extraction strategy either aiming at the total RNA fraction or at the poly(A) bearing mRNAs will result in different transcript abundance measurements due to the isolation of a different subpopulation of messengers. To select the total RNA extraction method, yield is an essential criterion. As shown, all column-based RNA extraction methods are not equal as yield was affected by the harshness of the cell dispersion buffer. Since the transcriptome is composed of various RNA types that differ in length and that ion-exchange columns have a size cut-off limit, we hypothesized that the higher yield with Trizol reagent could be caused by the presence of small RNA molecules (miRNAs, siRNA, piRNA, tRNAs, snRNAs, snoRNa, the small rRNAs) that most column-based extraction methods fail to isolate. To test if the mRNA populations were affected by the RNA extraction procedures, candidates were analyzed by quantitative RT–PCR. Quantification showed a close similarity between methods, indicating that the mRNA population is negligibly affected. These tests also revealed that column-based extractions methods provide stability when compared with the standard phenol–chloroform-based approach. Repeatability was negatively affected when the procedure necessitated a precipitation step. Sample concentration through liquid evaporation (Speedvac) was also tested but resulted in partial degradation in some samples (data not shown). Another parameter considered for the choice of the total RNA extraction method was the ease with which sample could be treated to eliminate contaminating genomic DNA. For all column-based extractions, the DNase treatment can be easily performed on the silica membrane, which allows for easy clean up of the reaction prior to elution. With the Trizol protocol, the DNase treatment is performed following resuspension of the RNA pellet. The DNAse can then be inactivated but complete removal would require another RNA extraction. From these tests, we implemented the PicoPure total RNA extraction procedure (Molecular Devices) in our microarray platform for its efficiency as determined by the yield, repeatability, ease of use for genomic DNA removal and low elution volume that eliminates the need for a sample concentration step. Three methods aiming at the isolation of specific RNA sub-fractions were also tested. We first noticed that the recommended input material and downstream reaction volumes were not suitable when working with oocytes and early embryos. Nevertheless, these methods were tested with somatic RNA samples and proved to perform marginally. Due to poor performance and the fact that isolation of specific RNA subfractions could impact the interpretation of the data set, since mRNAs are known to be found under diverse adenylation states that fluctuate during pre-hatching development, it was decided to work on total RNA samples to prevent the introduction of a bias related to the RNA extraction procedure. The sample handling step that has the most impact on compatibility between platforms is no doubt the global amplification step that is inevitable when working with mammalian oocytes and early embryos. The correlation between data sets produced from the diverse amplified samples was unexpectedly very low. These discrepancies are also represented by very different lists when ranking the top 20 candidates with the highest intensity signals for each amplification method (Supplementary data, Table S1). The weak similarities between data sets obtained from different global amplification strategies may arise from the very distinct approaches used by the reverse transcription step, in addition to the amplification strategy itself, be it either PCR or IVT based. This translates into the important compatibility issue that platforms face. It is worth mentioning that the choice of the amplification method must take into consideration the introduction of biases that may profoundly impact the value of downstream microarray data. The amplification step needs to preserve the relative abundance of transcripts within the sample so they stay similar to the unamplified sample. Several reports have compared unamplified and amplified samples from in vitro transcription-based methods (Pabon et al., 2001; Stirewalt et al., 2004; Patel et al., 2005; van Haaften et al., 2006; Zhu et al., 2006), as well as from PCR-based approaches (Puskas et al., 2002; Subkhankulova et al., 2006; Wilhelm et al., 2006; Gosselink et al., 2007). We recently showed that a more profound bias is introduced by the amplification step when samples naturally bear a different RNA content, which is often the case with oocytes and early embryos (Gilbert et al., 2009a). The bias arises from the homogenization of the output that occurs when the reaction reaches its plateau phase. Herein, we were able to follow the amplification kinetics for three of the four amplification strategies. As we reported previously, IVT-based methods reach a plateau phase relatively early following the start of the incubation. By contrast, the PCR-based method was clearly optimized to stay within the linear phase of the reaction. It is interesting that results for both DNA polymerase strategies somewhat correlated since the Transplex WT method utilizes random priming whereas RiboSPIA is based on the presence of a poly(A) tail and therefore targets mRNAs. It was expected that both approaches would amplify distinct RNA populations since the length of the poly(A) tail fluctuates for a proportion of mRNAs during development. It was rather expected that both IVT-based methods would provide better correlation values. However, our tests demonstrate otherwise. The exact reason for such low correlation is unknown. The other downstream steps, e.g. sample labeling, hybridization and microarray wash fall into the first category of impacts (reduction of variance through increased repeatability and quality) since they were not identified as potentially influencing the physiological value of microarray data. Optimization of these steps is inevitable and part of standard procedures for any microarray platform. Due to the considerable pitfall potential that is mining the process of microarray data generation, validation of gene lists must be conducted using a different approach. As such, quantitative RT–PCR has been shown to be a valuable tool. Although sample handling is much more limited in this quantification assay, common methodology is equally important, notably for RNA extraction and reverse transcription. We have tested different sources of reverse transcriptase and showed that abundance measurement can be influenced by the efficacy of the enzyme. The results obtained in the present study led to the selection of specific methods for our own microarray platform. Our first decision was to use the Picopure RNA isolation kit (Molecular Devices), which was specially designed for low RNA content samples and allows a low elution volume. Results using this kit have proven to be highly stable and this kit offers ease of use by allowing the DNAse I treatment to occur directly on the column. The hardest choice to make unquestionably involves the amplification strategy. No kits on the market are fully adapted for the study of early embryonic development. The use of random primers during the initial reverse transcription step has the benefit of overcoming the 3′ bias observed with oligo-dT priming. On the other hand, random priming is directly influenced by the amount of templates and the amount of short primers in the solution. With low template abundance, the addition of too many random primers will lead to very short cDNAs, which negatively influence downstream processing, whereas an insufficient amount will lead to sub-optimized yield. Thus, the efficacy of the random priming reaction follows a bell curve shape that depends on the amount of templates in the solution. To make matters worse, in studies involving across-stage comparisons, the fluctuation in RNA content would necessitate a titration assay for each stage. In addition, the use of random primers will target stored mRNAs, which are physiologically inert. Given all the issues related to the use of random primers, we decided to move away from this type of reverse transcription priming. We are therefore advocating the use of an amplification method using an oligo-dT to prime reverse transcription. To account for the 3′ bias created by this method, we design the oligos to be spotted on the microarray to be located at a definite distance from the poly(A) tail. The linearity of the reaction, another concern with oligo-dT priming, is only an issue when comparing samples of different nature, such as different stage embryos. With same stage samples, it is expected that the plateau phase will be reached fairly at the same time, therefore preventing the introduction of a large bias. It is however a very important aspect to consider for across-stage comparisons. We have tackled this issue by developing a means to follow the IVT reaction in real time (Gilbert et al., 2009a). Although downstream procedures are less prone to introduce biases and deeply influence results, we prefer the use of ULS for labeling for its yield and ease of use. The choice of hybridization buffer will vary depending on the microarray platform. Finally, for qRT–PCR validation, qScript (Quanta Biosciences) and Sensiscript (Qiagen) reverse transcriptases are excellent choices for working with limited samples. Conclusion In light of the herein results, it is clear that in order to compare microarray data, a careful analysis of the methods must be conducted to determine their level of compatibility. In our experience, the most problematic steps are RNA extraction and the priming strategy for the reverse transcription reaction (random versus oligo-dT), which can lead to the study of different mRNA populations. Global amplification is no doubt the step that raises the most concern. The loss of physiological relevance inherent to a procedure needs to be considered when appreciating the value of a microarray data set. The study of early embryo development would benefit from the integration of the numerous microarray data set that are being published. Per se, relative quantification should enable a basis of comparison across microarray platforms considering that, for any comparisons performed using a specific set of methods, both samples should be equally influenced by the methods thus preserving relative values. Keeping the impact of methodologies in perspective may enable the comparison of data set that share a common methodological basis, especially for the key steps identified herein. Keeping in mind the goal to eliminate compatibility issues between studies, a Canadian funded network (www.EmbryoGENE.ca) was recently established to study embryonic gene expression. Conclusions derived from these tests will not only provide our network with standardized methods for our new embryo-specific long oligo-microarray but will hopefully raise the awareness of the scientific community on the different determining steps of microarray sample preparation. Author roles' I.G. was involved in every aspects of this project, produced a large part of the results and prepared the manuscript. S.S. did the work to generate Fig. 4. E-L.S. performed the statistical analyses. I.D. worked on Fig. 8 and generated the results for Fig. 10. M.-A.S. was involved in the conceptualization of this project. C.R. was involved in the conceptualization of this project, the preparation of the manuscript and provided mentorship. Supplementary data Supplementary data is available at http://molehr.oxfordjournals.org/. Funding This project was supported by funds from the Natural Sciences and Engineering Research Council of Canada and from Fond québécois de la recherche sur la nature et les technologies. Acknowledgements The authors want to thank Isabelle Laflamme and Catherine Gravel for sample collection and technical assistance. We would also like to thank Julie Nieminen for critical review of the manuscript. References Albert TJ,  Norton J,  Ott M,  Richmond T,  Nuwaysir K,  Nuwaysir EF,  Stengele KP,  Green RD.  Light-directed 5'– > 3' synthesis of complex oligonucleotide microarrays,  Nucleic Acids Res ,  2003, vol.  31 pg.  e35  https://doi.org/10.1093/nar/gng035 Google Scholar CrossRef Search ADS PubMed  Bachvarova RF.  A maternal tail of poly(A): the long and the short of it,  Cell ,  1992, vol.  69 (pg.  895- 897) https://doi.org/10.1016/0092-8674(92)90606-D Google Scholar CrossRef Search ADS PubMed  Bachvarova R,  De Leon V,  Johnson A,  Kaplan G,  Paynton BV.  Changes in total RNA, polyadenylated RNA, and actin mRNA during meiotic maturation of mouse oocytes,  Dev Biol ,  1985, vol.  108 (pg.  325- 331) https://doi.org/10.1016/0012-1606(85)90036-3 Google Scholar CrossRef Search ADS PubMed  Barczak A,  Rodriguez MW,  Hanspers K,  Koth LL,  Tai YC,  Bolstad BM,  Speed TP,  Erle DJ.  Spotted long oligonucleotide arrays for human gene expression analysis,  Genome Res ,  2003, vol.  13 (pg.  1775- 1785) https://doi.org/10.1101/gr.1048803 Google Scholar CrossRef Search ADS PubMed  Braude P,  Bolton V,  Moore S.  Human gene expression first occurs between the four- and eight-cell stages of preimplantation development,  Nature ,  1988, vol.  332 (pg.  459- 461) https://doi.org/10.1038/332459a0 Google Scholar CrossRef Search ADS PubMed  Brown PO,  Botstein D.  Exploring the new world of the genome with DNA microarrays,  Nat Genet ,  1999, vol.  21  Suppl.(pg.  33- 37) [PubMed] Google Scholar CrossRef Search ADS PubMed  Carter MG,  Hamatani T,  Sharov AA,  Carmack CE,  Qian Y,  Aiba K,  Ko NT,  Dudekula DB,  Brzoska PM,  Hwang SS, et al.  In situ-synthesized novel microarray optimized for mouse stem cell and early developmental expression profiling,  Genome Res ,  2003, vol.  13 (pg.  1011- 1021) https://doi.org/10.1101/gr.878903 Google Scholar CrossRef Search ADS PubMed  Dobbin KK,  Kawasaki ES,  Petersen DW,  Simon RM.  Characterizing dye bias in microarray experiments,  Bioinformatics ,  2005, vol.  21 (pg.  2430- 2437) https://doi.org/10.1093/bioinformatics/bti378 Google Scholar CrossRef Search ADS PubMed  Fair T,  Hulshof SC,  Hyttel P,  Greve T,  Boland M.  Nucleus ultrastructure and transcriptional activity of bovine oocytes in preantral and early antral follicles,  Mol Reprod Dev ,  1997, vol.  46 (pg.  208- 215) https://doi.org/10.1002/(SICI)1098-2795(199702)46:2<208::AID-MRD11>3.0.CO;2-X Google Scholar CrossRef Search ADS PubMed  Gilbert I,  Scantland S,  Dufort I,  Gordynska O,  Labbe A,  Sirard MA,  Robert C.  Real-time monitoring of aRNA production during T7 amplification to prevent the loss of sample representation during microarray hybridization sample preparation,  Nucleic Acids Res ,  2009, vol.  37 pg.  e65  https://doi.org/10.1093/nar/gkp193 Google Scholar CrossRef Search ADS PubMed  Gilbert I,  Scantland S,  Sylvestre EL,  Gravel C,  Laflamme I,  Sirard MA,  Robert C.  The dynamics of gene products fluctuation during bovine pre-hatching development,  Mol Reprod Dev ,  2009, vol.  76 (pg.  762- 772) https://doi.org/10.1002/mrd.21030 Google Scholar CrossRef Search ADS PubMed  Gosselink JV,  Hayashi S,  Chau E,  Cooper J,  Elliott WM,  Hogg JC.  Evaluation of small sample cDNA amplification for microdissected airway expression profiling in COPD,  COPD ,  2007, vol.  4 (pg.  91- 105) https://doi.org/10.1080/15412550701246427 Google Scholar CrossRef Search ADS PubMed  Hamatani T,  Carter MG,  Sharov AA,  Ko MS.  Dynamics of global gene expression changes during mouse preimplantation development,  Dev Cell ,  2004, vol.  6 (pg.  117- 131) https://doi.org/10.1016/S1534-5807(03)00373-3 Google Scholar CrossRef Search ADS PubMed  Hamatani T,  Ko M,  Yamada M,  Kuji N,  Mizusawa Y,  Shoji M,  Hada T,  Asada H,  Maruyama T,  Yoshimura Y.  Global gene expression profiling of preimplantation embryos,  Hum Cell ,  2006, vol.  19 (pg.  98- 117) https://doi.org/10.1111/j.1749-0774.2006.00018.x Google Scholar CrossRef Search ADS PubMed  Hughes TR,  Mao M,  Jones AR,  Burchard J,  Marton MJ,  Shannon KW,  Lefkowitz SM,  Ziman M,  Schelter JM,  Meyer MR, et al.  Expression profiling using microarrays fabricated by an ink-jet oligonucleotide synthesizer,  Nat Biotechnol ,  2001, vol.  19 (pg.  342- 347) https://doi.org/10.1038/86730 Google Scholar CrossRef Search ADS PubMed  Lee M,  Trent JM,  Bittner ML.  Optimization of oligonucleotide microarray fabricated by spotting 65-mer,  Anal Biochem ,  2007, vol.  368 (pg.  61- 69) https://doi.org/10.1016/j.ab.2007.06.005 Google Scholar CrossRef Search ADS PubMed  Lipshutz RJ,  Fodor SP,  Gingeras TR,  Lockhart DJ.  High density synthetic oligonucleotide arrays,  Nat Genet ,  1999, vol.  21  Suppl.(pg.  20- 24) [PubMed] Google Scholar CrossRef Search ADS PubMed  Lu R,  Lee GC,  Shultz M,  Dardick C,  Jung K,  Phetsom J,  Jia Y,  Rice RH,  Goldberg Z,  Schnable PS, et al.  Assessing probe-specific dye and slide biases in two-color microarray data,  BMC Bioinformatics ,  2008, vol.  9 (pg.  314- 328) https://doi.org/10.1186/1471-2105-9-314 Google Scholar CrossRef Search ADS PubMed  Memili E,  Dominko T,  First NL.  Onset of transcription in bovine oocytes and preimplantation embryos,  Mol Reprod Dev ,  1998, vol.  51 (pg.  36- 41) https://doi.org/10.1002/(SICI)1098-2795(199809)51:1<36::AID-MRD4>3.0.CO;2-X Google Scholar CrossRef Search ADS PubMed  Mignone F,  Gissi C,  Liuni S,  Pesole G.  Untranslated regions of mRNAs,  Genome Biol ,  2002, vol.  3 (pg.  0004.1- 0004.10) Google Scholar CrossRef Search ADS   Nuwaysir EF,  Huang W,  Albert TJ,  Singh J,  Nuwaysir K,  Pitas A,  Richmond T,  Gorski T,  Berg JP,  Ballin J, et al.  Gene expression analysis using oligonucleotide arrays produced by maskless photolithography,  Genome Res ,  2002, vol.  12 (pg.  1749- 1755) https://doi.org/10.1101/gr.362402 Google Scholar CrossRef Search ADS PubMed  Pabon C,  Modrusan Z,  Ruvolo MV,  Coleman IM,  Daniel S,  Yue H,  Arnold LJJr.  Optimized T7 amplification system for microarray analysis,  Biotechniques ,  2001, vol.  31 (pg.  874- 879) [PubMed] Google Scholar PubMed  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, vol.  105 (pg.  331- 342) https://doi.org/10.1016/j.vetimm.2005.02.018 Google Scholar CrossRef Search ADS PubMed  Paynton BV.  RNA-binding proteins in mouse oocytes and embryos: expression of genes encoding Y box, DEAD box RNA helicase, and polyA binding proteins,  Dev Genet ,  1998, vol.  23 (pg.  285- 298) https://doi.org/10.1002/(SICI)1520-6408(1998)23:4<285::AID-DVG4>3.0.CO;2-W Google Scholar CrossRef Search ADS PubMed  Paynton BV,  Bachvarova R.  Polyadenylation and deadenylation of maternal mRNAs during oocyte growth and maturation in the mouse,  Mol Reprod Dev ,  1994, vol.  37 (pg.  172- 180) https://doi.org/10.1002/mrd.1080370208 Google Scholar CrossRef Search ADS PubMed  Pennie WD.  Use of cDNA microarrays to probe and understand the toxicological consequences of altered gene expression,  Toxicol Lett ,  2000, vol.  112–113 (pg.  473- 477) https://doi.org/10.1016/S0378-4274(99)00199-X Google Scholar CrossRef Search ADS PubMed  Pratt HP,  Bolton VN,  Gudgeon KA.  The legacy from the oocyte and its role in controlling early development of the mouse embryo,  Ciba Found Symp ,  1983, vol.  98 (pg.  197- 227) [PubMed] Google Scholar PubMed  Puskas LG,  Zvara A,  Hackler LJr,  Van Hummelen P.  RNA amplification results in reproducible microarray data with slight ratio bias,  Biotechniques ,  2002, vol.  32 (pg.  1330- 1340) [PubMed] Google Scholar PubMed  Schultz RM.  Regulation of zygotic gene activation in the mouse,  Bioessays ,  1993, vol.  15 (pg.  531- 538) https://doi.org/10.1002/bies.950150806 Google Scholar CrossRef Search ADS PubMed  Stirewalt DL,  Pogosova-Agadjanyan EL,  Khalid N,  Hare DR,  Ladne PA,  Sala-Torra O,  Zhao LP,  Radich JP.  Single-stranded linear amplification protocol results in reproducible and reliable microarray data from nanogram amounts of starting RNA,  Genomics ,  2004, vol.  83 (pg.  321- 331) https://doi.org/10.1016/j.ygeno.2003.08.008 Google Scholar CrossRef Search ADS PubMed  Subkhankulova T,  Livesey FJ.  Comparative evaluation of linear and exponential amplification techniques for expression profiling at the single-cell level,  Genome Biol ,  2006, vol.  7 pg.  R18  https://doi.org/10.1186/gb-2006-7-3-r18 Google Scholar CrossRef Search ADS PubMed  Vallee M,  Robert C,  Methot S,  Palin MF,  Sirard MA.  Cross-species hybridizations on a multi-species cDNA microarray to identify evolutionarily conserved genes expressed in oocytes,  BMC Genomics ,  2006, vol.  7 pg.  113  https://doi.org/10.1186/1471-2164-7-113 Google Scholar CrossRef Search ADS PubMed  Vallee M,  Dufort I,  Desrosiers S,  Labbe A,  Gravel C,  Gilbert I,  Robert C,  Sirard MA.  Revealing the bovine embryo transcript profiles during early in vivo embryonic development,  Reproduction ,  2009, vol.  138 (pg.  95- 105) https://doi.org/10.1530/REP-08-0533 Google Scholar CrossRef Search ADS PubMed  van Haaften RI,  Schroen B,  Janssen BJ,  van Erk A,  Debets JJ,  Smeets HJ,  Smits JF,  van den Wijngaard A,  Pinto YM,  Evelo CT.  Biologically relevant effects of mRNA amplification on gene expression profiles,  BMC Bioinformatics ,  2006, vol.  7 pg.  200  https://doi.org/10.1186/1471-2105-7-200 Google Scholar CrossRef Search ADS PubMed  Vigneault C,  Gilbert I,  Sirard MA,  Robert C.  Using the histone H2a transcript as an endogenous standard to study relative transcript abundance during bovine early development,  Mol Reprod Dev ,  2007, vol.  74 (pg.  703- 715) https://doi.org/10.1002/mrd.20665 Google Scholar CrossRef Search ADS PubMed  Vigneault C,  McGraw S,  Massicotte L,  Sirard MA.  Transcription factor expression patterns in bovine in vitro-derived embryos prior to maternal-zygotic transition,  Biol Reprod ,  2004, vol.  70 (pg.  1701- 1709) https://doi.org/10.1095/biolreprod.103.022970 Google Scholar CrossRef Search ADS PubMed  Wang Q,  Latham KE.  Translation of maternal messenger ribonucleic acids encoding transcription factors during genome activation in early mouse embryos,  Biol Reprod ,  2000, vol.  62 (pg.  969- 978) https://doi.org/10.1095/biolreprod62.4.969 Google Scholar CrossRef Search ADS PubMed  Wickens M.  In the beginning is the end: regulation of poly(A) addition and removal during early development,  Trends Biochem Sci ,  1990, vol.  15 (pg.  320- 324) https://doi.org/10.1016/0968-0004(90)90022-4 Google Scholar CrossRef Search ADS PubMed  Wilhelm J,  Muyal JP,  Best J,  Kwapiszewska G,  Stein MM,  Seeger W,  Bohle RM,  Fink L.  Systematic comparison of the T7-IVT and SMART-based RNA preamplification techniques for DNA microarray experiments,  Clin Chem ,  2006, vol.  52 (pg.  1161- 1167) https://doi.org/10.1373/clinchem.2005.062406 Google Scholar CrossRef Search ADS PubMed  Zhang P,  Zucchelli M,  Bruce S,  Hambiliki F,  Stavreus-Evers A,  Levkov L,  Skottman H,  Kerkela E,  Kere J,  Hovatta O.  Transcriptome profiling of human pre-implantation development,  PLoS One ,  2009, vol.  4 pg.  e7844  https://doi.org/10.1371/journal.pone.0007844 Google Scholar CrossRef Search ADS PubMed  Zhu B,  Xu F,  Baba Y.  An evaluation of linear RNA amplification in cDNA microarray gene expression analysis,  Mol Genet Metab ,  2006, vol.  87 (pg.  71- 79) https://doi.org/10.1016/j.ymgme.2005.08.007 Google Scholar CrossRef Search ADS PubMed  © The Author 2010. Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved. For Permissions, please email: journals.permissions@oxfordjournals.org
TI - Providing a stable methodological basis for comparing transcript abundance of developing embryos using microarrays
JF - Molecular Human Reproduction
DO - 10.1093/molehr/gaq038
DA - 2010-05-17
UR - https://www.deepdyve.com/lp/oxford-university-press/providing-a-stable-methodological-basis-for-comparing-transcript-pV9LFHj1wW
SP - 601
EP - 616
VL - 16
IS - 8
DP - DeepDyve
ER -