This study utilised 3 ES cell lines, IMT11 (derived from 129 mice), HM1 (which is Hprt negative) and SMHBl6 (derived from C57Bl6/J mice). The IMT11 line was selected for all investigations involving differentiation, as it is not genetically modified and is better characterised than SMHBl6. IMT11 cells showed the highest percentage of diploid cells after karyotyping and have been tested for germline transmission.

1) Undifferentiated ES cells were maintained at 37°C in a humidified atmosphere with 5% CO₂ on 0.1% gelatin in DMEM, with 2 mM L-glutamine, 50 U/ml penicillin, 50 μg/ml streptomycin (all from Gibco™, Invitrogen Ltd, Paisley, Renfrewshire, UK), 10^-4 M β-Mercaptoethanol (Merck KGaA, 64293 Darmstadt, Germany), 10^-3 U/ml murine LIF (ESGRO™, Invitrogen, Ltd, Paisley, Renfrewshire, UK), 10% FBS (foetal bovine serum) and 10% NBS (newborn bovine serum) (selected batches, PAA Laboratories GmbH, Linz, A-4020 Austria). All undifferentiated ES cell lines were karyotyped using standard protocols in order to test that the majority of cells showed a normal diploid chromosome number (40XY) prior to differentiation.

2) EB generation: a semi-confluent 100 mM dish of ES cells was trypsinized (0.25% trypsin/EDTA, Invitrogen), followed by trituration in additional ES medium to achieve a single cell suspension. ES medium was prepared as above for + LIF EBs, and without LIF for -LIF differentiations. Cells were counted using a haemocytometer and cell density was adjusted appropriately to the required number of cells per 10 μl. A multichannel pipette (Finnpipette 5–50 μl) was used to deposit approximately 200 10 μl drops on the floor of a 140 mm bacteriological dish (Sterilin). A smaller plate was filled with 1–2 mls PBS and placed in the lid of the bacteriological dish. The plate was inverted and incubated overnight to allow the EBs to aggregate. The following day dishes were righted and flooded with 20 mls of the appropriate differentiation medium, then grown in suspension culture until harvesting. EBs were formed by aggregation of 125, 250, 500, 750 and 1000 cells per 10 μl to determine optimal size. A 50% FCS, 50% NCS mix was prepared from ES batch tested serum samples (PAA), then added to serum free ES medium, at 0%, 5%, 10%, 15% and 20% final volumes. Hanging drops containing either 750 or 1000 cells were generated in order to test serum concentrations.

We also tested EBs that were allowed to aggregate randomly following dissociation. For brevity, we refer to these as "random EBs later in the text". Random aggregation gives rise to greater numbers of EBs per dish, that are of much more variable sizes and shapes than those generated using the "hanging drop" method. For random aggregation of EBs, 1 × 10⁶ cells were suspended in 10 mls media in a 100 mm bacteriological dish and allowed to aggregate spontaneously. Serum quantities were varied as above. Photographs were taken of EBs at day 4 of differentiation without LIF, at varying sizes and serum concentrations and saved in .tif format. Photographs were analysed and EB diameters measured using Scion Image (Scion Corporation). Diameter measurements were used to calculate radius and volume.

1) For array analysis, undifferentiated ES cells were washed twice with DPBS-A, treated with trypsin-EDTA (Invitrogen Ltd., Paisley, UK) and counted. Appropriate cell numbers were pelleted by centrifugation at 1000 g. The QIAgen RNeasy™ Midi Kit (Qiagen Ltd., Crawley, Sussex, UK) was used according to the manufacturer's protocol for RNA extraction, followed by OD 260/280 spectrophotometry (Camspec, Cambridge, UK) and gel electrophoresis using dissociating conditions (NorthernMax™ buffers; Ambion, Huntingdon, UK), used according to the manufacturer's protocol, to check RNA concentration and integrity.

2) For RNA extraction from individual embryoid bodies, the mini RNA isolation kit (Zymo) was used, followed by amplification using the RNA amplification kit (Arcturus), both according to the manufacturer's protocol. Following quantification by UV spectrophotometer (Camspec), amplified RNA samples were DNase treated using the Turbo DNAfree kit (Ambion) and reverse transcribed using the random hexamer protocol of the Superscript First Strand Synthesis System for RT-PCR (Invitrogen). (Note, a nuclease step is included in the Arcturus kit to remove residual DNA, but is not entirely sufficient to ensure blank negative control lanes, hence the extra DNase step).

3) For RNA extraction followed by RT-PCR, embryoid bodies were spun down, washed with PBS, then spun down again. EBs were treated with 0.25% trypsin/EDTA and washed with PBS a second time when extracting RNA from larger Day 3–4 EBs 100 ul of EBs+ residual PBS was resuspended in 1 ml TRIzol (Invitrogen); samples were homogenized by pipetting up and down using a 1 ml micropipette. RNA was extracted according to the manufacturer's protocol, quantified, DNase treated and reverse transcribed as above, except that the oligo-dT supplied with the kit was spiked with a 1/10000 dilution of an 18S rRNA gene specific primer, in order to allow 18S rRNA (which doesn't have a poly A tail) to be used as a housekeeping control. Where starting RNA quantity was low, we tripled the amount of RNA used and therefore the reaction volumes per sample. RT reactions were diluted with nuclease free water (Ambion) to 50 ul before PCR analysis. A "no RT" control corresponding to each sample was also produced for all RT-PCR experiments described in this paper; these were treated in exactly the same way as the samples except that reverse transcriptase was not added.

4) For RNA extraction prior to array analysis, EBs were spun down, washed, and trypsinized as above. The QIAgen RNeasy™ Midi Kit (Qiagen Ltd., Crawley, Sussex, UK) was used according to the manufacturer's protocol for RNA extraction. RNA samples were quantified and checked for quality as described in 1) above.

10 μg of total RNA was labelled with either Cy3 or Cy5 dyes using the CyScribe labelling system (GE Healthcare, Chalfont St. Giles, Bucks, UK), according to the manufacturer's protocol. 1 μl of labelled cDNA was combined with 2 ul 50% glycerol/50% TE mixture, run on a microscope slide sized, 1.5% agarose gel (mould manufactured in-house) and scanned using a GeneTac LS IV scanner (Genomic Solutions, Huntingdon, Cambs. UK). Control and experimental samples were then combined and prepared for hybridization.

Array slides were incubated in prehybridization buffer for 1 hour at 42°C (50% formamide, 5 × SSC, 0.1% SDS, 1% BSA). Targets were dried down via vacuum centrifugation then resuspended in 50 μl hybe solution (49.9% de-ionised formamide, 49.9% 20 × SSC, 0.2% SDS) with added 1 μl Cot1 DNA and 1 μl poly A oligo as blocking agents, heated to 95°C for 5 minutes and then added to the face of one slide. The printed face of the second slide of the pair was then placed face to face with the first, using the same probe. Slide pairs were then placed in a humidified container and incubated for 24–48 hours at 42°C. Following hybridization, slides were washed once in Wash solution 1 (1× SSC, 2% SDS, filtered autoclaved ddH2O) for 20 minutes, then twice in Wash solution 2 (0.1× SSC, 0.2% SDS, filtered autoclaved distilled deionised H2O (ddH2O) for 20 minutes each. Slides were dipped in nuclease free filtered water, then spray dried, finally, the backs of the slides were cleaned with filtered autoclaved ddH2O, then wiped with 100% EtOH, then wiped dry and scanned.

Scans were carried out at 12.5 μm, using the averaging setting (GeneTac LSIV scanner, (Genomic Solutions, Cambridgeshire, UK)). It is possible to carry out quick draft scans using this scanner. Gain and black settings, which affect image intensity and background, were varied slightly in order to optimize the signal/noise ratio for each channel and each slide before proper scans were initiated.

In adherence with MIAME standards [23], all data sets have been submitted to the GEO database and are fully MIAME compliant. Undifferentiated ES cell comparison data are described in GSE8625, while EB differentiation comparisons are described in GSE8766. The results from these experiments were compared with an array comparison (described elsewhere) between undifferentiated ES colonies and microdissected inner cell masses from embryonic day 3.5, 4.5, and delayed blastocysts and microdissected day 5.5 and day 6.5 embryonic ectoderm (GSE8881, [24]).

Undifferentiated ES cell array experiments (GSE8625) were carried out using NIA 15K slides printed in the Cardiff Microarray Facility. These consisted of 17136 spots, printed in 12 × 4 mini grids, each with 17 rows and 21 columns and are described more fully in GEO, platform accession number GPL5530. EB differentiation comparisons (GSE8766) were carried out using NIA 15K slides printed by the HGMP. These consist of 2 slides, 17280 spots per slide, including control spots, empties and landmarks. 2 duplicate spots are printed per slide for each clone. These slides are described more fully in GEO, platform accession number GPL5735. The array slides used for GSE8881 were also printed in Cardiff and contain 16128 spots, printed in 12 × 4 mini grids, each with 16 rows and 21 columns. The GEO platform accession number for these is GPL5771. Differences in spot number relate to differences in the number of landmarks and other controls printed; the core NIA 15K set is present on each of these platforms

Undifferentiated ES cell arrays were carried out using samples from undifferentiated IMT11, SMHBL6.3 and HM1 ES cells. EB arrays were carried out using samples from Day0 (undifferentiated), Day1, Day2, Day3 and Day 4 differentiated EBs, generated in the presence and absence of LIF (9 samples in total). Each experimental sample was hybridized on a slide with a pooled control derived from an equal amount of all experimental samples. ES arrays were repeated 12 times (this print run contained no duplicate spots). EB arrays were repeated twice. Duplicate spots were present on each array, giving four repeats total for each gene. Scanned images were stored and filtered, then analysed using ImaGene™ 5.5 (BioDiscovery). This series of array experiments was carried out using the NIA15K set [25]. ESTs comprising this set were isolated from a variety of embryonic stages and tissues. Arrays were hybridised with fluor switching, in order to counteract any issues of dye bias that may have arisen from direct labelling.

Array analyses for GEO entries GSE8625 and GSE8766 were carried out similarly to work described previously [26,27]. Output files from ImaGene were saved and analysed in MS Excel spreadsheet format. Each channel from each repetition was normalised via division by the mean intensity value. These data were collated and formatted for Significance Analysis of Microarrays (SAM; ). Genes that showed a fold change of 2 or above and were statistically significant above a delta value of 0.5 (which denotes an error rate of 5%) were selected for further appraisal. SAM, however, has a disadvantage; all replicates have to be in the same order, precluding any data filtration. Hence, we supplemented SAM analysis with a second analysis method as follows:

Following normalisation, we used approximately 700 blank spots per slide to calculate a mean background value + 2 standard deviations of that background value, for each channel. Genes that fell below this cut-off in BOTH control and experimental channels were removed, along with genes with a fold change of < 2. This filtered gene list was compared with that from SAM; genes appearing as differentially regulated using both methods were deemed significant (fold change > 2, above background + 2 SD in at least one channel, delta value of 0.5).

Samples were compared with pooled controls; sets of 4 experimental replicates were also compared with normalised samples from each of the other experimental samples. A "master list" of genes from all analyses was generated (see Figures 1 and 2 (all confirmed genes) and Additional File 1 (all unconfirmed genes)).

Significant ESTs were subjected to bioinformatic analysis via BLAST against the mouse genome, and, where available, against the reference sequence from the UniGene cluster to which they belonged. ESTs with sequences too short or too poor for primer design were rejected at this stage. Adequate EST sequences were used for primer design with Primer3 [28]), using sequence that matched the genome and/or RefSeq sequence. This analysis was carried out in order to rule out primer mismatches due to areas of erroneous first pass sequence. Primer sequences are given in Table 1.

Bioinformatic analysis of confirmed genes was also carried out in order to identify putative function in differentiation. Accession numbers were used to comprehensively search the NCBI databases . Data compiled from various NCBI databases (UniGene, Homologene, OMIM, LocusLink, PubMed etc.) are shown in Figures 1 and 2. Selected marker genes were chosen on the basis of a known role in maintenance of pluripotency, or early development, or because they are well known housekeeping genes. Primers were designed for these from the appropriate reference sequence, using Primer3 as above. Datasets derived from ES cell differentiation subtractive EST library studies [15,29] or microarrays [30-38] were also searched for the presence of the genes we confirmed in this study (Figures 1 and 2).

We used semi-quantitative RT-PCR to confirm differentially regulated genes. As most significant genes appeared to be differentially regulated at various different stages of differentiation, we tested all genes against Day 0, and Day1–4 EBs, + and - LIF. Three biological replicates of 1000 cell Day 1–4 EBs and 3 replicates of 750 cell EBs (+ and - LIF) were used for confirmations, including No RT controls. In order to be deemed confirmed, we asked that a gene demonstrate a consistent expression pattern in at least 4 out of 6 PCR tests. The number of cycles required for minimum visibility was identified and PCRs were optimised at the Tm indicated for each primer pair, such that only single bands appeared. PCR bands obtained for housekeeping controls were of even intensity at minimum visibility cycles before testing other genes. PCRs were carried out in 20 μl volumes using 0.025 μmol of each dNTP (Invitrogen) and 1 unit Taq DNA polymerase (Sigma) per reaction. PCRs were usually carried out using PCR buffer (Sigma) containing 15 mM MgCl₂ although magnesium concentrations, along with annealing temperature and cycle number, were varied where necessary in order to optimise results. Final PCR conditions are indicated for each gene in the appropriate figures (see Results). Primer sequences are given for all genes tested in Table 1 and Additional File 1. In order to confirm that semi-quantitative PCR was indeed representative of expression pattern changes, we studied the expression patterns of two confirmed genes, Hspa8 and BG063737, by quantitative PCR. Expression pattern changes achieved by q-PCR were very similar to the methods described above. Methods and results are given in Additional File 2[39].

We studied expression differences between 3 ES cell lines, IMT11 (derived from 129 mice), HM1 (which is Hprt negative) and SMHBl6 (derived from C57Bl6/J mice) (GEO, GSE8625). 21 genes, including Hprt, were identified from the arrays as possibly differentially regulated, four were confirmed by RT-PCR; Hprt, Mki67, Pter and Scmh1 (see Figure 3). Hprt was not expressed in HM1 cells, as expected. Mki67 and Scmh1 were upregulated in IMT11 cells, while Pter was downregulated in SHBl6.3 cells.

IMT11 ES cells were used to generate EBs of 125, 250, 500, 750 and 1000 cells via hanging drops, using ES medium - LIF and 20% serum (FBS + NBS). An upper limit of 1000 cells was chosen as we often observed substantial interior necrosis and RNA degradation in larger EBs by Day 4 of differentiation. Analysis of EB size measured at day 4 shows little variation in diameter between EBs initiated with a cell number of 500 or greater; size constraints may begin to apply once a certain diameter is reached (Figures 4a + b). RT-PCR showed that 125–500 cell EBs failed to show expression of genes such as brachyury by Day 4. Furthermore, the amounts of RNA retrieved were insufficient for array analysis at Days 1 and 2 (results not shown). 750 and 1000 cell EBs showed expression of markers of all 3 germ layers (see below), and gave a better RNA yield at days 1 and 2.

Higher serum concentrations are known to promote maintenance of pluripotency in cultures of undifferentiated ES cells, and also provide more nutrients and pH stability to the medium. We generated random, 750 cell and 1000 cell EBs in ES media without LIF, containing 0, 5, 10, 15 and 20% serum (1:1 mix of FBS + NBS). EBs cultured with no serum were dead after 24 hours and are therefore not shown in data analysis. Measurements of EB diameter showed that 20% serum promoted the largest EBs at Day 4 for 750 cell and randomly generated EBs, while 1000 cell EBs reached similar sizes with 15% and 20% serum concentrations (Figure 4b). RNA was extracted at days 1–4 of differentiation and analysed by RT PCR. Results are shown in Figures 5, 6 and 7. The greatest range of gene expression was noted on days 3–4 in concentrations of 15–20% serum (Figures 5, 6 and 7). Moreover, Fgf5 was only expressed in EBs with a starting size of 1000C. One should note that the higher serum concentrations also promote the maintenance within the EB of undifferentiated cells; the expressions of markers of pluripotency such as Oct4, Nanog and Rex1 diminish at Day 4 in 5% serum, but maintain and even increase expression levels at 15–20% serum. Given more varied gene expression profiles (suggesting that a greater range of cell types could be obtained eventually from such a protocol over time), size and greater RNA yield, we decided to use 1000 cell EBs grown in 20% serum for our subsequent array experiments.

Array results from Day 1–4 EBs, grown in both the presence and the absence of LIF, were compared at all stages with the pooled control (equal amounts of Day 0–4 ES cells + EBs, both with and without LIF), and also with each other, when analysing the arrays. 128 ESTs selected for follow-up were differentially regulated in all 4 repeats and were statistically significant according to SAM. Many ESTs were differentially regulated in more than one comparison. Removal of redundancy (more than one EST mapping to different parts of the same gene) and elimination of those ESTs where sequence quality was insufficient for primer design resulted in a master list of 104 genes. Notably, beta-actin and Gapdh were differentially regulated, so 18S rRNA was used as a housekeeping gene instead.

Given problems with reproducibility that have been noted with stem cell arrays [20-22], RT-PCRs were carried out on 3 sets each of 1000 cell and 750 cell EB samples; differentially regulated genes showed a reproducible expression pattern change in at least 4/6 samples. 18 genes were differentially regulated in -LIF samples; a further 8 were differentially regulated in + LIF samples. 3 genes appeared in both datasets (Tubb5 and 2 uncharacterised genes; BG063737 and BG069482, see Figures 8 and 9 and Figures 1 and 2). None of these 23 genes appeared in the list of 3 that were differentially regulated between different undifferentiated ES cell lines (GSE8625). A further 3 genes did not show the exact expression patterns predicted by the array, but were dramatically downregulated on induction of differentiation and are therefore also presented in Figure 8. This gave a total of 26 genes we deemed differentially regulated during ES cell differentiation. Q-PCR analysis of two confirmed genes, Hspa8 and BG063737, gave very similar results to semi-quantitative PCR (Additional File 2).

We then compared this dataset with that from another of our array experiments, which compared undifferentiated IMT11 ES cells with embryonic inner cell masses (the tissue from which ES cells are derived) and day 5.5 and day 6.5 embryonic ectoderm (see Figures 1 and 2, GSE8881, methods described [24]), Thirteen of the above 26 genes were also up- or down-regulated in this study, indicating their importance in early development and differentiation. Fourteen of the known genes were also noted in other array or subtractive EST library studies using material from different ES cell differentiation experiments (see Figures 1 and 2). Interestingly, two of the four genes that varied between undifferentiated cell lines were also identified by this study.

Our arrays have successfully identified 23 genes that show expression changes during EB differentiation (plus an additional 3 that did not show the exact expression changes indicated by the array). However, 23 genes represents a low confirmation rate (22%) given the initial 104 genes tested. The consistency of gene expression in individual embryoid bodies was therefore tested in order to assess how this variation might influence the reproducibility of ES differentiation protocols that are based on EBs. We studied 10 individual 1000 cell Day 4 EBs from the same tissue culture dish and 2 small, 2 medium and 2 large EBs from a similar plate of Day 4-L randomly aggregated EBs (medium represented the same size as those derived from the hanging drop method, large EBs were roughly double this size, while small ones were half this size). 18S rRNA was used as a housekeeping gene control. Expression of Gapdh and Hprt was noted in most EBs, albeit at varying levels. Expression of Nodal and Rex1 was also noted in a majority of samples. Nanog, Oct4, Goosecoid and beta-actin were expressed in 50% or less of EBs, while Afp, Kdr, Brachyury and Fgf5 were expressed in 1–3 EBs out of 16 only (Figure 10).