Dynamic transcription programs during ES cell differentiation towards mesoderm in serum versus serum-freeBMP4 culture

Background Expression profiling of embryonic stem (ES) cell differentiation in the presence of serum has been performed previously. It remains unclear if transcriptional activation is dependent on complex growth factor mixtures in serum or whether this process is intrinsic to ES cells once the stem cell program has been inactivated. The aims of this study were to determine the transcriptional programs associated with the stem cell state and to characterize mesoderm differentiation between serum and serum-free culture. Results ES cells were differentiated as embryoid bodies in 10% FBS or serum-free media containing BMP4 (2 ng/ml), and expression profiled using 47 K Illumina(R) Sentrix arrays. Statistical methods were employed to define gene sets characteristic of stem cell, epiblast and primitive streak programs. Although the initial differentiation profile was similar between the two culture conditions, cardiac gene expression was inhibited in serum whereas blood gene expression was enhanced. Also, expression of many members of the Kruppel-like factor (KLF) family of transcription factors changed dramatically during the first few days of differentiation. KLF2 and KLF4 co-localized with OCT4 in a sub-nuclear compartment of ES cells, dynamic changes in KLF-DNA binding activities occurred upon differentiation, and strong bio-informatic evidence for direct regulation of many stem cell genes by KLFs was found. Conclusion Down regulation of stem cell genes and activation of epiblast/primitive streak genes is similar in serum and defined media, but subsequent mesoderm differentiation is strongly influenced by the composition of the media. In addition, KLF family members are likely to be important regulators of many stem cell genes.


Background
Embryonic stem (ES) cells isolated from the inner cell mass of the early mammalian blastocyst-stage embryo retain pluripotency when cultured on mouse embryonic fibroblasts (MEFs) in the presence of leukemia inhibitory factor (LIF) [1]. In the absence of LIF and MEF attachment, ES cells spontaneously differentiate into multi-cellular aggregates termed embryoid bodies (EBs). Although the spatial complexity of organogenesis is not established during EB maturation, the dynamics of gene expression closely mimic those which characterize early stages of mouse development [2][3][4][5]. Thus, ES cell differentiation is an excellent model system for the discovery of genes involved in developmental processes.
Many studies have attempted to identify genes that define the stem cell state by mining for genes co-expressed in ES cells and various other stem cell populations such as hematopoietic and neural stem cells [6,7], or ES cells and trophoblast stem cells [8]. Some of these studies have been criticized because they were unable to define a similar cohort of stem cell genes [9]. Also, this approach is likely to find co-expression of 'housekeeping' genes that are irrelevant to the stem cell state. More recently, dynamic expression profiling during ES cell differentiation has been undertaken using various EB culture techniques [10,11] and the addition of exogenous growth factors [7]. Most of these studies followed differentiation for a short time period (up to 6-8 days) due to inherent limitations of liquid or adherent cultures cell differentiation systems.
Various platforms have been used to determine gene expression profiles, including differential display, SAGE [12] and slide micro-arrays, such as those provided by Affymetrix, Agilent and Compugen. In this study we employed Illumina ® Sentrix Mouse-6 oligo bead arrays [13], which have ~47 K probes that are largely based on the MEEBO set of transcripts, plus ESTs discovered as part of the FANTOM2 transcriptome project [14], some alternatively employed exons, and a small subset of non-coding transcripts. This is a new, rich and sensitive platform for expression profiling [15].
ES cells were differentiated in methylcellulose cultures, which allowed consistent EB development over 16 days in either serum [2,16] or defined serum-free media containing BMP4 (2 ng/ml), termed serum-free B4L hereafter [17]. The dynamics of loss of stem cell gene expression and activation of epiblast and primitive streak gene expression programs were similar in serum and serum-free B4L but QT clustering revealed a significant difference in mesoderm outcomes from day 6 onwards. In particular, cardiac gene expression was greater in serum-free B4L whereas activation of the blood program was enhanced in serum.
High expression levels of Kruppel-like factor (KLF) family members: Klf2, Klf4, Klf5 and Klf9 was detected in undifferentiated ES cells, and confirmed by quantitative real time RT-PCR, indirect immuno-fluorescence and DNAbinding assays. Expression was down regulated rapidly upon differentiation and a different set of Klfs: Klf3, Klf16 and Klf1/Eklf, were activated upon ES cell differentiation.
Also, KLF-binding sites were markedly over represented in the proximal promoters of most stem cell specific genes, suggesting possible direct regulation by KLFs. We suggest a model in which the balance between self-renewal and differentiation is regulated by competitive occupancy of the proximal promoters of key stem cell genes such as Zfp42/Rex-1, Nanog, Pou5f1(Oct4), Lefty1 and Lefty2, and others such as the Klf genes themselves.

Differentiation of ES cells in serum and serum-free B4L culture
The directed differentiation of embryonic stem cells into desirable cell types will depend upon the addition of specific growth factors to defined media and reproducible physical conditions of culture. To establish a baseline for such studies, we undertook a comparison of murine ES cell differentiation using an embryoid body (EB) methylcellulose culture system containing either 10% FBS [2] or a chemically defined BSA-based media containing 2 ng/ ml of BMP4, termed serum-free B4L [17]. Triplicate experiments were performed in which ES cells were harvested after feeder depletion, and set up in parallel EB cultures in serum versus serum-free B4L . RNA was collected at regular intervals up to 16 days, quantitative RT-PCR was performed on a panel of genes representing defined stages of development to establish the robustness of the culture system (Figure 1), and expression profiling was performed using Illumina ® Sentrix Mouse-6 bead arrays (see Methods and [13]).
Changes in gene expression were extremely reproducible across ES differentiation experiments with overall correlation of global gene expression for biological replicates in the order or 0.95-0. 96. Approximately 30% of the probes were differentially expressed at one or more stages of EB differentiation as determined by one-way Welch ANOVA with multiple testing corrections using the Benjamini and Hochberg False Discovery Rate algorithm (7,967 probes with a p-value of < 0.01 by Welch t-test) (Figure 2A).
To mine for genes with similar patterns of expression across the time course, hierarchical clustering was performed on the 7,967 probes and represented graphically. A large percentage of these (~30%) displayed highest expression in undifferentiated ES cells and subsequent reduced expression over time (labeled stem in Figure 2B). A second major cluster showed marked induction after day 4-6 of differentiation (labeled mesoderm in Figure  2B).

Putative stem cell genes sets
Previous studies have identified genes important to the stem cell state by comparing expression profiling data from different stem cell populations such as ES cells, hae-Dynamic gene expression during EB differentiation in serum and serum-free media containing BMP4 [2 ng/ml] Figure 1 Dynamic gene expression during EB differentiation in serum and serum-free media containing BMP4 [2 ng/ ml]. Expression of stem cell, primitive streak and late mesoderm genes were analysed by quantitative RT-PCR in undifferentiated ES cells (0, grey bar) and in EBs collected for up to 16 days in serum (white bars) or defined media (black bars) from the same starting ES cell populations. Bars represent the means of three (serum) or two (serum-free B4L ) biological replicates. The Y-axis represents a log scale normalized relative to the housekeeping gene HPRT. Error bars indicate standard deviation.
Gene expression profiles during 16 days of EB differentiation in serum and serum-free media containing BMP4 [2 ng/ml] Figure 2 Gene expression profiles during 16 days of EB differentiation in serum and serum-free media containing BMP4 [2 ng/ml]. (A) GeneSpring representation of the subset of genes from the 48 K Illumina array that showed significant differences in gene expression at 2 or more time points. Each line represents the mean normalised expression of an individual Illumina probe. Colours represent relative expression (red 4× increased; green 0.25× reduced) compared to a gene mean of 1 (black) with expression in undifferentiated ES cells (day 0) as the decision point for colour coding. The expression profiles of T (brachyury), Hba (α-globin), and Igf2 are indicated for reference. (B) Clustering (tree) of the genes that showed significant (p < 0.01) changes in gene expression upon differentiation by One-way Welch ANOVA. Expression levels are heat map colour coded (inset) with high expression at Day 0 coloured red and low expression coloured green. matopoietic stem cells (HSCs) and neural stem cells [6,18]. We used two alternative approaches to identify specific stem cell specific gene expression. Firstly, genes with dynamic expression patterns which correlated closely with Oct4, Sox2 or Nanog expression profiles (Pearson correlation > 0.9) were determined as previously described [6,7]. Second, genes rapidly down regulated ≥3 fold by day 3 of differentiation, a time point where ES cell clonogenicity decreases dramatically, were determined. Not surprisingly there was very little overlap between these gene lists, primarily because Oct4 and Nanog expression persists during EB differentiation longer than many other stem cell genes ( Figure 1). Full lists of genes ranked according to similarity of expression (Pearson correlation >0.9) with expression patterns for Oct4, Nanog or Sox2 are provided in Additional files 1, 2, 3, respectively.
Only fifty-nine genes were down regulated ≥3 fold by day 3 of EB differentiation ( Table 1). Many of these have been identified as stem cell markers by other groups using alternative strategies [6,7,19]. Information was compiled on their known expression patterns and/or gene knockout phenotypes via literature searches as referenced in Table 1. Insufficient space exists to discuss every gene and most have been previously identified as ES cell enriched transcripts. Osteopontin is secreted by osteoblasts in the bone marrow niche where it is thought to play an important role in the maintenance of HSC quiescence and maintenance of 'stemness' [20]. It has also been discovered by other groups as a strong marker of ES cells and is regulated directly by the key stem cell transcription factors, Oct4 and Sox2 [21]. Thus, osteopontin is likely to be a useful marker of ES cell pluripotency and may play autocrine or paracrine roles in maintenance of stemness. The second most rapidly down regulated gene was estrogen related receptor β, Esrrb, which had previously been identified as a stem cell specific gene and shown to be critical for maintenance of the stem cell state [19]. F-box 15 (Fbxo15), a member of the large F-box family of genes [22] was also rapidly down regulated. Fbxo15 is known to be highly expressed in ES cells, is regulated by Oct4 and Sox2, but is dispensable for ES cell self-renewal [23]. Recently, the stem cell specificity of this locus was used to screen for cDNAs capable of transforming mouse fibroblasts into ES cell populations [24]. Other notable rapidly down regulated genes included the nodal inhibitors Lefty-1 and Lefty-2, and Zfp42/Rex-1, which are also well established as stem cell markers [19].
Oct4, Nanog and Sox2 are not included in Table 1, even though each is required for the maintenance of ES cell pluripotency [25][26][27] due to their persistent expression during the first 3 days of EB differentiation (Figure 1 and 5A, and Additional file 1). Indeed, Oct4 expression is known to persist beyond the inner cell mass stage of development in vivo, with transcripts detected in the epiblast and later in the primitive streak, before it becomes further restricted to primordial germ cells [10,28]. 779 genes with a similar pattern of expression to Oct4 (Pearson correlation of > 0.9) were identified (Additional file 1). As expected this list also includes many well known stem cell genes such as Sal-like 1, nodal, chromobox homolog 7 (Cbx7), Dnmt3b and Sprouty homolog 4 [19]. We also mined for stem cell genes by looking for similarity (Pearson correlation of > 0.9) with Nanog and Sox2 gene expression profiles. These provided overlapping but not identical sets of putative stem cell specific genes (Additional files 2, 3). Full data for the 16 days of differentiation can be interrogated and mined for specific differentiation outcomes via Signet (Login_Bruce) (see Methods).

Tightly co-ordinated waves of gene expression suggest sequential activation of epiblast, primitive streak and mesoderm differentiation programs
Many alternative differentiation programs occur simultaneously during ES cell differentiation. Therefore, it is difficult to find gene syn-expression patterns that identify specific programs using standard clustering methods. An alternative approach is to search for expression patterns that closely resemble genes of known function. For example, the T box gene, brachyury, is a well characterized specific marker of the primitive streak [29]. Brachyury expression is transient in vivo, a feature recapitulated during ES cell differentiation [2]. Using qRT-PCR, brachyury was induced ~1000 fold from D0 to day 4 and then rapidly silenced to baseline levels by D6 ( Figure 1). The dynamics of brachyury expression were similar between serum and serum-free B4L media ( Figure 1 and 3A), but serum-free media was unable to support brachyury induction in the absence of BMP-4 (data not shown). Thus, BMP4 (or related factors in serum such as activin) is essential for mesoderm generation. Thirty genes were identified with a similar expression pattern to brachyury (Pearson correlation >0.9) ( Table 2) including many genes with established expression or critical functions during primitive streak formation, such as Mix1 [30], Lim1 [31], Sp5 [32] and eomesodermin [33]. Transient expression of many of these 'streak-specific' genes was validated by qRT-PCR ( Figure 3B and data not shown). In all cases there was a close correlation between the qRT-PCR results and the profiling data ( Figure 3A).
Interestingly, four novel transcripts were identified in this list ( Table 2). One of these, 8430415E04Rik, was detected by two independent probes, strongly supporting its 'streak-specific' expression; by qRT-PCR there was ~50 fold induction from day 2 to day 3 of differentiation and a decrease to near baseline levels by day 6 ( Figure 3B). The gene contains 24 exons, and the full length cDNA encodes  for a protein of 868 amino acids ( Figure 3D) which is highly conserved in vertebrates (Additional file 4). It has little homology to other proteins apart from three predicted HEAT domains or armadillo domains (grey shading in Figure 3D). Drosophila armadillo interacts with β-catenin via its HEAT/armadillo repeat [34], which suggests the protein encoded by 8430415E04Rik may also be involved in protein-protein interactions.
The list of primitive streak genes in table 2 is smaller than that previously published [10] due to the very stringent criteria used to determine brachyury-like patterns of expression. Thus, to find further genes characteristic of epiblast, early and late primitive streak programs, probes were identified which were transiently up and down regulated at least two fold with peak expression at day 2, 3 or 4 of EB differentiation. Gene lists were combined (102 Illumina probes) and clustered into epiblast, early, mid and late streak groups plus a unique group in which expression was very transient at day 3 of differentiation (designated 'brief' at the bottom of Figure 3C). Also, lists were generated by searching for similarity of dynamic expression profiles to known genes which are accepted specific markers of specific stages of mouse development. For example, Fgf5 is specifically expressed in the epiblast [35], and Wnt5a is expressed in the late primitive streak [36]. The Fgf5-like (epiblast) list includes the methyltransferase, Dnmt3b, which was massively up-regulated in the first 2 days of EB differentiation (Additional file 5) and is known to be essential for mesoderm differentiation [37]. This list also includes Pim2, Zic2, Wnt8a, the SP1 (and possibly SP5) cofactor Crsp2, and Irf1. The Wnt5a-like (late primitive streak) list contains 76 genes including many homeobox genes (Hoxb2, Hoxd1, Hoxb6 and Hox8/ Msx2), Cdx2, Hand1, Tbx3, Bmp4 and the VEGF receptor Flt1 (Additional file 6). Many are expressed at E10 of mouse development in the remnants of the primitive streak [36], suggesting the Wnt5a-like list includes many genes known to pattern the embryo following formation of the germ layers. There is a significant overlap between Mining for genes expressed in patterns consistent with primitive streak, and epiblast differentiation programs   [152] Myogenic regulator * More than one independent illumina probe for these genes were identified the genes we identified as being epiblast-, early and late streak-specific and those recently identified using a similar EB differentiation system and Affymetrix arrays [10], and by other groups [2,38]. In addition, there are addition novel RIKEN cDNAs in these lists, which are likely to encode for interesting proteins worthy of further study (Additional files 5, 6).
Following the primitive streak wave of gene expression there was a dramatic activation of many genes (Figure 2A). Many of the most highly induced genes, such as globin genes, heme synthesis enzymes and haematopoietic transcription factors, are known to be expressed in the blood. qRT-PCR for globin genes confirmed > 10,000 fold induction between days 3 and 8 with a steady decline thereafter ( Figure 1). Other mesoderm and endoderm programs are also activated at this stage, so it is difficult to mine for these different outcomes.

Enhanced cardiac gene expression and reduced erythropoiesis in serum-free B4L EB culture
Overall there were very strong similarities between ES cell differentiation in serum and serum-free cultures supplemented with 2 ng/ml BMP4. For example, all 29 of the primitive streak gene set, as defined from serum-containing cultures, were also expressed transiently and to a similar level in serum-free B4L media ( Figure 3A). Genes differentially expressed at each EB time point (Day 1-16) were pooled and subjected to QT clustering to find sets differentially expressed between serum and serum-free B4L EB culture. Using a minimal cut-off of five genes per group, nine gene sets were identified; seven showed higher expression in serum and two showed higher expression in serum-free B4L (Table 3 and Figure 4).
The genes in set 9 were expressed more highly in serumfree B4L media from day 6, and genes in set 8 showed up-regulation at day 12 and 16 in the absence of serum (Figure 4). Both sets are comprised almost exclusively of cardiac-specific genes such as myosin heavy chain 7 (Myh7), myosin light chain 7 (Myl7), the cardiac isoform of αactin (Actc1), troponin (Tnnc2), tropomysin (Tpm2), the nicotinic cholinergic receptor (Chrng), and cardiac muscle fast twitch 1 specific ATPase (Atp2a1) ( Table 3). Further interrogation of the array data revealed a cohort of cardiac transcription factors, including ISL1, MEF2C, Bop/Smyd1 and Hand2 [39], which were transiently up regulated 1.4 to 3.5 fold in serum-free B4L media compared to serum at day 6 (Signet Login_Bruce). Importantly, this elevated cardiac gene expression program correlated with increased numbers of spontaneously beating EBs in serum-free B4L media (data not shown). Enhanced expression (~100 fold) of Actc1, Myl4 and Myl7 in serum-free B4L media compared with serum-containing media was confirmed by qRT-PCR ( Figure 1). Taken together, this data strongly suggests cardiac muscle cell differentiation is either inhibited in serum or enhanced by the addition of BMP4 to serum-free culture.
The genes in set 2 were expressed in both serum and serum-free B4L conditions immediately following primitive streak formation, but expressed to higher levels and slightly earlier in serum (Table 3 and Figure 4). Many of these genes are globins (both embryonic and definitive) suggesting erythropoiesis is activated more robustly in the presence of serum (See Discussion). Alas2, a gene encoding the erythroid specific isoform of the first and rate limiting enzyme in the heme biosynthesis pathway, and the erythroid specific transcription factor, p45-NF-E2 are also in set 2 [40]. qRT-PCR for αand βh1-globin confirmed more rapid expression in serum versus SF B4L media (Figure 1), suggesting factors in addition to BMP4 enhance activation of the blood program from a similar primitive streak platform.
The genes in set 3 were expressed at a similar time in serum but expressed at very low levels or not at all in serum-free B4L media. Again, this set is highly enriched for erythroid specific genes such as alpha hemoglobin stabilising protein (Ahsp) [41], Glycophorin A, erythroid Kruppel-like factor (EKLF/Klf1) [42] and mitochondrial solute carrier protein (mitoferrin) [43] (Figure 4 and Table 3). Genes in sets 1 and 5 are activated more robustly at late time points in serum cultures. Most of these are specifically expressed in macrophages. Together, these results suggest delayed and less robust primitive haematopoiesis occurs in SF B4L culture compared with serum, an observation confirmed by less robust erythroid cell generation and 'redness' of the EBs from day 6-10 of differentiation (data not shown).

Rapid changes in Kruppel-like factor gene expression upon ES cell differentiation
Three Kruppel-like factor genes, Klf2, Klf4 and Klf5, were rapidly down regulated during the first few days of EB development (bold type in Table 1). All are members of the KLF family of transcription factors which are characterized by a similar C-terminal domain of three C 2 H 2 zinc fingers which confers binding to CACC box elements in promoters and more distant regulatory elements [44,45]. Klf2 and Klf4 have N-terminal transcriptional action domains and act primarily as transcriptional activators, whereas Klf5 is best known as a transcriptional repressor [46]. Until recently, Klf2 and Klf4 were considered to have restricted expression patterns [47,48] with gene knockout animals showing late developmental defects in vasculogenesis and skin differentiation, respectively [49,50].
Recently however, Klf4 expression was noted in ES cells [19,51], and forced expression was shown to enhance Oct4 expression and promote self-renewal [51].
qRT-PCR confirmed the rapid down regulation of Klf2, Klf4 and Klf5 in both serum and serum-free B4L differentiation conditions, although Klf5 expression was down regulated more slowly ( Figure 5A and 5B, and Table 4). Klf2, Klf4 and Klf5 were all re-induced from day 3-5 of differentiation as mesoderm and endoderm developmental pathways are activated ( Figure 5A and 5B). Sixteen of the 17 known KLFs were examined in detail throughout the entire differentiation time course (Table 4). Klf9 was also steadily down regulated (>10 fold) between ES cells and day 5 EBs in serum and serum-free B4L media ( Figure 5B). Klf3, a well characterized transcriptional repressor [52] via its ability to recruit the co-repressor, CtBP [53], displayed an interesting biphasic expression pattern with initial rapid down regulation then re-induction from days 5-6 ( Figure 5B). In contrast, Klf16 displayed little change in expression levels over the 16 days of EB maturation (Figure 5B).
Klf2 and Klf4 co-localize with OCT4 in the nuclei of ES cells ( Figure 6A). Interestingly, all three proteins were found to preferentially reside in a nuclear sub-compartment (possibly nucleoli for transcription factories), suggesting possible co-involvement in a protein complex or network. Also, endogenous KLF2 was detected by specific CACC box DNA-binding activity in undifferentiated ES cells, which was lost upon EB differentiation ( Figure 6B). In contrast, endogenous ES cell KLF4 DNA binding activity was not detected using the p18INK4c promoter CACC box sequence, although recombinant KLF4 was shown to bind to this probe (data not shown). SP1 and SP3 are ubiquitously expressed CACC-box binding proteins, which are dominant in ES cell nuclear extracts ( Figure 6B). We hypothesize KLF2, KLF3, KLF4, KLF5 and KLF6 compete with SP1 family members for DNA binding to key cis  QT clustering revealed sets of genes differentially expressed between serum and serum-free B4L EB culture Figure 4 QT clustering revealed sets of genes differentially expressed between serum and serum-free B4L EB culture. QT clustering revealed nine sets of genes with similar dynamic expression profiles (Pearson Correlation coefficient of > 0.9). In sets 1-7, expression is higher in serum (black sets). In sets 8-9, expression is greater in serum-free B4L EB culture (red sets). Some sets have been named based on the similarity of overall gene function during differentiation pathways (see Table 3).
elements in various stem cell genes to regulate expression in positive and negative fashions (see Discussion).

Enrichment of KLF-binding cis elements in the proximal promoters of stem cell specific genes
To investigate whether Kruppel-like factors might directly regulate stem cell gene expression, and therefore play a role in the maintenance or loss of 'stemness', the presence of KLF transcription factor binding sites (TFBS) within the proximal promoters of all genes in Table 1 was determined. Although a position weighted matrix (PWM) for KLF4 binding site specificity has been published [54], this site does not resemble the CACC box sequences known to bind other KLFs. Since KLF4 has identical DNA-binding residues in each of its three C-terminal C 2 H 2 zinc fingers to other members of the Kruppel-like factor family [45], it should bind to similar sequences. Based on the crystal structure of the zinc fingers of Zif268 and SP1 bound to DNA [55,56], and testing of specific binding of CACC box variants to recombinant KLF1, KLF3 and KLF17 [45,52,57,58], a generic KLF TFBS has been proposed. Although a SELEX experiment for KLF1 binding has not been published, experiments from our group suggest a C at position 1, an A or C at position 3, an A at position 5, and an A, T or C at position 9 of the 9 bp consensus provides enhanced affinity for KLF1 binding. Taken together, these studies have enabled the development of a PWM (called here KLF-A) that should predict KLF4, KLF2 and other KLF TFBSs (see Methods and Additional file 7).
Clover [59] was used to determine if defined TFBSs were statistically over-represented in the stem cell lists. We searched for PWMs for KLF-A, KLF4 [54], octamer, extended octamer, nanog, and Oct-Sox binary sites as well as negative control sites (gata-1 and E-box) in sequences 2 kb upstream of the TSSs of stem cell genes (Table 1), versus 2 kb of sequences upstream of the entire murine transcriptome (See Additional file 7 for details of PWMs). There was significant over-representation of KLF-A TFBS within the first 2 kb of promoter sequences from the stem list gene list (p value of < 0.01) ( Figure 7A), as well as over-representation of octamer sites and extended Oct4binding sites as defined from ChIP-PET data [60]. No statistically significant over-representation of KLF4 [54], NANOG, GATA1 or E-box TFBSs were identified. All genes in the rapidly down regulated gene list have at least one KLF-A type CACC site, or octamer or Oct4 [60] TFBS. A full list of the promoters with highlighted sites is available on request. Together, this data strongly suggests direct transcriptional regulation of many stem cell genes by KLFs.
The Zfp42/Rex1 gene promoter is presented as an example of output from the Clover program ( Figure 7B). It contains five KLF-A, one octamer and three OCT4 TFBS. The two CACC box elements within the first 250 base pairs of the TSS represent classical extended KLF-binding sites [42,61]. These regions are the same as previously reported [62] although their functional importance has not been determined. Interestingly, two of the three KLF-A TFBS were evolutionarily conserved between the mouse and rat gene promoters ( Figure 7C). Also, two further conserved CACC sites at -1.4 and -1.5 kb were identified within 500 bp of extended evolutionary conservation (>70% identity) ( Figure 7C). This might act as a KLF dependent enhancer. Most of the stem cell genes in list 1 have conserved CACC box elements in their promoters [63][64][65][66]. In some cases these have been reported to bind the ubiquitous SP1, but our Clover analysis suggests they are also likely to bind KLFs (see Discussion).

Discussion
Expression profiling of murine ES cell differentiation was undertaken over a 16-day time course. We compared gene expression in methylcellulose cultures containing serum versus chemically defined media containing LIF (1 U/ml) and low concentrations of BMP4 (2 ng/ml) [67]. The Illumina ® Sentrix Mouse 6 bead array provided a sensitive and detailed platform for analysis of dynamic gene expression. Using various data mining approaches, lists of stem cellenriched genes and genes that are induced during the in vitro equivalent of epiblast and primitive streak stages of differentiation were generated. In combination with other ES cell profiling studies [7], our detailed expression data provides a useful resource for future reverse genetic approaches (i.e. siRNA knockdown) to study the function of these genes during ES cell differentiation and in vivo development. We also found a number of previously uncharacterized cDNAs (RIKEN clones) which could play important roles during development.
Importantly, the loss of pluripotency, measured by Oct4 gene expression, was comparable between serum and serum-free B4L EB culture, following a predicted decrease over the first 6 days. Surprisingly, Oct4 gene expression gradually increased following day 6, an observation independent of the cell lines used (data not shown   Figure 5) delayed or arrested from entering the differentiation program.
We found significant differences in cardiac gene expression during EB differentiation in serum and serum-free B4L culture. Our current understanding of cardiac development provides possible insights into this observation. Briefly, in vivo studies have revealed BMPs secreted from the anterior lateral plate, and ill-defined signals from anterior primitive endoderm, are key inducers of cardiac development [71][72][73]. The administration of recombinant BMP2 or BMP4 to chick explant cultures induces cardiac differentiation in non-cardiogenic mesoderm [71] and Bmp2 knockout animals develop cardiac abnormalities [74]. Conversely, Wnts secreted by the neural tube are strong suppressors of cardiogenesis. Together, these opposing signals act to establish the borders of the heart field [71,75]. By mimicking the environment that establishes cardiogenesis in vivo, assays capable of directing cardiomyocyte production from ES cells have been established. BMPs can efficiently enhance the cardiac program when added to EB culture [76,77], whereas BMP inhibition drastically suppresses this outcome. In our hands cultures supplemented with BMP4 (2 ng/ml) sup-ported cardiomyocyte maturation (as determined by expression of cardiac specific genes) with greatest expression detected at Day 8-10. This correlated with increased numbers of spontaneously beating EBs, and the timing of initial spontaneous contractions during murine embryogenesis [78]. The cardiac program was significantly reduced in culture containing 10% serum, an observation supported by a number of other studies [79,80]. This suggests the constituents of serum are inhibitory to cardiomyocyte development. Although a detailed assessment of cardiomyocyte differentiation was not the focus of this analysis, the array output, qRT-PCR profiles and morphological observations described supports the usefulness of this assay in future investigations. In addition, the defined constituents of serum-free B4L media provides an excellent opportunity to identify additional recombinant factors required to further expand cardiac progenitor cell production from EBs.
In contrast, the hematopoietic program was more pronounced within EBs grown in 10% serum than BMP4 alone (Figure 1 and 4). In particular, EKLF (Klf1) expression was significantly reduced in serum-free media supplemented with BMP4 (2 ng/ml) compared with serum ( Figure 5A and Table 4). EKLF is essential for regulation of a large cohort of erythroid specific genes [81,82], many of which were identified in sets 2 and 3 ( Figure 4). Thus, it is likely that reduced EKLF expression in SF B4L media directly results in inhibition of a cascade of erythroid gene expression. Previous work showed the addition of BMP (at 5 ng/ ml) to serum-free ES cell culture induces EKLF expression and restores hematopoietic cell differentiation [83]. Although a weak hematopoietic program was observed in SF B4L culture, the concentration of BMP4 used was less than that used by Adelman et al. Thus, we suggest a robust cardiac (anterior-ventral mesoderm) gene expression program is induced by low conentrations of BMP4 (or other BMPs), whereas a robust blood gene expression program (posterior-ventral mesoderm) requires >2 ng/ml of BMP4 or additional growth factors. Recently, mesoderm derived progenitor cell populations for hematopoietic and cardiac lineages were studied during EB development using ES cells in which GFP is targeted to the brachyury locus [80]. Within day 3.25 EBs, GFP + Flk1 + cells were shown to represent hematopoietic precursors, whereas GFP + Flk1cells were significantly enriched for cardiac progenitor populations. It would be interesting to determine if the ratio of Flk1 + /Flk1within the brachyury positive population is altered between our two culture conditions.
A number of the Kruppel-like factor family of transcription factors were dynamically expressed in the first few days of ES cell differentiation. It was initially surprising to find highly enriched Klf2 and Klf4 expression in ES cells, since both are considered markers of terminally differen-tiated cell types such as skin [50], gut [84], vascular smooth muscle [49,85] and lymphocytes [86]. However, recent evidence suggests KLFs may regulate stem cell function, since Klf4 is enriched in ES cells [19] and forced overexpression within these cells can maintain pluripotency in the absence of LIF [51]. Furthermore, Klf4 can bind the Lefty1 gene core promoter co-operatively with Oct4 and Sox2 [65]. Although Lefty 1 and 2 are best characterized as repressors of nodal, acting to regulate left-right patterning [87], our results and those of others, suggest a possible redundant role for lefty proteins during the maintenance of ES cell pluripotency [65]. Similarly to the Oct4 expression profile, many of the KLF family members also increased in expression late in the differentiation program. As mentioned, the KLFs are expressed in diverse tissues during development. It is therefore expected that the profiles obtained reflect the generation and early specification of mesoderm and endoderm cells following primitive streak gene activation at day 4.
Based on the likely identical binding specificity of the KLF family, established transcriptional activation and repression roles of certain family members [44], and bioinformatic evidence of a high prevalence of KLF TFBSs in many stem cell genes, we propose KLF competition for occupancy of these CACC box elements might determine selfrenewal versus differentiation of ES cells. According to this model, high levels of KLF2 and KLF4 expression in undifferentiated ES cells would lead to occupancy of CACC box elements in promoters of stem cell genes such as Pou5f1/ Oct4, Nanog, Esrrb, Zfp42/Rex1, Lefty1 and Lefty2. Furthermore, the Klf2 and Klf4 genes themselves have CACC box elements in their proximal promoters ( Figure 7A), suggesting a positive feedback loop within ES cells is likely. Interestingly, gene knockout and lentiviral shRNA gene knockdown of Klf2 or Klf4, does not lead to an obvious stem cell defect [19,49,50,85]  selves. Therefore, a combination of cross regulation of transcriptional outputs, and competition for occupancy of key stem cell promoters by KLF proteins with differing biochemical properties but identical DNA-binding capacities, may determine self-renewal versus differentiation outcomes. Again validation of this model requires functional analysis. It is important to note we have not proven occupancy of stem cell gene promoters by the KLFs in vivo.
Chromatin immuno-precipitation (ChIP) experiments in ES cells and differentiated populations are essential to validate this model, however ChiP grade antibodies are not currently available.
This scenario is not necessarily limited to ES cell differentiation. A similar program may take place in many adult stem and progenitor cells populations. Certainly competition between Klf4 and Klf5 has been suggested to regulate cell growth [88], and Klf1 and Klf3 compete for binding and have opposing effects in blood cells [89]. Furthermore, an imbalance between occupancy of key stem cell or proliferation gene promoters by KLFs, with differing activating and repressing functions, could be responsible for the development or progression of many common forms of cancer. For example, loss of hetero-zygosity for Klf4 and Klf5 has been found in colon, stomach, breast, prostate and liver cancers [90,91].
In addition to the proposed complex transcriptional interplay between KLFs and stem cell gene promoters, direct protein-protein interactions and protein networks involving KLFs are possible. We showed sub-nuclear localization of KLF2 and KLF4 with OCT4 in ES cells ( Figure 6A), and KLF4 can co-operate with OCT4 and SOX2 to drive expression of the Lefty1 core promoter [65]. Also, very recent biochemical purification of NANOG-interacting proteins discovered KLF4 as part of a NANOG-OCT4 network in ES cells (Stuart Orkin, personal communication). This is also consistent with the ability of KLF4 to co-operate with OCT4, SOX2 and c-MYC to drive de-differentiation of fibroblasts into ES-like cells [24].

Conclusion
Defining the genetic regulation of ES cell self renewal and differentiation will be instrumental for the development of future cell based therapies. Using

RNA extraction, cDNA synthesis and real-time RT-PCR data analysis
Total RNA was made using TRIzol (Invitrogen), and cDNA was synthesized from 2 μg of DNase1-treated total RNA using Superscript III (Invitrogen) and oligo-DT [12][13][14][15] (Promega) according to the manufacturer's instructions. Quantitative RT-PCR (qRT-PCR) was carried out using the Applied Biosystems SYBR-green dye system and 7500 Real Time Cycler in 96-well plates. Cycling variables were as follows: 50°C for 2 minutes, 95°C for 10 minutes, then 40 cycles of 15-second denaturation at 95°C and 1minute at appropriate annealing temperatures, optimized for each set of primers based on dissociation curves (Additional file 8). Expression levels were normalized to HPRT as determined from the ratio of delta CT values. Mean of relative expression ± SD was determined from three biological replicates. Sequencing of gel-purified amplicons was performed to ensure correct product amplification.

Micro-array hybridization and analysis
Expression profiling was performed using Sentrix Mouse-6 Expression BeadChip arrays from Illumina ® . RNA was assessed for integrity using the Agilent Bioanalyzer 2100 and RNA integrity (RIN) scores above 9.5 were present in all samples. Amplification was performed with 500 ng of total RNA using the Illumina TotalPrep RNA Amplification kit (Ambion) with a 12 hour in vitro transcription reaction period. The quantity and quality of biotin-UTP incorporated cRNA was also assessed on the Agilent Bioanalyzer 2100. Amplified cRNA (1500 ng per array) was hybridized to Mouse-6.v1 BeadChip arrays according to the manufacturer guidelines and detected with Fluorolink Streptavidin-Cy3 (Amersham Biosciences). Arrays were scanned using the Illumina BeadStation Scanner. The raw intensity values obtained for the scanned array images were compiled using the proprietary BeadStudio v1.5.1.3 software and imported into GeneSpring GX v7.3.1 (Agilent). A mouse Illumina probe set was defined in the GeneSpring Workgroup using the Illumina targetIDs as the unique identifiers and annotated according to array content files supplied by illumina ® . Data normalization was performed by first setting all measurements less than 0.01 to 0.01, then applying per chip normalization to the 50 th percentile, and per gene normalization to the median.
From an interpretation which included the three FBS experiments, a non-parametric Welch ANOVA (where variances were not assumed equal) was performed on all 46,120 probes to find a subset of genes whose expression varied significantly throughout the differentiation time course. A Benjamini and Hochberg False Discovery Rate multiple testing correction was applied to reduce the number of false positives. This yielded a set of 7,967 probes that showed statistically significant (p < 0.05) differences in expression by Welch t-test. This approach simplified running the data mining algorithms to find synexpression patterns by removing genes such as housekeeping genes which did not display differential expression, or genes which did not reach a threshold of expression at any time points.
Genome interpretations were generated in GeneSpring GXv7.3.1 (Agilent) in which the mean relative expression level for each gene was represented at each time point relative to the average of all time points. A number of statistical approaches were used to find syn-expression patterns. A Pearson correlation (between 0.9 and 1.0) was performed to find genes with similar expression profiles to Pouf1/Oct4, nanog, sox2, fgf5, wnt5 and brachyury. Hierarchical clustering was performed using the gene by gene and group by group 'tree' algorithm within the Gene-Spring (Agilent) program. This compares each probe with every other in the lists to generate similarity statistics and a tree representation of similarity based on a Pearson correlation. Also, quality threshold (QT) clustering [92] was used to define distinct subsets of co-expressed genes. The entire data set for the experiments is available via GEO as well as via our instance of Genespring (contact corresponding author).

Bioinformatic searches for CACC sites and other over represented elements in promoters of stem cell genes
The data-mining tool Biomart (accessed via Ensembl release 36 of the mouse genome), was used to collate 2 kb of sequences upstream of the transcriptional start sites (TSSs) of all genes listed in Table 1. The Clover program, run via the MotifViz web interface [59], was used to search for over-representation of position weighted matrices (PWMs) for octamer [93], Oct4 [60], Oct-Sox [60], KLF4 [54], E-box, GATA-1 [94] and CACC (KLF-A) elements in these promoter sets. The PWMs are listed in Additional file 7. The ECR Browser [95] was used to determine phylogenetic conservation of cis elements between mouse, rat and human genomes.

Competing interests
The author(s) declares that there are no competing interests.