Global Expression Analysis
For each of three transfections harvested at 24 and 72 hours, triplicate RNA samples were labeled with either Cy3 or Cy5, including dye swap. Transcription profiles were then generated by using a cDNA microarray (Ensembl Chip) consisting of 15,529 resequenced and annotated clones. The overall correlations between the OCT4 and EGFP RNAi expression data were 0.993 (24 hours) and 0.943 (72 hours), indicating that the expression levels of the vast majority of genes remain unaltered by the procedures. To judge whether a gene is expressed in the respective cells, we computed a background (BG) tag for its signal. This number reflects the proportion of background signal lower than the actual signal . Typically, a BG tag of 0.9 indicates detectable probe signals (Fig. 2B, bottom panel). Using this criterion, we found that 1,038 (6.63%) of genes represented on the chip (probes) were expressed solely at the 24-hour stage, whereas 518 (3.31%) were expressed exclusively at the 72-hour stage, consistent with a restriction of gene expression and therefore developmental competence upon stem cell differentiation . The vast majority of genes were either expressed at both time points (5,923 genes; 37.83%) or not expressed at all (8,178 genes; 52.23%). The full lists of genes expressed solely at 24 and 72 hours after transfection and also at both time points, together with the corresponding ratios, are presented in supplemental online Tables 1, 2, and 3, respectively. For a global overview of the transcriptional changes resulting from the loss of OCT4 function and insights into the physiological state of human ESCs (hESCs) resulting from OCT4 depletion, we combined the expression data at the 24- and 72-hour time points in an online database for interrogating the expression levels of specific genes and their related gene ontologies (http://goblet.molgen.mpg.de/cgi-bin/stemcell/pluripotency.cgi).
Figure Figure 2.. Global data analysis. (A): Effect of LOWESS normalization. Plotted are the ratios of the red and green signals for each spot (log scale, y-axis) and the signal range (log scale, x-axis) of a typical experiment. Whereas the raw data show a nonlinear bias in particular at the extremes of the signal area (top graph), after LOWESS normalization, this bias was eliminated (bottom graph). (B): Venn diagram of genes expressed at the different time points of OCT4 knockdown. Gene expression was judged by a numerical value (BG tag) computed from a negative control sample (bottom panel). (C): Cluster of genes that show expression patterns most similar to that of OCT4 across the experimental conditions. Colors correspond to normalized signals. For each gene, signals were divided by the average gene signal across all conditions (log scale). Red boxes indicate that the signal in the particular condition is higher than the average signal, whereas green boxes indicate the opposite. Hierarchical clustering was performed on 199 genes that showed high variation across the four conditions and a significant difference in the OCT4 and EGFP RNAis using Pearson correlation as a pairwise similarity measure and average linkage as an update rule. The analysis was done using J-Express Pro 2.6 software (MolMine AS, Bergen, Norway http://www.molmine.com). Abbreviations: BG, background; EGFP, enhanced green fluorescent protein; hrs, hours; RNAi, RNA interference.
Download figure to PowerPoint
To identify genes with altered expression levels as a result of OCT4 knockdown, we compared the 24- and 72-hour replicates, with EGFP knockdown at the same time points using statistical tests for differential expression (described in Materials and Methods). Three distinct tests (Student's t test, the Welch test, and Wilcoxon's rank sum test) were used to minimize individual bias . By adopting this approach, we identified a subset of 1,104 marker genes from the 72-hour time point. Of these marker genes, 399 (36%) were downregulated, and another 705 were upregulated. Both sets of marker genes were at a significance level of .05. For example, among these differentially regulated marker genes are previously characterized downstream targets of OCT4, such as the pluripotency markers NANOG and SOX2 and markers of undifferentiated stem cells ZFP42, LEFTY1, LEFTY2, DPPA4, THY1, FLJ10884, and TDGF1. The negatively regulated genes included EOMES, BMP4, fibroblast growth factor 8 (FGF8), DKK1, HLX1, GATA2, GATA6, ID2, and DLX5, which are implicated in differentiation processes [14, 15], as well as a large number of novel genes. These differentially regulated genes are listed in supplemental online Table 4. Furthermore, we computed Q values for each of the genes to assess statistical significance by the false discovery rate (FDR) . Using an FDR level of 0.05 identifies 721 of the more than 1,104 genes as significant. However, it should be stressed that FDR assessment can also increase the false-negative rate. For example, ID2, a differentiation marker directly regulated by OCT4 (supplemental online Table 6; Fig. 3) and verified as significant by p value computation, was rejected by FDR assessment. At an FDR level of 0.1, all genes were marked significant.
Figure Figure 3.. Confirmation of gene expression changes of selected genes by real-time PCR (A) and Western blotting (B). (A): SYBR green real-time reverse transcription-PCRs were carried out on RNA samples harvested 72 hours after OCT4 knockdown using two independent siRNA molecules in separate transfection experiments (designated siRNA 1 and siRNA 2). Error bars refer to technical variation. Ratios are represented as log ratio (base 2), with values above 0 denoting overexpression and values below 0 denoting repression of gene expression. (B): Western blots using duplicate samples of protein extract of the same set of transfection experiments. The specificities of the primary antibodies used are indicated on the left. Abbreviations: EGFP, enhanced green fluorescent protein; PCR, polymerase chain reaction; siRNA, small interfering RNA.
Download figure to PowerPoint
By comparing gene expression profiles from 24- and 72-hour-post-transfection samples, we selected genes that show changes in expression similar to those of the selected key genes OCT4, EOMES, and BMP4. The OCT4 hierarchical cluster is shown in Figure 2C and includes genes already implicated in the maintenance of pluripotency and self-renewal, such as NANOG; markers of undifferentiated stem cells LEFTY1, LEFTY2, DPPA4, and THY1; and novel genes.
To have an overview of negatively regulated downstream targets of OCT4, we repeated the profile analysis for EOMES and BMP4, respectively (supplemental online Figs. 1 and 2). EOMES encodes a T-box containing transcription factor expressed in the trophectoderm of human and mouse blastocyst and has been shown to be required for mouse trophoblast development and mesoderm formation [20, 26, 27]. The gene encoding bone morphogenetic protein 4 (BMP4) is also highly expressed in the human trophectoderm  and has been shown to promote human embryonic stem cell differentiation into trophoblast . The critical trophoblast stem cell regulator CDX2 could not be been included in this analysis because the clone is not represented on our array. Nonetheless, real-time polymerase chain reaction (PCR) analysis showed that its expression is induced upon depletion of OCT4 (Fig. 3).
Included in this set are genes implicated in ESC differentiation, such as (a) the transcription factors ID3, TBX18, GATA6, and HLX1 ; (b) signal transduction pathways crucial for the maintenance of pluripotency, such as WNT (DKK1, FRZB, FZD2), transforming growth factor β (TGFβ) (TGFB2), and FGF (FGF8); and (c) cytoskeletal and extracellular matrix-associated genes, such as MFAP5, COL4A1, KRT14, LOXL4, and CLDN4, which encodes a tight junction protein crucial for trophectoderm formation. These changes in gene expression overlap with those identified in BMP4-treated hESCs  and may reflect the morphological changes seen upon differentiation of hESCs per se and adoption of trophoblast fate (Fig. 1A, 1D).
Expression patterns of a selection of the differentially expressed genes were verified independently using real-time PCR and Western blotting of samples prepared from cells transfected with either of two different OCT4 siRNAs (Fig. 3). The results shown in Figure 3A confirmed the downregulation of key pluripotency-controlling genes, OCT4, SOX2, and NANOG, and ES-associated genes LEFTY1, LEFTY2, TDGF1, DNMT3B, and ZFP42.
Expression of mesodermal (T/Brachyury) and ectodermal (PAX6) markers was also downregulated, re-emphasizing the lineage-restricted differentiation of OCT4-depleted cells. The induction of CDX2 is in agreement with results obtained in mouse and human ESCs [8, 10], but it differs from our previous results with hESCs grown in mouse embryonic fibroblast (MEF)-conditioned medium, presumably reflecting the improved sensitivity of the current reverse transcription (RT)-PCR assays and inherent variability in MEF-conditioned medium preparations . Confirmation of gene expression characteristic of trophoblast lineage is supported by the upregulated expression of key markers, such as EOMES, CDX2, and BMP4 [9, 10, 17, 20]. The list of primer sequences used for this assay is given in supplemental online Table 5. Western blot analysis (Fig. 3B) shows downregulation of the transcription factors NANOG and SOX2 proteins upon OCT4 knockdown, consistent with their cooperation with OCT4 as part of an interdependent core hES transcriptional regulatory circuit [14, 15].
Downstream Targets of OCT4
Our experimental approach will identify both direct and indirect targets downstream of OCT4. To distinguish between these possibilities, we cross-checked the OCT4-regulated genes with the recently compiled set of transcription targets for OCT4, SOX2, and NANOG identified in hESCs using chromatin immunoprecipitation (ChIP) coupled to promoter microarrays and ChIP-pair end diTag (PET) using mouse ESCs [14, 15]. For the analysis, we defined direct targets as those regulated by OCT4 alone or in combination with NANOG and/or SOX2. We found poor overlap between the three data sets, with 60 genes common to our data set and that of Boyer et al. , 49 common to ours and the mouse data set , and nine common to all three data sets (supplemental online Fig. 3). Included in the nine genes are OCT4, NANOG, and SOX2 and the trophoblast inducer EOMES. Interestingly, CDX2 was a conserved target in human  and not mouse .
The list of direct targets of OCT4 in ESCs is given in supplemental online Table 6. An additional 162 are bound by NANOG and/or SOX2 but not by OCT4, indicating that these are regulated indirectly by OCT4 through its regulation of these downstream effectors (supplemental online Table 7).
In line with the finding that OCT4 regulates its own expression and that of other core hES transcription factors [14, 28, 29], it was not surprising that the subset of directly downregulated genes included NANOG, OCT4, SOX2, HMGB2, and NR2F2 (supplemental online Table 6). Interestingly, the promoters for HMGB2, a coactivator of OCT4 activity, and NR2F2, a regulator of OCT4 expression [30, 31], are positively and negatively regulated, respectively, by OCT4 alone. Additional markers of undifferentiated stem cells identified as positively regulated direct targets were LEFTY2, DPPA4, and TDGF1, whereas ZFP42, LEFTY1, and FLJ10884 were classified as indirect targets. As anticipated, components of signal transduction pathways implicated in the maintenance of pluripotency, such as WNT (DKK1, FZD2), TGFβ (NODAL, LEFTY1, LEFTY2, MADH3 ID2 and PITX2), FGF (FGF8 and FGF2), and Hedgehog (PTCH), are regulated by OCT4.
The remaining genes that are either up- or downregulated upon OCT4 knockdown but do not appear in the data sets of Boyer et al.  and Loh  may represent either previously undiscovered novel OCT4/SOX2/NANOG targets or genes regulated by downstream targets of OCT4 other than SOX2 and NANOG or simply not included in the promoter analysis (supplemental online Table 4). Moreover, this set of genes would also be expected to include a large number of genes involved in trophoblast differentiation.
Signaling and Metabolic Pathways Crucial for the Maintenance of Pluripotency
ESC self-renewal and pluripotency requires inputs from extrinsic factors and their downstream effectors . We therefore analyzed the 24- and 72-hour data set for components of signaling pathways by assigning p values using Wilcoxon's matched pair signed rank test to compare groups of genes associated with particular pathways instead of conventional gene-wise analysis . The analysis, summarized in Table 1, identifies changes in key components of the WNT, transforming growth factor (TGF)β, fibroblast growth factor (EGF), mitogen-activated protein kinase, NOTCH, Hedgehog, JAK/STAT, and extracellular matrix signaling pathways, as well as regulators of the cytoskeleton, apoptosis, cell cycle, and metabolic processes, such as oxidative phosphorylation, methionine metabolism, and folate biosynthesis. The list of complete KEGG annotated pathways identified as operative in ESCs is given in supplemental online Table 8.
Table Table 1.. Pathways of which gene components show significant expression changes between the OCT4 and EGFP knockdowns at 72 hours after transfection
Epigenetic Control of Pluripotency and Trophoblast Lineage Specification
Mutant mouse ESCs lacking DNA methyltransferase activity [45, 46], methyl DNA binding protein function , or histone acetylase activity  exhibit impaired differentiation, highlighting the critical role chromatin modification plays in regulating embryonic differentiation. To investigate whether the expression of chromatin and epigenetic modifiers is affected by OCT4 depletion, we identified differentially regulated genes involved in methyl/folate cycle, DNA methylation, methyl DNA binding, and histone modification communication (supplemental online Fig. 4). Downregulation of MAT2A (methionine adenosyltransferase II α) and MTRR (5-methyltetrahydrofolate-homocysteine methyltransferase) was observed at 72 hours of OCT4 depletion (supplemental online Fig. 4A), supporting the involvement of folic acid metabolism in the maintenance of pluripotency . The de novo methyltransferase DNMT3B was downregulated upon OCT4 depletion, contrasting with the maintenance methylase DNMT1, which showed no significant change. The, histone lysine methyltransferase (H3-K4-HMTase) SET7 was dramatically upregulated upon OCT4 knockdown, contrasting with downregulation of the histone lysine methyltransferase EZH2 (H3-K27-HMTase) (supplemental online Fig. 4B). The acetylases H2AFY and H2AFY2 and the deacetylase HDAC6 also show significant upregulation on OCT4 knockdown (supplemental online Fig. 4C, 4D). These expression patterns perhaps indicate that the hES-to-trophoblast transition is accompanied by significant changes in histone acetylation and methylation patterns.
The nonhistone chromatin-associated proteins HMGB1, HMGB3, DPPA4, NASP, chromatin assembly factor 1 (CHAF1A), PHF17, PHF5A, POLE3, and retinoblastoma binding protein 7 (RBBP7) are all significantly downregulated as a consequence of OCT4 depletion (supplemental online Fig. 4B). Of these, HMGB1, DPPA4, and HMGB3 have previously been shown to be highly enriched in undifferentiated stem cells and isolated ICM cells . The RBBP7-encoded protein has been shown to interact with MBD3 and may have a role in the regulation of cell proliferation and differentiation .
Interestingly, we also observed downregulated expression of PARP1 (Fig. 3), which encodes a chromatin-associated enzyme, poly(ADP-ribosyl) transferase, capable of modifying nuclear proteins. This DNA-dependent ribosylation has been shown to regulate cell proliferation, transformation, and differentiation . Indeed, Parp1-deficient mouse embryonic stem cells differentiate more readily into trophoblast-like cells, implying that one of the functions of this protein is to restrict differentiation in this lineage .
Altered expression of imprinted genes was also observed (upregulation of CALCR and GNAS and downregulation of KIP2 and UBE3A; supplemental online Fig. 4E). KIP2 (a cyclin-dependent kinase inhibitor) acts as a key regulator of embryogenesis through regulation of cell cycle by blocking the activity of G1 cyclin/Cdk complexes and the regulation of actin dynamics through binding to LIMK-1 [53, 54]. Differential gene expression of other cell cycle-related genes was also observed (supplemental online Fig. 5), including the upregulation of CDKN2B, a cyclin-dependent kinase inhibitor, and downregulation of CDC25A, which is required for progression from G1 to the S-phase of the cell cycle, consistent with the limited G1 phase and rapid cell cycle characteristic of undifferentiated ESCs [55, 56].
RNAi-Mediated Suppression of OCT4 Function in Human ESCs Recapitulates Primary Differentiation at the Blastocyst Stage of Development
To examine how closely the transcriptomes of an ICM and trophectoderm (TE) cell matches that of an undifferentiated ESC and OCT4 RNAi-mediated trophoblast cell, we compared our current data set with that derived from the blastocyst . We identified potential candidate genes that were overexpressed in the TE- and OCT4-deficient ESCs compared with the ICM and undifferentiated ESCs when a cluster analysis of genes coregulated in the same manner as BMP4 was performed (Fig. 6). Genes upregulated in the TE- as well as the OCT4-depleted hESCs are involved in the organization of the extracellular matrix (PDLIM3), cell growth and differentiation (AKR1C3 and PLXND1), transcriptional regulation (PME-1, MGC11349/ZXDC, and KIAA1245/COAS1), signal transduction processes (ARL7, PIP5K1C, RASL12, ARHGAP8, SELM, and DKK3, an inhibitor of the WNT pathway), and novel genes (KIAA1949, C14orf173, and FLJ20507/TMEM127). We do not present detailed analysis of this data set here for the simple reason that the vast amount of data is beyond the scope of this study. In summary, these results would imply that these TE marker genes could serve as additional factors required for inducing trophoblast differentiation and further propagation of these cells.
Figure Figure 6.. Cluster analysis of 15 genes bearing expression profiles the most similar to that of the trophoblast marker BMP4. The comparison was made between this study and the previously published expression data on human ICM and trophectoderm cells . Colors correspond to normalized signals. For each gene, signals were divided by the average gene signal across all conditions (log scale). Red boxes indicate that the signal in the particular condition is higher than the average signal, whereas green boxes indicate the opposite. Hierarchical clustering displays subgroups using Pearson correlation as a pairwise similarity measure and complete linkage as an update rule. The analysis was done using J-Express Pro 2.6 software (MolMine AS). Abbreviations: EGFP, enhanced green fluorescent protein; ICM, inner cell mass; MAX, maximum; MIN, minimum; RNAi, RNA interference.
Download figure to PowerPoint