|Home | About | Journals | Submit | Contact Us | Français|
Molecular regulation of embryonic stem cell (ESC) fate involves a coordinated interaction between epigenetic1–4, transcriptional5–10 and translational11,12 mechanisms. It is unclear how these different molecular regulatory mechanisms interact to regulate changes in stem cell fate. Here we present a dynamic systems-level study of cell fate change in murine ESCs following a well-defined perturbation. Global changes in histone acetylation, chromatin-bound RNA polymerase II, messenger RNA (mRNA), and nuclear protein levels were measured over 5 days after downregulation of Nanog, a key pluripotency regulator13–15. Our data demonstrate how a single genetic perturbation leads to progressive widespread changes in several molecular regulatory layers, and provide a dynamic view of information flow in the epigenome, transcriptome and proteome. We observe that a large proportion of changes in nuclear protein levels are not accompanied by concordant changes in the expression of corresponding mRNAs, indicating important roles for translational and post-translational regulation of ESC fate. Gene-ontology analysis across different molecular layers indicates that although chromatin reconfiguration is important for altering cell fate, it is preceded by transcription-factor-mediated regulatory events. The temporal order of gene expression alterations shows the order of the regulatory network reconfiguration and offers further insight into the gene regulatory network. Our studies extend the conventional systems biology approach to include many molecular species, regulatory layers and temporal series, and underscore the complexity of the multilayer regulatory mechanisms responsible for changes in protein expression that determine stem cell fate.
We applied a single well-defined perturbation to murine ESCs by downregulating Nanog, a key pluripotency factor13–15. A lentiviral-based complementation system was introduced into mouse ESCs in which short hairpin RNA (shRNA) depletes endogenous Nanog mRNA, and normal levels of Nanog expression are restored in a doxycycline-dependent manner from an shRNA ‘immune’ version7 (Fig. 1b). Previously, we showed that this engineered ESC clone is fully pluripotent in vitro and in vivo when maintained in the presence of doxycycline7. After doxycycline removal, Nanog mRNA and protein levels rapidly decline (Fig. 1c), and both pluripotency and self-renewal capacities of ESCs diminish with time. We collected data from four molecular layers. Specifically, we performed: (1) chromatin-immunoprecipitation microarray (ChIP-chip) analysis of histone H3 lysine 9 and 14 acetylation (acH3K9/14) at gene promoter regions to assess chromatin modification (designated as HIS); (2) ChIP-chip analysis of RNA polymerase II localization at 3′ exons of gene coding regions to reveal active transcription (designated as POL); (3) gene expression microarrays to quantify mRNA abundance (designated as RNA); and (4) protein mass spectrometry to measure nuclear protein abundance (designated as PRO) (Fig. 1a). Fold changes were calculated for each gene by comparing the expression levels of a molecular layer on days 1, 3 and 5 (doxycycline absent, Nanog depleted) to day 0 (doxycycline present, Nanog expressing), allowing for comparisons across the different experimental platforms (Supplementary Fig. 1). To estimate experimental noise, a significance threshold in each experiment was determined based on the experimental replicates of all measured genes at a false discovery rate (FDR) of 5% (Fig. 1d and Supplementary Fig. 2).
Although changes between different gene expression steps are generally correlated (Supplementary Fig. 3), both concordances and discordances exist on the individual gene level. The discordances show regulatory events that alter gene expression. We performed a supervised gene/protein classification to identify the key regulatory step that is most responsible for changes in protein levels, which directly determine cellular phenotype. We anchored our analysis on observed changes in protein levels and assessed the concordance of changes in the other three layers by comparing PRO to RNA, then RNA to POL, and finally POL to HIS (Fig. 2a). Proteins with significant changes were assigned to one of four categories at each time-point: category 1 proteins exhibit discordant PRO and RNA changes in expression, which is indicative of translational and posttranslational regulation; category 2 proteins exhibit concordant PRO and RNA changes in expression, but discordant RNA and POL changes in expression, which is indicative of post-transcriptional regulation; category 3 proteins exhibit concordant PRO, RNA and POL changes in expression, but discordant POL and HIS changes in expression, which is indicative of transcriptional regulation; and category 4 proteins exhibit concordant changes in expression across all four layers, which is indicative of regulation through chromatin modification. Proteins tend to stay in the same category over time (Supplementary Fig. 4). Category 1 constitutes 43–52% of all the genes with significant changes in protein levels, indicating that translational and post-translational regulatory mechanisms have important roles in ESC fate decisions11,12,16,17. However, it is unclear whether this is specific to stem cells or whether it is characteristic of other biological systems.
In addition to providing a genome-wide perspective of ESC fate change, our concordance analysis also provides useful information on the level of individual genes (Fig. 2b). For example, the ESC transcriptional regulator Esrrb7 falls into the category 2 concordance pattern at all time points. This indicates that ultimate levels of Esrrb protein are primarily regulated post-transcriptionally, at least under our experimental conditions, and not by direct Nanog regulation at the transcriptional level. It has been proposed that Esrrb and Nanog mutually regulate each other by a positive feedback circuit6,18. Our concordance pattern analysis of Esrrb indicates that at least one other component is likely to be involved in this circuit, which is responsible for the post-transcriptional regulation of Esrrb, possibly a microRNA19,20.
Gene-ontology analyses across the four molecular layers suggest a complex interaction between different molecular regulatory mechanisms in cell fate regulation (Fig. 2c and Supplementary Fig. 5). For example, differentiation- and development-related genes are over-represented among the genes that only show changes in acH3K9/14 levels, but not on the other three layers (Fig. 2c). Furthermore, chromatin- and nucleosome-assembly-related genes are overrepresented among the genes upregulated on the RNA polymerase II binding layer but not on any of the other three layers (Fig. 2c), suggesting that the chromatin modifiers are primarily regulated at the transcription step. Therefore, reconfiguration of chromatin structure, although an important factor in ESC fate alteration, may have a secondary role to primary regulation by transcription factors5,6,8,21–23.
To gain further insight into systems-level regulatory control of changes in ESC fate, we combined our data with that of previous stem cell regulatory network studies to form a new synthesis (Fig. 3)6,8,24. A core protein–protein interaction network was previously identified in murine ESCs involving 26 proteins centred around Nanog24. We found that this interactome is enriched in proteins that decreased in expression after downregulation of Nanog (Supplementary Fig. 6). On day 5, 8 out of the 26 interactome proteins are at significantly reduced levels (Supplementary Fig. 7). These are: Sall4, Rnf2, Oct4 (also known as Pou5f1), Ilf2, Nanog, Mybbp1a, Sall1 and Esrrb. Of these eight proteins only one (Rnf2) does not directly interact with Nanog (Fig. 3a). This suggests interdependence between the Nanog interactome and the network of genes under Nanog transcriptional control.
Nanog protein binds to thousands of genomic locations in undifferentiated ESCs5,6. Our data show that approximately 20% of the previously identified Nanog-binding genes change their transcription levels (POL) during the first 5 days after Nanog downregulation. Of those that changed, approximately 50% also exhibit changes in protein levels (PRO) (Fig. 3b and Supplementary Fig. 7). To determine how the changes in expression develop after the downregulation of Nanog, we analysed the temporal alterations of mRNAs in the context of an extended transcriptional regulatory network proposed previously8 (Fig. 3c). Our data show that most genes in this network are downregulated after the removal of Nanog. In particular, downregulation of Oct4 and Sox2 (protein levels shown in Supplementary Fig. 7) occurred later than downregulation of Klf4 or Rex1. This suggests that decreases in Oct4 and Sox2 expression are not responsible for decreases in Klf4 and Rex1 expression under our experimental conditions. The temporal sequence of changes in gene expression is loosely correlated with the chromatin-binding data6,8. These two sources provide independent and complementary information about the ESC gene regulatory network. Using the same principle that later molecular events cannot regulate earlier events, we can extract new sets of useful information concerning the gene regulatory relations from the temporal order of the network reconfiguration (Fig. 4 and Supplementary Fig. 8).
To facilitate comparisons and visualization of the multilayered time series, we generated interactive movies to display our data (Fig. 4 and Supplementary Fig. 8; http://amp.pharm.mssm.edu/ronglu). Expression changes for 400 genes with the most significant changes in protein levels on day 5 were projected onto two-dimensional hexagonal arrays (Fig. 4a). Individual hexagons representing specific genes are dynamically coloured according to the fold changes in each of the four molecular layers. This approach facilitates genome-wide and temporal comparisons among the different molecular layers, and allows clustering of genes with similar dynamics on multiple gene expression regulatory layers. We have also generated interactive scatter plot movies to help visualize concurrent changes across the different molecular layers (Fig. 4b). In these movies, individual genes can be selected to illustrate the concurrent changes between pairs of molecular layers. For instance, Fig. 4b demonstrates that changes in Esrrb mRNA and protein expression are monotonically related, whereas Sall1 and Oct4 both show increased mRNA levels without any corresponding increase in protein levels during the early stage of ESC differentiation. Similar dynamics are also exhibited by several other previously identified essential ESC factors25 (shown as red dots in Fig. 4b). These genes are regulated on different regulatory layer(s) compared to Esrrb, and suggest that the transcription layer undergoes an early cell fate reconfiguration without significant accompanying changes in protein production. Recent studies proposed that fluctuating levels of Nanog may discriminate between alternative pluripotent states of ESCs, in which high or low Nanog levels render ESCs resistant or susceptible to differentiation inducing stimuli, respectively15,26–29. In our system, the early time point of Nanog downregulation is comparable to the ‘low’ Nanog state from these studies. Thus, the absence of changes in protein levels during the mRNA layer reconfigurations could reflect the nature of these distinct pluripotent states. Collectively, the variety of the multilayered expression patterns underscores the complexity of the molecular regulation of ESC fate and suggests an intricate regulatory network involving several molecular regulatory layers.
In this study we have provided a dynamic multimolecular layer view of a murine ESC fate change in response to the downregulation of Nanog. In our experimental system the transcription of Nanog is regulated by exogenous manipulation and not by the endogenous regulatory circuit. This disrupts the balance of mutually regulated ESC molecular circuits15,26–29, and allows for rapid and synchronous cell fate changes within the population. However, our results nonetheless represent the average of a large cell population, as we have shown previously that removing Nanog results in a complex mixture of cell lineages7. In this work, our primary goal was to analyse the molecular dynamics that are associated with the transition away from the pluripotent state as it occurs in most of the cells. In vivo, cell fate alteration is probably triggered by several perturbations and inputs dynamically. The single gene perturbation that we have used does not reflect the natural signals that pluripotent cells are subjected to in vivo. However, it is a powerful tool to dissect the complex regulatory networks that underpin ESC fate changes and offers an initial window into the dynamic complexity of ESC fate regulation across multiple molecular levels.
AcH3K9/14 levels were assayed using ChIP-chip. Acetylated regions in a 1-kilobase window around the transcription initiation position were identified to generate acetylation profiles (Supplementary Figs 9 and 10). ChIP-chip was also used to measure RNA polymerase II localization on 3′ exons to directly assess transcriptional activity (elongation). Changes in mRNA levels were monitored using Agilent two-colour microarrays. Nuclear protein levels were measured using peptide isobaric tagging followed by two-dimensional liquid chromatography mass spectrometry (LC-MS/MS)16. We chose to measure nuclear protein levels because cell fate determination is largely controlled in the nucleus. For technical reasons, attempts to measure the entire proteome would have significantly decreased the sensitivity of the nuclear protein measurements, as these only constitute approximately 20% of all proteins in ESCs. All experiments were conducted in triplicate except for the acH3K9/14 measurements, which were performed in duplicate. Reliability of all data sets was verified using independent experimental assays, including conventional chromatin immunoprecipitation (ChIP), quantitative PCR (qPCR), and western blot assays for key pluripotency regulator genes (Supplementary Figs 11 and 12). Experimental reproducibility was also verified using a linear analysis of variance (ANOVA) model30. After data pre-processing and normalization, we were able to validate 1,627 nuclear proteins and 12,488 genes (HIS/POL/RNA) with high confidence. For 1,212 nuclear proteins, we were able to jointly obtain high-quality data across all four layers (HIS/POL/RNA/PRO). Supplementary Fig. 1 provides an overview of the entire data processing pipeline and the results of the quality-control procedures (ANOVA analysis). The significance of change is determined at a FDR of 5% using an empirical Bayes’ model with Benjamini–Hochberg correction on the basis of experimental replicates.
We would like to thank D. Storton for technical support, and E. Wieschaus, Y. Shi, S. Tavazoie and N. Slavov for constructive discussions. We also acknowledge the laboratories of the following people for providing antibodies for western blot: A. Okuda, J. Flint and Y. Kang. This work was supported by the NIH, and in part supported by the BBSRC and Leukaemia Research UK. O.G.T., F.M. and E.M.A. were partially supported by the NIH and US National Science Foundation.
Full Methods and any associated references are available in the online version of the paper at www.nature.com/nature.
Author Contributions R.L. and I.R.L. designed the experiments. R.L. prepared the cell samples for all the experiments, performed the RNA polymerase II ChIP-chip, the mRNA microarray, and verification experiments such as western blot, ChIP and quantitative PCR. R.D.U. and A.D.W. performed the proteomic experiments and primary analysis on proteomic data. L.A.B. performed the histone acetylation ChIP-chip experiments. R.L., F.M., E.M.A., R.R. and O.G.T. performed general data processing and statistical analyses. R.L. and F.M. plotted Figs 1–3. A.L., B.D.M. and A.M. developed and plotted interactive Fig. 4a. A.L. and A.M. developed and plotted interactive Fig. 4b. R.L., J.L., F.M. and I.R.L performed network analysis shown in Fig. 3. R.L. and I.R.L. wrote the paper.
Reprints and permissions information is available at www.nature.com/reprints.