The chromatin landscape of both mouse and human ESCs has been intensively investigated. In the pluripotent state, many developmental loci are marked with both activating H3K4me3 and repressing H3K27me3, and are thus termed “bivalent” (Azuara et al., 2006
; Bernstein et al., 2006
; Pan et al., 2007
). The notion is that these genes are simultaneously suppressed but poised for activation should the cell receive appropriate cues. Bivalent promoters have also been found at developmental loci in mouse embryos, both in the inner cell mass and trophectoderm, and also in zebrafish embryos (Dahl et al., 2010
; Lindeman et al., 2010
; Vastenhouw et al., 2010
). Other studies have shown that bivalent promoters are present in progenitor and adult stem cell populations, including neural progenitors, mesenchymal stem cells and hematopoietic stem cells, and that these ultimately resolve to either active or inactive upon differentiation (Collas, 2010
; Cui et al., 2009
; Mazzarella et al., 2011b
; Mohn et al., 2008b
We sought to characterize the epigenetic changes that occur during cardiovascular differentiation from human ESCs by performing genome-wide mapping of three histone modifications, H3K4me3, H3K27me3 and H3K36me3, at five key developmental timepoints. Our study shows that the temporal trajectories of H3K4me3 and H3K27me3 during differentiation are more complex than a simple `resolution from bivalency' model. As an example of this, FGF19 and NODAL are highly transcribed in human ESCs with high levels of H3K4me3 and low levels of H3K27me3 (Figure SD). They subsequently lose H3K4me3 and gain H3K27me3 over time. If one were only to have taken time points T3 (sometime between T2 and T5) and T14, one might conclude that the genes began in a bivalent state and then resolved towards repression, while in reality the `bivalent' appearance was merely an artifact of the complete reversal from H3K4me3 to H3K27me3. Yet another example is the set of genes involved in mesodermal differentiation (Fig S7), which are highly expressed despite being heavily marked with H3K27me3. These and other examples point to a complex regulatory relationship than cannot be described by a simple `resolution from bivalency' model.
Transcription factors and signaling molecules known to play critical roles in cardiovascular development, such as NKX2.5, showed a unique chromatin signature that consisted of high enrichment for H3K27me3 in pluripotent ESCs that gradually decreased as H3K4me3, H3K36me3 and RNA expression increased over time. In contrast, structural proteins like alpha-myosin heavy chain (MYH6) demonstrated markedly increased H3K4me3 enrichment and RNA expression at later timepoints, without early H3K27me3 repression. The differences in chromatin markings between genes encoding developmental regulators and structural proteins are consistent with previous studies comparing pluripotent and differentiated cells. However, our study is the first to recognize that the complex temporal chromatin patterns over a time course of differentiation actually contain a far richer amount of information regarding the exact function of the genes they are marking.
To use the information contained within the temporal chromatin signatures to identify novel regulators, we developed a classifier to rank genes according to the likelihood that a given gene would modulate cardiogenesis. Interestingly, several genes that ranked highly were transcription factors that had not been previously studied in cardiac development. We therefore tested our hypothesis that these genes were in fact previously unappreciated key regulators of heart development by utilizing morpholino knock-down technology in zebrafish embryos. Indeed, knockdown of meis2b resulted in severe defects in heart looping in early stage zebrafish embryos. This provides in vivo evidence that our methodology of coupling stem cell differentiation with histone modification pattern identification can identify regulators of development. It is worth noting that several of these novel regulators could be identified using the rank list of the top 50 genes based on expression alone (Supplemental Table 1). However, these regulators usually rank much lower down the list, and could easily get lost amidst the noise of so many structural genes. For example, MEIS2 ranked 4th at T5, 4th at T9 and 5th at T14 using the chromatin signature ranking but 49th at T5, 37th at T9 and 48th at T14 based on expression alone. Our dataset will thus serve as a resource for scientists and clinicians across multiple fields. The list of novel cardiac regulators is extensive and each potential candidate gene warrants further investigation to flesh out its role in differentiation and/or morphogenesis. One can imagine that perturbations of these regulators could alter cardiogenesis during human development. Thus, screening for genetic etiologies of congenital heart disease could be expanded to include our list of regulators.
It is interesting to consider why the chromatin dynamics of genes involved in developmental regulation are so different from structural genes that regulate the function of differentiated cells. The principal difference in chromatin signatures is the high degree to which developmental regulators are repressed by H3K27me3 prior to their expression, while structural genes show no such modification. We propose that the consequence of inappropriate activation of developmental regulators is more deleterious, e.g. by inducing the wrong cell type, proliferative state or survival/death signals. These genes therefore require both loss of repression and gain of activation to be expressed. Conversely, inappropriate activation of a gene encoding contractile proteins, ion channels or metabolic enzymes may have less severe consequences for development, and chromatin regulation through activation mechanisms achieves sufficient fidelity.
The establishment of human ESC technology opened the doors to analyses that can provide insights into human development that have not been possible before. Directed differentiation of human ESCs into numerous cell types has been one of the major advances in the field over the past 10 years. In addition to the cardiovascular directed-differentiation model system we utilized in this study, similar protocols exist for the generation of other cell types that are particularly relevant to regenerative medicine, including neurons (Lee et al., 2007
) and pancreatic beta-cells (Phillips et al., 2007
). Thus, our approach of mapping chromatin states over the course of differentiation could easily be applied to other cell lineages, thereby facilitating the identification of novel regulators of differentiation and development of other organ systems.
Several limitations of the approach used in this study are important to note. First, while the directed differentiation model is very robust, it is not perfect. We are able to obtain highly enriched populations of cardiovascular cells; however, non-cardiac cells are also present in our cultures. Therefore, we cannot exclude the possibility that some of our chromatin patterns might be due to the presence of other cell types, such as non-cardiac mesoderm and endoderm. That said, the cell populations analyzed in this study contained at least 80% progenitors at T5 and 50% cardiomyocytes at T14, so the majority of the chromatin patterns are likely informative for cardiac development. Another issue is that the efficiency of directed differentiation is often dependent on the particular ESC cell line used, or even the batch of such cells used. Thus, it will be critical for these experiments to be repeated in other cell lines beyond the H7 ESC line used in our laboratory. Lastly, although human ESCs provide a platform to model human development in vitro, it is presently unclear how well this differentiation system mimics in vivo development in terms of expression patterns and epigenetic changes, such as histone modifications. These limitations are common to many directed differentiation systems, and reflect the current standard issues shared by stem cell biologists worldwide. On the other hand, there are currently no other ways to study early events in human development, and we were able to utilize our cardiovascular directed-differentiation system to identify novel regulators of heart development. Additionally, our study showed that the temporal chromatin profiles along cardiomyocyte differentiation contained enough information to identify genes with likely unappreciated roles in neurectodermal development, even though the vast majority of cells in the population split from that cellular-fate quite early in differentiation. As the chromatin states along other differentiation pathways are measured, the information from all of these model systems can be integrated using methods similar to those described in this study to provide even richer insights into all branches of developmental regulation.