|Home | About | Journals | Submit | Contact Us | Français|
During development, diverse cellular identities are established and maintained in the embryo. Although remarkably robust in vivo, cellular identities can be manipulated using experimental techniques. Lineage reprogramming is an emerging field at the intersection of developmental and stem cell biology in which a somatic cell is stably reprogrammed into a distinct cell type by forced expression of lineage-determining factors. Lineage reprogramming enables the direct conversion of readily available cells from patients (such as skin fibroblasts) into disease-relevant cell types (such as neurons and cardiomyocytes) or into induced pluripotent stem cells. Although remarkable progress has been made in developing novel reprogramming methods, the efficiency and fidelity of reprogramming need to be improved in order increase the experimental and translational utility of reprogrammed cells. Studying the mechanisms that prevent successful reprogramming should allow for improvements in reprogramming methods, which could have significant implications for regenerative medicine and the study of human disease. Furthermore, lineage reprogramming has the potential to become a powerful system for dissecting the mechanisms that underlie cell fate establishment and terminal differentiation processes. In this review, we will discuss how transcription factors interface with the genome and induce changes in cellular identity in the context of development and reprogramming.
During development, cell fate is established and maintained by complex regulatory networks of transcription factors that promote expression of cell-type specific gene products and repress regulators of other lineages. Once established, cellular identity is remarkably stable despite numerous intrinsic and extrinsic perturbations. This stability is likely the result of a combination of multiple molecular features including cis-acting epigenetic modifications, such as DNA methylation, post-translational modifications of histone tails, nucleosome positioning, incorporation of histone variants into nucleosomes, and trans-acting regulatory factors, such as sequence-specific DNA-binding transcription factors, transcriptional co-activators, non-coding RNAs and chromatin remodeling complexes (Graf and Enver, 2009; Ho and Crabtree, 2010; Yamanaka and Blau, 2010). Although generally stable in vivo, under certain experimental conditions cell fate can be dominantly reprogrammed by forcing expression of transcription factors involved in the establishment and maintenance of a distinct cellular lineage (Figure 1). Identifying the relevant stimuli that can reprogram one cell into another cell type of interest and understanding how this process occurs are two key goals for the reprogramming field.
In this review we will summarize the critical discoveries to date but only briefly discuss applications of cellular reprogramming technologies for understanding human disease and regenerative medicine (for more detailed reviews on these topics see (Graf, 2011; Gurdon, 2006; Holmberg and Perlmann, 2012; Saha and Jaenisch, 2009; Vierbuchen and Wernig, 2011)) and instead focus on selected discoveries that have helped to identify the mechanisms that promote successful lineage reprogramming. Furthermore, we propose that the non-physiological direct lineage reprogramming approaches will be useful for studying physiological mechanisms of transcriptional reprogramming, such as the establishment of cellular identity, as well as the transcriptional regulatory networks that drive terminal differentiation and functional maturation (Bussmann et al., 2009; Graf and Enver, 2009; Vierbuchen and Wernig, 2011).
Seminal work by Briggs, King, and Gurdon in the 1950’s demonstrated that the stability of the differentiated state is not the result of irreversible genomic changes that occur during differentiation (Briggs and King, 1952; Gurdon et al., 1958). This was demonstrated by somatic cell nuclear transfer (SCNT), a technique in which intact Xenopus nuclei from embryonic or adult cells are transferred into an enucleated oocyte. Gurdon used this system to demonstrate that nuclei from endoderm cells taken from tailbud stage frog embryos could successfully control the development of new tadpoles. Later work showed that even nuclei from terminally differentiated adult cells (e.g. blood cells, skeletal muscle, kidney, and others) could generate Xenopus larvae following nuclear transfer, albeit at reduced efficiency compared to nuclei from embryonic cells (Gurdon, 2006; Pasque et al., 2011). These results indicated that the oocyte contained powerful trans-acting reprogramming factors that could effectively erase somatic epigenetic marks and return nuclei from differentiated cells to a pluripotent state. However, it was not clear whether these results were a testament to the unique molecular properties of oocytes or to the inherent plasticity of epigenetic modifications acquired during development.
Experiments by Taylor and Jones provided initial evidence that somatic cells could also exhibitcell fate plasticity (Taylor and Jones, 1979). They found that treatment with 5-azacytidine, an inhibitor of DNA methylation, caused fibroblasts to spontaneously differentiate into muscle and fat cells. This suggested that DNA methylation is important for preventing expression of genes that regulate differentiation into alternative lineages. In the early 1980’s Blau and colleagues demonstrated that multinucleated myotubes could dominantly reprogram nuclei from other cell types to express muscle-specific gene products in heterokaryons (artificially fused cells that maintain distinct nuclei), suggesting that reprogramming activity was not unique to the oocyte and that the terminally differentiated state was actively maintained by specific groups of trans-acting factors (Blau et al., 1983). In 1987 Weintraub and colleagues showed that the basic helix-loop-helix (bHLH) transcription factor MyoD is sufficient to convert fibroblasts into contracting myocytes (Davis et al., 1987). However, when MyoD was expressed in cell types from different germ layers (i.e. retinal pigment epithelium, melanocytes, hepatocytes) activation of muscle markers was sometimes observed but complete reprogramming failed (Weintraub et al., 1989). This suggested that a single transcription factor can be sufficient to initiate and control the differentiation of a specific cell type, which provided a possible mechanism for the control of terminal differentiation processes during development (Weintraub, 1993). Gehring and colleagues also provided dramatic proof of this principle by showing that ectopic expression of the transcription factor eyeless (Pax6 in mammals), a master regulator of eye development, could generate functional eyes at various sites on the body (Halder et al., 1995).
In 1996, Wilmut and colleagues successfully generated live offspring from the nucleus of a mammalian somatic cell (Campbell et al., 1996). This important breakthrough re-ignited interest in the field of epigenetic reprogramming. Nuclear transfer was quickly accomplished in a variety of other species, but the pairing of mouse genetics and nuclear transfer technology proved to be especially fruitful, leading to a variety of insights into the biology of pluripotency and the epigenetic control of cell type specification (Hochedlinger and Jaenisch, 2006; Wakayama et al., 1998). For example, mice were generated by SCNT from the nuclei of adult lymphocytes and olfactory neurons using a modified two-step nuclear transfer procedure that involved first creating NT-ES cells followed by injection into tetraploid blastocysts (Figure 1B) (Eggan et al., 2004; Hochedlinger and Jaenisch, 2002).
Encouraged by work in the SCNT field, as well as the demonstration that fusion of embryonic stem cells and fibroblasts could activate pluripotency markers in somatic nuclei (Cowan et al., 2005; Tada et al., 2001), Yamanaka and colleagues hypothesized that reprogramming factors could be identified by their specific expression in pluripotent cell types (Mitsui et al., 2003; Takahashi and Yamanaka, 2006). Surprisingly, the combined expression of 24 ES cell-specific genes in mouse fibroblasts yielded colonies of cells with pluripotent properties. After systematic elimination, the four transcription factors Oct4, Sox2, Klf4, and c-myc (OSKM) were shown to be sufficient for this process and further studies proved that these “induced pluripotent stem (iPS) cells” were molecularly and functionally equivalent to ES cells including their capacity to contribute to the germline (Maherali et al., 2007; Okita et al., 2007:Wernig, 2007 #19). Again a small group of transcription factors was able to recapitulate complex developmental processes, similar to the MyoD experiments mentioned above. However, an unresolved issue was whether reprogramming to pluripotency was fundamentally different than reprogramming to other somatic cell types. The pluripotent state has been conceptualized as the “ground state” of cellular identity, and thus the pluripotent state could represent a default response to the erasure of somatically acquired epigenetic marks (Silva and Smith, 2008). However, reprogramming from one somatic cell state to another would theoretically require a highly specific erasure of the epigenetic marks of one lineage, followed by the establishment of a new set of epigenetic features characteristic of the new cell state. It is hard to conceive how a transcription factor could directly control such a process, as it seems unlikely that, for example, promoter and enhancer elements of neuron-specific genes would be accessible for transcription factor binding in fibroblasts or hepatocytes. For these reasons it was assumed that lineage reprogramming was possible between closely related cell types, such as fibroblasts-myocytes, lymphocytes- macrophages, or astrocytes-neurons, because they are likely to share some epigenetic features as a result of their recent descent from a common progenitor cell, and would thus provide a chromatin landscape that was permissive for reprogramming factor binding and activity (Graf and Enver, 2009; Vierbuchen and Wernig, 2011; Zhou and Melton, 2008). For example, Graf and colleagues had provided convincing evidence for direct conversion of mature B cells into macrophages (Xie et al., 2004), Slack and colleagues demonstrated acquisition of hepatic properties in pancreatic cells (Shen et al., 2000), Gotz and colleagues showed induction of neuronal traits in glial cells (Berninger et al., 2007; Heins et al., 2002), and Melton and colleagues provided evidence for the conversion of exocrine to endocrine pancreatic cells following in vivo delivery of three transcription factors (Zhou et al., 2008). However, like previous studies with MyoD, the transcription factors used by Melton and colleagues were insufficient to reprogram cells representing a different germ layer such as embryonic fibroblasts (in vitro) or skeletal muscle cells (in vivo) to endocrine pancreatic cells, suggesting again that reprogramming between distantly related somatic cells might not be possible.
In an attempt to clarify this issue, we decided to test whether fibroblasts (representing a mesodermal lineage) could be reprogrammed into neurons (representing an ectodermal lineage). Remarkably, we observed that the forced expression of the three transcription factors Brn2, Ascl1, and Myt1l (BAM) was sufficient to convert fibroblasts to fully functional neuronal cells that we termed induced neuronal (iN) cells (Vierbuchen et al., 2010). In the following year other groups used a similar approach to show that fibroblasts can be reprogrammed into functional cardiomyocyte-like and hepatocyte-like cells (Huang et al., 2011; Ieda et al., 2010; Sekiya and Suzuki, 2011) (Figure 2). These data confirmed that reprogramming between distantly related somatic lineages is possible without passing through the pluripotent state, and suggested that reprogramming fibroblasts into pluripotent stem cells might represent an active conversion process rather than a return to the default cellular state following erasure of chromatin marks associated with cellular specification.
As cells differentiate during development, they acquire lineage-specific patterns of epigenetic modifications that reinforce cell fate decisions and promote the faithful transmission of cellular identity during cell division (Ho and Crabtree, 2010). Thus, the pre-existing chromatin state of a cell represents a potential roadblock to lineage reprogramming. Experiments performed in a variety of reprogramming systems suggest that lineage conversion tends to be more difficult the more distantly related the two cell types are. For example, Gurdon and colleagues showed that the efficiency of deriving swimming tadpoles decreases dramatically when nuclei from more differentiated cells are used as donors for nuclear transfer (Gurdon, 2006; Pasque et al., 2011). Similarly, nuclei from mouse ES cells generate NT-ES cells following nuclear transfer at much higher efficiency than more differentiated cells, and generation of live pups from terminally differentiated cell types (e.g. olfactory neurons and T cells) by NT is extremely inefficient without passage through a NT-ES cell state (Hochedlinger and Jaenisch, 2006; Yamanaka and Blau, 2010). However, iPS cells have been derived from most tissue types tested, albeit with variable efficiency, and in some cases specific experimental modifications were required for success (Hochedlinger and Plath, 2009).
Cell fusion experiments also indicated that cellular identity affected the transcriptional response to ectopic trans-acting factors. For example, although myotube heterokaryons could activate muscle-specific genes in nuclei of cells derived from all three developmental germ layers, nuclei from endodermal and ectodermal lineages exhibited slower kinetics of transcriptional activation, which suggested that the lineage-specific patterns of epigenetic information determined the response to the reprogramming factors found in myotubes (Blau et al., 1985). Building on the experiments of Taylor and Jones (discussed above), Blau and colleagues determined that pre-treatment of fusion partners with 5-azacytidine could elicit expression of muscle-specific genes from previously non-responsive HeLa cell nuclei following fusion with myotubes, which suggested that pre-existing DNA methylation in fibroblast nuclei is one barrier that prevents ectopic induction of muscle genes during reprogramming (Chiu and Blau, 1985).
More recent studies have established heterokaryons between somatic cells and pluripotent stem cells. In this system, nuclei from somatic cells rapidly activate genes associated with pluripotency, albeit at low levels compared to the ES cell nuclei (Bhutani et al., 2010; Pereira et al., 2010; Piccolo et al., 2011). However, due to the nature of heterokaryon formation it is difficult to analyze the extent and stability of fusion-mediated reprogramming. Heterokaryons have also been used to investigate the requirement of specific genes for reprogramming activity in the pluripotent fusion partner. For example, transcriptional activation of some pluripotency genes required AID-mediated DNA demethylation (Bhutani et al., 2010), and Oct4 (but not Sox2) and polycomb complex activity were also required for complete activation of pluripotency genes in human lymphocytes (Pereira et al., 2010; Pereira et al., 2008). Recently, Lahn and colleagues generated stable, dividing hybrid cell lines from fused rat fibroblasts and mouse ES cells but were unable to detect AID expression in this system, perhaps as a result of the different methods used (Foshay et al., 2012). Further studies are certainly needed to clarify the details of reprogramming kinetics in hybrid cell lines and heterokaryons, and to what extent the ratio of nuclei in fusion products and the species of donor nuclei affects reprogramming activity (e.g. (Palermo et al., 2009)).
In myotube-fibroblast heterokaryons it has been proposed that not all relevant myogenic genes are activated in fibroblast nuclei (Gaetz et al., 2012). This suggests that some genes can be readily activated by trans-acting factors while others might be kept silent by specific cis-acting epigenetic modifications (referred to as “occluded” genes) (Lahn, 2011). Surprisingly, when bacterial artificial chromosomes (BAC) containing these “occluded” genes (e.g. Myf5) with their associated regulatory regions are transfected into mouse fibroblasts, they tend to be transcriptionally active. However, in fibroblasts from Myf5/6 BAC-transgenic mice the Myf5-BAC reporter was not activated. This suggests that cell-type specific cis-acting epigenetic marks acquired during development prevent the expression of occluded genes in heterokaryons and that fibroblasts do not initiate epigenetic silencing of these loci de novo. No specific DNA methylation status, histone modification pattern or binding of chromatin factors was found in promoters of “occluded” genes, raising the possibility that enhancer elements may be more informative or that other as yet unidentified marks could provide an explanation for the differential gene activation (Lahn, 2011). Unbiased genome-wide analysis of histone modifications may provide further insight into the chromatin dynamics of those two classes of loci.
Given its muscle inducing activity in fibroblasts MyoD is certainly a strong candidate for a trans-acting factor responsible for reprogramming in fibroblast-myotube heterokaryons. It is clear though that there are other factors involved in heterokaryon reprogramming, For example, MyoD was not sufficient to activate muscle genes in cells derived from different germ layers, such as hepatocytes, whereas cell fusion was(Miller et al., 1988). Furthermore, heterokaryon formation between MyoD-expressing hepatocytes and fibroblasts could also activate muscle-specific genes in the hepatocyte nucleus showing that fibroblasts contain additionally required trans-acting factors (Schafer et al., 1990).
It is important to point out that in cell fusion and nuclear implantation experiments cells are exposed to much lower (i.e. closer to physiological) levels of the reprogramming factors, which are limited to genes expressed in the fusion partners (Figure 1). Although iPS cell reprogramming factors are all expressed in ES cells, the transcription factors used for lineage conversion in some other tissues are not always expressed in the target cell type. For example, the key iN cell reprogramming factor Ascl1 is not expressed in mature, differentiated neurons, and therefore fibroblast-neuron heterokaryons might fail to activate neuronal genes in fibroblast nuclei (although to our knowledge this experiment has not yet been performed). Thus, it is important to consider that in this context a failure to reprogram is not necessarily due to insurmountable barriers encoded in chromatin, when it can also be explained by the absence of essential reprogramming factors or co-factors (as was demonstrated in MyoD-expressing hepatocytes fused to fibroblasts).
As described above, it is now well established that transcription factors can induce distantly related cell fates (Figure 2) (Vierbuchen and Wernig, 2011). We believe these new experimental systems provide an excellent opportunity to investigate the mechanisms of reprogramming. Compared to cell fusion methods, reprogramming with transcription factors is a much simpler experimental system because the reprogramming factors are more clearly defined. Currently, insights into direct reprogramming between distantly related cell types have been gained exclusively by studying the mechanism of reprogramming cells to pluripotency using the OSKM factors. These studies have benefited from rigorous functional assays and the fact that the transcriptional networks that control pluripotency are relatively well understood. However, the utility of this system for understanding the mechanisms controlling reprogramming is limited by the low efficiency of reprogramming, the prolonged period required to generate pluripotent cells, the stochastic nature of certain events required for the acquisition of pluripotency and the multitude of cell divisions that are required (Hanna et al., 2010). For these reasons, direct conversion of fibroblasts to neurons and other distantly related cell types provide a perfect complement to the iPS cell system for understanding reprogramming processes. For example, iN cell reprogramming using mouse embryonic fibroblasts can reach efficiencies of 20% and generally occurs with a maximum of one cell division (Vierbuchen et al., 2010). When more closely related neonatal astrocytes were used as donor cells, neuronal induction required fewer factors and was more efficient and rapid (Heinrich et al., 2010). Also, MyoD-mediated induction of the muscle fate was reported to be up to 50% efficient and dramatic morphological changes (e.g. formation of multinucleate cells) occurred within 3 days after induction (Davis et al., 1987). Similarly, in vitro conversion of a pre-B cell line into macrophages by forced expression of C/EBPα can induce phenotypic characteristics of macrophages as early as 10 hours after induction with nearly 100% efficiency (Bussmann et al., 2009). The complementary nature of these tools should provide novel insights into the transcriptional and epigenetic changes that occur during reprogramming and can begin to elucidate general principles of reprogramming that are shared between diverse systems. In summary, both the epigenetic state of the donor cells and trans-acting factors regulate the activation of previously silent genes in the context of lineage reprogramming. However, it is also clear that specific experimental manipulations, such as treatment with histone deacetylase inhibitors (HDAC inhibitors), 5-azacytidine or other inhibitors of epigenetic maintenance can help to increase the efficiency and fidelity of reprogramming in various reprogramming protocols (Xu et al., 2008).
For reprogramming to occur it is probably necessary that transcription factors be able to access a majority of their relevant binding sites (i.e. the sites that they bind to in order to regulate gene expression during normal differentiation processes from tissue-specific stem cells). However, due to the preponderance of potential binding sites in the genome for most transcription factors, binding is thought to be constrained to a specific subset of potential sites by both the cell type-specific chromatin context in which the transcription factor is normally expressed and/or by a requirement for cell type-specific co-factors that bind to the adjacent regulatory elements and promote stable interaction with DNA. For example, the bHLH transcription factor Scl/Tal, which recognizes the generic E-Box motif CANNTG, binds to a different set of sites in the different hematopoietic cell types in which it is expressed (Palii et al., 2011; Wilson et al., 2010). Similarly, certain bHLH transcription factors also control differentiation more than one type of cell (e.g. Ngn3, Ascl2) during development, which could also involve binding to different regulatory regions in the genome and/or interaction with cell-type specific co-factors (Bertrand et al., 2002). With respect to use of transcription factors in reprogramming, these observations suggest a potential paradox; how is specificity of DNA binding achieved in the absence of some or all context-specific cues? As we will discuss in greater detail below, one potential explanation is that the reason multiple factors are required for reprogramming between distantly related cell types in order to provide the cooperative interactions necessary to confer binding specificity or stability lacking in the cell type to be reprogrammed. A second possibility is that the reprogramming factors can act as “pioneer” transcription factors meaning that they can recognize their target sites irrespective of the pre-existing chromatin state (Zaret and Carroll, 2011). These scenarios are not mutually exclusive, and could provide a conceptual starting point to study how successful reprogramming is accomplished and why under many conditions reprogramming fails (See Figure 3 for a summary). It also raises important questions about how transcription factors regulate cellular differentiation during development. For example, how does cell lineage affect the activity of lineage-specific transcription factors (Tapscott, 2005)? More specifically, how do transcription factors initiate cell-type specific differentiation programs when initially activated in a stem/progenitor cell type?
It is generally believed that cells sequester unneeded genes into densely packed heterochromatin, which is thought to inhibit binding of transcription factors, thus preventing activation of transcription at that locus (Beisel and Paro, 2011). The combination of classic methods to study accessibility, such as DNAse sensitivity, with high-throughput sequencing have made it possible to map accessible regions genome-wide, which has begun to provide insight into the relationship between transcription factor binding and DNA accessibility on a global scale (Bell et al., 2011). Careful studies of the PHO5 promoter in yeast have demonstrated the functional consequences of DNA accessibility on transcriptional activation. Nucleosomes positioned in the PHO5 promoter region limit Pho4 binding to a single accessible site in the promoter, which allows Pho4 to to recruit chromatin remodeling proteins that remove or shift these nucleosomes and make additional Pho4 binding sites in the promoter accessible (Almer and Horz, 1986; Bell et al., 2011). Similarly, on a genome-wide scale, cell type specific patterns of DNA accessibility appear to be important for limiting binding of the glucocorticoid receptor (GR) to a subset of its response elements (John et al., 2011). These data provide compelling evidence that nucleosome positioning plays an important role in restricting transcription factor binding to specific sites. However, during reprogramming transcription factors need to gain access to (presumably) occluded cis-regulatory regions. Thus, the success of reprogramming suggests that these cis-acting roadblocks can be overcome by forced expression of transcription factors (Figure 3).
In the “immediate access” scenario, transcription factor access to the critical cis-regulatory elements of a gene is rapid and unimpeded by cis-acting chromatin modifications. This could happen in one of two ways. First, target regions could be in a heterochromatic, “repressive” state (e.g. trimethylated histone H3 lysine 27 (H3K27me3 enriched), bound by polycomb complexes, nucleosome-enriched). Alternatively, the region of interest could be in a pre-existing “permissive” chromatin state (e.g. H3K4me1 enriched, nucleosome-depleted) that allows unimpeded binding of reprogramming factors. The first scenario would suggest that reprogramming can only be accomplished by pioneer transcription factors (Zaret and Carroll, 2011). Pioneer transcription factors are thought to be the first to bind to lineage-specific regulatory elements during organogenesis, which can help to displacenucleosomes (either through active or passive mechanisms) and create a permissive binding environment for other regulatory factors that do not have pioneer activity. This model is supported by the fact that many transcription factors used for lineage conversion are also thought to act as pioneer factors during development and have been shown to be capable of binding to nucleosomal DNA in vitro (e.g. GATA and FOXA family transcription factors, see Figure 1) (Cirillo et al., 2002). The idea that many sites are in fact permissive for binding is intriguing because it would be a simple molecular explanation for why closely related cell types are more amenable to lineage-conversion. Jones and colleagues recently provided experimental support for this idea (Taberlay et al., 2011). Using a well characterized MyoD enhancer and promoter as an experimental starting point they examined how the pre-existing chromatin configuration of these autoregulatory elements affects MyoD binding in various cell lines that do not express MyoD. They found that a minimal MyoD enhancer element could exist in a “permissive” state for MyoD binding (nucleosome-depleted, H3K4me1-enriched, flanked by H2AZ containing nucleosomes, not bound by polycomb complex components or enriched for H3K27me3) even when the promoter region exhibited a “repressive” chromatin state (see Figure 3). This “multivalent” epigenetic state of the MyoD locus allowed for MyoD binding to the enhancer element in the first 24 hours afterectopic MyoD expression, which caused nucleosome displacement at the promoter region and the acquisition of H3K4me3 enrichment by 48 hours, but was insufficient to induce transcription from the MyoD locus. Conversely, in a colorectal cancer cell line in which both the MyoD promoter and proximal enhancer were nucleosome-bound, ectopic MyoD binding was not observed. Thus, the presence of enhancer elements in a permissive state might be able to predict whether a gene can be activated by reprogramming factors. Using computational approaches and publicly available ChIP-seq data, the authors found that this type of permissive enhancer is present at a substantial fraction of polycomb-repressed genes across a wide range of human cell types, providing a potential explanation for global epigenetic remodeling induced by ectopic transcription factors. Of note, using ChIP-seq, MyoD was recently shown to bind predominantly to accessible sites in fibroblasts (Fong et al., 2012).
Interestingly, the MyoD enhancer also contains an Oct4 binding site. When Oct4 is expressed in cell lines with a permissive MyoD enhancer, Oct4 binds almost immediately to the enhancer and promotes formation of a bivalent chromatin state at the MyoD promoter, which is similar to the chromatin state of the MyoD promoter in ES cells. This suggests that diverse reprogramming factors can gain immediate access to critical regulatory elements at individual genes through these permissive enhancer elements and catalyze reprogramming of the chromatin status of the promoter to promote changes in cell fate.
It will likely be informative to examine how these permissive enhancers are established and maintained during normal differentiation processes, and to determine to what extent their locations vary between different cell types. Recent work has identified a chromatin signature for “active” enhancer elements, which are distributed across the genome in highly cell type specific patterns (Creyghton et al., 2010; Rada-Iglesias et al., 2011). The identification of permissive enhancers genome-wide could help to provide a rational method for determining the potential for one cell type to be converted into another. These observations could also provide a mechanistic explanation for the “occluded” genes observed in heterokaryon experiments (Lahn, 2011). For example, a transcriptionally silent gene would be occluded if its enhancers are not permissive for DNA binding. However, ChIP-seq studies of MyoD binding in fibroblasts and myotubes showed no evidence of dramatic differences in MyoD occupancy between these two cell types, consistent with the fact that MyoD can efficiently reprogram fibroblasts to functional muscle cells (Cao et al., 2010). The immediate access model predicts that reprogramming factor binding at critical target genes should be detectable shortly after introduction into the cells. Most studies of reprogramming have focused on later timepoints, so surprisingly little is known about global changes in transcription and chromatin modifications during the earliest phases of reprogramming. Work by Tapscott and colleagues detailing the kinetics of gene activation following MyoD expression in fibroblasts indicated that MyoD targets exhibit gene-specific kinetics of activation (Bergstrom et al., 2002). For example, a subset of MyoD targets can be activated within 6 hours of MyoD induction, and in some cases these genes can be activated in the absence of protein synthesis, suggesting that they are activated directly by MyoD (potentially with help from fibroblast trans-acting factors). At genes that exhibit slower kinetics of activation, MyoD binding is not detected at early timepoints, but appears to increase concomitantly with transcriptional activation, suggesting that MyoD either does not have access to these regions at early timepoints or that it can access the DNA but is missing co-factors that might stabilize its interaction with these sites. It is also possible that binding to enhancers of late stage genes precedes binding to the promoter regions assessed for MyoD binding in this paper, as suggested by Jones and colleagues (discussed above) (Taberlay et al., 2011). Repeating these experiments and measuring genome-wide MyoD binding at similarly early timepoints could begin to clarify this issue. These data also demonstrate the idea that feed-forward activation of gene expression by MyoD helps to pattern the temporal activation of its target genes even when it is expressed ectopically in fibroblasts (Tapscott, 2005). Because the regulatory regions of each of its target genes have specific requirements for activation, correct temporal activation of MyoD downstream genes thought to be controlled by different requirements for cooperating factors (e.g. bFGF signaling, p38 and MAPK signaling, and MEF2, Pbx/Meis and Six family transcription factors), some of which are downstream targets of MyoD themselves (Aziz et al., 2010).
Graf and colleagues observed even more rapid kinetics of reprogramming during in vitro lineage conversion between B cells and macrophages by forced expression of the transcription factor C/EBPα (Bussmann et al., 2009). In these studies, induction of C/EBPα in a highly homogeneous B cell line leads to dramatic gene expression changes within 3 hours. Further analysis of C/EBPα binding and the pre-existing chromatin state at the macrophage-specific gene regulatory elements in B cells might clarify the influence of chromatin on reprogramming factor binding and transcriptional activation in this reprogramming system.
The idea of immediate access is also consistent with the finding that short term expression of iPS cell reprogramming factors predominantly induces the activation of genes whose promoters are marked by H3K4me3 in fibroblasts (Koche et al., 2011). Surprisingly, out of 5 histone modifications examined, only the H3K4me2 mark appeared to be dynamic at early stages of reprogramming, but these changes were not associated with transcriptional activation. Instead H3K4me2 enrichment was found largely at putative enhancer elements (defined in this case as non-promoter elements with H3K4me2 enrichment), which showed a shift away from fibroblast-specific enhancers to ES-cell specific enhancers. In a related study, Plath and colleagues examined promoter binding of the iPS cell reprogramming factors in both partially reprogrammed and fully reprogrammed mouse iPS cells (Sridharan et al., 2009). During reprogramming, stable lines of partially reprogrammed cells can be established that share some morphological, functional and molecular characteristics with ES cells, but have failed to activate the complete program of pluripotency (Meissner et al., 2007). It is also known that in clonal populations of partially reprogrammed cells, complete reprogramming can occur spontaneously at a low frequency, suggesting that stochastic events are required for the transition to a fully pluripotent state (Meissner et al., 2007). Interestingly, the OSKM factor binding sites in partially reprogrammed cell lines were substantially different than those in ES cells. The authors proposed that differential binding might be explained by the absence of Nanog (and potentially other factors) in the partially reprogrammed cells, as the Nanog DNA binding motif was enriched in sites bound in ES cells but not in partially reprogrammed cells. These data are consistent with reprogramming occurring as a step-wise process, in which access to certain DNA regulatory regions is dependent on specific-cofactors that are induced during the process of reprogramming and argue against a pioneering mechanism for the iPS cell reprogramming factors.
In another model, a transcription factors gains access to important regulatory regions by means of a non-physiological mechanism. For example, high levels of reprogramming factor expression, which are typical of lineage reprogramming experiments, might allow for limited and transient access to important binding sites during normal nucleosome turnover or stochastic unwinding of nucleosomal DNA, which causes the nucleosome to be displaced and allows for the subsequent recruitment of chromatin modifying complexes to stably remodel the chromatin at this locus (Li et al., 2005). Indeed, kinetic studies of reprogramming to pluripotency have suggested that stochastic events, presumably related to the cell-cyle, are likely to be rate-limiting during certain phases of reprogramming (Hanna et al., 2009). For example, the increased efficiency of reprogramming in p53-deficient cells is cell cycle dependent, and targeted chromatin remodeling seen during early phases of reprogramming seems to be enhanced in cells that have undergone multiple rounds of cell division (Koche et al., 2011). How might cell division promote reprogramming? During cell division, nucleosomes must be partitioned to newly synthesized DNA, and it is not known to what extent histone modifications are replaced when newly synthesized histones are incorporated into nucleosomes (Probst et al., 2009). Presumably, new and recycled histone subunits are distributed randomly among the two double helixes. Thus, cell division could provide a window for transcription factor access to otherwise occluded cis-regulatory regions, which could prevent the subsequent re-establishment of repressive marks or positioned nucleosomes at these loci.
Although cell cycle obviously plays an important role in iPS cell reprogramming, many somatic lineage reprogramming experiments do not require cell division such as iN cell reprogramming, B-cell to macrophage, and astrocyte-to-neuron conversion (Di Tullio and Graf, 2012; Heinrich et al., 2010; Marro et al., 2011; Vierbuchen et al., 2010). In these cases, transient transcription factor binding could also be mediated by stochastic, localized alterations of histone marks due to random activity of chromatin modifying enzymes, referred to as chromatin “breathing”. In agreement with this idea is the observation that pharmacological HDAC inhibition has been shown to promote reprogramming in a variety of systems (Huangfu et al., 2008; Xu et al., 2008). HDAC activity might be important for reinforcing specific chromatin states at gene regulatory elements, which when disrupted could potentially lead to global changes in the patterns of occluded genes. It is also possible that HDAC inhibition could cause transcriptional activation of previously silenced genes, which could provide additional co-factors necessary for reprogramming.
Heterokaryon experiments provided the first evidence that cellular identity is established and actively maintained by trans-acting factors (Yamanaka and Blau, 2010). Furthermore, fusion of keratinocytes and muscle cells demonstrated that gene activation in heterokaryons is in fact bi-directional, and the ratio of regulatory factors between the two cell types (heterokaryons often have more than 2 nuclei) determines which fate is dominant (Palermo et al., 2009). During transcription factor mediated reprogramming, high expression levels of reprogramming factors (often from strong artificial promoters) are likely to help facilitate cell fate conversion. However, in some cases high expression alone is not sufficient for reprogramming factors to override the transcriptional program of the host cell. For example, mature B lymphocytes could only be reprogrammed to pluripotency by the OSKM factors with the concomitant knockdown of the B cell transcription factor Pax5 or overexpression of C/EBPα, suggesting that pre-existing lineage-specific transcriptional programs are an impediment to reprogramming (Hanna et al., 2008). In addition to cell type-specific factors there may be more general repressors of cell fate changes. The REST/NRSF complex, which blocks expression of neuronal genes in non-neural cell types, may be one such example (Chong et al., 1995). For example, recent work has shown that conditional ablation of REST in fibroblasts causes an upregulation of some neuronal genes, but does not appear to cause overt neuronal conversion (Aoki et al., 2012).
Cell type specific microRNAs are a second mechanism employed by various cell types to prevent translation of lineage-inappropriate transcripts (Hornstein and Shomron, 2006). Accordingly, the brain-specific microRNAs miR-9 and miR-124 could promote the generation of iN cells from human fibroblasts when combined with the transcription factors ASCL1, NEUROD2, and MYT1L (Yoo et al., 2011). These miRNAs have been previously shown to downregulate Baf53a, REST, Co-REST and PTBP1, which are all thought to actively prevent neuronal fate acquisition (Yoo et al., 2009). Even more surprising was the claim that iPS cell reprogramming could be achieved by miRNAs in combination with HDAC inhibition (Anokye-Danso et al., 2011; Miyoshi et al., 2011). It remains unclear how these miRNAs, which are not thought to be capable of directly activating gene expression, can induce pluripotency genes. One explanation may be the inhibition of potential repressors of pluripotency associated genes analogous to REST for the neuronal lineage. Further experiments are certainly required to help elucidate how miRNA-mediated reprogramming can occur without directly activating gene expression.
Another potential mechanism for interference with transcriptional activity is the competition of exogenous and endogenous transcription factors for shared generic cofactors. This has been proposed as a potential mechanism helping to regulate neuronal versus glial cell fate choice in cultured neural stem/progenitor cells (Sun et al., 2001). In this system, forced expression of the bHLH transcription factor Ngn1 efficiently induces neurogenesis and also blocks acquisition of astrocyte fate. The induction of neurogenesis requires Ngn1 transcriptional transactivation whereas inhibition of astrocyte fate is DNA binding independent. Instead of directly suppressing astrocyte genes, Ngn1 acts to sequester CBP-Smad1 activating complexes from genes that promote astrocytic fate, such as STAT transcription factors. Furthermore, lineage-specific bHLH factors (such as Ascl1 and MyoD) require heterodimerization with widely expressed E-proteins (Tcf3) for DNA binding (Bertrand et al., 2002). Thus, high levels of ectopic bHLH factor expression might limit the pool of E-proteins available to endogenous bHLH transcription factors, which could lead to downregulation of some elements of the host transcriptional program, and thus facilitating changes in cellular identity.
“Epigenetic memory” refers to remnants of transcriptional properties or chromatin features typical of the starting cell type after reprogramming. The persistence of epigenetic memory is a critical issue in the reprogramming field because it has the potential to modify the behavior of reprogrammed cells, which could have large consequences for in vitro disease modelling studies and any future clinical applications of reprogrammed cells. Gurdon and colleagues demonstrated that there is epigenetic memory in cloned Xenopus embryos derived from embryonic nuclei that had initiated expression of lineage-specific regulators. For example, 81% of embryos derived from neuroectodermal nuclei (Sox2+) inappropriately expressed Sox2 in endoderm (Ng and Gurdon, 2005). This suggests that an epigenetic mark of transcriptional activity persisted through the reprogramming process, leading to lineage inappropriate Sox2 expression. Similarly, epigenetic memory of MyoD expression in nuclear transfer embryos derived from muscle cell nuclei could persist for 24 cell divisions during reprogramming (in the absence of MyoD transcription) and correlated with K4-trimethylated H3.3 retention at the MyoD promoter (Ng and Gurdon, 2008).
There is also evidence that iPS cells exhibit specific transcriptional and epigenetic signatures associated with their cell type of origin (Kim et al., 2010; Kim et al., 2011; Polo et al., 2010). This epigenetic memory decreased by treatment with HDAC inhibitors, continued passage in vitro or by differentiation followed by subsequent reprogramming. In contrast, early passage ES cells derived from nuclear transfer exhibited less evidence for epigenetic memory, perhaps because of differences in the kinetics of DNA demethylation that occurs during reprogramming in these two systems (Kim et al., 2010). Similarly, iN cells derived from hepatocytes were shown to efficiently downregulate the hepatocyte-specific transcriptional program but small remnants of hepatic gene expression were detectable on the single cell level (Marro et al., 2011). Further work will be required to determine the extent to which epigenetic memory affects the functional properties of reprogrammed cells. It will also be interesting to determine the molecular basis of epigenetic memory, whether different reprogramming methods lead to more or less epigenetic memory and to determine whether certain loci are more resistant to complete reprogramming than others, as has been seen during reprogramming to pluripotency under certain experimental conditions (Carey et al., 2011; Stadtfeld et al., 2012).
Cellular reprogramming research has been energized by its potential as a critical tool for the next generation of medical diagnostics and cell-based regenerative therapies, as well as the study of human embryonic development and organogenesis. It is now possible to generate a large variety of cell types in vitro from induced pluripotent stem cells derived from human patients. This provides unprecedented access to rare populations of genetically matched cells such as motor neurons or cardiomyocytes that are nearly impossible to obtain from live patients (Saha and Jaenisch, 2009). These cell types can then be rigorously characterized and compared to matching cell types from healthy individuals, with the hope that specific manifestations of the disease process are recapitulated in vitro (Han et al., 2011; Ming et al., 2011). The recent progress in direct lineage reprogramming also suggests that readily-available human fibroblasts can be directly converted into cells resembling several types of neurons found in the central nervous system (Caiazzo et al., 2011; Pang et al., 2011; Pfisterer et al., 2011; Yoo et al., 2011). These human iN cells have even shown some potential to demonstrate specific in vitro manifestations of disease (Qiang et al., 2011) (see Figure 1). The combination of these approaches should provide a powerful toolkit to study human disease processes and the developmental biology of human tissues in a culture dish, which could have transformative consequences for regenerative medicine.
Our work is generously supported by the National Institutes of Health (NIH grants RC4 NS073015-01, R01MH092931, AG010770-18A1), the California Institute for Regenerative Medicine (CIRM grants DR1-01454, RT2-02061), the Department of Defense (PR100175P1), the Ellison Medical Foundation, the Stinehart-Reed Foundation and the Baxter Foundation. M.W. is a New York Stem Cell Foundation-Robertson Investigator and T.V. is a California Institute for Regenerative Medicine predoctoral fellow (TG2-01259). We would also like to thank Karen Jann for generating all of the figure illustrations.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.