Cycling through developmental decisions: how cell cycle dynamics control pluripotency, differentiation and reprogramming

The identity of a cell can be defined by its specific, metastable program of transcription and by the activity of cell type-specific transcription factors. Cell type-specific patterns of chromatin organization and epigenetic modifications are crucial for the establishment and maintenance of these transcriptional programs. In order to transition from one state to another, cells must modify their transcriptome, epigenetic landscape and chromosome architecture in a highly coordinated way. Over the last quarter of a century, numerous observations have established a role for the cell cycle in broad aspects of cell fate decisions, and have shown that the expression of cell fate ‘decision’ genes is often coupled to cell cycle regulatory mechanisms (Fig. 1). These studies show that the cell cycle machinery impacts chromosome architecture, the epigenome and transcriptional programs required for cell identity in multiple contexts including differentiation, reprogramming and trans-differentiation.

Reprogramming differentiated cells was first demonstrated by exposing the nucleus of a somatic cell to the cytoplasm of an enucleated egg using somatic cell nuclear transfer (SCNT), and this technique has been used to generate animals ranging from frogs to primates (Briggs and King, 1952; Campbell et al., 1996; Gurdon, 1962; Tachibana et al., 2013; Wakayama et al., 1998). Similarly, it is possible to trans-differentiate cells from one differentiated type to another by exposing the nucleus of a donor cell to the cytoplasm of another cell. An example of this is the fusion between human fibroblasts and mouse muscle cells, which results in the induction of muscle-specific genes in the human genome (Blau et al., 1983). Pre-existing trans-acting factors in the recipient cytoplasm can therefore reset the transcriptional program of a donor somatic genome. Shortly after this discovery, ectopic expression of MYOD (also known as MYOD1) was shown to be sufficient for the conversion of fibroblasts to myoblasts, demonstrating that specific transcription factors can redirect cell identity (Davis et al., 1987). More recently, Takahashi and Yamanaka showed that OCT4 (POU5F1), SOX2, KLF4 and MYC (together abbreviated as OSKM) are sufficient to convert somatic cells to the pluripotent state, thereby generating induced pluripotent stem cells (iPSCs) (Takahashi and Yamanaka, 2006). Together, these studies reveal that cell type-specific transcription factors are central to the cell fate decision-making process (Graf and Enver, 2009; Xu et al., 2015). Numerous observations have identified important connections between these transcriptional master regulators, cell state transitions and the cell cycle, although the molecular mechanisms that connect these processes are only starting to be elucidated. This Review will begin by broadly summarizing the role of the cell cycle in differentiation, reprogramming and trans-differentiation in developmental models ranging from yeast to humans. Then, attention will specifically focus on the role of cell cycle regulatory mechanisms in cell fate decisions made by pluripotent stem cells (PSCs). Finally, we put forth our views on how placing cell fate decisions within the context of cell cycle will have implications for a broad spectrum of developmental decisions and is likely to change our current methods of manipulating cell identity for clinical purposes and for understanding human disease.

Progression through the cell cycle involves a sequence of events in which chromosomes are replicated during S phase and then segregated to daughter cells during M phase (Morgan, 1995). These key events are separated by gap phases that serve as regulatory windows to ensure that cell cycle events occur at the correct time and in the right order. All of these events are orchestrated by the activity of cyclin-dependent kinases (CDKs) that phosphorylate substrates required for the different cell cycle transitions. The gap phase separating M phase from S phase is known as G1 phase and marks the time when cells make the decision to exit the cell cycle or continue through further rounds of division. This decision is classically thought of as being controlled through the phosphorylation of retinoblastoma (RB) family proteins by CDKs, thereby establishing a binary switch mechanism known as the restriction (R-) point that gates cell cycle progression in G1 phase (Blagosklonny and Pardee, 2002). Recent evidence suggests that the CDK-dependent G1 decision point in cycling cells may actually initiate upon mitotic exit and may precede the classically defined R-point (Cappell et al., 2016; Spencer et al., 2013). The R-point mechanism links the cell cycle machinery to mitogenic signals and, under the appropriate signaling conditions, genes required for G1-S progression are activated. This mechanism has enormous implications for control of normal cell growth and de-regulated proliferation in cancer. In a stem cell context, R-point control is crucial for determining the balance between self-renewal, quiescence and differentiation of stem cell populations (Li and Clevers, 2010; Tetteh et al., 2015). For example, proliferative control of hematopoietic stem cells is a critical determinant that distinguishes normal and cancer-related hematopoietic function (Pietras et al., 2011). The developmental state of pluripotent cells can also be regulated by exit from the cell cycle. For example, PSCs exit the cell cycle and enter a ‘dormant’ developmental state that mimics diapause following MYC depletion (Scognamiglio et al., 2016). This is likely to be related to the ability of MYC to control CDK activity and is an interesting example of how proliferation and developmental status are coupled.

Another broad example of cell cycle decisions being coupled to cellular decisions is exemplified by size control mechanisms in a wide range of organisms (Ginzberg et al., 2015). In principle, cells are required to achieve a critical volume in G1 phase before entering S phase and committing to another round of cell division. If a cell is too small to sustain itself, it will delay progression into S phase in an attempt to acquire sufficient volume by growth before it commits to another round of DNA replication and cell division. If size homeostasis is de-regulated, a cell could potentially overgrow or reduce in size to the point where its function would be severely compromised. This process has been studied extensively in budding yeast where size-regulated commitment to the cell cycle occurs at a point in G1 phase known as ‘Start’ (Jorgensen and Tyers, 2004). This control point is generally considered to be the equivalent of the mammalian R-point and involves a regulatory mechanism whereby nutrient- and biosynthesis-regulated signaling pathways converge with the cell cycle machinery to coordinate cell volume with progression into S phase (Ferrezuelo et al., 2012). The cell cycle machinery is also linked to cellular decisions that occur following environmental stress such that cells undergo a checkpoint arrest or, alternatively, undergo apoptosis (Carvajal and Manfredi, 2013). These fundamental examples of coordination between cellular decisions and the cell cycle have broad relevance to the function of multipotent cells in development.

Early work in yeast provided a mechanistic link between cell cycle machinery and changes in cell identity through double-strand break-induced recombination (Haber, 1998). In this study, haploid budding yeast were found to switch between two different alleles of the mating-type (MAT) locus; a and α, leading to mixtures of a and α cells that can mate and form a/α diploids. In each cell cycle, haploid cells could potentially undergo a mating type switch driven by expression of the HO endonuclease, which acts on the MAT locus in G1 phase. In Dictyostelium development, amoeba decide to become pre-spore cells if they sense starvation conditions in G1 phase but choose a pre-stalk fate under the same conditions in S and G2 phases (Gomer and Firtel, 1987). This example is interesting because it indicates that commitment to different cell fates is determined at different stages of the cell cycle in multipotent cells. A similar conclusion has emerged from studies of Caenorhabditis elegans vulval development (Ambros, 1999). Here, the point at which lin-12 acts in the cell cycle impacts cell fate choice. Overall, these reports show that linking the cell cycle to cell fate decisions is a common theme in multipotent cells and is not restricted by species boundaries (Fig. 1). This principle also broadly applies to mammalian embryogenesis. During pancreatic development, for example, endocrine progenitor cells adopt different fates depending on whether they are exposed to differentiation signals in early or late G1 phase (Kim et al., 2015). If exposure to signals occurs in early G1 phase, cells differentiate and exit the cell cycle through an asymmetric cell division. By contrast, if pancreatic progenitors are programmed in late G1 phase they complete the cell cycle and generate two differentiated endocrine cells. The time at which pancreatic progenitors receive induction signals in G1 phase is therefore crucial for determining how they respond. This concept is reiterated in studies of murine neocortical development (McConnell and Kaznowski, 1991). Here, multipotent cortical progenitors respond to local induction cues generating different cell fate outcomes depending on where they are in the cell cycle at the time of induction. In murine fetal erythropoiesis, entry and progression through S phase is required for activation of the erythroid differentiation program through the erythroid master regulator GATA1 (Pop et al., 2010). Downregulation of the cyclin-dependent kinase inhibitor (CDKI) KIP2^p57 (CDKN1C) and the GATA1 antagonist PU.1 (also known as SPI1) are key requirements of this cell cycle-dependent regulatory mechanism. Linking S-phase progression to cell fate decisions in multipotent cells has also been reported in the Drosophila central nervous system (Weigmann and Lehner, 1995).

So far, examples of cell fate decisions being initiated during G1 and S phase have been described, but G2 phase is also potentially important for cellular decisions. During bristle patterning in Drosophila, Notch signaling controls transition through late stages of the cell cycle such that cells with elevated Notch signaling divide first and those with lower signaling extend their G2 phase and delay division (Hunter et al., 2016). This G2-phase transition time is crucial for determining the decision of sensory organ precursor cells to progress to a microchaete fate or a neural fate. The timing of mitotic entry and duration of G2 phase, during which cells are exposed to differentiation cues, therefore impacts Notch-mediated lateral inhibition and, consequently, cell fate decisions. Studies in zebrafish (Bouldin and Kimelman, 2014), sea squirts (Ogura et al., 2011), frogs and flies (Davidson et al., 2009) reinforce this concept. Together, these studies show that although there are variations to the central theme, the mechanistic coupling of the cell cycle to cell fate decisions is a strong recurring biological theme in multipotent cells from diverse origins.

Exit from the cell cycle in G1 phase is frequently required for terminal differentiation of cells during development. The mechanisms underlying this have been reviewed elsewhere (Buttitta and Edgar, 2007; Ruijtenberg and van den Heuvel, 2016) and so only the coordination between developmental programs and terminal differentiation will be highlighted here. In most cases, terminal differentiation is linked to the upregulation of CDKIs, ensuring inhibition of CDK during G1 as well as hypophosphorylation of the RB tumor suppressor protein family, which serves to repress the E2F target genes required for further cell cycle activity. It has been difficult to establish unequivocally the precise mechanism linking terminal differentiation to cell cycle exit but the two processes seem to be linked at several levels and are often mutually antagonistic. During skeletal myogenesis, CDKs inhibit the activity of myogenic transcription factors such as MYOD, thereby maintaining cells in a proliferative, immature state (Guo and Walsh, 1997; Rao et al., 1994; Skapek et al., 1995). A similar scenario has been described during neurogenesis, during which CDKs inhibit pro-differentiation transcription factors such as NGN2 (NEUROG2) (Hardwick and Philpott, 2014). It is also true, however, that MYOD counteracts the impact of CDKs by activating the expression of genes for CDK inhibitors such as CIP1^p21 (CDKN1A) and KIP2^p57 (Busanello et al., 2012; Halevy et al., 1995; Parker et al., 1995). Inhibition of CDKs then leads to cell cycle arrest in conjunction with the activation of terminal myogenic events. This theme is also seen during terminal differentiation of Drosophila neuroblasts (Choksi et al., 2006; Li and Vaessin, 2000). Here, the homeo-domain transcription factor Prospero (Pro; also known as Pros) activates genes required for differentiation but also inhibits transcription of key cell cycle regulatory genes, such as Cyclin E and string, and promotes transcription of the CDKI gene dacapo (Choksi et al., 2006; Li and Vaessin, 2000). These and other studies (Ruijtenberg and van den Heuvel, 2016) indicate an inverse mechanistic relationship between the cell cycle and terminal differentiation in a broad spectrum of cell types. These events depend on the activity of G1-specific CDKs and their regulation of transcription factors required for developmental decisions. Conversely, transcription factors required for cell fate decisions serve to modulate CDK activity and drive exit from the proliferative state. The balance between CDK activity and transcription factor activity therefore serves as a cell fate decision ‘tipping point’.

The examples considered so far cover the relationship between cell cycle stage and cellular decisions during the process of differentiation. It is important, however, to emphasize the requirement for cell cycle controls in other decision-making contexts such as reprogramming, trans-differentiation and regeneration, and to establish the similarities and differences among these contexts. Interestingly, several reports indicate that cell proliferation per se is not a crucial determinant for changes in cell identity. In addition to terminal differentiation (Falcone et al., 1984), trans-differentiation of fibroblasts to myoblasts (Chiu and Blau, 1984) and reprogramming by SCNT (Halley-Stott et al., 2010; Jullien et al., 2010, 2011) have no requirement for active cell division. This observation may also apply to some examples of transcription factor-induced cell fate changes. This includes the switch in exocrine to endocrine pancreatic identity following ectopic expression of NGN3 (NEUROG3), PDX1 and MAFA in mice (Zhou et al., 2008) and the generation of neurons from fibroblasts following ectopic expression of ASCL1, BRN2 (POU3F2) and MYT1L (Vierbuchen et al., 2010). Interestingly, ASCL1 was found to drive somatic cells to exit the cell cycle in trans-differentiation (Treutlein et al., 2016), whereas ASCL1 seems to drive quiescent adult hippocampal stem cells to re-enter the cell cycle during differentiation (Urbán et al., 2016). This suggests that the mechanistic coupling between transcription factors and the cell cycle machinery is context dependent and may be determined by the specific level of the protein as well as the pre-existing molecular landscape specific to the starting cell (Masserdotti et al., 2016). C/EBPα-induced conversion of pre-B cells to macrophage-like cells, however, can occur under proliferative and non-proliferative conditions (Di Tullio and Graf, 2012), but whether this involves two separate mechanisms is unclear. A caveat to some studies is that limited molecular characterization of trans-differentiated cells has been performed, making it unclear whether a complete cell fate switch was accomplished in the absence of cell division (Cahan et al., 2014; Morris et al., 2014). A growing amount of evidence, however, does show a key role for active cell division in cell fate switching. During regeneration in Medusae, trans-differentiation of mono-nucleated muscle cells to other cell types, including sensory cells, requires active cell division (Schmid et al., 1988). Cardiac regeneration in zebrafish is also intimately linked to cell cycle regulators such as polo-like kinase 1 (plk1) in proliferating cardiomyocytes (Jopling et al., 2010; Poss et al., 2002). It is also clear that reduced proliferative capacity represents a major barrier for reprogramming to the pluripotent state (Hong et al., 2009; Kawamura et al., 2009; Ruiz et al., 2011; Utikal et al., 2009) and there are several cases where active cell division is an important requirement for directed trans-differentiation (Feng, 2016; Jiang et al., 2015) in response to reprogramming factors.

Although continued cell division is a requirement for cell fate changes in many instances, it is unclear why it may not be necessary in all situations. Subtle differences in the barriers that cells face as they undergo identity changes could explain variations in cell cycle dependency. For example, reprogramming of fibroblasts to the pluripotent state is generally associated with extensive erasure and re-establishment of a DNA-methylation signature (Lister et al., 2009, 2011) but in B-cells, no major changes in DNA methylation are required (Di Tullio and Graf, 2012). If the erasure of DNA methylation or other epigenetic signatures requires active cell division, this could explain some of the discrepancies described above. It is also possible that DNA replication rather than cell division is mechanistically important for reprogramming (Lavagnolli et al., 2015). Also, the dependency of a lineage-specifying transcription factor on CDK for activation could explain differences in various stem/progenitor cells.

Pluripotent cells that exist during peri-implantation development have short generation times but the cell cycle lengthens significantly as cells differentiate along the germ layer lineages (Lawson and Pedersen, 1992; Mac Auley et al., 1993; Snow and Bennett, 1978). Similar trends have been described for PSCs cultured in vitro (Boward et al., 2016). Rapid cell division is associated with a truncated G1 phase and only a short delay before cells enter S phase after exiting M phase. The absence of fully formed gap phases establishes a situation wherein PSCs spend 50-65% of their time in S phase. As PSCs commit to one of the three embryonic germ layers their progeny acquire an extended G1 phase, resulting in increased cell division times. This can be accounted for by a fundamental change in the regulation of CDK activity (Faast et al., 2004; Stead et al., 2002; White et al., 2005). It has been assumed, mainly for anecdotal reasons, that the low G1-phase/high S-phase cell cycle structure of PSCs supports pluripotency by limiting the time cells are exposed to specification signals. As differentiation initiates, an elongated G1 phase would then make cells more susceptible to irreversible germ-layer commitment. Several reports have now established this concept experimentally. For example, if the length of G1 phase is increased through inhibition of CDK activity, PSCs spontaneously differentiate (Neganova et al., 2008; Ruiz et al., 2011). More recently, the strategic advantage of having a cell cycle with a short G1 phase has been demonstrated at the molecular level (Boward et al., 2016).

Although multiple laboratories showed that PSCs respond to induction signals in G1 phase over two decades ago (Mummery et al., 1987; Pierce et al., 1984; Wells, 1982), this general observation was not fully explored until recently, when the fluorescence ubiquitin cell cycle indicator (Fucci) reporter system was used to explore this phenomenon (Sakaue-Sawano et al., 2008). In a seminal report, Pauklin and Vallier (2013) confirmed that PSCs initiate cell fate decisions when in G1 phase, but they also identified an unanticipated mechanism whereby mesoderm and endoderm commitment occurs in early G1 phase and ectoderm commitment is restricted to late G1 phase (Fig. 2). This partitioning of G1 phase along germ layer boundaries is related to the elevated activity of SMAD2,3 in early G1 and its inactivation in late G1. The former is dependent on activin A signaling and the latter dependent on the removal of SMAD2,3 from target genes by cyclin D activity. Because mesoderm and endoderm differentiation requires SMAD2,3 activity and because ectoderm requires pan-SMAD inhibition, this mechanism provides an insightful explanation for how cell fate commitment is partitioned into different stages of G1 phase. It also provides a mechanistic link between the cell cycle-regulated activity of G1-phase cyclins and developmental genes required for cell fate decisions.

An additional report using the Fucci system to dissect cell cycle events has since been published, focusing on the links between epigenetic events and the cell cycle (Singh et al., 2015). In this study, Singh and co-workers characterized the epigenetic changes that occur in self-renewing hPSCs and found that H3K4 trimethylation within bivalent domains of developmental genes increases in G1 phase whereas H3K27me3 repressive marks remain constant (Fig. 2). This establishes that bivalent domains of H3K4me3 and H3K27me3 are enriched in G1 phase and that throughout the remaining part of the cell cycle developmental genes are marked primarily by H3K27me3. This is an intriguing observation because it establishes a level of dynamic epigenetic regulation at developmental genes that was not previously appreciated. Consistent with changes in H3K4me3 in G1 phase, developmental genes become transcriptionally competent (Singh et al., 2013). The study suggests that developmental genes are primed for activation each time they go through G1 phase but are not activated unless the appropriate signaling networks are also active. In support of this, chromatin conformation-capture (4C) assays showed that G1-specific epigenetic changes at developmental genes coincide with the establishment of DNA loops that bridge distal enhancers with proximal promoters (Fig. 2). Chromosome architectural changes in G1 phase require increased H3K4me3 at the bivalent domain along with increased CDK2 activity, indicating a functional link between epigenetic remodeling, chromosome architectural changes and the cell cycle machinery. Although the mechanism by which CDKs control chromosome architecture at bivalent genes has not yet been established, these studies indicate that in each G1 phase the epigenetic landscape and chromosome architecture changes so that it puts developmental genes in a ‘lineage-primed’ state. The potential mechanisms by which CDKs remodel chromatin and activate developmental genes in G1 phase could be direct or indirect, but potentially could involve the direct phosphorylation of chromatin remodeling enzymes or sequence-specific transcription factors. In summary, a combination of epigenetics, chromosome architecture and transcription factor recruitment appears to be involved in priming developmental genes for G1-specific differentiation. CDK activity has been implicated in control of these regulatory steps (Pauklin and Vallier, 2013; Singh et al., 2015) (Fig. 2), establishing a link between cell fate decisions and the cell cycle machinery.

The idea that the transition from mitosis to G1 phase might establish conditions in which switching of lineage-specific transcriptional programs is possible is supported by studies showing that this window of the cell cycle represents a hyperactive, dynamic transcriptional state (Hsiung et al., 2016). This is consistent with earlier reports showing that developmental genes are primed for transcription in G1 phase (Singh et al., 2013, 2015). Increased transcription as cells transition into G1 phase is thought to localize at genomic regions pre-marked with H3K27ac. This hyper-transcriptional activity is not uniform and might also account for cellular heterogeneity (Hsiung et al., 2016). A recent study has revealed another interesting connection between exit from pluripotency and early cell fate decisions, this time involving S phase and G2 phase. In this study, a high-throughput RNA interference (RNAi) screen performed in human embryonic stem cells (hESCs) identified cell cycle genes involved in DNA replication and G2-phase progression that restrict exit from pluripotency or ‘pluripotent state dissolution’ (Gonzales et al., 2015). This is another line of evidence that supports the idea that the cell cycle state of PSCs is related to maintenance of pluripotency. By devoting most time to S phase and minimizing the time spent in G1 phase, cells have little opportunity for dissolution of the pluripotent state. As G1 phase lengthens during differentiation, this would presumably cause an irreversible breakdown of the pluripotency network and enhance germ layer commitment. This is further evidence indicating that G1 phase represents a gateway for the initiation of cell fate decisions.

By integrating what we know so far, it is possible to envisage a model in which the initiation of differentiation consists of at least two phases. The first phase would be pluripotent state dissolution, during which the pluripotency network is inactivated, and the second phase would be lineage commitment, involving the establishment of new transcriptional programs corresponding to germ layer formation. This latter step could involve a combination of bookmarking events during mitosis (see below and Box 1) and lineage priming in G1 phase. Pluripotent state dissolution is then attenuated in S phase and G2 phase by activities in these respective cell cycle phases, maintaining high levels of pluripotency factors such as NANOG and OCT4. This increase of NANOG and OCT4 is tightly controlled and more homogenous at the transcriptional level after DNA replication (Skinner et al., 2016). These findings support the model that G1 phase represents a window of time when cells are predisposed to initiate fate decisions. Thus, entry and exit from G1 phase is mechanistically linked to cell fate decisions in PSCs.

Paulkin and Vallier previously found that CDK4,6-cyclin D complexes impact activin/NODAL signaling and promote human embryonic stem cell differentiation through phosphorylation of SMAD2,3 (Pauklin and Vallier, 2013). More recently, however, the same laboratory reported that cyclin D can influence cell fate decisions independently of SMAD2,3 inhibition and independently of its association with CDK4,6 (Pauklin et al., 2016). This second mechanism of action requires the ability of cyclin D to recruit transcriptional co-activators and co-repressors to developmental target genes as it accumulates during G1 phase (Fig. 2). For example, when cyclin D1 is absent in early G1, the co-activator p300 (EP300) is recruited to endoderm genes. However, when cyclin D1 accumulates in late G1, it recruits histone deacetylases (HDACs) in place of p300, which functions as a histone acetyltransferase (HAT), resulting in loss of the active histone acetylation and an increase in the repressive histone methylation H3K27me3 at these genes. Consistent with the G1-phase partitioning model, neuroectoderm genes are activated in late G1 by cyclin D-dependent recruitment of p300 and increased H3K4me3 at these genes. Recruitment of cyclin D to ectoderm genes is dependent on SP1 whereas cyclin D recruitment to endoderm genes is E2F dependent. This work reinforces a number of principles that help us to understand how pluripotent stem cells initiate cell fate decisions from G1 phase. Importantly, cyclin D proteins can interact with transcription factors bound to developmental genes in G1 phase and have vastly differing effects on their regulation via their ability to recruit co-repressors or co-activators. The principle that CDK activities coordinate cell fate decisions at developmental genes in G1 phase is underscored by these observations.

DNA replication and mitosis represent two events that could facilitate global restructuring of chromatin during the cell cycle. During S phase, re-establishing the chromatin state on newly synthesized DNA is potentially important for maintaining or switching cell identity. It is therefore conceivable that this represents a time when new, cell type-specific epigenetic landscapes can be founded. The inheritance of this epigenetic state through the S phase can use the mother chromatin state as a template or be directly coupled to the DNA replication machinery (for recent reviews, see Ma et al., 2015; Probst et al., 2009). It is mitosis, however, that has recently attracted most attention in this area, in part because of studies in which asymmetric cell divisions have been implicated in cell fate changes (Arsenio et al., 2015; Congdon and Reya, 2008; Tran et al., 2012). The hallmarks of mitosis include chromosome condensation and nuclear envelope breakdown, both of which are under control of CDK activity. During M phase, most transcription-associated factors, including RNA polymerases, dissociate from chromatin and the cell type-specific transcription programs temporarily halt (Egli et al., 2008; Gottesfeld and Forbes, 1997; Spencer et al., 2000). Furthermore, histone modifications, such as global histone acetylation, which are associated with active gene expression, are generally diminished during mitosis (Kruhlak et al., 2001; McManus and Hendzel, 2006). Loss of histone tail acetylation, such as H4K16, has been found to be directly linked to mitotic-specific histone phosphorylation (H3S10P), promoting chromatin fiber condensation (Wilkins et al., 2014). However, residual amounts of histone acetylation, which bookmarks a select group of gene promoters, has also been reported (Dey et al., 2009; Kouskouti and Talianidis, 2005; Valls et al., 2005; Zhao et al., 2011). Despite many repressive marks such as H3K9me3 and H3K27me3 being retained in mitotic chromatin (Follmer et al., 2012; Li et al., 2006; McManus et al., 2006; Peters et al., 2002; Singh et al., 2015), the functional proteins – known as ‘writers’ and ‘readers’ – that deposit and associate with these marks, such as SUV39H1, HP1 (HP1 gamma also known as CBX3) and BMI1, are generally excluded from mitotic chromatin (Egli et al., 2008; Kellum et al., 1995; Minc et al., 1999; Voncken et al., 1999). Histone modifications during mitosis have been reviewed in more detail elsewhere (Wang and Higgins, 2013). In addition, the precise nucleosome positioning and histone variant distribution that usually mark cell type-specific promoters and enhancers are also lost during mitosis (Deniz et al., 2016; Kelly et al., 2010; Komura and Ono, 2005). This culminates in the erasure of cell type-specific, three-dimensional genomic organization (Dileep et al., 2015; Naumova et al., 2013) and the loss of genome-nuclear lamina interactions (Kind et al., 2013). Epigenetic erasure and chromosome remodeling could therefore explain the requirement for transition through mitosis in order to enhance cellular reprogramming, as discussed in more detail below.

Despite the global reorganization of the epigenome during mitosis, mitotic memory of active gene expression programs is retained. In effect, this may allow newly divided cells to pick up where they left off, either by preventing mitotic compaction of previously active DNA loci, or by facilitating reassembly of transcription complexes on the promoter, or both. This phenomenon is referred to as mitotic bookmarking (see Box 1), and has been shown to be important for the rapid reactivation of certain genes upon entry into G1 phase. Thus, it is possible that the selective retention of certain marks over others could represent a means to instructing cell fate change in the newly divided cell.

It is well-established that chromosome condensation and nuclear envelope breakdown during mitosis are important for successful SCNT reprogramming of mammalian cells (Campbell et al., 1996). This general model is also supported by studies of SCNT in amphibians in which somatic cells are most responsive to reprogramming factors present in the cytoplasm of unfertilized, metaphase II (MII)-arrested oocytes (Halley-Stott et al., 2014). Histone H2A de-ubiquitylation on mitotic chromatin but not interphase chromatin seems to contribute to this reprogramming responsiveness. Surprisingly, no other histone modification or epigenetic mechanism has been identified that, together with H2A de-ubiquitylation, enhances SCNT reprogramming in M phase. It has been postulated that histone de-ubiquitylation and chromatin condensation might enhance SCNT reprogramming not through the eviction or exclusion of factors from mitotic chromatin but, instead, by facilitating factor exchange (Halley-Stott et al., 2014). Hyper-dynamic protein exchange is a feature of pluripotency that decreases following exit from the pluripotent state (Meshorer et al., 2006). This is also consistent with the dynamic binding of pioneer factors to mitotic chromatin (Caravaca et al., 2013). This pattern of dynamic protein exchange on chromatin might therefore be crucial for establishing the pluripotent state, but it could also establish conditions that predispose PSCs to differentiation or reprogramming following mitotic exit under the appropriate signaling conditions.

Mitosis might serve to provide a window of opportunity for reprogramming because reprogramming factors have preferential access to condensed chromatin and therefore face less competition from other factors that are excluded from target loci. It has been shown that OCT4, SOX2 and KLF4 (together termed OSK) can access closed chromatin by acting as pioneer factors during iPSC reprogramming (Soufi et al., 2012, 2015). However, whether these pioneer factors interact with mitotic chromatin is yet to be examined. Nevertheless, one can hypothesize that reprogramming factors have a unique ability to bookmark mitotic chromatin, enabling gene priming immediately following exit from mitosis and prior to gene reactivation (Fig. 3). Another mitotic advantage might be due to the exclusion of HP1 and SUV39H1 from heterochromatin, which blocks access of the reprogramming factors to key pluripotency genes during interphase (Soufi et al., 2012) (Fig. 3). The association and dynamic exchange of reprogramming factors with mitotic chromatin functions to bookmark the genome, both specifically and non-specifically (Fig. 3). This landscape is then reset upon mitotic exit and potentially establishes a new epigenomic state for cell fate conversion.

The mitotic advantage in reprogramming is not restricted to unfertilized oocytes but is applicable to other advanced embryonic stages as well. For example, mouse zygotes and electro-fused blastomeres arrested in mitosis can acquire a reprogramming capacity in SCNT (Egli et al., 2007; Riaz et al., 2011). However, a recent study has shown that even the cytoplasm of interphase two-cell mouse embryos can reprogram somatic cells in SCNT, if the donor nucleus and recipient cytoplasm are synchronized (Kang et al., 2014). This suggests that the mitotic advantage is due to effective cell cycle coordination, and not necessarily the presence of special reprogramming proteins only present in the recipient mitotic cytoplasm or a special mitotic chromatin configuration of the donor nucleus. It has long been known that the cytoplasm of an MII-arrested oocyte retains high activity of CDK that can efficiently lead to nuclear envelope breakdown, chromatin condensation and subsequent DNA replication of the donor nucleus from all cell cycle stages (Campbell et al., 1993). However, the cytoplasm of S-phase cells is less effective at synchronizing a G2-phase donor nucleus, unless the nuclear envelope is chemically permeabilized (Blow and Laskey, 1988). Thus, the nuclear envelope might act as a barrier for cell synchronization, which is eliminated during mitosis. Taken together, these seemingly conflicting reports seem to agree on the idea that resetting the somatic epigenome to pluripotency or totipotency can only be tolerated if supported by active cell division.

Throughout this Review, the link between cell fate decisions and the cell cycle has been emphasized in a developmental context using examples ranging from yeast to humans. The emerging general themes from this work indicate that cell fate decisions are context dependent with regard to their requirement for cell division. Both cell cycle-dependent and cell cycle-independent mechanisms have been highlighted. For example, terminal differentiation of muscle cells from immature precursors requires cell cycle exit whereas differentiation towards pancreatic and erythroid progenitors requires active division and phase-specific cell cycle activities. Similar variations have been reported in reprogramming to the pluripotent state and trans-differentiation across lineages. Irrespective of the requirement for cell division, the cell cycle machinery impacts these cellular decisions (Table 1). For example, CDKI-regulated CDK inhibition is central to cell fate decisions made in a cell cycle-independent context, whereas elevated CDK activity is a requirement for most cell cycle-dependent cell fate decisions. Clearly then, it is important to improve our understanding of the molecular basis underpinning cell fate decisions in each scenario in order to place the cell cycle in a broader developmental perspective.

We used PSCs as an example to highlight mechanisms that coordinate cell fate commitment with the cell cycle. In this scenario, cells transitioning through G1 phase are highly responsive to differentiation cues that target developmentally regulated transcription factors and chromatin remodelers. In conjunction with developmental signals, the G1-CDK machinery collaborates with signal-regulated transcription factors such as SMADs to recruit co-repressors and co-activators to developmental genes. Altogether, these events result in chromosome remodeling, enhancer recruitment and the coordinated activation of a transcriptional program required for cell fate decisions. We refer to this G1-specific mechanism as lineage priming. Interestingly, SMADs are known to act on loci primed by master transcription factors (Mullen et al., 2011), indicating the existence of a more elaborate mechanism for cell fate specification. Along with lineage priming, G1 phase represents a time when the pluripotency network is vulnerable to inactivation, also called dissolution, whereas in S phase and G2 phase it is stabilized. As pluripotent cells spend most time in S phase, this further suggests a connection between cell cycle stage and pluripotency wherein S phase supports pluripotency whereas G1 phase represents a period of differentiation competency. Establishing a cell cycle in which cells spend most time in S phase and a brief time in G1 phase therefore minimizes the opportunity for pluripotent cells to switch state. Partitioning networks that sustain pluripotency or promote differentiation between different cell cycle phases seems a logical strategy to activate developmental genes and silence pluripotency genes in a coordinated manner. Another mechanism implicated in marking chromatin for future decisions in pluripotent cells is mitotic bookmarking. These observations indicate that mitosis is required to lay the epigenomic foundations for a cell fate switch in G1 phase. In total, these observations imply that the entire cell cycle is part of a coordinated network that orchestrates cell fate decisions.

It is important to note that the cell cycle machinery components present in the cytoplasm must coordinate with the chromatin configuration of the genome in the nucleus in order to make a cell fate decision effectively. This is highly relevant to reprogramming and trans-differentiation as the field is moving towards generating cell types for clinical applications and disease modeling, especially when the efficiency and the fidelity of the current protocols represent the biggest challenge in the field. In order to achieve these prerequisite requirements, future research must not only concentrate on how to manipulate chromatin and gene expression but also how these changes fit within the cell cycle. Defining in greater detail the mechanistic aspects that connect cell fate switches to the cell cycle machinery will be instrumental in developing novel and effective methods to control cell types, as well as understanding whether this mechanism is a general feature of cell fate decisions or is restricted to specific cases.

We thank Amar Singh, Tianming Wu, Martin Pera and Kersti Jääger for helpful comments on the manuscript.