A Conceptual Framework for Cell Identity Transitions in Plants

Abstract

Multicellular organisms develop from a single cell that proliferates to form different cell types with specialized functions. Sixty years ago, Waddington suggested the ‘epigenetic landscape’ as a useful metaphor for the process. According to this view, cells move through a rugged identity space along genetically encoded trajectories, until arriving at one of the possible final fates. In plants in particular, these trajectories have strong spatial correlates, as cell identity is intimately linked to its relative position within the plant. During regeneration, however, positional signals are severely disrupted and differentiated cells are able to undergo rapid non-canonical identity changes. Moreover, while pluripotent properties have long been ascribed to plant cells, the introduction of induced pluripotent stem cells in animal studies suggests such plasticity may not be unique to plants. As a result, current concepts of differentiation as a gradual and hierarchical process are being reformulated across biological fields. Traditional studies of plant regeneration have placed strong emphasis on the emergence of patterns and tissue organization, and information regarding the events occurring at the level of individual cells is only now beginning to emerge. Here, I review the historical and current concepts of cell identity and identity transitions, and discuss how new views and tools may instruct the future understanding of differentiation and plant regeneration.

Cell Differentiation: A Concept in Crisis

Plant cells are traditionally ascribed pluripotent potential (Steward et al. 1970) but, during normal plant development, cellular trajectories are limited in scope and most possible transitions do not occur (Sachs 1991). Cell files usually originate from specialized stem cells and, following predictable and gradual cell identity transitions, differentiate to form stable tissue-wide patterns (Mähönen et al. 2014, Drapek et al. 2017). In a prime, well-described example, progenitor meristemoid cells in the leaf epidermis undergo a sequential differentiation process with multiple stable intermediate states, on their way to form mature stomata. This multistep process is guided by a complex gene regulatory cascade, orchestrated by the sequential activity of a group of basic helix–loop–helix transcription factors (Han and Torii 2016). High-resolution transcriptional analysis of this developmental lineage has revealed specific gene expression patterns at different stages of the developmental cascade, suggesting that they represent distinct identity steps (Adrian et al. 2015). The notion that plant cells undergo a gradual and ‘smooth’ differentiation during normal development is further supported by whole-tissue profiling of different plant organs that revealed consistent, temporally regulated gene expression patterns across developmental gradients (Birnbaum et al. 2003, Brady et al. 2007, Efroni et al. 2008, Park et al. 2012).

This view of gradual differentiation fits well within the classical visual metaphor proposed by Waddington, wherein cell differentiation is analogous to progression through an ‘epigenetic landscape’, with cells gradually acquiring distinct features and losing their development potential, until arriving at one of the possible final fates. This model provides a mental image linking a cell’s identity with its competence (Fig. 1A; Waddington 1957). While never meant to be taken rigorously, this model has nevertheless shaped our understanding of the differentiation process, and the term ‘cell differentiation’ concurrently refers to the loss of developmental potential and to acquisition of cell type-specific gene expression and morphological markers.

Fig. 1.

Models for cell identity and differentiation. (A) Waddington’s original representation of the cell as a ball gliding through the epigenetic landscape. (B and C) Defining identity using single-cell transcriptomics, using unbiased identity reduction (B) or using a pre-defined identity marker (C). Each spot represents a cell in identity space, with blue and red marking different identities. Multicolor spots are cells in transition. Gray arrows in (B) define the preferred trajectory for transitional cells, generating specific attractors. (D) The dynamic landscape model, according to which the epigenetic landscape can change its shape in different developmental scenarios. (E) The epigenetic disc model lacks hierarchy. Different possible identities are represented by different colors. Cell movement is determined by external factors tilting the disc. (F) Schematics of a chemical reaction with an intermediate high-energy, unstable transition state. (G) Cell differentiation according to the transition state model. Cells start from one identity, ‘excite’ to an unstable high energy state and then ‘collapse’ to one of two new identities. Illustrations in (A), (D), (E) and (G) are from Waddington (1957), Rajagopal and Stanger (2016), Ladewig et al. (2013) and Moris et al. (2016), respectively, and are reprinted with permission.

During normal plant development, many processes follow the Waddington scheme, with clear lineages of differentiation. However, there are multiple developmental contexts where cells in the plants can undergo non-canonical identity transitions. For example, cells in the leaf epidermal layer originate from a protoderm source that divides in an anticlinal manner to form a single cell layer. Cells in this layer undergo differentiation gradually to acquire the stereotypical morphology of epidermal pavement cells. In rare cases, epidermal cells can divide periclinally to invade an underlying layer in the leaf. Using genetic chimeras, it was shown that displaced cells are able to switch their identity and adopt the morphology of their new layer (Stewart and Burk 1970). Clonal analysis in multiple plants and in different developmental contexts has shown that, in general, when plant cells are dislocated, they rapidly alter their identity to match their new position, even when the dislocation occurs late in development (Szymkowiak and Sussex 1996, Berger et al. 1998, Kidner et al. 2000). This flexibility in fate acquisition is also evident in genetic perturbation experiments. Overexpression of the final regulator of the stomatal differentiation cascade, FAMA, is sufficient to transform plant cells including cells from non-epidermal lineages, directly into guard cells. Such cells do not form well-patterned stomata, suggesting that the gene expression cascade during stomatal differentiation may be more important for tissue pattern formation rather than for the acquisition of guard cell identity per se (Ohashi-Ito and Bergmann 2006). Thus, it seems that plant cells are not required to undergo a gradual differentiation process in order to adopt a final fate but are relatively free to change identity across lineages (Szymkowiak and Sussex 1996, van den Berg et al. 1997). This remarkable plasticity is difficult to reconcile with Waddington’s landscape metaphor.

Another challenge to the Waddington view is regeneration, during which cells undergo rapid and radical identity transitions (Sánchez Alvarado and Yamanaka 2014). Plants possess broad regeneration ability, and injured organs can repair themselves by the activity of multiple cell types (Feldman 1976, Reinhardt et al. 2003, Xu et al. 2006, Sena et al. 2009, Melnyk et al. 2015). At the most extreme end of such plasticity is the ability to generate de novo organs and meristems readily from various differentiated tissues (Perianez-Rodriguez et al. 2014, Ikeuchi et al. 2016). Furthermore, hormonal treatment can even cause a complete organ conversion where forming root meristems can be directly converted to shoot meristems, a process involving identity switching in many cells (Chatfield et al. 2013, Kareem et al. 2016, Rosspopoff et al. 2017). As will be discussed below, what underlies this capacity is the plasticity of plant cells and their ability to change their identity rapidly.

Early transplanting experiments in animals, followed by the introduction of induced pluripoent stem cell (iPSC) technology and advances in tissue regeneration have shown that such cellular plasticity is not unique to plants but is a universal feature of development in multicellular organisms. This realization has triggered several attempts to rethink the concepts of cell identity and cellular differentiation (Ladewig et al. 2013, Sánchez Alvarado and Yamanaka 2014, Adler and Sánchez Alvarado 2015, Moris et al. 2016, Rajagopal and Stanger 2016). These approaches try to readdress old questions in light of the apparent widespread cellular plasticity. How should we view cell identity and cell identity transitions? What allows and inhibits identity transitions and what underlies cellular pluripotency? How different are identity transitions during normal development from those during regeneration? At the heart of all these questions is the concept of cell identity itself, a concept that is currently being rethought and evaluated.

In this review, I will present an overview of the current debate regarding cell identity, differentiation and pluripotency, and show how these apply to plant regeneration. A distinction will be made between events occurring at the level of a single cell and those occurring at the level of entire tissues. This distinction is important to analyze processes occurring at different scales and allows reconciliation of conflicting descriptions of the regeneration process.

Toward a Formal Definition of Cell Identity

Earlier generations of biologists were fortunate to have only a handful of morphological and cytological markers with which to characterize individual cells. The observation that some of these cell-inherent markers correlated with each other and with cellular function allowed cells to be classified into distinct types and to construct cellular taxonomies. However, the introduction of modern genomic techniques and, most recently, single-cell genomics enabled the measurement of a myriad of cellular properties, such as the copy number of tens of thousands of mRNAs, the modifications of many millions of histones, and much more (Schwartzman and Tanay 2015, Wagner et al. 2016). This highly expanded characterization of cell properties makes the number of cell states practically endless—even when assuming only 20,000 genes, and allowing them to be either on or off (ignoring expression levels, splice forms, transcriptional and post-transcriptional modifications, etc.), the number of possible cell states is a mindboggling 10⁶⁰²⁰. While clearly only theoretical, this immense possibility space ensures that no two cells are ever alike, and hinders the ability to define and distinguish between cell types. As a result, the concepts of cell identity and cell types are coming under considerable scrutiny (Clevers et al. 2017).

Currently, cell identity is mostly assigned using a small number of key marker genes or morphological features. However, the somewhat arbitrary selection of marker genes may not accurately represent functional division into cell types. Further, cells can express multiple markers for different distinct identities, which makes classification problematic. The introduction of single-cell-level transcriptomics brought an opportunity to define cell types and identify identity genes in a less biased manner (Trapnell 2015). Multiple methods have been developed that attempt to classify cells from single-cell genomic data (mostly gene expression data) and derive cellular taxonomies. Application of these methods in plants is currently at its very early stages but, nevertheless, these methods are of importance as they shape the view of cellular identity and differentiation at large.

There are many computational challenges when dealing with single-cell genomic data. These include a high level of technical noise, finding clusters in sparse high-dimensional data and the identification of rare small clusters. Multiple algorithms have been designed to address these unique difficulties. These have been recently reviewed (Wagner et al. 2016) and will only be broadly discussed here.

Regardless of algorithmic details, the common output of such classification methods is a definition of cell identity as coordinates in an abstract low-dimension identity space, usually made up by reduction of high-dimensional gene expression space. The concept of cell identity as a coordinate in an identity space maps well onto Waddington’s landscape metaphor, adding a quantitative, measurable quality to it. Cells are not uniformly located in the identity space, but rather group into discrete positions, viewed as ‘attractors’. These attractors represent classical cell identities, or types (Fig. 1B; Trapnell 2015). All classifiers depend on collected data and include subjective parameters such as the granularity of clustering. Thus, different experimental set-ups can result in different clustering and cell type assignments. Importantly, these methods classify cells based on similarity, but not all similarity in gene expression represents cell identity. The cell’s transcriptome may change in a co-ordinated manner in response to physiological stress, the circadian clock, external or internal environment of the organism, and more. A distinction was therefore made between a cell state, reflecting transient responses to environmental changes, and cell type, which refers to long-living stable properties of the cell (Wagner et al. 2016). This distinction is not a clear one and does not apply to developmental scenarios characterized by rapid identity changes, such as regeneration. As temporal behavior is difficult to measure in destructive single-cell transcriptomic experiments, most of these methods implicitly assume that cell type is the major factor of cell to cell variability.

An alternative approach relies on identification of discrete marker genes based on the averages of cell populations of a known cell type. Such a supervised approach utilized a priori knowledge to disentangle molecular markers of cell type from other features of the cell, such as its physiological state (Birnbaum and Kussell 2011, Efroni et al. 2015). These methods also define identity as a coordinate in identity space, but the vectors of this space correspond to pre-defined biological cell types. Cells can be classified as ‘pure identity’, where a cell’s gene expression matches that of the representative for a give cell type, or be at different levels of mixed or transitional identities (Fig. 1C). Such a classification can be more readily translated into biological insight, but cannot identify new cell types.

Common to these methods is that the exact coordinates in identity space rely on experiment-dependent features, such as the physiology of the organism. Thus, it is difficult to derive a universal molecular definition for biological cell types. However, for practical purposes, such a definition may not be needed at all (Clevers et al. 2017). The ability to classify cells into functional groups and identify the transitions between these groups within the context of a given experimental condition provides a measure of relative or context-dependent identity. This context-dependent quantity by itself is a useful tool to study how cells maintain and switch identities and how different cell types co-ordinate to perform a specific biological function.

It is important for the study of regeneration that all these formal definitions of cell identity leave out the notion of an ‘undifferentiated cell’. Every cell has a unique position in identity space and thus is different from other cells, while a ‘basal state’ is left undefined. The term ‘undifferentiated cell’ may refer to cells with pluripotent potential but, as will be discussed below, cellular pluripotency is widespread and context dependent, and thus the term seems too vague to be useful.

Cell Identity Transitions and the ‘Transition State’

Observations of cellular plasticity obtained from regeneration and iPSC studies are difficult to reconcile with the notion of a ‘landscape’ of gradual and hierarchical differentiation, and modern attempts were made to modernize Waddington’s metaphor. Rajagopal and Stanger (2016) have suggested to keep the landscape metaphor, but expand it by proposing that the landscape itself may be highly dynamic, and that it's hills and valleys, connecting different cellular identities, can change in development and in response to challenges such as regeneration (Fig. 1D). As the landscape is flexible, this notion, in effect, abolishes the hierarchy of cell identities. Other suggestions discard the landscape analogy all together. In one possible analogy, different cellular states are represented along the edges of a disc. In this abstract model, genetic and epigenetic factors act as rails along which cells, free to move between different states, glide. Injury or introduction of strong external factors act to ‘tilt’ or ‘shake’ the disc, allowing cells to move from one fate to another (Fig. 1E; Ladewig et al. 2013). Both these models let go of the implied hierarchy of cellular identities in the original Waddington scheme. However, identity transitions are still thought to be a gradual, smooth and predictable occurrence.

Observations of cell identity transitions in multiple experimental systems, including hematopoietic cell differentiation (Hu et al. 1997, Laslo et al. 2006), cellular reprogramming (Buganim et al. 2012) or plant regeneration (Efroni et al. 2016), have shown that cells can transiently express identity markers for multiple cell types in the same cell. Thus, prior to formation of a stable cell identity, cells can exist in a state marked by stochastic identity marker expression. To accommodate these observations, a different model was proposed, in which identity transitions are analogous to chemical reactions with a high-energy, unstable transition state separating stable, low-energy ones. The movement between these states is stochastic, and cell identity transitions are abrupt, with the highly unstable ‘transition state’ collapsing to an individual identity (Fig. 1F, G; Moris et al. 2016). The identity instability of such a transition state can allow pluripotency and can be a characteristic of stem cells (Trapnell 2015). A similar idea of identity transitions and pluripotency as an unstable stochastic state was proposed to explain planaria regeneration. During planaria regeneration, specialized cNeoblast cells possess pluripotent ability and are able to restore regeneration ability to irradiated animals. Their apparent scarcity seems incompatible with the regeneration of very small animal pieces. A proposed hypothesis is that cNeoblasts are not a cell type, but rather represent an unstable pluripotent state that cells can move in and out of in a stochastic manner (Adler and Sánchez Alvarado 2015). Interestingly, the existence of a possible transition state, which plant cells go through when radically changing their fate, was raised >25 years ago by Sachs, when trying to reconcile plant cells’ apparent pluripotency with the relative deterministic differentiation process (Sachs 1991).

It follows from these theories of stochastic identity acquisition that discordant expression of identity genes should be common in cell states of pluripotent potential. Indeed, single-cell mRNA-Seq of cells isolated from the root quiescent center (QC), a uniform tissue located at the heart of the stem cell niche (Bennett and Scheres, 2010), uncovered significant variability in gene expression with occasional discordant expression, like that of epidermis identity and even pollen-specific genes (Lin and Schiefelbein 2001, Brennecke et al. 2013, Efroni et al. 2015). Further, careful observation of the expression of the atrichoblast identity marker GLABRA2 in early stages of epidermis differentiation has detected stochastic expression of this transcription factor that did not always correspond to morphological identity transitions (Costa 2016). This view is also consistent with many stochastic identity transitions occurring in plants, for example in the variable number of pericycle cells undergoing identity transitions during the formation of a new lateral root meristem (Von Wangenheim et al. 2016). However, transcriptome-level data of cell identity transitions are still scant, and the nature of this hypothetical ‘transition state’ remains to be elucidated.

These new views of cell identity and differentiation are undergoing rapid development and are likely to change. However, the concept of a rigid hierarchy of cell states leading from an immature to a differentiated cell is being phased out and replaced by a more fluid and flexible view of cell identity transitions and differentiation. According to these views, many so-called differentiated cells have the capacity for broad identity transitions, which raises the question of what does it mean for a cell to be pluripotent.

Cellular Pluripotency

The best example of broad pluripotency during plant regeneration is callus. This tissue can undergo differentiation to form both roots and shoots, and thus it was suggested that callus cells are in a pluripotent state (Ikeuchi et al. 2013). Callus initiates following injury or by the application of high levels of the plant hormones auxin and cytokinin. As callus was thought to arise from mature tissue, it was assumed that cells must ‘dedifferentiate’ when they form callus in order to acquire pluripotency. However, studies in tissue culture have shown that when induced by external hormone application, callus originates specifically from specialized pericycle-like cells found throughout the plant (Atta et al. 2009, Sugimoto et al. 2010). In this case, no such pluripotency acquisition, or dedifferentiation, step is required as these specialized cells may already be in a highly competent state (Sugimoto et al. 2011).

However, under non-tissue culture conditions, callus can arise from tissues other than the pericycle. The induction of the AP2-like transcription factor gene WOUND INDUCED DEDIFFERENTIATION1 triggers the production of callus from epidermal tissues (Iwase et al. 2011). During wounding of tree barks, callus is formed from multiple vasculature-associated tissues and can generate a variety of new ones, suggesting that it has some pluripotent potential (Stobbe et al. 2002). Other examples of non-canonical identity transitions appear in studies of adventitious root production, where roots are generated following injury from a non-pre-patterned tissue. There, root meristems are derived from the pericycle, but also from xylem or phloem parenchyma cells, cambium or from the stem endodermis (Falasca et al. 2004, Bellini et al. 2014). In fact, a proliferating cell mass that can form entire plants can be derived from isolated phloem cells (Steward et al. 1958). This indicates that while the pericycle, with its putative specialized properties, is the main contributor to tissue culture-based regeneration, pluripotency can be widespread amongst plant cells.

It is possible that certain cell types, like the pericycle, are already primed and can easily acquire pluripotency, while cells originating from other tissues need to undergo a competence acquisition stage before their pluripotent potential becomes apparent. Indeed, identity transitions during regeneration are not necessarily immediate, and studies of adventitious root initiation have noticed a delay between the wound response and the appearance of cytological markers associated with root meristematic cells. This suggested that cells require a stage of ‘competence acquisition’ or ‘dedifferentiation’, taking between 24 and 48 h, before identity switch and de novo formation of root meristems can occur (de Klerk et al. 1999).

The idea that plant cells may need to dedifferentiate before adopting a completely new identity goes back to Steward, who commented on the delay in regeneration of carrot phloem cells: ‘…it may be said that the differentiated phloem tissue must first ‘de-differentiate’-whatever that may mean’ (Steward et al. 1958). However, the tone of voice suggests he was skeptical of this hypothesis. Indeed, there are reasons to doubt the existence of such a cellular dedifferentiation process during identity transitions, and the concept has been criticized before (Sugimoto et al. 2011). Crucially, the appearance of cytological markers and induction of cell division is a slow process, taking several hours before it can be observed, and thus may not be suitable to track rapid identity transitions. Recent studies utilizing transcriptional response and expression of marker genes have revealed a much more rapid response.

When the tip of the root meristem is decapitated, removing distal tissues and the entire stem cell niche, existing cells in the stump undergo reorganization to regenerate the root tip within around 48–72 h (Feldman 1976, Sena et al. 2009). While morphological features take time to be apparent, single-cell transcriptional assay of this process has shown that cell identity transitions occur extremely fast. Within <3 h, multiple cells in the stump have already switched their fate and adopted identities of regenerated tissue, as indicated by the expression of hundreds of cell identity markers (Sena et al. 2009, Efroni et al. 2016). These fast transitions are not unique to the regeneration of injured meristems. When cut Arabidopsis stems are placed on an auxin-rich medium, adventitious root meristems are initiated from vasculature-associated cells (Welander et al. 2014). Under these conditions, it took about 2 d for the appearance of morphological markers, but, remarkably, root primordia-specific genes, such as Lateral Root Primordium1 (Smith and Fedoroff 1995), were induced in detached stems as early as 1–5 h following root induction (Welander et al. 2014), indicating that identity changes can be very rapid, even in cases of de novo organ formation.

Curiously, during early stages of root tip regeneration, cells shifting identities were shown to express markers of multiple cells types within individual cells (Efroni et al. 2016). This phenomenon was also observed during induced reprogramming (Buganim et al. 2012). While speculative, it is consistent with the hypothesized ‘transient state’, where multiple identity genes are co-expressed, placing the cell in an unstable state, and allowing dramatic cell identity shifts (Moris et al. 2016). Thus, it is possible that what underlies cellular pluripotency is not a specific factor or cell type, but rather the presence of an internal instability that collapses to a stable identity once proper positional signals have been established.

Identity Transitions and Pattern Acquisition During Plant Regeneration

Individual cells in the plant appear to be able to change their identity very quickly, but recovery of proper tissue patterning following injury is a slow process. This is demonstrated by the gradual recovery of cell identities marked by proteins from the WUSCHEL (WUS) family. Cells expressing WUS and WUS-related genes are located at the center of meristems and regulate stem cell activity in shoot, root, leaf and cambial meristems (Schoof et al. 2000, Sarkar et al. 2007, Forzani et al. 2014, Alvarez et al. 2016).

In the Arabidopsis root, the WOX (WUSCHEL-LIKE HOMEOBOX) gene WOX5 is strongly expressed in the cells of the QC, normally 4–6 cells located at the heart of the stem cell niche (Haecker et al. 2004), and found weakly in the surrounding vasculature and ground tissue initials (Long et al. 2017). When the QC cells of the Arabidopsis root are ablated, expression of WOX5 is lost, and the dead cells are replaced by proximal cells from the stele. At 16 h following the ablation, multiple cells located proximally to the injury were found to express WOX5. After 2 d, more cells were expressing the marker, but, by 72 h following the ablation, this ectopic expression was lost and WOX5 was confined only to central cells, corresponding to the normal position of a QC (Fig. 2A; Xu et al. 2006). In another set of experiments, the distal part of the root meristem was completely abscised, including the entire QC and surrounding stem cells. Here, a group of cells located proximally to the cut site began to express the WOX5 marker at 6 h post-cut, with transcriptional activation of the same gene observed at merely 3 h post-cut. The number of cells expressing the marker increased until 36 h post-cut, but by 72 h expression of the marker was repressed in most of these cells and retained only by cells occupying the normal position of the QC in the root (Fig. 2B). Transcriptome analysis of regenerating roots and of individual cells isolated from the regenerating region of the stump has revealed that cells did not express just WOX5, but also many QC-specific transcripts, indicating that the gradual establishment of the pattern involved rapid and dynamic identity shifts in many cells (Sena et al. 2009, Efroni et al. 2016).

Fig. 2.

Gradual pattern acquisition in tissue regeneration. Cartoon generalization of temporal expression patterns of WUSCHEL-family proteins in different regeneration scenarios. (A) Laser ablation of QC cells (Xu et al. 2006). (B) Decapitation of the root tip (Sena et al. 2009, Efroni et al. 2016). (C) Ablation of the shoot meristem center (Reinhardt et al. 2003). (D) Direct conversion of root primordium to shoot meristems (Kareem et al. 2016, Rosspopoff et al. 2017). (E) Formation of shoot meristems in tissue culture (Gordon et al. 2007, Kareem et al. 2015, T.-Q. Zhang et al. 2017). Black marks the damaged area; green marks the expression of WOX5 in (A and B) and WUS in (C–E). Gray dashes indicate missing data.

The time required to recover normal expression patterns, either following QC ablation or due to the complete removal of the distal tip of the root, seems very similar. However, a delay in appearance of WOX5 patterns can be apparent when the distal root tip dissection is performed high above the QC, near the end of the meristem. When these high-cut roots regenerate, they require an additional 12–16 h before WOX5 appears, but stable patterns still arise by 72 h post-cut (Sena et al. 2009). It is yet unclear what causes this delay in identity switching. There is a considerable decline in the regeneration efficiency of root meristems when cut high above the QC, and it is possible that it represents low competence for identity transitions (Sena et al. 2009). Employing the analogy of a chemical reaction (Moris et al. 2016), it may be said that cells higher up in the meristem may reside in a lower energy state, and thus require higher excitation energy to allow them to reach a transition state and adopt other identities. An alternative explanation is that this apparent loss of competence does not arise from inherit cellular traits, but rather from tissue-wide properties and lack of signals to initiate the regeneration process. The INTERACTOR of CONSTITUTIVELY active ROP1 protein is rapidly degraded in roots in response to high auxin levels. When cut at a low position, near the QC, the protein was rapidly degraded, but when the meristem was cut at a distal location, which did not lead to regeneration, no degradation was observed, indicating that auxin does not form a concentration peak near the wound site (Hazak et al. 2014). Auxin response is required for regeneration and, during normal regeneration, the auxin reporter DR5 is rapidly induced following injury (Xu et al. 2006, Sena et al. 2009, Efroni et al. 2016). Thus, the delay in cellular identity transitions at higher cuts could also be due to a delay in auxin accumulation (Xu et al. 2006; Sena et al. 2009).

In another system of regeneration, detached Arabidopsis leaves are incubated on a hormone-less medium, to generate roots from cambium-associated cells. WOX5 is induced in a small set of root founder cells that will go on to generate a root. Unlike other regeneration systems, here it takes up to 72 h from the time of injury for cells to start expressing WOX5, but expression of another member of the WUSCHEL family, WOX11, is detected at 48 h following dissection (Liu et al. 2014, Hu and Xu, 2016). Similarly to high-cut root tips, the delay in this system may be due to the slow accumulation of auxin in potential cambium founder cells, as such accumulation requires 1–2 d in this system (Chen et al. 2016). However, other factors could be involved, for example the requirement for cell division or some intermediate patterning stage occurring in this regeneration system. When supplied with auxin, cambial cells were shown to be competent to express root primordia-specific identity genes at just 1–5 h following treatment (Welander et al. 2014), so it is unclear whether this delay represents a ‘dedifferentiation’ or competence acquisition stage for individual cells (Sugimoto et al. 2011).

In the shoot meristem, WUS is expressed at the organizing center, a group of cells in the central zone of the meristem (Schoof et al. 2000). When this zone was ablated in the tomato meristem, cells in the immediate vicinity of the damaged region began to express WUS <24 h after the ablation. At 48 h after the damage, the number of cells expressing the marker increased, but by 72 h the signal was confined to a group of cells at the center of the meristem, representing the normal tissue pattern (Fig. 2C; Reinhardt et al. 2003).

Thus, it seems that despite differences in developmental context and inflicted injury, the process of pattern recovery in meristems is remarkably consistent, taking around 72 h to recover from the time of its initiation. During this time, the number of WOX-expressing cells is first increased and then diminished to finally occupy the proper number of cells, while individual cells readily transition from a WOX-expressing to a non-WOX-expressing state, and vice versa. As the WOX gene products are transcription factors, this change of expression likely induces large changes in the cell transcriptome, as observed in transcriptomic studies (Sena et al. 2009, Wang et al. 2014, Efroni et al. 2016).

Tissue Culture Versus Wound-Induced Regeneration

Organ regeneration following wounding (Reinhardt et al. 2003, Xu et al. 2006, Sena 2014, Efroni et al. 2016) and normal organogenesis (Dubrovsky et al. 2000, Von Wangenheim et al. 2016) both occur during roughly the same time frame, requiring about 3 d to achieve stable tissue patterning. In contrast, plant regeneration in tissue culture conditions requires significantly more time, up to 12 d (Skoog and Miller 1957, Gordon et al. 2007, Sugimoto et al. 2010, Sugimoto and Meyerowitz 2013, Kareem et al. 2015).

Regeneration in tissue culture initiates with callus tissue, whose development is promoted by the application of auxin and cytokinin (Sugimoto and Meyerowitz 2013). Callus itself was regarded as a uniform undifferentiated cell mass. However, analysis of gene markers and transcriptomic assays have revealed that in tissue culture, callus expresses many root meristem-specific markers and is made up of distinct domains, suggesting that callus may be similar in many ways to lateral roots (Sugimoto et al. 2010, Ikeuchi et al. 2013). However, it should be noted that callus induced by wounding, or by overexpression of WOUND INDUCED DEDIFFERENTIATION 1, does not seem to express root markers. Additionally, mutants of ABERRANT LATERAL ROOT FORMATION 4, which are perturbed in formation of lateral roots and callus in tissue culture, are able to form wound-induced callus (Iwase et al. 2011, Melnyk et al. 2015), suggesting that wound-induced callus formation is mediated by distinct genetic mechanisms from those of callus in tissue culture. Loss of multiple members of the PLETHORA (PLT) family of transcription factors cause callus to lose both its lateral root-like identity markers and its ability to regenerate shoots on high cytokinin media (Kareem et al. 2015). However, further study is required to clarify whether the lateral root-like character of tissue culture callus has a role in its pluripotent properties.

When calli are incubated on shoot induction media (SIM), containing high levels of cytokinin, shoot meristems are generated. Using a promoter fusion of the WUS gene, it was shown that cells expressing WUS appear at 3–4 d following transfer to SIM in a diffuse pattern (Gordon et al. 2007, Kareem et al. 2015, T.-Q. Zhang et al. 2017). In situ hybridization of WUS transcript and a new version of the WUS reporter reveals a somewhat different pattern, with WUS expression initiating in disperse isolated cells at 3 d following transfer. Both markers show that after an additional 4 d, WUS-expressing cells are found in stable patterns at the center of initiating shoot meristems. Despite some discrepancies between the methods, common to all of them is an initial transient spread of WUS to multiple cells, followed by confinement to cells located at distinct foci (Fig. 2D; Li et al. 2011, Kareem et al. 2015, Meng et al. 2017, T.-Q. Zhang et al. 2017).

Thus, while pattern recovery is significantly slower than during injury-induced regeneration, it follows similar principles of weak and expanded expression that becomes restricted to a defined domain within 2–4 d. During this time, many cells transition from a WUS-expressing to a non-WUS-expressing state, and vice versa. It follows that a large number of cells in the callus are competent to undergo identity transitions, but only where stable patterns are formed is cell identity stabilized (Fig. 2).

Apart from regeneration via callus in tissue culture, plants are capable of other hormone-induced direct transitions. When primed lateral root primordia at an early stage of development are cultured on media containing a high level of cytokinins, they are converted into shoot primordia within a few days without an intermediate callus stage (Fig. 2E; Chatfield et al. 2013, Kareem et al. 2016, Rosspopoff et al. 2017). In these converting lateral roots, expression of the root-specific WOX5 is replaced by the shoot meristem variant WUS (Kareem et al. 2016, Rosspopoff et al. 2017). WUS expression by itself can drive the formation of shoot meristems (Meng et al. 2017, Negin et al. 2017) but, as WUS and WOX5 were shown to be interchangeable (Sarkar et al. 2007), it is unclear whether the expression conversion between the two WUS family members is the cause of the identity transitions, or whether is it a marker for that event. Lateral roots lose WOX5 expression within 24 h of cytokinin treatment, while expression of WUS is induced at the same time (Rosspopoff et al. 2017). By 19 h, expression of the WUS marker appears sporadic and in many cells, but between 24 and 48 h only a few cells at the center of the converting organ continue expressing the marker. Formation of a shoot meristem requires an additional 72 h, during which expression of WUS remains stable (Chatfield et al. 2013, Rosspopoff et al. 2017). As co-expression of WOX5 and WUS was not examined, it is not known whether the intermediate stages contain cells with mixed identity, expressing both WOX genes.

It is unclear why formation of stable patterns is delayed in tissue culture conditions. One possibility is that these delays are caused by slow epigenetic modifications required for cell identity switches (Birnbaum and Roudier 2016). Indeed, there are documented effects of chromatin modifications on cell identity transitions, including during regeneration (Ikeuchi et al. 2015, T.-Q. Zhang et al. 2017). Loss of DNA methylation in the dna methyltransferase1 mutant caused WUS to appear 2 d earlier than the wild type in regenerating shoots on a high level of cytokinins (Li et al. 2011). In addition, loss of H3K27 histone methylation at the WUS promoter was detected during induction of shoot regeneration in culture (T.-Q. Zhang et al. 2017). Apart from epigenetic modification, it is possible that other developmental stages are required. An interesting recent study has identified a crucial intermediate step during regeneration requiring multiple PLT genes (Kareem et al. 2015). Similar members of the PLT family were not required for root tip regeneration (Sena et al. 2009), but not all plt mutant combinations were tested in both systems. How these PLT proteins promote tissue competence for regeneration and how universal is their activity is still unclear.

It is important to note that tissue culture studies are conducted under very high and growth-inhibiting concentrations of the plant hormones auxin and cytokinin. The formation of a balance between auxin and cytokinin is intimately linked to stable pattern formation in the plant (Schaller et al. 2015), and their high level in the media may greatly inhibit the formation of these growth-promoting patterns. Indeed, hormonal treatment of regenerating root tips induced a delay of about 24 h in pattern stabilization (Efroni et al. 2016). Of note is that many of the mutants that disrupt regeneration from callus also affect, either directly or indirectly, the response sensitivity to auxin and cytokinin (Celenza et al. 1995, Buechel et al. 2010, Furuta et al. 2011, Melnyk et al. 2015, Iwase et al. 2016, Meng et al. 2017).

The Future of Regeneration Studies: Focusing on the Cell’s Point of View

The prevailing notion of early investigators of plant morphogenesis was that the division and arrangement of cells were secondary to organ-wide polarity and growth rates (Barlow 1982, Kaplan and Hagemann 1991). As a result, the distinction between cell identity transition and tissue patterning is often conflated. This review emphasizes the difference between cellular competence and tissue competence. In many cases, references to cellular dedifferentiation or transdifferentiation (Ikeuchi et al. 2013, Kareem et al. 2016, Rosspopoff et al. 2017) can better be understood as referring to the reactivation of dormant tissue-wide patterning mechanisms, rather than to the identity switches or changes in potency of any individual cell. For example, during the recovery of the meristem following the removal of its distal tip, broad spatial patterning programs were shared with the formation of the root in the embryo (Efroni et al. 2016). Thus, while individual cells were rapidly changing their identities in complex ways, the whole tissue was ‘de-differentiating’ in the sense that it was replaying a program from an earlier stage in development. In contrast, during de novo shoot meristem regeneration in tissue culture, no activation of WOX2, a gene involved in the formation of the shoot in the embryo, was observed, suggesting no activation of embryonic programs (T.-Q. Zhang et al. 2017, Z. Zhang et al. 2017). Plants produce new organs throughout their lives, and the embryo is just one of the possible sources for early developmental programs. Indeed, multiple root initiation programs exist, depending on the developmental context (Bellini et al. 2014, Verstraeten et al. 2014, Birnbaum 2016) and the tissue-culture-induced transition of aerial tissue to callus involved the activation of programs that are specific to the formation of lateral roots, but not embryonic roots (Sugimoto et al. 2010).

Both reactivation of organogenesis patterning programs and plasticity in cell identity transitions are crucial for successful recovery from damage, but it is important to distinguish between the two, if not only for the different time scales that govern these two processes. It is clear that while cell identity transitions can occur extremely rapidly, within the scale of hours (and probably less than that), the emergence of stable patterns can takes several days. Hence, an individual cell embedded within the context of a regenerating tissue may undergo many complex identity transitions before adopting a stable fate. How these rapid changes contribute to the complete regeneration of the tissue remains to be discovered.

A limitation of current regeneration studies is their focus on Arabidopsis, while a broader perspective may be gained from examining other plant species. Examination of cell identity transitions in multiple species had mixed results, with the conservation in the expression of some, but not all, key identity markers (Lim et al. 2000, Fan et al. 2008, Iwase et al. 2013, Ron et al. 2014). Cross-species transcriptomic studies have uncovered common transitions during animal embryonic development (Levin et al. 2016). Comparative genomic and transcriptomic studies in plants are still in their infancy but, as the ability to regenerate is highly conserved in plants, such comparisons can help to better define and understand cell identity transitions and pattern emergence during regeneration.

The apparent widespread ability of plant cells to acquire new fates raises interesting questions. How do cells remove or ‘forget’ their old fate? What is the role of epigenetic modifications in the process? Are unstable transitional states required for identity transitions? Recent technical breakthroughs, such as live imaging that allows the tracing of individual cellular histories (Prunet et al. 2016, Von Wangenheim et al. 2016, T.-Q. Zhang et al. 2017), the introduction of new fluorescent markers for multicolor imaging of individual cells and the utilization of single-cell transcriptomics (Efroni and Birnbaum 2016), make this topic of research accessible to current researchers. Plant cells are embedded in a rigid cell matrix that poses technical challenges to addressing these questions at the level of individual cells, but overcoming these challenges promises rich rewards in our understanding of these crucial developmental processes.

Funding

The Israeli Science Foundation [ISF966/17] and the Howard Hughes Medical Institute [International Research Scholar Grant 55008730].

Glossary

Abbreviations

iPSC

induced pluripotent stem cell

PLT

PLETHORA

QC

quiescent center

SIM

shoot induction media

WUS

WUSCHEL

WOX

WUSCHEL-LIKE HOMEOBOX