Outstanding questions in developmental ERK signaling

Animal development relies on a small set of signaling systems acting in combination to guide pattern formation and tissue morphogenesis (Martinez-Arias and Stewart, 2002). By now we have a nearly complete parts lists of at least the core elements of these systems and are studying them at multiple levels of biological organization. However, we are still far from understanding what makes signaling systems robust and how a single pathway can have such diverse outputs, and also from being able to explain how relatively subtle perturbations to signaling transduction can cause developmental abnormalities (Tidyman and Rauen, 2012; Rauen, 2013). Here, we focus on the extracellular signal-regulated kinase (ERK) cascade, an essential regulator of animal development (Fig. 1) (Gabay et al., 1997; Dorey and Amaya, 2010; Corson et al., 2003). Using three extensively studied experimental models of developmental ERK signaling, we highlight some of the key outstanding questions that must be addressed to achieve the next level of understanding. In each of these models, ERK signaling is triggered by a well-defined ligand source and, via an intracellular phosphorylation cascade, induces spatial patterns of gene expression in a field of responding cells. Although this scenario is certainly not the only mode of developmental ERK signaling (Molotkov et al., 2017; Kang et al., 2017; Reim et al., 2012; Kadam et al., 2012; Stathopoulos et al., 2004), its relative simplicity makes it especially attractive for discussing the most crucial unanswered questions. These questions, and the insights we can gain into them from simple systems, should also be relevant to more complex scenarios.

We start by discussing unanswered questions related to the processes at the input layer of the ERK cascade, focusing on the spatiotemporal control of receptor activation. We then turn to the transcriptional interpretation of ERK activation. The following section focuses on the mechanisms that ensure robust signaling and discusses the origins of ERK-dependent developmental defects. We close by proposing directions for future studies and discuss the relevance of the stated questions for other signaling systems.

As summarized above, we focus here on three major areas where we still have much to learn about the ERK pathway and its effects. The first set of questions is related to the quantitative understanding of mechanisms that are already well studied at the molecular and cellular levels, such as signal initiation – when ligands bind to transmembrane receptor tyrosine kinases (RTKs) (Lemmon and Schlessinger, 2010). Despite decades of study, we still have a poor understanding of how the absolute concentrations of ligand and receptor, and the kinetics of their interactions, impact both quantitative and qualitative aspects of signal output. How many ligand-receptor complexes are required to initiate a signal? How are ligand-receptor complexes spatially distributed in a field of responding cells? We therefore need to be able to quantify the numbers of active RTKs required to trigger intracellular pathways to provide an absolute measure of signaling inputs. These numbers can be readily estimated in cultured cells; however, to the best of our knowledge, they have yet to be obtained in a single developmental context (Schoeberl et al., 2002; Stockmann et al., 2017).

The second set of questions addresses the mechanisms by which active ERK controls gene expression to influence developmental pattern formation. In comparison with studies of ERK activation by upstream components of the pathway, the mechanisms by which active ERK alters the activities of downstream transcription factors and basal transcription machinery are relatively unexplored (Kim et al., 2011; Kolch, 2000; Hollenhorst et al., 2011). What physical changes to transcription factors are induced by ERK activity? Where do these changes occur in the cell? We need a better understanding of how the ERK pathway regulates transcriptional activators and repressors to alter the gene expression profile of a cell – and how it can induce diverse transcriptional outputs in different contexts.

Finally, a third set of questions is motivated by studies of a large class of genetically based human developmental abnormalities associated with deregulated ERK signaling (reviewed by Jindal et al., 2015). Although it is believed that these abnormalities reflect quantitative changes in the spatiotemporal distributions of developmental ERK signals, the magnitudes of pertubation are essentially unknown and we have a poor understanding of the limits within which the pathway must operate to achieve normal development (Goyal et al., 2017). What are the maximum and minimum amplitudes and durations of ERK activity that still lead to normal patterning outcomes? To what extent must ERK activity be restricted in space? Quantifying these parameters is necessary for probing the intrinsic robustness of developmental ERK signals.

We discuss these questions in the context of the three well-studied inductive events in Ciona intestinalis, Drosophila melanogaster and Caenorhabditis elegans. In the 32-cell stage Ciona embryo, fibroblast growth factor (FGF) is secreted from 16 vegetal cells and activates the ERK pathway in immediately adjacent animal cells (Fig. 2A) (Bertrand et al., 2003). The downstream target of the ERK pathway, the neural marker otx, is consequently expressed in a four-cell pattern in the animal hemisphere – specifically in the a6.5 and b6.5 pairs of cells that have the highest surface contact with the vegetal cells (Hudson and Lemaire, 2001; Tassy et al., 2006). otx ultimately specifies part of the neural lineage of the Ciona embryo (Wada et al., 2002; Imai et al., 2006). ERK activity also specifies part of the neural lineage in the early Drosophila embryo. In this case, two ventrolateral stripes of the epidermal growth factor (EGF) in the 3 h-old syncytial embryo activate the ERK pathway in an autocrine manner (discussed further below), inducing expression of its downstream target: the homeobox transcription factor ind (Fig. 2B) (Lim et al., 2015; Von Ohlen and Doe, 2000). In the final example, a gradient of EGF ligand, released from a single cell, induces a stereotypic cell fate pattern in the underlying epidermis that always includes a single cell bearing primary (1°) fate in the vulval precursor cells (VPCs) of the C. elegans larva (Fig. 2C) (Katz et al., 1995). Each of these systems exhibits clear loss-of-function phenotypes in the absence of ERK activation: in Ciona and Drosophila, the nervous system does not develop properly without ERK signaling, while in C. elegans, insufficient ERK signaling leads to a ‘vulvaless’ phenotype – the worm cannot lay eggs (Moghal and Sternberg, 2003). Together, these three examples have served as significant insights into how ERK signaling functions, and provide valuable testing grounds to further probe the questions laid out above.

ERK signaling is initiated by ligand/receptor binding at the cell surface (Fig. 1). In all three examples chosen for this Review, locally produced ligands reach their target receptors either by diffusing away from the source (paracrine signaling; Fig. 3A), or by acting at short range with diffusion being either insignificant or nonexistent (juxtacrine or autocrine signaling; Fig. 3B,C). We still have an incomplete understanding of what defines ligand diffusivity in different contexts, but it is clear that ligand range and concentration will be important determinants of signal response.

The patterning of vulval precursor cells (VPCs) in C. elegans provides what seems to be a case of pathway activation by diffusible ligands (Fig. 3A). In this system, the anchor cell (AC) secretes a diffusible ligand, EGF, towards the undifferentiated vulval precursor cells (VPC), named P3.p to P8.p (Moghal and Sternberg, 2003; Sternberg, 2005). EGF seems to activate the ERK pathway such that three distinct cell fates, primary (1°), secondary (2°) or tertiary (3°) are induced in a distance-dependent manner (Katz et al., 1995). The gradient of ERK activation normally peaks at the VPC situated closest to the AC (P6.p), which is the VPC that always assumes 1° fate while the cells adjacent to P6.p take on 2° fate. The remaining peripheral cells become 3° cells (Sternberg and Horvitz, 1986). Although EGF forms a gradient, VPC induction may actually occur sequentially such that P6.p is induced with 1° fate first, before inducing 2° fate in the VPCs adjacent to P6.p (Simske and Kirn, 1995). EGF secreted by the AC induces expression of the rhomboid protease ROM-1 in proximal VPCs, which then cleaves and activates a variant of EGF to which more distal cells are sensitive (Dutt et al., 2004). Signal relay from proximal to distal cells also increases the range of EGF.

The distribution of receptors available to bind and sense the extracellular EGF on the membranes of the VPCs will also control the ERK inputs. In VPCs, EGF receptor (EGFR) localization and mobility modulates the number of ligand-receptor complexes formed. For one, localization complexes target EGFR to the basolateral membranes of VPCs, which face the EGF-secreting ACs (Simske et al., 1996; Kaech et al., 1998). Moreover, once targeted to the basolateral membrane, an actin-binding protein, ERM1, can sequester EGFR in an inactive compartment. It is thought that attachment to the cortical F-actin via ERM1 restricts EGFR mobility and therefore its access to other proteins that are required for receptor activation. This sequestration of a pool of EGFR permits long-lasting sensing of the external EGF gradient during the course of VPC induction (Haag et al., 2014).

The responses of the VPC pattern to variations in the strength of signaling inputs are inconsistent with the idea that minimum thresholds of ERK activation define cell fate. For example, partially reducing EGFR expression can lead to multiple VPCs with 1° fate, which results in a multivulva phenotype (Aroian and Sternberg, 1991). Reduced signaling levels would not lead to more 1° fate induction if a minimum threshold of ERK activity were required. A threshold model for cell fate induction suggests that the external ligand gradient serves as a morphogen, providing positional cues for cell fate among the VPCs. When the source of available ligand is increased, this model would predict that more cells should be induced with 1° fate. Instead, the wild-type 3°-3°-2°-1°-2°-3° pattern is maintained over an approximately threefold range of EGF levels (Barkoulas et al., 2013). This robustness is in stark contrast to an analogous example of paracrine ERK signaling the Drosophila oocyte, which is highly sensitive to the inductive ligand gradient. In this system, extra genetic doses of EGF are not tolerated and strongly dorsalize the Drosophila egg (Neuman-Silberberg and Schupbach, 1994).

As discussed below, the emerging patterns of cell fates can be attributed to a variety of regulatory feedback mechanisms. In addition, a first level of control – ensuring specification of exactly one cell with 1° fate – could be provided by the rapid sequestration of the diffusible EGF ligand by P6.p. Such a model is reminiscent of other systems involving long-range induction by a diffusible morphogen, in which interactions between the ligands and their cognate receptors that lead to self-enhanced ligand degradation are important for generating robust patterns (Eldar et al., 2003). Indeed, induction of distal VPCs with 1° fate occurs only when cells closest to the AC are ablated, supporting the idea that the VPCs sensing EGF at its highest concentration are capturing ligand before it can diffuse away (Sternberg and Horvitz, 1986; Sulston and White, 1980). When EGFR expression is partially reduced, less EGF sequestration by P6.p could lead to increased ligand diffusion to more distal VPCs, explaining the apparent hypersensitivity in these mutants (Hajnal et al., 1997). Quantitative understanding of these effects requires accurate measurements of the spatiotemporal distribution of EGF/EGFR complexes as well as a framework for connecting the information about input layer of the patterning network to the emerging cell fate patterns.

In contrast to the VPC patterning system, in which one cell secretes a diffusible ligand, multiple ligand-producing cells collectively contribute to the total ERK input that induces otx in a subset of cells of the 32-cell stage C. intestinalis embryo. In this system, prior stages of asymmetric cell division result in a unique set of contacts with adjacent cells (Fig. 2A) (Ohta and Satou, 2013; Rothbächer et al., 2007). Asymmetric maternal factors both predispose the animal hemisphere to express the transcription factors required for otx expression (Oda-Ishii et al., 2016; Bertrand et al., 2003; Rothbächer et al., 2007) and are responsible for FGF secretion from the vegetal hemisphere (Imai et al., 2002). In these embryos, the intercellular space is too small to permit significant ligand diffusion from the source to the responding cell, so direct cell-cell contacts between the plasma membranes of ligand-producing and -responding cells are needed for ERK activation (Tassy et al., 2006). As a consequence, the ERK inputs are contact dependent and can be thought of as juxtacrine (Fig. 3B). The animal cells that have the highest surface contact area with the FGF-secreting vegetal cells, named the a6.5 and b6.5 pairs, always express otx (Tassy et al., 2006). It is still unclear, however, whether ligand secretion is uniform on all faces of an inducing cell, and whether receptors are evenly distributed on the plasma membranes of responding cells. Addressing these issues is difficult without the ability to monitor ligand-receptor interactions, but such techniques will be required if we are to understand in detail the mechanisms underlying the defined spatial pattern of ERK activation.

Although the VPC and otx patterning systems are examples of ligand molecules diffusing from one source cell to another, ligands can also bind to receptors on the same cells that produce them (Fig. 3C). For example, during ind induction in the Drosophila embryo, EGF signaling appears to work in an autocrine regime, whereby ligand-producing cells also express cognate receptors. At this point, cellularization is not complete and the Drosophila embryo is still a syncytium with multiple nuclei lining the periphery. Ligands produced in the syncytial embryo are secreted into the perivitelline space, which surrounds the common plasma membrane that contains EGF receptors. A transcriptional circuit downstream of the well-characterized ventral-dorsal Dorsal (Dl) morphogen gradient establishes a two-striped pattern of the expression of Rhomboid (Rho), an intracellular protease that controls the secretion of the EGF ligand Spitz (Spi). An incoherent feed-forward loop is established when Dl induces both Rho and a repressor of Rho, Snail (Rushlow and Shvartsman, 2012). Rho is therefore induced only in lateral cells in which intermediate levels of Dl are strong enough to induce Rho whereas Snail repression is minimal. Strongly overlapping spatial profiles of Rho expression and active dually phosphorylated ERK (dpERK) suggest that diffusion of secreted ligand is negligible, most likely reflecting capture of secreted ligands by the very same cells that produce them (Lim et al., 2015). Notably, prepatterned transcription factors limit ind expression to a subset of cells within the domain of dpERK activity. Interestingly, both Spi and EGFR are controlled by Zelda (Zld), a uniformly expressed activator of early zygotic transcription (Liang et al., 2008). A mathematical model of these processes, based on the assumptions that signaling levels are proportional to the number of ligand-bound EGF receptors and that ligand diffusion is negligible, predicts that signaling levels should rise as the cube of time, measured from the onset of ligand production. The time-resolved measurements of the levels of dpERK support this prediction, suggesting that the signaling cascade connecting cell-surface receptors and ERK activation indeed functions as a linear system (Lim et al., 2015).

Studies of some of these systems have led to the formulation of mathematical models that have ligand-receptor complexes as their key variables (Giurumescu et al., 2006). At the same time, the absolute values of active complexes are currently unknown, preventing direct testing of model predictions. Several existing tools may enable quantitative analysis of ligand binding in a developing embryo. In particular, fluorescence correlation spectroscopy (FCS) of fluorescently tagged ligands can measure ligand concentrations and diffusivities in vivo. This technique may be useful for quantifying ligand concentrations in systems in which the ERK-activating ligands diffuse away from a localized source. For example, FCS analysis of the FGF8 morphogen in the zebrafish embryo has been shown to allow quantification of local concentrations of ligand with high precision (Yu et al., 2009). Monitoring complexes and ligand-receptor interactions could also be enabled by live imaging of quantum dot-labeled ligands and fluorescently tagged receptors (Lim et al., 2016). Quantum dot ligands can be detected at the single nanoparticle level and have already been used to study ligand binding and transport phenomena during RTK signaling in cells (Lidke et al., 2004). Such techniques will allow us to monitor the evolution of ligand-receptor complexes, and directly measure quantitative aspects of signal initiation such as the lifetime and turnover rate of a complex at the membrane. Moreover, these experiments have the potential to reveal important roles of receptor trafficking in regulating ERK activation in time and space. As illustrated in the VPC induction example, receptor localization and mobility influence the duration of an ERK signal. When and where ligand-bound and unbound receptors are trafficked within a cell could therefore dramatically alter how a cell senses external signaling cues in other systems.

ERK controls gene expression by phosphorylating transcription factors and components of the basal transcription machinery. The reported effects on transcription factors are diverse and include potentiation of the effects of existing activators and antagonism of repressor functions. During VPC patterning in C. elegans, phosphorylation by ERK induces structural changes in at least two transcription factors involved in the regulation of lin-39, a key gene responsible for the 1° vulval fate (Clark et al., 1993; Wagmaister et al., 2006) (Fig. 4A). This gene is initially repressed by a complex formed by the forkhead transcription factor LIN-31 and the ETS transcription factor LIN-1, which represses lin-39 by interacting with nucleosome remodeling and deacetylation complexes (Guerry et al., 2007; Maloof and Kenyon, 1998; Leight et al., 2005; Miller et al., 1993; Tan et al., 1998). ERK phosphorylates the C-terminal domain of LIN-1, disrupting the repressor complex and potentially turning LIN-1 into an activator of lin-39 (Fig. 4A, bottom) (Jacobs et al., 1998; Tiensuu et al., 2005; Leight et al., 2015; Wagmaister et al., 2006). The role of LIN-1 as an activator or a repressor depends on its phosphorylation status, and on the particular target gene. Although ERK-mediated phosphorylation of LIN-1 converts it to an activator of 1° fate genes (Leight et al., 2015), for other target genes, LIN-1 may act solely as a repressor. For example, transcription of lateral Notch ligand genes (see below) does not require LIN-1 activation (Underwood et al., 2017). Furthermore, dissociation of the LIN-1/LIN-31 complex exposes a phosphorylation site in the transactivation domain of LIN-31 such that it also becomes an activator when phosphorylated by ERK (Fig. 4A) (Tan et al., 1998). In the current model of the 1° vulval fate induction, active ERK first disrupts the LIN-1/LIN-31 complex that represses the key target gene, and then converts one or both components of this complex into a direct activator of the same gene (Tan et al., 1998; Sundaram, 2013). Thus, ERK signaling both relieves repression of a target gene and promotes its activation, as if first releasing the brakes of a car and then pressing on the accelerator. This model is not unique to the specific example of VPC induction in C. elegans. In Drosophila, the ETS factors Pointed (Pnt) and Yan collaboratively repress target gene transcription. Gene expression is induced when an ERK signal leads to disruption of the repressive complex followed by Pnt-mediated activation (Webber et al., 2018).

In contrast to the VPC system, ERK works only by relieving repression during the induction of ind in the early fly embryo. In this case, ERK phosphorylates Capicua (Cic), an HMG-box repressor that acts as a sensor of ERK signaling in Drosophila and other organisms (Jimenez et al., 2012). In the absence of ERK signaling, Cic is localized predominantly to the nucleus, where it represses ind through the highly conserved Cic-binding sites within the ind enhancer (Ajuria et al., 2011). One of the mechanisms proposed for the signal-dependent relief of gene repression by Cic involves the ERK-dependent nuclear export of Cic, followed by its degradation in the cytoplasm (Fig. 4B) (Grimm et al., 2012). The expression of ind, however, can be detected before any significant reduction in the nuclear levels of Cic, suggesting that the ERK-dependent relief of gene repression can be achieved while Cic is still in the nucleus (Lim et al., 2013). Presumably, ERK phosphorylates Cic while it is still bound to its target enhancers. One possibility is that, similar to the disruption of the protein complex that represses lin-39 in the VPC system, phosphorylation by ERK causes rapid disruption of Cic interactions with its binding partners involved in gene repression. These events would be followed by dissociation from DNA and slower export from the nucleus. Interestingly, ERK also phosphorylates Groucho (Gro), a broadly expressed co-repressor involved in the regulation of ind as well as a number of other genes (Hasson et al., 2005; Helman et al., 2011; Cavallo et al., 1998). Gro acts as a non-DNA-binding co-repressor that interacts with other DNA-binding transcription factors, such as Cic, that are crucial for silencing gene expression. Phosphorylation interferes with the co-repressor function of Gro but does not lead to its degradation (Cinnamon et al., 2008). The phosphorylated form of Gro persists in the nucleus even after the nuclear levels of Cic are re-established after termination of ERK signaling, potentially providing a long-term memory mechanism for the transcriptional interpretation of the transient ERK signal (Helman et al., 2011). Although it is known that Gro is required for Cic repression based on genetic perturbations, there is no direct evidence that they form a complex on the regulatory DNA of target genes.

The nature of the transcriptional response during the ERK-dependent otx expression in Ciona is more poorly understood than in the two systems described above. What is known is that ERK phosphorylates the Ciona ETS1/2 factors that act as essential activators of otx expression (Farley et al., 2015; Bertrand et al., 2003). It remains to be determined whether these factors are converted from repressors into activators, such as LIN-1 and LIN-31 in C. elegans, and other ETS factors found in vertebrates and invertebrates or behave completely differently (Maki et al., 2004; Rebay and Rubin, 1995; Sharrocks, 2001). Although we know that ETS factors are required for otx expression, it is still not even clear in the Ciona system whether ERK directly phosphorylates ETS1/2, or whether it, for example, inactivates their repressors.

Note that phosphorylation of transcription factors that are already expressed in the cell is only one of the strategies by which ERK can control its transcriptional targets. Cascade-type mechanisms, whereby a protein product of a gene induced by ERK activation can function as a new regulator of additional downstream target genes, add another level of complexity to transcriptional control. This mechanism is well documented for the Drosophila ETS factors, which are expressed upon relief of their repression by Cic, and control multiple genes involved in ERK-dependent cell functions (Dissanayake et al., 2011; Jin et al., 2015). In addition to targeting the transcription factors that work at the level of regulatory DNA, active ERK can also affect gene expression more directly, by phosphorylating the components of basal transcription machinery. Recently, it has been shown in human cells that ERK phosphorylates INTS11, a catalytic subunit of an RNA polymerase-associated complex called the integrator (Yue et al., 2017). Interestingly, ERK can also act to modulate the chromatin landscape at its target genes. For example, in cancerous prostate cells, ERK-mediated phosphorylation of an ETS factor causes dissociation of components of the polycomb repressive complex from the chromatin, which then creates a permissive environment for transcription (Kedage et al., 2017).

Biochemical characterization of the full repertoire of mechanisms available for transcriptional interpretation is essential to complete our understanding of inductive ERK signaling. We can begin by asking how ERK interacts with transcription factors in space and time, and at different levels of pathway activation. When ERK targets a transcriptional repressor, what structural changes lead to de-repression (as occurs for Cic) versus conversion to the activator state (as occurs for LIN-31)? Does phosphorylation simply switch off a repressive function, change subcellular localization, involve a co-factor or change the function of the protein altogether? Although dynamic changes in subcellular localization are most readily pursued by live imaging, questions related to changes in the interaction partners can be addressed using tools such as chromatin immunoprecipitation (ChIP), which can identify how active ERK interacts with proteins in the nucleus. For example, ChIP-seq analysis was recently used to show that ERK-induced activation of the transcription factor Elk1 also leads to histone modifications that promote transcription in mouse embryonic fibroblasts (Esnault et al., 2017). Structural analysis will then be essential to describe how ERK-induced changes between transcription factors and DNA or other components of transcriptional machinery lead to gene expression. In the case of C. elegans VPC induction, these approaches could help clarify whether LIN-1 and LIN-31 dissociate on or away from the DNA. The answer may be dependent on whether LIN-1 is simply de-repressed or whether it also becomes an activator.

Defects in tissue patterning and morphogenesis can be caused by both gain- and loss-of-function genetic perturbations of the ERK pathway and its inputs (Runtuwene et al., 2011; Visser et al., 2012; Newbern et al., 2008; Xing et al., 2016; Pucilowska et al., 2012; Vithayathil et al., 2017; Tartaglia et al., 2007; Pandit et al., 2007). This conclusion stems from studies of both human genetic diseases and of developmental processes in model organisms (see Box 1) (Jindal et al., 2017). For example, ERK activation by locally secreted FGF plays a key role in patterning of the mammalian forebrain, a process that can be disrupted by both loss of ligand and by extending the duration of ligand production (Nonomura et al., 2013; Meyers et al., 1998). The effects of gain-of-function mutations on ERK activation in vivo appear to be context dependent. Indeed, recent studies demonstrate that constitutively active mitogen-activated protein kinase kinase (MEK) can in fact have divergent effects on ERK activity in different regions of the embryo (Goyal et al., 2017). Clearly, both lower and upper bounds on the doses, durations and spatial extents of ERK signals must exist during development. How strict are these limits? What are the mechanisms that establish them and ensure that they are obeyed during embryogenesis? In all cases studied so far, robust developmental outcomes require coordinated control of ERK signaling at multiple levels, from ligand production to transcriptional interpretation.

Open-loop control mechanisms, which by definition do not involve feedback, are important for ensuring robust signaling outcomes. For an open-loop control system, the downstream effects of activated ERK do not affect the input signal or change signal transduction. Rather, orthogonal, ERK-independent inputs influence the emerging spatiotemporal patterns of ERK activation by altering signaling transduction at multiple levels. For example, an open-loop control mechanism can originate from other signaling systems that disrupt protein interactions in the ERK signaling cascade when activated. In addition, expression of the target genes of ERK are often controlled by a number of transcription factors that do not all necessarily interact with active ERK. ERK-independent transcriptional repressors and activators therefore offer an open-loop control mechanism at the level of transcription.

The expression of ind in the early Drosophila embryo is an example of a patterning event that is robust with respect to significant variations in the dose, duration and spatial extent of ERK activation (Fig. 5). The two-striped pattern of ind, which is established by a transient pulse of ERK signaling, persists when the amplitude of this pulse is reduced to a quarter of its value in the wild-type embryo and when the pulse is delayed by almost 1 h (Lim et al., 2015; Rogers et al., 2017). Furthermore, the expression of ind remains essentially unperturbed when the duration of the ERK signaling pulse is extended well beyond the normal 1 h time window and when it is expanded to the entire blastoderm (Johnson et al., 2017). Part of this impressive robustness can be attributed to the fact that, as discussed above, in this context, ERK works by relieving repression. As a consequence, the ERK-independent activators can initiate the expression of ind as long as the strength of the provided ERK signal exceeds a threshold value (shown as θ in Fig. 5). The value of this threshold appears to be significantly lower than the signaling level in the wild-type embryo, which means that the inductive signal can be reduced significantly and still elicit normal transcriptional response (Lim et al., 2015). Robustness with respect to perturbations in the opposite direction relies on the effects of ERK-independent factors. In particular, the repressors Snail (Sna) and Ventral nervous system defective (Vnd) ensure that ind cannot be activated in the ventral and ventrolateral part of the embryo, even if ERK activation is expanded beyond its normal domain (Rogers et al., 2017; Stathopoulos and Levine, 2005). Furthermore, diminishing levels of the nuclear localization of the transcriptional activator Dorsal limits ind expression on the dorsal side of the embryo. Thus, robust induction of ind relies on several regulatory strategies, including threshold-dependent responses and combinatorial effects of ERK-independent activators and repressors (Samee et al., 2015).

The ERK-independent repressors restraining the wild-type expression of ind act at the level of the regulatory DNA. Similarly, in the C. elegans VPC model, the transcriptional repressors REF-2 and the MAB-5 repress lin-39 in posterior Pn.p cells to render only a subset of exactly six VPCs, P3.p to P8.p, competent to respond to ERK signals (Alper and Kenyon, 2002). Viewed more broadly, this strategy amounts to prepatterning of a cellular response to an extracellular cue and can be implemented in many ways. For example, during the induction of otx in the Ciona embryo, the effect of FGF is spatially restrained by the Eph/ephrin pathway (another juxtracrine signaling system) that provides negative control of processes leading to ERK activation and is required to restrict otx expression to exactly four cells (Haupaix et al., 2013; Ohta and Satou, 2013). Similar to what happens in the ind regulation circuit, the ephrin signals are established by spatially nonuniform maternal inputs and are independent of the ERK activation level. This open-loop control strategy appears to be especially suited for the rapidly developing early stages of embryogenesis, where inductive signals operate under strict temporal constraints. Indeed, cell fate patterning in the early Ciona embryo and ind induction in Drosophila take place on the order of minutes, within the time scale of a cell cycle (Lim et al., 2015; Nakatani and Nishida, 1994). For patterning processes that work on a longer time scale, negative regulators of ERK activation and transcriptional responses can be subordinated to the inductive signals, leading to a variety of feedback control strategies.

Rather than rely on ERK-independent control mechanisms, feedback enables a signaling system to self-regulate, fine-tuning the spatiotemporal patterns of input signals with their downstream responses. Feedback mechanisms have been extensively studied in the C. elegans VPC model, and contribute significantly to the robust patterning outcomes. One important mechanism is that high levels of ERK activity in the VPC nearest the EGF source activate the expression of Notch ligands that signal laterally to the neighboring cells (Chen and Greenwald, 2004; Hoyos et al., 2011; Sternberg, 1988). Notch signaling inhibits ERK activation in those cells by upregulating negative regulators of the ERK pathway, including an ERK phosphatase, lip-1, that counteracts the effect of ligand-dependent ERK phosphorylation (Berset et al., 2001; Yoo et al., 2004). The resulting combination of lateral signaling and negative feedback ensures that the cells adjacent to the VPC with 1° fate are less sensitive to the EGF input (Zand et al., 2011). Positive-feedback regulation also takes place in the form of ERK-induced EGFR expression in P6.p, which receives the most EGF. Moreover, endocytosis-mediated downregulation of Notch receptors in P6.p desensitizes this VPC to lateral Notch signaling, further amplifying the all-or-nothing response in terms of 1° fate to the locally secreted inductive signal from the AC (Shaye and Greenwald, 2002). In addition to these signaling processes, EGF induces migration of VPCs towards the AC to realign displaced cells. The VPC induced with 1° fate migrates up the EGF gradient towards the AC. During this migration, this VPC is exposed to increasing concentrations of EGF that further promote the 1° fate. In summary, multiple levels of control confer robustness that cannot be described by a gradient model alone (Hoyos et al., 2011). These regulatory networks also involve cell cycle control, possibly relaxing the time constraint of VPC differentiation (Euling and Ambros, 1996; Miller et al., 1993). The downstream targets of ERK signaling, LIN-1 and LIN-31, also promote expression of cki-1, which inhibits cell cycle progression during VPC differentiation (Clayton et al., 2008). The same ERK signals involved in cell fate induction are therefore simultaneously regulating fate specification and cell cycle progression.

Thus, each of the systems discussed uses several concurrent regulatory strategies to ensure timely and correct responses to inductive signaling by the ERK pathway. Although the joint effects of the open-loop and feedback mechanisms have been documented in multiple developmental contexts (O'Connor et al., 2005; Rogers and Schier, 2011; Rogers et al., 2017; Ribes and Briscoe, 2009), we are still far from assigning the differential contributions of multiple mechanisms in any given system. Understanding these differential contributions is needed for establishing the quantitative constraints that govern ERK signaling at multiple stages of development, which is in turn essential for probing the origins of the ERK-dependent developmental abnormalities – where ERK signaling has been disrupted beyond the limits that permit normal development. Such experiments will be greatly enabled by new tools that permit external manipulation of signaling pathways in vivo as well as live readouts of pathway activity – as discussed further below.

When presented with the large and steadily growing number of publications about ERK signaling, one may wonder whether we have already reached saturation and satisfied our curiosity about this highly conserved signaling system. After all, the majority of publications revolve around the same set of components and frequently pose very similar questions about the specificity, dynamics and robustness of ERK regulation and function. Our comparative review of three canonical examples of inductive ERK signaling argues that although we are still far from answering these questions, even in some of the most advanced experimental models, doing so should be an important future goal for the field.

Some of the outstanding questions may be addressed by existing approaches, such as fluorescence correlation spectroscopy for measuring concentrations and ChIP for studying interactions at the target gene loci. At the same time, the field is in crucial need of new techniques for visualizing and manipulating ERK signaling in vivo. Live reporters of ERK activation have been used quite extensively in cultured cells, but their applications in developmental contexts are only beginning to emerge (Regot et al., 2014). For example, a recent study (de la Cova et al., 2017) used signal-dependent changes in the nucleocytoplasmic ratio of an engineered ERK substrate to monitor ERK signaling in the VPC system. This study revealed unexpected oscillatory dynamics of ERK activity in the VPCs. Importantly, the higher resolution visualization of ERK dynamics showed that different levels of the EGF gradient leads to frequency-modulated, rather than amplitude-modulated, VPC specification. In theory, oscillations in ERK activity may be caused by overexpression of an exogenous ERK substrate, arising as a consequence of competition between ERK substrates and phosphatases (Liu et al., 2011). However, if these oscillations reflect the endogenous dynamics induced by locally secreted EGF, we might have to revisit the current models of cell fate specification in this extensively studied model of inductive ERK signaling.

In addition to live monitoring of ERK activity, new tools to manipulate ERK signaling inputs in vivo are much needed and are rapidly being developed. Optogenetic systems allow independent control of the spatial extent, dose and duration of signaling with high precision in an embryo (Toettcher et al., 2011, 2013). Recently, the optoSOS system, comprising components of the ERK pathway engineered to respond to light inputs, was shown to strongly activate ERK in various contexts during Drosophila embryogenesis (Johnson et al., 2017). This study found that the same, strongly activating optogenetic perturbations to the ERK pathway applied at different stages of embryogenesis produced drastically different phenotypes. Early embryogenesis is highly sensitive to levels of ERK activity, whereas later stages are more robust (Johnson et al., 2017). This sensitivity remains true for perturbations in space, as ectopically activating the ERK pathway in only a few cells in the middle of the embryo is lethal, whereas overactivation at the poles, where there is the endogenous signal, is not. Much like developmental disorders involving hyperactivation of the ERK pathway (Aoki et al., 2013; Runtuwene et al., 2011), the results of studies using this optogenetic approach demonstrate that the consequences of deregulated signaling depend on the developmental context.

Many of the unanswered questions that we highlight with examples of inductive ERK signaling must be asked for the handful of other signaling systems that together generate complexity during development (Housden and Perrimon, 2014). For example, Hedgehog (Hh) signaling in the developing Drosophila wing disc and abdominal epidermis relies on filopodial extensions called cytonemes that carry concentrated Hh signaling components across several cell diameters (Bischoff et al., 2013; González-Méndez et al., 2017; Chen et al., 2017; Kornberg, 2017). This mode of bringing ligands and their cognate receptors together lies in between the limiting regimes demonstrated by the VPC and otx patterning systems, respectively. Whereas contact-mediated ERK signaling in the early Ciona embryo depends on arrangement of cells in a tissue, the Hh signaling contacts are controlled by dynamic cytoskeletal structures. These cytoplasmic extensions can also be biased to break the symmetry of ERK activity in a field of responding cells (Peng et al., 2012). Generally, dynamic morphology of inducing cells, such as protrusive structures carrying ligand, and of responding cells, as observed in the VPC system (Grimbert et al., 2016; Huelsz-Prince and van Zon, 2017), may serve as an important control mechanism of the absolute numbers of ligand-receptor complexes formed.

Quantitative limits to the signaling parameters of developmental pathways other than the ERK pathway are also important. For example, temporal modulation (through optogenetic techniques) of the signals provided by the Nodal pathway can lead to different patterning outcomes in the early zebrafish embryo (Sako et al., 2016): different durations of Nodal activity induce qualitatively distinct responses. In fact, Nodal signaling in the zebrafish acts in concert with ERK signaling to specify endoderm and mesoderm (Poulain et al., 2006), having opposing effects on activation of the transcription factor Casanova, which is required for the induction of a stereotypic pattern of endodermal cells at the zebrafish blastoderm margin (Aoki et al., 2002). This system may be an ideal application of dual-input optogenetics to study how the combinatorial actions of signaling systems control multiple aspects of tissue patterning and morphogenesis (Martinez-Arias and Stewart, 2002).

We have presented three canonical examples of inductive ERK signaling in Ciona, Drosophila and C. elegans to demonstrate the important unanswered questions related to multiple aspects of ERK dynamics and function. There are a number of issues that need to be resolved to explain how a single pathway, like the ERK pathway, can have such diverse effects during embryogenesis. We need a quantitative understanding of signal initiation, as there may be important ligand-receptor dynamics that shape the inputs to signaling pathways. The interpretation of incoming signals ultimately determines the downstream transcriptional responses. In many cases, it is still not known how active ERK interacts with downstream targets and ultimately alters their functions. Moreover, we must now quantify the context-dependent limits on signaling parameters such as spatial extent, duration and signaling strength to understand the origins of the remarkable robustness observed in differentiating tissues. Accomplishing these tasks is crucial for laying down the foundation for a quantitative picture of developmental ERK signaling and is impossible without well-studied experimental systems, such as those discussed in this Review.

The authors thank Yogesh Goyal, Granton Jindal, Meera Sundaram, Emma Farley and Patrick Lemaire for helpful discussions.