Most cells in our body communicate during development and throughout life via Notch receptors and their ligands. Notch receptors relay information from the cell surface to the genome via a very simple mechanism, yet Notch plays multiple roles in development and disease. Recent studies suggest that this versatility in Notch function may not necessarily arise from complex and context-dependent integration of Notch signaling with other developmental signals, but instead arises, in part, from signaling dynamics. Here, we review recent findings on the core Notch signaling mechanism and discuss how spatial-temporal dynamics contribute to Notch signaling output.
Notch is a cell surface receptor that provides Metazoans with an exquisitely simple cell-cell communication device (Artavanis-Tsakonas et al., 1999; Bray, 1998; Kopan and Ilagan, 2009). Its ectodomain can read information about the state of neighboring cells, as it recognizes ligands that are expressed at their surface, whereas its intracellular domain acts, upon activation, as a transcription regulator that adjusts the cell state according to the state of neighboring cells. Signaling from the cell surface to the genome is direct, linear and devoid of signal amplification (Fig. 1). It is also transient, as each receptor is used only once and the released intracellular domain is short-lived. Despite this inherent simplicity, the study of Notch signaling is often viewed as discouragingly complex and it is not uncommon to hear scientists having to reluctantly consider the role of Notch in their favorite experimental system. Moreover, Notch signaling is known to function throughout development in a plethora of cells, tissues and organs, regulating many cellular behaviors in a context-dependent manner, ranging from differentiation, proliferation, death to migration. This pleiotropy can often mask the operation of a simple signaling mechanism. In addition, and perhaps not surprisingly, many factors have been shown to interact in a context-specific manner with Notch. Although this may suggest the existence of a very large number of context-specific regulators of Notch signaling, blurring the simplicity of this signaling device, a simpler interpretation may be that Notch signaling dynamics are highly sensitive to perturbations in a wide range of cellular processes. In this Review, we discuss recent findings that support this perspective and are likely to be of general relevance. This Review will not cover the many specific roles that Notch signaling plays in developmental processes (Artavanis-Tsakonas et al., 1999; Bray, 1998). Likewise, we will not discuss the consequences of Notch alterations in human disease, but refer the reader to other reviews on these topics (Mašek and Andersson, 2017; Siebel and Lendahl, 2017).
Receptor activation by mechanical allostery
The core mechanism of Notch receptor activation is now relatively well understood (Bray, 2006; Gordon et al., 2008; Kopan and Ilagan, 2009) (Fig. 2A). Receptors in a given cell are activated in a juxtacrine manner by cell surface ligands from neighboring cells in a process known as ‘trans-activation’ (‘trans’ refers here to the fact that signaling ligands and activated receptors are present in distinct cells). In the absence of ligand binding, receptors are in an inactive state. However, receptor trans-activation following ligand binding leads to a structural change in the Notch receptor that renders an otherwise buried cleavage site (S2) accessible to metalloproteases of the ADAM/TACE family (Gordon et al., 2007, 2015; Tiyanont et al., 2011; Weng et al., 2004). Cleavage at this S2 site then generates a membrane-tethered form of Notch that is further cleaved by the γ-secretase complex (Mumm et al., 2000; Struhl and Adachi, 2000) to release the Notch intracellular domain (NICD). NICD localizes to the nucleus, in which it associates with the sequence-specific DNA-binding protein CSL to regulate gene expression (Lecourtois and Schweisguth, 1998; Schroeter et al., 1998; Struhl and Adachi, 1998) [note that CSL is an acronym for the names of this conserved protein in humans, flies and worms: CBF1 (also known as RBPJ), Su(H) and LAG-1]. Thus, the stepwise proteolytic cleavage of Notch provides a simple, direct and irreversible mechanism to relay an extracellular signal to the genome.
A key step in pathway activation is the ligand-induced structural change in the Notch receptor to make the S2 site accessible. However, although ligand binding is necessary for Notch trans-activation, it is not sufficient, and it is thought that a mechanical force is crucial to unmask the S2 site (Fig. 2A). This idea of a pulling force exerted onto Notch receptors by signal-sending cells can be traced back to experiments using Drosophila S2 cells, designed to test whether Notch interacts at the cell surface with its ligand Delta (Fehon et al., 1990). These cell-cell adhesion assays not only showed that Notch and Delta mediate heterophilic adhesion, but further revealed that the receptor is transferred from Notch-expressing cells into Delta-expressing cells. This transfer is dependent on Delta endocytosis, indicating that Notch is trans-endocytosed. This led to the proposal that ligand endocytosis drives receptor dissociation and activation via mechanical pulling (Parks et al., 2000). Soon afterwards, structural, modeling and human genetics studies revealed that the S2 cleavage site is buried within the negative regulatory region (NRR) of Notch, and that receptors with mutations that are predicted to destabilize the NRR, hence exposing the S2 site at the protein surface, are hyperactive in a ligand-independent manner (Gordon et al., 2007, 2015; Tiyanont et al., 2011; Weng et al., 2004). In addition, biophysical experiments demonstrated that applying pulling forces in the range of those produced by endocytosis (∼5-10pN) is sufficient for receptor activation, and that ligand-mediated allostery and oligomerization are not essential for force-dependent shedding (Gordon et al., 2015; Luca et al., 2017; Meloty-Kapella et al., 2012; Seo et al., 2016; Wang and Ha, 2013). More recently, the lifetime of Notch-ligand complexes was found to increase as a function of applied force, reaching a maximum at ∼10 pN for jagged canonical Notch ligand 1 (Jag1). This ‘catch-bond’ property, whereby dissociation lifetime increases when force is applied to the bond, suggests that complex stability increases as the ligand becomes endocytosed (Luca et al., 2017). At the structural level, the pulling force may first induce a rotation within the Notch-binding region of the ligand, which adopts an elongated conformation that increases the ligand-receptor contact interface. The resulting increased binding may then stabilize the complex under tension and transmit force to the NRR (Luca et al., 2017). Together, these data provide strong evidence that Notch activation is a mechanosensitive process that requires a pulling force to expose an otherwise buried S2 cleavage site, and that this force might arise from the endocytosis of bound ligands (Lovendahl et al., 2018) (Fig. 2A).
Testing the mechanical pulling model in vivo has remained challenging. However, an elegant genetic mosaic strategy was recently designed to probe force generation by ligand-dependent endocytosis using chimeric ligand-receptor binding pairs in Drosophila (Langridge and Struhl, 2017). In this assay, signaling between heterologous ligand-receptor pairs was monitored along artificial clone borders. One such chimeric ligand-receptor pair involved the follicle-stimulating hormone (FSH; fused to Delta) and the extracellular domain of its receptor (FSHR, fused to Notch) and with the NRR of Notch replaced with a force sensor domain derived from von Willibrand factor (vWF) (Fig. 2B). The unfolding of this domain, which is termed the A2 domain, requires a force greater than 8 pN. When tested in vivo, the chimeric FSHR-A2-Notch receptor was not activated by Delta-FSH. This suggests that the A2 domain does not allow for ligand-dependent cleavage of the receptor, presumably because the force produced by the endocytosis of Delta-FSH is below 8pN. In contrast, chimeric receptors with mutant versions of the A2 domain that display lower thresholds for unfolding appeared to be processed in a ligand-dependent manner (Fig. 2B). This observation strongly suggests that the endocytosis of Delta-FSH provides a pulling force sufficient to unfold these mutant versions of the A2 domain. Although activation of the FSHR-A2-Notch receptor must involve cleavage, it is not clear, in the absence of the NRR domain replaced by the A2 domain, whether cleavage also involves Kuzbanian, the fly ADAM/TACE family protease. Nevertheless, these results provide strong in vivo evidence for a mechanical pulling force that is generated by the endocytosis of Delta in the range required for Notch activation (Langridge and Struhl, 2017).
In sum, studies over the past thirty years have revealed that Notch mediates cell-cell communication between direct neighbors via a very simple and direct mechanism from the cell surface to the genome. This simple mechanism can only operate when core components (e.g. ADAM/TACE metalloproteases and the γ-secretase) are active. In this regard, we note that the context-specific regulation of these core components has been reported (Dornier et al., 2012; Seegar et al., 2017; Upadhyay et al., 2013). Furthermore, this simple mechanism can be modulated at various levels, starting with receptor-ligand interactions that are well known to be modulated by sugar modifications (Bruckner et al., 2000; Moloney et al., 2000; Okajima and Irvine, 2002; recent review by Harvey and Haltiwanger, 2018). Indeed, recent work has revealed that fucose modification of EGF8 and EGF12 by Lfng enhances Notch1 binding to and activation by the Notch ligand delta like canonical Notch ligand 1 (Dll1) but not the Notch ligand Jag1 (Kakuda and Haltiwanger, 2017; Luca et al., 2017), suggesting new ways to modulate Notch activation (Schneider et al., 2018). Along the same line, the recent identification of a lipid-binding domain in the C2-DSL domain of Notch ligands, which is involved in Notch binding, suggests that lipids may stabilize ligand-receptor complexes and modulate signaling (Suckling et al., 2017). Finally, an alternative ligand-independent mechanism of receptor activation has also been reported (review by Schnute et al., 2018). It involves the trafficking of full-length Notch to late endosomes and its processing at the limiting membrane of maturing endosomes upon fusion with lysosomes (Schneider et al., 2013). In this context, the shedding of extracellular Notch is thought to be force- and ADAM/TACE-independent. However, this activation mechanism is mostly observed when Notch is abnormally stabilized at the limiting membranes of maturing endosomes upon genetic perturbations, and its contribution to Notch-mediated cell-cell communication in living organisms appears to be limited.
Cis-inhibition: a simple regulation of general relevance
Notch receptors can also interact with their ligands in cis, i.e. within the same cell. Cis- and trans-ligands are thought to compete for binding to the same region of Notch, but cis-ligands likely fail to activate Notch because no cellular force can pull onto the ligand/receptor bridge to change the NRR conformation and expose the S2 site. Thus, cis-ligands inhibit Notch through a non-catalytic process that results from simple sequestration of Notch receptors (review by del Álamo et al., 2011). This view is supported by the recent analysis of orthogonal ligand/receptor pairs in which the extracellular domains of Delta/Notch were replaced with heterologous ectodomains: in all cases studied, cis-inhibition was observed, arguing that it requires no special property other than cis-binding (Langridge and Struhl, 2017). Thus, cis-inhibition of Notch appears to be a simple, general and intrinsic property of Notch signaling.
Cis-inhibition may confer specific properties to Notch-based decisions. First, cis-ligands can act in vivo as a buffering mechanism that limits ligand-independent activation of Notch. Second, cis-ligands were proposed to confer ultra-sensitivity to the switch between two mutually exclusive signaling states in a cell, the ON state (low cis-ligand levels, hence trans-activation) and the OFF state (high cis-ligand levels). This switch allows a sharp transition between the two signaling states (Sprinzak et al., 2010), and modeling studies have indicated that this property may be relevant for patterning dynamics (Sprinzak et al., 2011) and signaling precision (Barad et al., 2010). Finally, cis-inhibition could contribute to reducing the strength of Notch signaling by competing with trans-ligands for interaction with Notch. Although this prediction appears to be rather obvious, testing it experimentally in vivo is challenging, as cis- and trans-binding properties cannot be easily uncoupled (but see Miller et al., 2009 for a nice exception). Thus, despite its likely broad relevance, this interesting regulatory mechanism is most often overlooked. In this regard, developing tools to differentially detect Notch in an inactive/auto-inhibited, cis-inhibited or trans-interacting state would greatly advance the analysis of this general regulatory mechanism.
Cell-cell contact and signaling sites
As mentioned above, ligand endocytosis is generally required for receptor activation; it is thought to provide the necessary force to unfold the NRR in vivo. This in turn implies that S2 cleavage usually takes place at the cell surface. Thus, ligand-receptor trans-interaction, ligand endocytosis and S2 cleavage all likely co-occur at the cell surface. Various cell-cell contact sites at which these processes might occur have been proposed. These range from apical cell junctions in epithelia (Hatakeyama et al., 2014; Sasaki et al., 2007) to apical cell membranes (Lopez-Schier and Johnston, 2001), basal protrusions (Cohen et al., 2010) and/or long and specialized cellular extensions (Eom et al., 2015). However, in none of these cases is the exact location of Notch receptor activation defined. Determining where receptor activation takes place at the subcellular level is relevant for three main reasons. First, co-localization of ligands and receptors does not predict signaling, as cis-interactions might dominate. Second, the size and shape of these signaling cell-cell contacts might regulate signaling strength (Shaya et al., 2017). Third, the spatial range of signaling may be extended through long cellular protrusions that reach beyond immediate neighbors, as has been proposed for long cellular extensions in zebrafish (Eom et al., 2015) and basal filopodia in Drosophila (Cohen et al., 2010).
Recently, a new approach has been developed and applied to determine the membrane domain at which Notch is activated (Trylinski et al., 2017). This method relies on receptors that are intracellularly tagged by fluorescent proteins such that NICD can be detected in the nucleus. To determine where NICD comes from, the fluorescence properties of Notch at a particular cell surface site can be specifically modified, e.g. by photo-bleaching or photo-conversion, allowing receptor activation to be monitored by tracking the nuclear accumulation of modified NICD fluorescence over time. This approach was applied in the context of a binary fate decision that immediately follows asymmetric division of neural precursor cells in Drosophila. It revealed that only a specific subset of Notch receptors, located basal to the midbody, contribute to NICD release in this context (Trylinski et al., 2017). Restricting signaling to this new cell-cell interface may be key for intra-lineage fate decisions. Given the importance of membrane trafficking in Notch signaling, it is tempting to speculate that the regulated targeting of Notch to specific cell-cell contacts at which signaling takes place may, in some cases, underlie the functional requirement for membrane trafficking in Notch signaling.
Of note, although this method can identify signaling pools of Notch at the membrane, it does not directly identify where S2 and S3 cleavage take place. Indeed, S3 cleavage may occur at both the plasma membrane and at the limiting membrane of endomembranes, following S2 cleavage and internalization of the membrane-tethered form of intracellular Notch (Vaccari et al., 2008; Narui and Salaita, 2013; Sorensen and Conner, 2010). This trafficking step introduces a time delay and allows for further regulation. In addition, the cellular location of S3 cleavage appears to also have an impact on the precise peptide bond that is cleaved by the γ-secretase, resulting in NICD fragments with distinct amino-termini. Indeed, cleavage at the cell surface was reported to produce a moderately stable NICD with an N-terminal valine, whereas cleavage at the limiting membrane of endosomes produces short-lived NICD (Tagami et al., 2008) owing to the N-end rule protein degradation pathway (Varshavsky, 2011). Signaling strength might also be downmodulated by targeting S2-cleaved receptors for degradation before S3 cleavage. Therefore, further analysis of the state of Notch at signaling sites is required and would benefit from new tools that could be used to monitor and locally manipulate force-dependent S2 cleavage and further processing by γ-secretase.
Notch signaling and tissue mechanics
Recent studies have highlighted the interplay between cell-cell signaling and tissue mechanics in pattern formation and organogenesis (Shyer et al., 2017). As a ligand-dependent mechanosensitive receptor, Notch integrates mechanical and chemical/molecular cues. However, an open question is whether Notch integrates molecular cues (Notch ligands) with force acting at the supracellular scale, and not merely at the subcellular scale, as is the case during endocytosis. For example, could it be that shear stress along cell-cell contacts contributes to ligand-dependent receptor activation, possibly by lowering the force threshold for mechanical allostery? The observation that mechanical stress generated by the high-mobility of Notch-expressing T-cells on a plastic surface that is covered with immobile ligands leads to receptor activation is, at least in this artificial context, consistent with this possibility (Varnum-Finney et al., 2000). Similarly, ligands at the surface of a supported lipid membrane surface are more active when their lateral mobility is restricted (Narui and Salaita, 2013). Moreover, force-induced destabilization of the NRR (Notch activation) may not necessarily require the force to be induced by ligand binding. For example, EDTA treatment appears to destabilize the NRR of Notch in Drosophila S2 cells, which results in ligand-independent S2 cleavage (Rand et al., 2000). In addition, physiological ligand-independent activation of Notch through S2 cleavage has recently been reported in the Drosophila follicular epithelium (Palmer et al., 2014) and in mammalian T-cells (Steinbuck et al., 2018). Of note, these signaling events differ from those that result from the mis-trafficking of Notch towards late endosomal compartments, in which it is proteolytically cleaved by a mechanism that does not involve S2 cleavage and may not involve force (Schneider et al., 2013). How the S2 site becomes unmasked in a ligand-independent manner in fly follicular cells and in mouse T-cells is still unknown and it will be interesting to decipher whether and how mechanics also contribute to the destabilization of the NRR in these contexts.
Interestingly, Notch signaling is required in adult blood vessels to maintain endothelial cell barrier integrity in response to flow/hemodynamic fluid stress (Lagendijk et al., 2018), and it has also been shown that shear stress is required for the activation of Notch1 in endothelial cells in vitro, raising the possibility that blood flow contributes directly to receptor activation (Fang et al., 2017; Loerakker et al., 2018). Alternatively, it is possible that shear stress indirectly regulates Notch signaling, for example via the endocytosis of delta like canonical Notch ligand 4 (Dll4), the ligand that activates Notch1 in endothelial cells through mechanical effects on the endothelial cell cytoskeleton. Whether forces exerted by fluid shear stress on Notch1 receptors at endothelial cell membranes contribute to NRR unfolding remain to be tested. Whatever the role of shear stress on Notch1 signaling, recent in vitro work points to an additional mechanism through which Notch might help strengthen the endothelial cell barrier in response to fluid shear stress, without involving its transcriptional activity (Polacheck et al., 2017). This mechanism is mediated by a fragment of the Notch1 receptor that includes its transmembrane domain (TMD) and that is produced following Dll4-dependent S2 and S3 cleavages and release of the extracellular domain (ECD) and NICD fragments. The predicted TMD fragment is proposed to promote the assembly of a complex in the plasma membrane that consists of VE-cadherin (cadherin 5), the phosphatase Lar and the Rac1 guanidine exchange factor trio; this complex triggers an increase in Rac1 activity to stabilize adherens junctions and promote barrier integrity (Polacheck et al., 2017). However, the predicted TMD fragment that is proposed to interact with VE-cadherin was shown in previous work to produce, following further intramembrane cleavage, secreted extracellular Notch Aβ like-peptides (Nβ) (Okochi et al., 2006), which implies that assembly of this complex must be remarkably efficient to compete with release of the Nβ peptide. In parallel to this proposed non-transcriptional mechanism, flow and shear stress in the same microvasculature model result in NICD-mediated induction of target genes such as Hey1 and Hes1. It will therefore be important to determine the relative contribution of these two mechanisms to the response of endothelial cells to flow-mediated shear stress. Thus, whereas Notch1 senses force and regulates tissue integrity in endothelia (and possibly other tissues), the extent to which chemical/molecular and mechanical cues are integrated at the level of the receptor itself or up- or downstream of receptor activation awaits further analysis.
Dynamic encoding at the cell surface
Notch receptors have more than one ligand. For example, two ligands (Delta and Serrate) are found in Drosophila and five (DLL1, DLL3, DLL4, JAG1 and JAG2) are found in humans. The existence of multiple ligands accounts, in part, for the complex regulation of Notch activity in time and space, and may also contribute to modulating signal levels and duration. For example, signaling by a strong trans-ligand can be reduced upon expression of a second weaker trans-ligand, as the latter competes for binding to Notch (Benedito et al., 2009; Petrovic et al., 2014). Perhaps less intuitively, different ligands activating the same receptor may trigger different responses at the genome level. For example, Notch1 activation by Dll1 in the myotome of chick embryos leads to strong Hes1 expression, whereas activation by Dll4 results in Hey1 expression, with opposite effects on myogenesis (Nandagopal et al., 2018). As these different ligands have a single and shared primary output, i.e. NICD, it is not clear how the information associated with ligand identity is transduced by NICD. A recent study (Nandagopal et al., 2018) provides a solution to this problem and suggests that information is encoded through distinct signaling dynamics (Nandagopal et al., 2018; Purvis and Lahav, 2013). A first hint at this solution comes from the observation that Hes1 and Hey1 genes respond differently to Notch dynamics in cultured cells that have been engineered to conditionally express a constitutively active form of Notch (Nandagopal et al., 2018). Specifically, a short pulse of Notch activation is sufficient to induce a transient peak of Hes1 expression, the amplitude of which does not significantly vary with the duration of the pulse. However, sustained Notch activation leads not only to Hes1 expression but also to Hey1 expression. Activation of different genes by Dll1 and Dll4 might thus derive from distinct dynamics of NICD production caused by each ligand. To test this hypothesis, the NICD production rate was calculated at the single cell level in an ex vivo assay, by measuring the output of a GAL4-driven promoter coupled to a stable fluorescent reporter, in response to a ligand-activated chimeric Notch1-Gal4 receptor (Nandagopal et al., 2018). By design, this system preserves Notch activation dynamics, allows signal amplification to facilitate detection and prevents cross-talk with the endogenous signal transduction machinery. ‘Receiver’ cells were cultured together with ‘sender’ cells, and the results revealed that when these cells express Dll4, the activity of the NICD-responsive promoter increases steadily to reach a plateau, the level of which depends on Dll4 levels (Fig. 3A). Thus, Notch is activated in a sustained manner by Dll4, and signaling strength depends on Dll4 levels. By contrast, discrete and transient pulses of promoter activity are observed when ‘receiver’ cells are in contact with Dll1-expressing cells (Fig. 3A). These pulses were estimated to last ∼1 h, and their frequency, but not amplitude, depends on the levels of Dll1. Thus, in this ex vivo assay, Dll1 levels regulate pulse frequency whereas Dll4 levels modulate the amplitude of the response (Nandagopal et al., 2018). Interestingly, this difference in Notch dynamics elicited by Dll1 or Dll4 maps to their intracellular domains (ICD): whereas a chimeric ligand containing Dll1 ICD linked to the ECD of Dll4 [Dll4(ECD)-Dll1(ICD)] produces a pulsatile Notch activity, the reverse Dll1(ECD)-Dll4(ICD) chimera generates a sustained Notch response, as with Dll4 itself (Nandagopal et al., 2018). Furthermore, how each ICD confers the observed ligand-specific dynamic activation may be related to its ability to regulate trans-endocytosis of Notch: Dll1 appears to trigger trans-endocytosis when Notch receptors are clustered, possibly resulting in pulses of NICD production, whereas Dll4 triggers signaling from dispersed Notch receptors, possibly resulting in a continuous, steady NICD production rate (Nandagopal et al., 2018). Whether the ICDs of Dll1 and Dll4 interact with different molecules that contribute to these distinct capacities is a possibility that remains to be explored. We note, however, that although Dll1 function cannot be replaced by Dll4 in mesodermal tissues of the mouse embryo, the ICD of Dll1 can be functionally substituted by the ICD of Dll4 in the same tissues (Tveriakhina et al., 2018). This suggests that a possible alteration in Notch dynamics by the Dll4 ICD (which causes a sustained NICD production) might be buffered in vivo by other components of the Notch response. It would therefore be important to monitor signaling dynamics in vivo and correlate them with the phenotypic consequences of altering ligand structures. In addition, as many tissues express more than one receptor as well as more than one ligand, it will be important to study the response dynamics of different Notch receptors to their various ligands, and to investigate how dynamic encoding is employed in a combinatorial manner in vivo.
Decoding Notch dynamics in the nucleus
The findings above identify a novel mechanism – based on temporal dynamics of receptor activation at the cell surface leading to distinct rates of NICD production – through which Notch transduces information from ligand-expressing cells to the nucleus. However, how distinct rates of NICD production are mechanistically decoded at the genome level to direct expression of distinct sets of Notch targets remains to be studied in vivo. In the case of Hes1 and Hey1, their different expression profiles in response to Notch activation by different ligands may, in principle, involve a downstream circuitry of auto- and cross-inhibition among these genes (Fig. 3B,C), similar to that observed for other genes of this family (Fior and Henrique, 2005; Schröter et al., 2012; Trofka et al., 2012). We speculate that, although both genes are initially upregulated following Notch activation, Hes1 accumulates faster than Hey1, for example because of optimal CSL/NICD-binding sites in the Hes1 promoter, repressing Hey1 gene expression. Then, as Hes1 levels decrease because of protein instability and auto-repression (Hirata et al., 2002), two temporal patterns of Hey1 expression could emerge. In the case of Notch activation by Dll1, the proposed bursts of NICD and Hes1 production are coupled such that Hes1 keeps Hey1 repressed whenever NICD is induced. However, when Notch is activated by Dll4, sustained NICD production allows for Hey1 expression when Hes1 levels drop, resulting in a delayed wave of Hey1 gene activation (Fig. 3C). It will be of interest to study the extent to which a regulatory circuit such as this underlies the contrasting response of Notch targets to different ligands.
In addition to the information that is encoded in Notch ligands, each responding cell is predicted to display a set of accessible Notch-responsive elements (NREs) that allow it to mount a specific transcriptional program following Notch activation. NRE accessibility likely depends on the state of the cell, which itself is influenced by the local environment and developmental history of the cell. Several studies have attempted to define the chromatin environment that is associated with open or closed NREs in various cell types by mapping chromatin conformation and modifications in relation to the presence of CSL and NICD. For example, in Drosophila cell lines, as well as in mammalian C2C12 cells and T-lymphoblastic leukemia (T-LL) cell lines, CSL-bound loci are preferentially enriched in a particular array of histone modifications (H3K4me1, H3K27ac and H3K56ac), but this combination alone is not enough to predict which of the CSL-bound sites are capable of recruiting NICD (Skalska et al., 2015; Wang et al., 2011). Other factors are therefore required to establish the appropriate regulatory environment that allows specific NREs to become competent for CSL/NICD binding. By searching for transcription factor (TF)-binding sites that appear to be frequently associated with NREs, various candidate TFs that might play crucial roles in priming NREs for enhancer activity, in a lineage-specific manner, have been identified. A good example is the Drosophila Runt-domain TF Lozenge, which is required to make NREs competent for Notch activation in fly hemocytes (Terriente-Felix et al., 2013). Binding sites for Runx1, the mammalian homologue of Lozenge, were also found to be enriched close to CSL-Notch1 binding sites, both in T-LL cells and in the mouse kidney cell line mK4 (Wang et al., 2014; Hass et al., 2015). It was further shown that RUNX1 cooperates with NOTCH1 in activating the expression of the target gene IL7R through an enhancer that contains adjacent CSL and RUNX1 binding sites (Wang et al., 2014). These examples are likely just the tip of the iceberg, and there may be other TFs that cooperate with CSL-NICD in a lineage-specific manner to regulate which genes are competent to respond to Notch activation. Further work is therefore needed to identify these TFs and understand how their interaction with CSL-NICD regulates target gene accessibility and responsiveness.
Another regulatory layer involves the different sensitivities to NICD levels that are displayed by each individual NRE, which vary according to the number, type and affinity of CSL-binding sites that each NRE contains. These sites can occur as single sites or paired sites, known as Su(H)-Paired Sites (SPS), which consist of two high-affinity binding sites that are separated by 15-17 base pairs and are found in a head-to-head orientation (Bailey and Posakony, 1995; Ong et al., 2006). Diverse combinations of CSL-binding sites can be found in individual NREs and these, together with neighboring TF-binding sites, are likely to define unique regulatory architectures for each Notch target gene that allow it to decode nuclear NICD levels and modulate its transcriptional activity accordingly (Severson et al., 2017). Assuming rapid and commensurate kinetics for both receptor activation/NICD production and NICD degradation, varying nuclear NICD levels may reflect different rates of NICD production, which, as described above, can be elicited by distinct ligands. Thus, the challenge is to understand how the dynamic interaction between CSL, NICD and associated proteins can mechanistically interpret the information that is received from neighboring cells. In the current view, when signaling is inoperative and NICD is not produced, CSL dynamically binds its target DNA sites and recruits a repressive complex that maintains Notch targets in a silenced state. When Notch is active and NICD levels rise in the nucleus, a switch from repression to activation is proposed to occur at this subset of ‘open’ Notch targets. Two possible mechanisms may underlie this switch. First, NICD may interact with DNA-bound CSL and displace the corepressors while recruiting coactivators. This view is, however, difficult to reconcile with the finding that corepressors and NICD have comparable high affinities for CSL (Oswald and Kovall, 2018), making a competitive displacement mechanism unlikely to generate robust and sensitive responses. Alternatively, NICD may assemble an activation complex outside of DNA and this complex may dynamically compete with repression complexes for target sites. To test these two possibilities, a recent fluorescence recovery after photobleaching-based study examined the dynamic behavior of CSL in vivo, both in repressive conditions (Notch-OFF) and in the presence of NICD (Notch-ON), using Drosophila salivary glands as a cellular system (Gomez-Lamarca et al., 2018). The analysis of CSL in Notch-OFF conditions revealed that it exhibits fast dynamics in the nucleus: only about one third of CSL molecules are predicted to be bound at each time to DNA, showing a very short residence time of 0.5-2 s. Upon constitutive expression of NICD (Notch-ON), CSL retains its fast dynamics globally, but is seen to accumulate at a specific location on polytene chromosomes that corresponds to the E(spl)-C locus, which contains a cluster of Notch target genes with multiple high-affinity CSL-binding sites. This accumulation reflects an increased residence time (of up to 10-15 s) of CSL at E(spl)-C, and is indeed dependent on the interaction of CSL with NICD; notably, a mutant CSL that is unable to bind NICD exhibits a shorter residence time and shows no visible accumulation at the E(spl)-C locus (Gomez-Lamarca et al., 2018). It was further shown that the CSL/NICD complex establishes a more open chromatin environment at the E(spl)-C locus, as measured by the ATAC method as well as by the increased detection of H3K27Ac and H3K4me1, which are two histone modifications normally associated with open, active enhancers. The increase in H3K27Ac is likely mediated by recruitment of the histone acetyltransferase CBP/p300 to the CSL/NICD complex, whereas the H3K4 mono-methylase KMTD2 (Trithorax-related, Trr, in Drosophila) was shown to be responsible for the observed increase in chromatin accessibility in Notch-ON cells (Gomez-Lamarca et al., 2018). Strikingly, in Notch-ON cells depleted for Trr, no CSL accumulation at the E(spl)-C locus is observed, and the locus remains silent. This indicates that Trr is required to modify chromatin at Notch target genes and promote transcriptional activation by NICD. As KMTD2 interacts with a CSL corepressor in mammals (Oswald et al., 2016), it appears that the CSL repressor complex might prime Notch targets for NICD activation. Conversely, the increased chromatin accessibility promoted by the CSL-NICD complex appears to also facilitate access of the CSL corepressor complex to Notch targets. Such an ‘assisted’ loading mechanism might therefore allow NICD to indirectly stimulate chromatin opening and boost transcriptional activity (Gomez-Lamarca et al., 2018).
In summary, the observed dynamic turnover of CSL and NICD at target sites suggests that CSL-NICD occupancy levels should be highly sensitive to NICD concentration in the nucleus. Indeed, short residence times imply that high local concentrations of NICD and CSL at target genes have to be attained in order to extend occupancy for enough time to recruit other components and assemble a functional, transcriptionally active complex. The dynamic CSL hub thus provides a system that may be tunable and responsive to fluctuations in NICD concentrations. This might account for the observed heterogeneity in transcriptional activation of target genes in Notch-ON cells (Lee et al., 2016), in which some of these genes are reported to be asynchronously activated within the same cell, possibly reflecting the dynamic turnover of CSL complexes and the inherent fluctuations in NICD concentration at these targets.
Notch in context: distinct outcomes within a single lineage
Given the relative simplicity of the core Notch signaling mechanism, it is remarkable how this cell-cell communication pathway participates in so many diverse cell fate decisions. A simple and striking example of this versatility is provided by the intestinal stem cell (ISC) lineage in the Drosophila midgut epithelium (Micchelli and Perrimon, 2006; Ohlstein and Spradling, 2006). In this lineage, multipotent ISCs divide to self-renew and to produce a transient progenitor that either differentiates into an absorptive enterocyte (EC) or divides once to produce two secretory enteroendocrine (EE) cells (Chen et al., 2018; Guo and Ohlstein, 2015) (Fig. 4A). Remarkably, Notch is required for both ISC-EC and ISC-EE fate decisions (Micchelli and Perrimon, 2006; Ohlstein and Spradling, 2006), but signaling acts in opposite directions in the two decisions, relative to in the self-renewed ISC (Guo and Ohlstein, 2015). In ISC divisions that produce an EC, Delta signals from the newly forming ISC and activates Notch in the sister cell that will become an EC. On the contrary, when an ISC divides to produce an EE progenitor, Delta signals from this progenitor to activate Notch in the sister cell that will become a new ISC (Fig. 4B). Thus, the same Delta signal is employed in different settings within the same lineage to regulate either EC differentiation or ISC self-renewal.
How this occurs is unclear. One possibility is that these two lineage decisions may involve different strengths of Notch activity, and the Notch antagonist Numb may be one of the factors required to modulate the Notch response in the ISC-EE decision (Sallé et al., 2017). Context is therefore key to explain how Notch regulates these two seemingly opposite outcomes. In addition, recent work suggests that ISCs fluctuate between two states, associated with distinct levels of the pro-EE transcription factor Scute (Chen et al., 2018). When an ISC reaches high Scute levels, it starts expressing Prospero (Pros), a downstream target of Scute that is required to implement an EE fate. At division, Pros is proposed to be unequally segregated into the daughter cell that will become an EE progenitor (Guo and Ohlstein, 2015). In this context, Delta is expressed in the EE progenitor and the signal-receiving cell is the self-renewed ISC, in which Notch activity is required to switch off the pro-EE factor Scute and thereby prevent EE differentiation. These low Scute and Pros-negative ISCs may then undergo several rounds of divisions, each producing a self-renewed ISC and an EC progenitor (Fig. 4B, dashed arrow). In this context, Notch signaling also becomes directional, with Delta-expressing ISCs activating Notch in EC progenitors (Micchelli and Perrimon, 2006; Ohlstein and Spradling, 2006). Here, Notch activity may function both to inhibit the pro-EE determinant Scute and to antagonize the transcription factor Daughterless, which has previously been shown to be required to maintain ISC identity (Bardin et al., 2010). We therefore suggest that a common and key function of Delta/Notch signaling in this lineage is to restrain the dynamic expression of Scute in ISCs, likely through repressor E(spl) proteins. These Notch targets were shown to also be induced by Scute in ISCs, and it has been proposed that reciprocal regulation between Scute and E(spl) may keep Scute expression at low levels in most ISCs (Chen et al., 2018). However, when Scute autoregulation becomes active in a given ISC, this causes a switch to a high-Scute state, leading to the production of EEs, before reverting back to a low Scute state upon Notch activation. Thus, the interplay of Notch with Scute dynamics in the ISC lineage may provide a simple mechanism to understand how two seemingly distinct cell fate decisions can be governed by a single molecular process. As other stem cell lineages may follow the same lineage logic, e.g. the airway epithelium (Pardo-Saganta et al., 2015) in which Notch regulates a Scute-related factor (Borges et al., 1997; Mori et al., 2015; Rock et al., 2011), one may wonder whether fate decisions seen as distinct might be explained by a unified mechanism as Notch dynamics become unraveled.
Self-organized Notch dynamics in patterning
A classic function of Notch is to single-out individual cells from groups of cells with similar developmental potential via a process termed ‘lateral inhibition’. In this process, the selected cells send an inhibitory signal to prevent other cells in the group from adopting the same fate. Lateral inhibition can operate in time to limit the number of cells adopting a given fate in response to a general inductive signal, thereby maintaining a pool of uncommitted progenitors (Chitnis et al., 1995). It can also operate in space to mediate patterning via the differentiation of regularly spaced cells (Simpson, 1990). For example, during peripheral neurogenesis in flies, Notch signaling is known to play a role in selecting sensory organ precursor cells (SOPs) from groups of equipotent cells that are endowed to become neural via the expression of proneural factors (Simpson, 1990). This occurs as part of a two-step process (Ghysen and Dambly-Chaudiere, 1989). First, in response to developmental signals that convey positional and temporal information, proneural factors are expressed in a restricted manner within the neuroepithelium. Second, Notch-mediated inhibitory cell-cell interactions single-out a limited number of regularly spaced neural progenitors from these proneural clusters. This two-step model also applies to the formation of the large mechanosensory organs (macrochaetae) that are found at stereotyped positions on the Drosophila thorax (Gómez-Skarmeta et al., 2003).
In contrast with this two-step model, a recent study on the patterning of smaller microchaetae in Drosophila suggests that Notch acts in a continuous manner to both establish proneural domains and select neural progenitors in the dorsal-central region of the pupal notum (Corson et al., 2017). In this context, Notch appears to regulate what is usually seen as two distinct steps in sensory organ formation (i.e. establishment of proneural clusters and selection of neural progenitors with these clusters) via a single developmental operation through its interplay with proneural factor dynamics. Briefly, a stable activity pattern of Notch at the onset of patterning provides a negative template to define where proneural genes can be switched on at metamorphosis, such that the two stripes of cells with initially high Notch activity become flanked by three proneural stripes. Then, as the ability of cells to produce and respond to a Notch signal is modulated by proneural factors, the pattern of Notch activity dynamically evolves. These dynamic changes define two new regions of cells with low Notch activity, intercalated between existing proneural stripes, to produce two additional proneural stripes. This self-organized process therefore creates a pattern of five proneural stripes. Each of these stripes then progressively resolves to produce singled-out SOPs, eventually giving rise to a final pattern of five rows of sensory bristles (Corson et al., 2017). The geometry and dynamics of this patterning process can be faithfully produced in a simple mathematical model based on bistability that involves a single variable describing the state of cell, when given appropriate initial and boundary conditions (Corson et al., 2017). Remarkably, both stripe patterning and the resolution of proneural stripes into regularly spaced neural progenitors are the sequential outputs of a single logical operation with a unique set of parameters (Fig. 5). Thus, modeling suggests that what is often viewed as a two-step process, with first the specification of proneural clusters and then the singling out of neural progenitors, may actually involve a single continuous self-organized process. This therefore strongly suggests that signaling by Notch can mediate both proneural patterning and SOP selection via a single logical operation deployed in time and space to robustly produce a complex stereotyped pattern (Corson et al., 2017). Whether Notch-mediated self-organization similarly operates at the tissue scale to regulate the patterning of other differentiated cells is not yet known. We note, however, that the constraints and assumptions of the mathematical model are generic, suggesting that this model may be of general relevance. Again, this illustrates how versatility in Notch function in vivo may not necessarily arise from the complex and context-dependent integration of Notch signaling with other developmental signals but may, in part, arise from self-organized dynamics.
Conclusions and future directions
Notch receptors relay information from the cell surface to the genome in a simple, linear and stoichiometric manner. At the core of the activation mechanism is the ligand- and force-dependent change in the structure of the NRR, which allows for the processing of the receptor, and the subsequent release of the intracellular domain from the membrane. Although we have witnessed impressive recent progress in our understanding of Notch signaling mechanisms, key challenges remain. For example, the 3D architecture of the complete extracellular ligand-receptor complex remains to be determined. Recent advances in determining the structure of large protein complexes at the cell surface, using Cryo-Electron Transmission Microscopy, may help to address this challenge. In addition, visualizing receptor activation in real-time, possibly in association with ligand endocytosis, and determining the organization of signaling trans-complexes, discriminating them from inactive cis-complexes, are important challenges that remain to be addressed. Interestingly, the central signal processing region of the receptor is flanked by an ECD with signal input activity and an ICD that has output activity, and this modularity in design makes it possible to swap input/output functions (Gordon et al., 2015; Lecourtois and Schweisguth, 1998; Struhl and Adachi, 1998; Vooijs et al., 2007). This approach was recently used to build synthetic cell-cell communication circuits using synthetic Notch receptors (synNotch) in which both the input and output domains were replaced with heterologous protein domains, keeping only the mechanosensitive and cleavable part of Notch (Morsut et al., 2016). The resulting novel cell-cell signaling devices can thus function orthogonally to Notch as well as to one another. By combining different input/output functions, user-defined responses could be obtained to produce diverse multicellular patterns (Morsut et al., 2016). As noted by the authors, however, these synNotch receptors appear to function in the absence of any endocytic signal, suggesting that endocytosis is unlikely to provide the force to render the S2 site accessible for cleavage (Morsut et al., 2016). Thus, the ligand-dependent mechanism responsible for displacing the NRR of synNotch is not clear. It is possible that shear stress contributes to synNotch activation, in which case these synNotch receptors might become very useful tools to report on shear stress and tissue mechanics in living organisms. More generally, this approach opens up a wide range of possible applications for multicellular synthetic biology (Roybal et al., 2016a,b; Toda et al., 2018). Along a similar line, light has emerged as a powerful tool for manipulating the activity and/or localization of target proteins using light-sensitive domains. The rational design of light-regulated ligands and/or receptors will be of considerable interest to control the timing, pattern and strength of signaling. Combined with recent advances in CRIPSR-based genome engineering, it should be possible to move beyond the tagging of endogenous molecules (Corson et al., 2017; Trylinski et al., 2017) and towards manipulating the dynamics of endogenous Notch pathway components with light. The future should be bright!
We thank A. Bardin, K. Storey, M. Trylinski and our anonymous referees for useful comments and suggestions. We thank F. Corson for Fig. 5.
Our research received specific grants from the Agence Nationale de la Recherche (ANR16-CE13-0003-02 and ANR-10-LABX-0073) and Fondation pour la Recherche Médicale (DEQ201180339219) to F.S.
The authors declare no competing or financial interests.