|Home | About | Journals | Submit | Contact Us | Français|
Cell signaling is dominated by analyzing positive responses to stimuli. Signal activation is balanced by negative regulators that are generally considered to terminate signaling. Rather than exerting only negative effects, however, many such regulators play important roles in enhancing cell-signaling control. Considering responses downstream of selected cell-surface receptors, we discuss how receptor internalization affects signaling specificity and how rapid kinase/phosphatase and GTP/GDP cycles increase responsiveness and allow kinetic proofreading in receptor signaling. We highlight the blurring of distinctions between positive and negative signals, recasting signal termination as the response to a switch-like transition into a new cellular state.
The path of discovery in cell signaling has necessarily focused largely on the activation—or “switching on”—of signaling pathways by activating ligands. Genetic and biochemical approaches have encouraged the conceptual division of large and interconnected cellular signaling networks into sets of linear pathways. Understanding activation of these individual pathways is relatively straightforward—an activating ligand binds to a receptor that transmits the signal inside the cell. In essence, this communication process allows information from the extracellular environment to shape cellular processes. The question then becomes how to put the “brakes” on the response—when and how should the lines of communication be shut down?
Within the cell, propagation of the signal through an integrated network in which multiple different branches interact through positive and negative feedback (and feedforward) loops makes signal termination or deactivation far more complicated than simply flipping a switch to “off.” Events that appear to constitute signal termination mechanisms in isolated signaling pathways can actually be very important for propagating signals—or for defining the nature of signals—in a network context. Indeed, the molecular mechanisms that control signal activation and termination are essentially the same, they are just used differently. One illustration of this is seen when two adjacent cells with equivalent differentiation potential are forced to adopt opposite fates through Notch-mediated lateral inhibition (Sancho et al., 2015). The two cells function as one bistable system (Figure 1), and stochastic initial differences between their levels of Notch signaling become amplified. This process leads in turn to complete suppression of the Notch pathway in one cell (the “winner”, which also produces more of the ligand, Delta) and elevation of Notch signaling in the other (the “loser”), which downregulates Delta production. Depending on the organismal context, the winner might then differentiate, while the loser remains a stem cell. Molecularly, the same mechanisms that drive Notch signaling in the loser cell are also responsible for terminating Notch signaling in the Delta-producing winner cell (Fior and Henrique, 2009; Sancho et al., 2015).
In this review, we will discuss several negative regulatory processes in cell signaling that are frequently considered to function as mechanisms for signal termination. We will cover responses to a variety of extracellular stimuli, but will focus substantially on receptor tyrosine kinase (RTK) and G protein coupled receptor (GPCR) signaling since it is impossible to be comprehensive and these reflect our own interests. Nonetheless, as will be evident from the additional examples that we discuss, the same basic principles appear to apply in most other signaling systems. Among the most well studied negative regulators in RTK and GPCR signaling are those that reversibly modify proteins and other signaling molecules (e.g., by phosphorylation/dephosphorylation or binding of guanosine triphosphate [GTP] versus guanosine diphosphate [GDP]) or promote receptor internalization. Additional negative regulation arises from induction of inhibitory proteins and microRNAs (miRNAs) in response to signals, constituting apparent negative feedback loops. In many cases, these and other negative regulators keep signals “in check” in the absence of a stimulus, providing local reversibility in the network when signaling inputs are incomplete or partial. Far from terminating signals once they are initiated, however, the negative regulators typically play important roles in defining the nature and quality of the signal. They can also dramatically enhance signaling responsiveness and/or specificity, as discussed below for phosphatases and GTPase-activating proteins (GAPs). The ability of negative signaling regulators to provide local reversibility and/or to stop pulses of signaling activity should be distinguished from actual signal termination, which only occurs once the cell has irreversibly committed to a phenotypic response—be it differentiation, cell-cycle entry, or apoptosis—with the wholesale transcriptional and other changes that ensue. The focus of our discussion in this article will be on the shaping and sensitization of cell signaling responses by the most well-studied negative regulators.
At its simplest level, a typical profile for a signaling pathway studied in the laboratory might look like the curve shown in Figure 2A, where the stimulus under study (a growth factor in this case) is applied acutely to a cell that was previously starved of that stimulus, and response is monitored over time. Naturally, after the initial “rise” in signal and response, there is a signal decay or “dark side”—as marked in Figure 2A. Studies of this signal-decay phase, along with analyses of desensitization, have revealed a range of negative regulatory events. There is no doubt that these events do terminate the signal monitored in this particular case (as in Figure 2A) and that they can keep the signal “off” in the absence of stimulus, but they also play very important roles in defining the nature of the response.
Activation of the extracellular signal-regulated kinase (Erk) pathway by growth factors provides one of the starkest examples of how differences in the decay phase of a signaling curve can dramatically alter the cellular outcome of receptor activation. Both epidermal growth factor (EGF) and nerve growth factor (NGF) activate RTKs in the neuroendocrine PC12 cell line (EGF receptor and TrkA, respectively), which in turn activate Erk through the Ras pathway. Whereas Erk activation promotes cell proliferation in the case of EGF treatment, the response to NGF is instead terminal differentiation into neuron-like cells (Marshall, 1995). Contrary to initial suspicions that these diametrically opposed outcomes would reflect engagement of distinct signaling pathways by the two growth factor ligands, few qualitative differences could be detected (Chao, 1992). Instead, quantitative differences in the signal decay phases were found to correlate with response (Marshall, 1995). Transient Erk activation in response to EGF leads to cell proliferation (Figure 2B; left), whereas more sustained Erk activation by NGF leads to terminal differentiation (Figure 2B; right). Thus, the nature of the signal-decay phase can define the signaling outcome—although it is important to stress that the phenotypic responses occur many hours after the Erk-signaling response monitored experimentally. The distinction between transient and sustained Erk signaling has been shown to reflect differences in the engagement of feedback and feedforward loops downstream of the EGF and NGF receptors (Ryu et al., 2015; Santos et al., 2007; Sparta et al., 2015). Moreover, the cellular interpretation of the differences in Erk-activation dynamics appears to involve additional layers of signal termination that follow immediate early transcriptional responses (Murphy and Blenis, 2006; Nakakuki et al., 2010). More recent analysis of this same phenomenon over longer periods has shown pulsatile Erk activation following EGF stimulation, with EGF concentration modulating the frequency of the pulses (Albeck et al., 2013). Equivalent pulsatile Erk activation is not seen following NGF activation of TrkA (Sparta et al., 2015), presumably reflecting a different combination of feedback and feedforward effects.
The shape or pattern and the dynamics of all cell-signaling responses are determined by feedback and feedforward loops. Several excellent reviews have described how such feedback and feedforward loops define the behavior of signaling systems (Alon, 2007; Brandman and Meyer, 2008; Ferrell, 2013; Kholodenko, 2006), and the nature of these signaling-network motifs will not be discussed explicitly here. Key negative feedback mechanisms in EGF receptor signaling include endocytosis, GTPase activation, dephosphorylation, and inhibitory phosphorylation events—marked in red in the simplified network representation shown in Figure 3A (Lemmon and Schlessinger, 2010). Most of these feedback loops are rapid and provide local reversibility, helping to keep the cell “alert” to new signals and switched “off” without them. Several positive feedback loops are also marked in blue in Figure 3A, and a number of possible feedforward loops are evident (Alon, 2007). Together, depending on their relative strengths and timing, these network motifs determine responses at different nodes within the network with a variety of characteristics, including transient, adaptive, sustained, or pulsatile responses that show frequency modulation. Additional positive and negative feedback loops are also shown in Figure 3B, as communicating between elements of a robust “hourglass” or “bow-tie” network (Kitano, 2004). In this representation for RTKs, the “input” (receptor) layer communicates with a set of “core processes” including signaling by MAP kinases, Ras, phosphoinositides, Ca2+, and other kinases, which in turn communicate with an “output” layer defined by changes in transcriptional responses, in epigenetic events, and in others that have longer-term consequences. In general, positive and negative feedbacks that emanate from the output layer of this bow-tie network occur over a longer timescale than those within the input layer or core processes, since they involve transcriptional responses and can contribute to actual termination of cell signaling.
Perhaps the most obvious and well known mechanism for downregulation of signaling—when a cell surface receptor is involved—is to reduce the levels of that receptor at the plasma membrane. Ligand (agonist)-induced endocytosis and subsequent degradation of activated receptors has been well studied for GPCRs, RTKs, and other classes of cell-surface receptors and was initially recognized as a desensitization and/or signal termination event (Goh and Sorkin, 2013; Irannejad and von Zastrow, 2014). Engagement of GPCRs by agonists leads to activation of the associated heterotrimeric G protein by receptor-induced GTP/GDP exchange and downstream signaling by both the GTP-bound Gα subunit and the released Gβγ subunit. The activated GPCR itself is then phosphorylated, promoting association with β-arrestins and recruitment of the GPCR/β-arrestin complex into clathrin-coated pits for endocytosis (Kang et al., 2014). The GPCR becomes ubiquitylated in the process, through an E3 ligase associated with β-arrestin. Following internalization, the receptor may either be rapidly dephosphorylated and deubiquitylated—and recycled back to the cell surface— or it may proceed further in the endocytic pathway to lysosomal degradation. Whether recycling or degradation dominates depends on the specific GPCR, the activation context, and the ligand.
RTKs are subject to similar ligand-induced endocytosis (Goh and Sorkin, 2013), although the specific molecular mechanisms, and how they differ from RTK to RTK, are still not completely defined. Autophosphorylation of RTKs, as a result of trans-autophosphorylation within receptor dimers, is a key characteristic of their activation and signaling (Lemmon and Schlessinger, 2010), and dephosphorylation is the first step in RTK deactivation. It has been suggested that endocytosis of ligand-activated RTKs serves to deliver them to protein tyrosine phosphatases (PTPs) located on internal membranes in order to promote receptor inactivation by dephosphorylation (Haj et al., 2002). Clathrin-mediated endocytosis of EGFR requires its tyrosine kinase activity, recruitment of the Grb2 adaptor protein and the E3 ubiquitin ligase Cbl (leading to receptor ubiquitylation), and/or the participation of key internalization motifs that engage clathrin adaptor proteins (Fortian et al., 2015). Once internalized, RTKs—like GPCRs—can be deactivated and recycled to the plasma membrane or degraded after processing to lysosomes. Intriguingly, the “decision” on recycling versus degradation depends on how the receptor is activated. In the case of EGFR, for example, activating ligands such as EGF and betacellulin promote receptor internalization and degradation, whereas receptors activated by transforming growth factor-α, amphiregulin, or epiregulin are recycled (Roepstorff et al., 2009). This difference appears to reflect the pH sensitivities of the ligand/receptor complexes; those that are most pH sensitive are recycled—presumably because of ligand dissociation early in the endocytic pathway. The degree of receptor recycling versus degradation can also depend on ligand concentration. EGFR is internalized primarily through clathrin-mediated endocytosis when activated with low levels of EGF, directing the receptor toward a recycling fate. By contrast, high concentrations of EGF promote clathrinin-dependent processes that lead to receptor degradation (Sigismund et al., 2008).
Despite its reputation as a signal termination event, it is now appreciated that endocytosis does not always result in signal cessation. Indeed, there is substantial evidence for signaling of GPCRs, RTKs, and other cell-surface receptors from endosomal and other intracellular compartments. For example, GPCR-initiated signaling through Gα activation of the class III phosphatidylinositol 3-kinase (PI3K) Vps34p in yeast endosomes has been reported (Slessareva et al., 2006). Moreover, antibody-based fluorescent probes have convincingly revealed the presence of activated β2-adrenoceptors (βAR) in endosomes after ligand stimulation (Irannejad et al., 2013). Associated functional studies have also shown that blocking receptor internalization attenuates (rather than enhances) late-stage cAMP signaling in response to βAR agonists.
There are also now numerous studies showing that RTKs can signal after internalization into intracellular compartments (Miaczynska, 2013), although internalization does not seem to be absolutely required for any key signaling events (Sousa et al., 2012). Modulating intracellular localization can alter the nature as well as the extent and timing of the signal. For example, misfolded mutated RTKs such as Flt3-ITD can signal from the endoplasmic reticulum (Choudhary et al., 2009) and do so with different pathway biases than seen for the same receptor at the plasma membrane (preferring STAT5 over Erk and Akt). Similarly, activated TrkA signals quite differently in neurons depending on whether it is activated by NGF or by neurotrophin-3 (NT-3), as a result of differences in how endosomes containing TrkA activated by these two ligands are trafficked within the cell (Harrington et al., 2011).
Perhaps one of the best examples of the importance of signaling from intracellular compartments—and where the route of receptor internalization can define the nature of the signal—is seen with the serine/threonine kinase transforming growth factor-β (TGF-β) receptor (Di Guglielmo et al., 2003; Moustakas and Heldin, 2009). Activated TGFβ receptors can be internalized through clathrin-mediated or clathrin-independent pathways and are delivered to different downstream mediators depending on their route of internalization. Receptors internalized by the clathrin-mediated pathway reach early endosomes where the Smad anchor for receptor activation (SARA) serves as a scaffold for receptor-regulated Smad2/3 (R-Smad2/3) phosphorylation. This step promotes subsequent signaling through transcriptional regulation by nuclear Smad complexes, and TGFβ receptors are recycled to the plasma membrane. Receptor internalization has a direct positive effect on TGFβ signaling in this case, and inhibiting clathrin-mediated endocytosis prevents TGFβ-induced Smad activation (Di Guglielmo et al., 2003). By contrast, receptors internalized through caveolae are delivered to intracellular compartments that contain the inhibitory Smad (I-Smad), Smad7, which recruits the Smurf2 E3 ubiquitin ligase to TGFβRs. This route of internalization thus promotes TGFβ receptor degradation and negative regulation of its signaling, although it may also facilitate non-Smad-mediated signaling pathways. One key distinction compared with GPCRs or RTKs is that TGFβ receptor internalization is not accelerated by ligand binding (Mitchell et al., 2004). As with GPCRs and RTKs, however, it seems clear that receptor internalization of TGFβ receptors can be just as frequently involved in promoting—or defining the nature of—receptor signaling as it is in termination. It is certainly not the signal-termination mechanism that it was initially thought to be.
G proteins play a crucial role in signaling downstream of RTKs and GPCRs, with the GTP-bound forms of Ras (Figure 3A) and of Gα subunits mediating much of the signaling from these classes of receptors. Binding of GAPs to GTP-bound Ras or Gα increases their intrinsic GTPase activity by two to four orders of magnitude, promoting the hydrolysis of bound GTP to GDP and conversion of the G protein to its inactive state. Regulators of G protein signaling (RGS proteins) and other effectors also have this effect on heterotrimeric G proteins. The timescales of GTP hydrolysis are important to consider when discussing the functions of these negative regulators. In the presence of their activators (a GAP and an RGS respectively), N-Ras and Gαs hydrolyze GTP at rates of 15 s−1 and 3 s−1 (Nixon et al., 1995; Sprang, 1997), corresponding to half-lives of the GTP-bound active state in the 50–200 ms range (substantially less in some cases). As discussed by Ross (Ross, 2008), such high rates of GTP hydrolysis—and thus of signal deactivation—are crucial for rapid response to the removal of stimulus, in vision for example, where perceiving sudden onset of darkness requires such rapid termination. For a rapid signal, such as in vision, this makes sense. For slower signaling responses such as Ras signaling or response to glucagon, for example, the need to deactivate the G protein in 10–100 ms is less clear. In these cases, while stimulus is applied to the relevant GPCR or RTK, the G protein continually hydrolyzes GTP (stimulated by the GAP) and has its GTP replaced (stimulated by the receptor or signaling pathway)—with the half time of each cycle being in the 10–100 ms range. This rapid cycling throughout the duration of the signal is typically not considered in the frequently promulgated view of G protein GTPase activity as a simple timer to “switch” the signal off as a negative regulator and has numerous important implications for signaling responsiveness that we will discuss below.
Protein phosphatases are key negative regulators in cell signaling. In particular, PTPs are poised to terminate activation of RTKs by reversing the autophosphorylation required for activity of the receptor, and by dephosphorylating the phosphotyrosines that recruit signaling molecules with Src homology 2 (SH2) and/or phosphotyrosine binding (PTB) domains (Lemmon and Schlessinger, 2010). In the particular cases depicted as negative feedback loops in Figure 3A, the SH2-domain-containing phosphatases Shp1 (PTPN6) and Shp2 (PTPN11) are recruited to the activated EGFR in a ligand-dependent manner and promote dephosphorylation of the receptor. A classical view of dephosphorylation of activated RTKs such as EGFR and platelet-derived growth factor (PDGF) receptor involves their endocytosis and delivery to internal membranes at which PTP1B is located (Haj et al., 2002). EGFR phosphorylation diminishes with approximately the same time course as receptor internalization (Figure 4; black curve), approximately resembling the time course of transient EGF-induced Erk activation seen in Figure 2B. Receptor dephosphorylation thus appears to resemble a simple negative feedback loop that terminates signaling to generate a transient response.
This simple sense of signaling timescale is misleading, however, as with the case of GDP/GTP cycling. Several recent studies have shown that phosphotyrosine turnover on RTKs is in fact remarkably rapid (Kleiman et al., 2011; Monast et al., 2012). Indeed, overlaid on the plot in Figure 4 are data reflecting that fact that acute blockade of the EGF receptor kinase (responsible for its own autophosphorylation) with a specific small-molecule kinase inhibitor causes very rapid dephosphorylation with half-lives (t1/2) in the 10–15 s range—although this is an upper limit because the experimental approaches employed could not reasonably have measured smaller half-lives. This dephosphorylation takes place at the plasma membrane and is unaffected by inhibition of endocytosis (Monast et al., 2012). Although the phosphatases responsible for this rapid and chronic receptor dephosphorylation are not known, a PTP called DEP-1 (density-enhanced phosphatase-1) or PTPRJ has been identified as one contributor to EGFR dephosphorylation at the plasma membrane (Tarcic et al., 2009). Taken together, these studies show that, even prior to endocytosis, phosphotyrosines on EGFR turn over on a timescale of seconds, rather than minutes or hours, and that a given EGF receptor must go through hundreds to thousands of cycles of phosphorylation/dephosphorylation in the course of a typical EGF receptor response (Kleiman et al., 2011). Moreover, these observations have been extended to several other RTKs across multiple cell types. Importantly, associated modeling studies suggest that the EGF receptor also cycles rapidly between phosphorylated and dephosphorylated states in the absence of activating ligand (Kleiman et al., 2011), as predicted by the well-known observation that phosphatase inhibition using pervanadate or hydrogen peroxide rapidly enhances autophosphorylation (and activation) of EGFR and other RTKs. These rapid phosphorylation/dephosphorylation cycles are not limited to the signal-initiating RTKs. By analyzing downstream events, Kleiman et al. (2011) showed almost equally rapid dephosphorylation of Akt (at serine 473) and Erk (at threonine 202 and tyrosine 204) following acute EGFR inhibition. In these cases, protein phosphatase activity is clearly playing a signal-damping role. The frequency of the kinase/phosphatase cycles is much greater than one might anticipate, however, with important implications that are discussed below. The rapid cycling also seems unexpectedly “expensive” from an ATP-consumption point of view, suggesting that the kinase/phosphatase cycles might play more complicated additional roles in regulating signaling dynamics, crosstalk, and amplification (Heinrich et al., 2002).
There is no doubt that GAPs negatively regulate G protein signaling and silence it in the absence of stimulus. Similarly, dephosphorylation of RTKs and other phosphoproteins plays an essential role in negative regulation and signal damping. The importance of GAPs and phosphatases as negative regulators is evident in human disease. Several function as tumor suppressors, loss of which causes cancer because of unabated signaling. Examples include the neurofibromatosis type 1 (NF1) gene product, which is a Ras-GAP called neurofibromin (Ratner and Miller, 2015) and the EGFR phosphatase DEP-1. Similarly, in the GPCR field, patients with RGS9 mutations show impaired inactivation of transducin and have defects in adaptation to sudden light changes (Nishiguchi et al., 2004).
The high rates of GTP/GDP and kinase/phosphatase cycling outlined above, however, suggest that GAPs and phosphatases might function as more than simple negative regulators. As pointed out 35 years ago (Goldbeter and Koshland, 1981), an activation/deactivation cycle such as seen for kinase/phosphatase pairs and GTP/GDP cycles (Figure 5) can display zero-order ultrasensitivity if the enzymes performing both reactions are operating close to saturation, which makes the signaling step much more responsive. Readers are referred to an excellent review by Ferrell and Ha (2014) for a thorough analysis of this phenomenon. The consequences of zero-order ultrasensitivity are illustrated in Figure 5 for a kinase/phosphatase cycle (Gomez-Uribe et al., 2007). When both kinase and phosphatase are far from saturation (Figure 5A), the response (level of phosphoprotein) is hyperbolic as kinase concentration is increased (as a proxy for receptor activation). When only the kinase is saturated, the response is almost linear with kinase concentration until essentially all of the protein is phosphorylated (Figure 5B). This curve exceeds the hyperbolic response in its sensitivity (Hill coefficient of 2) and becomes profoundly sigmoidal with increasing saturation. When the phosphatase is saturated but the kinase is not, a threshold emerges (Figure 5C), beyond which the response is near hyperbolic (but the response is still ultrasensitive). When both reactions are saturated, however, the curve is steeply sigmoidal (Figure 5D), with an effective Hill coefficient of ~26 for the parameters used in this figure (Ferrell and Ha, 2014). It is important to note that zero-order ultrasensitivity is only achieved when the substrates are in excess over the converting enzymes. Since the EGF receptor phosphorylates itself, the 10- to 100-fold excess of substrate over enzyme required for an ultrasensitive response is not achieved (although each receptor does have multiple phosphorylation sites). However, the rapid turnover of phosphorylation sites in RTKs is thought to enhance sensitivity through this general mechanism. Indeed, performing their initial analysis of zero-order ultrasensitivity around the time of the discovery of tyrosine phosphorylation, Goldbeter and Koshland presciently commented that “Because phosphorylation has been identified with the src gene, it is intriguing to ask whether a change in sensitivity may be important in the loss of control identified with cancer cells.” (Goldbeter and Koshland, 1981).
Zero-order ultrasensitivity presumably also operates for GTP/GDP cycles when the guanine nucleotide exchange factor (GEF) and/or the GAP are operating near saturation (Roth et al., 2015). Indeed, optical manipulations of GPCR signaling in single cells were recently used to reveal such switch-like ultrasensitivity in Gi-mediated activation of PI3K in macrophages (Karunarathne et al., 2013). There are also data suggesting that GAPs can potentiate G protein activation by GPCRs and, further, that GPCRs can indirectly inhibit GAP activity (Ross, 2008), both of which will enhance the switch-like characteristics of GPCR signaling. The latter of these two phenomena is paralleled by an interesting feedback mechanism in RTK signaling, depicted in Figure 3A. One of the consequences of RTK activation is the production of H2O2 and other reactive oxygen species (ROS), mediated by PI3K- and Rac-dependent activation of NADPH-oxidase (Bae et al., 1997). These ROS transiently inhibit PTP activity by oxidizing a catalytically critical cysteine in the active site of the phosphatase (Tonks, 2006). This process constitutes inhibition of a negative regulator and therefore generates a positive feedback loop that results in a switch-like response that could drive lateral signaling propagation through diffusing ROS (Tischer and Bastiaens, 2003). In all of these cases, signaling responsiveness is increased as a result of reactions that expend ATP by cycling rapidly between phosphorylated and dephosphorylated protein, or between GTP and GDP.
Another possible significance of the rapid dephosphorylation and GTP hydrolysis steps characteristic of RTK and G protein signaling is kinetic proofreading. Kinetic proofreading was first proposed as a mechanism for reducing error in protein synthesis and DNA regulation (Hopfield, 1974). During translation, a nonequilibrium/energy-consuming event (GTP hydrolysis by the G protein EF-Tu) follows initial binding of the aminoacyl-tRNA/EF-Tu-GTP complex to the ribosome. Most non-cognate tRNAs dissociate without GTP hydrolysis by EF-Tu. However, some near-cognate tRNAs bind sufficiently strongly that EF-Tu proceeds to GTP hydrolysis. In these cases, the tRNA is released from EF-Tu on the ribosome to be accommodated in the peptidyl transferase center. Non-cognate aminoacyl-tRNAs dissociate rapidly during this relatively slow step, whereas the cognate aminoacyl-tRNA remains bound to be incorporated into the growing polypeptide (Rodnina and Wintermeyer, 2001). Several cycles of GTP hydrolysis can occur for each correct aminoacyl-tRNA incorporation, coordinated by this proofreading step.
A conceptually related process was later invoked as a mechanism for antigen discrimination by T cell receptors (McKeithan, 1995). The basic idea is that productive receptor signaling can only take place if the cognate antigen is bound and is bound (like cognate aminoacyl-tRNA) for sufficient time to allow the multiple steps required for downstream signaling by the receptor to take place. Antigen binding to the T cell receptor induces tyrosine phosphorylation of residues in the ITAM (immunoreceptor tyrosine-based activation motif)—an energy-requiring reaction— and the phosphotyrosines are constantly dephosphorylated by PTPs (this is a requirement of the kinetic proofreading here). If the bound antigen is non-cognate, it dissociates rapidly, and PTPs dephosphorylate and reverse the activation of the receptor— before downstream signaling has been initiated. There are numerous other relevant steps in building the activated T cell receptor that also contribute to this kinetic proofreading (Goldstein et al., 2004), but this brief description highlights the role of the phosphatases.
RTKs such as the EGF receptor may use kinetic proofreading in a similar way to discern cognate from non-cognate ligands, and GTP/GDP cycles may play a similar role in ensuring GPCR fidelity. As stated by McKeithan (1995), cycles of ligand association and dissociation appear to result in the “waste” of metabolic energy as phosphorylation (or GTP-binding) events that have been initiated are rapidly reversed by phosphatases (or GTPase activity). For kinetic proofreading to be effective, nonspecific ligand/receptor complexes (or adventitiously dimerized receptors) must dissociate before the receptor complex has the opportunity to initiate signaling. Dephosphorylation (or GTP hydrolysis) must deactivate the receptor complex similarly rapidly to avoid initiation of signaling after dissociation of the non-specific ligand—explaining why dephosphorylation or GTP hydrolysis proceeds at rates substantially beyond those needed for simple negative regulation. Despite these high rates, it is estimated that rapid phosphorylation/dephosphorylation cycling in EGFR, for example (Monast et al., 2012), accounts for less than 0.01% of the ATP used by the cell (noting that a human turns over their body weight in ATP every day, or ~108 ATP molecules per minute per cell).
It has also been suggested that high levels of phosphatase activity might be important for specificity and proofreading on the “output” side for RTKs, by damping spurious signaling through SH2 domains with limited specificity and high dwell times (Oh et al., 2012). Moreover, theoretical studies have suggested the importance of rapid substrate recycling by phosphatases as a proofreading mechanism in downstream activation of MAP kinases (Swain and Siggia, 2002).
Ubiquitylation plays roles in the function of each of the major degradative pathways in the cell, including proteasomes, lysosomes, and autophagosomes (Clague et al., 2012). As mentioned previously, ubiquitylation is important in initiating GPCR and RTK endocytosis and degradation following ligand activation of these receptors. It is also key for TGFβ receptor degradation following its internalization via clathrin-coated pits. The machinery for ubiquitylation and deubiquitylation is quite well characterized, and the cycle of modification shares conceptual similarities with phosphorylation/dephosphorylation.
Ubiquitylation also influences many aspects of cell signaling (Clague et al., 2012) and, like phosphorylation, can contribute both to negative regulation and signal propagation. One of the most well studied signaling events involving ubiquitylation occurs in NF-κB signaling induced by tumor necrosis factor-α (TNFα), interleukin-1 (IL-1), and Toll-like receptors (TLRs). Following ubiquitylation of different combinations of intermediate proteins in each case (Clark et al., 2013; Grabbe et al., 2011), each of these ligands causes the IκBkinase (IKK) complex to become activated and to phosphorylate the inhibitory IκBα protein that associates with and blocks activity of p50/p65 NF-κB dimers. Phosphorylated IκBα is then subjected to K48-linked polyubiquitylation and is degraded by the proteasome, liberating NF-κB to be translocated into the nucleus where it activates transcription of genes involved in inflammation and cell survival. As part of its response, NF-κB also activates the transcription of key genes that terminate the inducible signal. The gene products responsible for this negative feedback include IκBα itself, production of which serves to re-establish NF-κB inhibition. Another is the deubiquitylating enzyme (DUB) A20, which reverses several of the ubiquitylation events (K48-linked and K63-linked) that precede NF-κB activation (Renner and Schmitz, 2009). Indeed, germline mutations that cause haploinsufficiency of A20 were recently reported to cause an early onset autoinflammatory disease (Zhou et al., 2016). A20 has also been reported to interact directly with the IL-17 receptor IL-17RA to provide a more direct negative feedback loop (Garg et al., 2013) and can inhibit interaction of several of the E3 ubiquitin ligases involved in promoting NF-κB signaling with their E2 enzymes (Shembade et al., 2010). As we will see below, the upregulation of a DUB in this way, following an ubiquitylation-dependent signal, shows interesting parallels with the upregulation of dual-specificity phosphatases (DUSPs) downstream of Erk signaling—extending the analogy between kinase/phosphatase and ubiquitylation/deubiquitylation cycles. It remains to be determined, however, whether the turnover rate of ubiquitin at modified sites is as rapid as seen for phosphates on EGFR, for example. Indeed, this will be an important question for all of the growing number of reversible protein modifications employed in cell signaling.
Key elements of the negative regulation described above for NF-κB signaling employ a DUB and also replace an ubiquitylated inhibitory protein (IκBα). This occurs through transcriptional responses—events in the “output layer” in Figure 3B that lead to production of proteins that inhibit steps within the core processes of the bow-tie network. There are several other notable examples in which signals are terminated or suppressed through transcription of negative regulators, particularly in immune cell signaling. The suppressor of cytokine signaling (SOCS) and cytokine-induced SH2 (CIS) proteins are well-studied examples (Endo et al., 1997; Starr et al., 1997). STAT-activated expression of SOCS and CIS proteins is induced following activation of many cytokine receptors and RTKs, and SOCS proteins function as negative regulators of cytokine signaling in several different ways (Palmer and Restifo, 2009). Each SOCS protein contains an SH2 domain that targets it to a subset of proteins that become tyrosine phosphorylated in cytokine or RTK signaling and can also compete with the SH2 domains of STATs for binding to activated cytokine receptors. They also all contain a C-terminal SOCS box that binds elements of the ubiquitylation machinery (Zhang et al., 1999), allowing SOCS proteins to function as adaptors that promote ubiquitin-dependent proteasomal degradation of their binding targets. SOCS1 and 3 additionally have a KIR motif that specifically inhibits Janus kinases (JAKs), thus exerting direct negative feedback on cytokine-induced STAT activation (Babon et al., 2012). Several of the SOCS proteins are also upregulated through STAT function in response to RTK activation (Kazi et al., 2014) and have been reported to interact with— and inhibit—several RTKs. This may be a primary function of SOCS4 and 5, notably with the EGF receptor (Kario et al., 2005). The upregulation of SOCS proteins by RTKs, and their promotion of RTK degradation represents a direct negative feedback between the output and input layers of the bow-tie network in Figure 3B, as well as linking RTK and cytokine signaling.
Transcriptional activation of negative regulators is universally important in the termination of signaling from cell-surface receptors. The transcriptional events that follow EGF receptor activation provide a useful illustration. Transcriptional responses begin around 45 min after EGF simulation with the immediate early genes (IEGs), including FOS, JUN, and EGR1. The IEGs are followed by the delayed early genes (DEGs) at around 45–120 min and then by the secondary response genes (SRGs) as the outcome-defining step (Avraham and Yarden, 2011). The SRGs and arguably the DEGs would constitute output layer events in Figure 3B. Avraham and Yarden (2011) have called the DEGs “superintendents of IEGs”, since some of their products function as feedback inhibitors of the IEG products (Amit et al., 2007). Examples include dual-specificity phosphatases (DUSPs) that inhibit sustained Erk signaling and several known transcriptional repressors such as ID2, NAB2, FOSL1, andJUNB, all of which attenuate EGF-driven transcription (including that driven by IEG products). The DEGs also include KLF2, KLF6, and MAFF (which all repress EGF-dependent transcription events) and tristetraprolin (TTP), an mRNA-destabilizing RNA-binding protein that enhances the decay of mRNAs (Brooks and Blackshear, 2013), including a subset of IEG transcripts. Interestingly, EGF-induced transcription of Mig6/RALT is also increased on the same timeframe as the DEGs, leading to the production of a soluble protein that becomes tyrosine phosphorylated by EGFR and directly inhibits the intracellular tyrosine kinase domain of the receptor (Park et al., 2015). This process will not serve as an acute negative feedback mechanism of EGF receptor signaling, however, since the activated receptor is degraded within the timeframe of Mig6 production. It could nonetheless serve to terminate continuous receptor activation or to prevent repeated activation.
It should be noted that the IEG products are also crucial for interpreting the differences in duration of Erk activation in Figure 2B, as mentioned previously. The IEG product c-Fos is an unstable protein that becomes multiply phosphorylated— and thus stabilized—when Erk activity is sustained but is instead rapidly degraded when Erk activation is only transient and ceases before Erk has the opportunity to phosphorylate c-Fos (Murphy and Blenis, 2006; Nakakuki et al., 2010). The resulting differences in AP1-mediated transcriptional activation contribute to the difference in phenotypic consequences.
Many studies have shown that miRNAs control levels of intracellular signaling molecules through post-transcriptional repression (Lui et al., 2015). These miRNAs have been termed “master controllers” since most influence multiple targets (frequently hundreds) in different parts of the signaling network. These miRNAs can therefore alter the behavior of a signal network dramatically—depending on the cellular context, since levels of both positive and negative regulators are influenced by miRNAs. There have been several reports of receptor control of miRNA levels, providing the opportunity for feedback regulation mediated by miRNAs. For example, EGF stimulation of mammary epithelial cells leads to upregulation of around 20 miRNAs and immediate degradation of a similarly sized set (Avraham et al., 2010). The set that is degraded includes several miRNAs that otherwise repress IEG products and thus damp inappropriate signaling. By relieving this damping effect, miRNA degradation following EGF receptor activation constitutes a double-negative feedback—reducing a negative regulatory influence (Avraham and Yarden, 2011). There are also several cases in which receptor-induced miRNA induction has been reported to have a direct negative feedback effect. For example, production of miR-146a is induced following NF-κB activation downstream of certain TLRs or cytokine receptors (Taganov et al., 2006) and represses levels of IRAK1 and TRAF6. Since these are both positive regulators of NF-κB in responses downstream of TLRs and cytokine receptors, miR-146a induction constitutes a direct negative feedback loop. Similar situations have also been described for miR-155 and miR-302b as negative feedback loops in NF-κB signaling (Ceppi et al., 2009; Zhou et al., 2014). Certain miRNAs have also been implicated in negative feedback loops in RTK signaling. Activation of MET, the receptor for hepatocyte growth factor, leads to increased miR-27a levels (through Elk-1), and miR-27a in turn represses MET and EGFR levels (Acunzo et al., 2013). Activation of the RTK Axl also exerts negative feedback on its own expression by upregulating miR-34a—again through Elk-1 activation (Cho et al., 2016). Similarly, miR-146b is induced following activation of the PDGF receptor with PDGF BB (but not AA) in several cancer cells. Induction of this miRNA depends on c-Fos and leads to suppression of EGFR, providing a cross-RTK inhibitory feedback mechanism (Shao et al., 2011). These few examples only scratch the surface of the complexity of negative (and positive) feedback mediated by miRNAs in cell signaling, and it is clear that miRNAs must be accounted for when we consider quantitative aspects of signal termination, propagation, or aberration in disease—and this will require a great deal more data. It is important to appreciate, however, that since each miRNA can have hundreds of targets, all of the negative feedback loops noted above will ensue in the context of many other changes. Although the data for the negative feedback events cited above seem quite robust, production of the same miRNAs in a different cell type or context could alternatively exert positive effects. For example, the miR-26 family has been reported to play opposing tumor-suppressor and tumor-promoting roles depending on the cell type (Mendell and Olson, 2012)—again emphasizing the fine distinction between “off” and “on” cues in cell signaling.
The phenomenon of “dependence receptors” (Goldschneider and Mehlen, 2010; Mehlen and Bredesen, 2011) provides a vivid example of how our assumptions about what constitute “off” and “on” signals need revision in some cases. The traditional view of all cell-surface receptors, including RTKs and others, is that their steady state is “off” in the absence of ligand. The notion that these receptors do not signal in the absence of ligand is challenged by the dependence receptors, which include several RTKs (such as MET, Trks, ALK, and insulin receptor family members) plus several netrin receptors, integrins, and patched. There is no doubt that these receptors are switched to active (or canonical) signaling states by their cognate ligands and signal according to their type—initiating both stimulatory signals and signals that terminate the initial ligand-induced stimulatory response. Far from being signaling-inactive in the absence of ligand, however, it is clear that the unoccupied dependence receptors promote proapoptotic responses or control expression of imprinted genes and miRNAs, and these effects are reversed—or switched “off”—by the cognate ligand (Boucher et al., 2014; Goldschneider and Mehlen, 2010). One key feature of most of the dependence receptors is that they are subject to cleavage by caspases in the absence of ligand, with the resulting fragments mediating proapoptotic signaling. A second key feature is that ligand binding inhibits this cleavage, leading to the termination of proapoptotic signaling. Thus, in addition to their best-known roles, the ligands for dependence receptors can function as crucial survival factors in a variety of developmental and other processes. It is also intriguing to note that expression of one dependence receptor in the absence of its ligand (UNC5D, a netrin receptor) induces apoptosis in neuroblastoma and is highly expressed in tumors that spontaneously regress (Zhu et al., 2013)—implying a tumor-suppression function.
From the point of view of this review, the studies of dependence receptors blur the distinction between signal termination and initiation in an important way (Figure 6). Ligand binding switches off, or terminates, the proapoptotic signal of the unoccupied dependence receptors. At the same time, ligand binding switches on, or initiates, the canonical signals of the receptor, such as Ras/MAP kinase signaling and the other events summarized in Figure 3A, including the negative and positive feedback events discussed above.
Recent results in preclinical studies of targeted cancer therapeutics have illustrated how several of the negative regulators and negative feedback loops in RTK signaling maintain local reversibility and “plasticity” in the network (Graves et al., 2013). Kinome-wide analyses of protein kinase activation and levels have revealed extensive dynamic reprogramming of the whole network following the selective inhibition of a single kinase, negating the effectiveness of that single inhibitor in blocking cancer-cell proliferation through any number of “bypass” mechanisms. For example, subjecting triple-negative breast cancer cells to a specific MAPKK inhibitor led to changes in the apparent activity levels of over 100 different protein kinases, almost equally divided between inhibition and activation (Duncan et al., 2012), leading to resistance. One notable change was the upregulation of transcription of several RTKs through the induction of c-Myc degradation following MAPKK blockade by the inhibitor. This finding in turn suggested the possibility of combining the MAPKK inhibitor with RTK inhibitors to block proliferation of these cancer cells more effectively by targeting multiple points in the network to combat resistance. The same concept has emerged from many other studies. Inhibiting Akt in breast cancer cells was also shown to promote upregulation and activation of several RTKs (Chandarlapaty et al., 2011). This effect was found to arise from relief of a negative feedback loop in which active Akt phosphorylates and inhibits FOXO transcription factors and from blockade of TORC1 signaling downstream of Akt, relieving additional negative influences on RTK signaling. Inhibition of BRAF in BRAF-driven melanoma similarly relieves negative feedback, leading to upregulation of RTK signaling and bypass of/resistance to the desired inhibition (Lito et al., 2012). Again, these findings point to the need to combine any given targeted agent with other inhibitors that can counter the reactivation events arising from severed negative feedback loops when trying to terminate signaling synthetically with inhibitors. Upregulation of Met in acquired resistance to EGF receptor inhibitors (Engelman et al., 2007) also reveals how cross talk between pathways at the level of negative feedback, or possibly the miRNA influences described above, can have similar consequences. Indeed, numerous RTKs and their activating ligands have been found to be upregulated in cases of acquired resistance to EGFR inhibition in lung cancer (Niederst and Engelman, 2013) and other cancers. The ability of the signaling networks to adapt in this way argues that the feedback mechanisms involved are quite reversible and lead to wholesale reprogramming of the network, in turn functioning to stymie synthetic signal termination in cancer cells.
In considering which processes terminate cellular signaling once it has been initiated, it seems reasonable to focus on the plethora of negative feedback loops and negative regulators. We have discussed some of them here but have by no means been exhaustive, focusing largely on RTK and GPCR signaling. There is no doubt that full exertion of all the negative regulators in the signal networks we have discussed can keep a signal switched off and/or terminate it. Indeed, many of the negative regulators that we have discussed here play important roles in stabilizing basal levels of signaling. Truly terminating the signaling once it has started, however, appears to be a more defining “decision” for the cell. It typically involves a switch-like response into a new cellular state or fate, frequently on a much longer timescale than that used for studying molecular aspects of cell signaling. Examples include commitment to differentiation, cell-cycle entry, and apoptosis. The dark side of cell signaling, or how signals decay, contributes profoundly to these decision-making processes. The nature and strength of the negative regulatory events define the shape and timing of the responses, and these in turn help define the behavior and fidelity of the whole system. Importantly, nature also exploits the negative regulators in cell signaling to enhance responsiveness and specificity, with frequencies of GTP/GDP and phosphorylation/dephosphorylation cycling being significantly greater than one might expect for straightforward negative regulation, with a resulting unexpectedly high cost in terms of ATP consumption (Qian, 2007). Given the characteristics that these cycles impart to the responses, it can be argued that dephosphorylation and GTP hydrolysis events, while capable of damping signals, play as much of a role in defining how signals are propagated as they do in negative regulation. These and other examples, including the dependence receptors, serve to blur the distinction between positive and negative signals and to emphasize the importance of network properties in defining outcome.
This work was supported in part by NIH grants R01-GM099891, R01-GM107435, and R01-CA198164 to M.A.L., F32-GM109688 to D.M.F., and PSOC@Penn (U54-CA193417) to D. Discher. We thank Kathryn Ferguson, Boris Kholodenko, Daryl Klein, John Ladbury, Mitchell Lazar, Matthew Lazzara, Robert Phair, Ravi Radhakrishnan, John Scott, and Alex Toker for discussions and for valuable comments on the manuscript.