The Hedgehog (Hh) signalling cascade is highly conserved and involved in development and disease throughout evolution. Nevertheless, in comparison with other pathways, our mechanistic understanding of Hh signal transduction is remarkably incomplete. In the absence of ligand, the Hh receptor Patched (Ptc) represses the key signal transducer Smoothened (Smo) through an unknown mechanism. Hh binding to Ptc alleviates this repression, causing Smo redistribution to the plasma membrane, phosphorylation and opening of the Smo cytoplasmic tail, and Smo oligomerisation. However, the order and interdependence of these events is as yet poorly understood. We have mathematically modelled and simulated Smo activation for two alternative modes of pathway activation, with Ptc primarily affecting either Smo localisation or phosphorylation. Visualising Smo activation through a novel, fluorescence-based reporter allowed us to test these competing models. Here, we show that Smo localisation to the plasma membrane is sufficient for phosphorylation of the cytoplasmic tail in the presence of Ptc. Using fluorescence cross-correlation spectroscopy (FCCS), we also demonstrate that inactivation of Ptc by Hh induces Smo clustering irrespective of Smo phosphorylation. Our observations therefore support a model of Hh signal transduction whereby Smo subcellular localisation and not phosphorylation is the primary target of Ptc function.
Hedgehog signalling has been implicated in crucial developmental and physiological processes in both Drosophila and vertebrates. Despite these important roles, there are still large gaps in our mechanistic understanding of Hh signal transduction (Jiang and Hui, 2008; Ingham et al., 2011). In the absence of ligand, the Hh receptor Ptc (Ingham et al., 1991; Marigo et al., 1996; Zheng et al., 2010) inhibits the GPCR-like signal transducer Smo (Alcedo et al., 1996; van den Heuvel and Ingham, 1996). Smo inhibition occurs without direct protein interaction (Taipale et al., 2002) and presumably involves a small-molecule intermediate, which is probably a lipid (Bijlsma et al., 2006; Khaliullina et al., 2009). However, it is not known where in the cell Ptc acts to inactivate Smo. In addition, although the components of the Hh signalling cascade are largely conserved across evolution, there are also clear differences in the way the vertebrate and fly cascades operate. This is illustrated by the different sensitivities of fly and mammalian Smo to Cyclopamine and the differences in the relative importance of Cos2 and Su(Fu) proteins for pathway activity. These differences might be associated with the absence of primary cilia, which play a central role in mammalian Hh signalling, from most Drosophila cells (Ingham et al., 2011; Briscoe and Thérond, 2013).
Recently, Drosophila Ptc was postulated to control Smo localisation and activation indirectly through a wide range of intermediate protein players acting on Smo. These include lipid-modifying enzymes (Yavari et al., 2010), cAMP and protein kinase A (PKA) (Ogden et al., 2008), and the phosphatases PP1, PP4 and PP2A (Jia et al., 2009; Su et al., 2011). Regardless of the precise mechanism, Hh binding alleviates the inhibitory activity of Ptc towards Smo. In its inactive state Smo resides on internal cell membranes (Denef et al., 2000; Zhu et al., 2003; Nakano et al., 2004) that are presumably a mixture of early and late endosomes, and lysosomes (Nakano et al., 2004; Li et al., 2012; Xia et al., 2012). Electrostatic interactions between four positively charged arginine clusters collectively termed Smoothened autoinhibitory domain (SAID) in the membrane proximal part of the C-terminal cytoplasmic domain and negatively charged distal patches keep the cytoplasmic tail in a closed conformation (Zhao et al., 2007).
Upon pathway activation, Smo is phosphorylated by protein kinase A (PKA) and casein kinase 1 (CK1) at multiple serine residues within the cytoplasmic tail (Jia et al., 2004; Zhang et al., 2004; Apionishev et al., 2005; Jia et al., 2010; Su et al., 2011). Their phosphorylation or phosphomimetic replacement masks the positive charge of the SAID, releasing the C-terminal tail into an open conformation and inducing Smo clustering (Zhao et al., 2007). Smo phosphorylation also activates downstream signal transduction (Jia et al., 2004; Zhao et al., 2007) by the recruitment of a protein complex centred on Costal2 (Jia et al., 2003; Ogden et al., 2003) (Fan et al., 2012). Assembly of this complex is thought to suppress the proteolytic processing of the Gli family transcription factor Cubitus interruptus (Ci) into its repressor version, thereby stabilising the full-length activator form (Jiang and Hui, 2008; Ingham et al., 2011). Finally, Smo phosphorylation is sufficient for the redistribution of Smo to the plasma membrane associated with pathway activation (Denef et al., 2000; Zhu et al., 2003; Nakano et al., 2004). A possible mechanism is provided by the observations that SAID phosphorylation prevents the ubiquitylation of adjacent lysine residues, which promotes Smo internalisation (Li et al., 2012). In addition, Hh promotes the recruitment of a deubiquitylating enzyme to the Smo cytoplasmic tail, thereby suppressing recruitment of Smo into early endosomes (Xia et al., 2012).
Together, these observations can be condensed into a model of Hh pathway organisation whereby Ptc primarily controls the phosphorylation state of Smo. Ptc inactivation in response to Hh then leads to Smo phosphorylation and subsequently to conformational change, clustering and accumulation at the plasma membrane, where Smo becomes active as a signal transducer. However, Smo localises to the plasma membrane in a Hh-independent manner when cellular phosphatidylinositol-4 phosphate (PI4P) levels are experimentally increased. Intriguingly, Ptc itself directly or indirectly downregulates PI4P accumulation (Yavari et al., 2010). This suggests an alternative order of events, whereby Ptc inactivation by Hh first drives Smo membrane localisation by modulating membrane phospholipids, with Smo phosphorylation and clustering occurring downstream.
To improve our understanding of Hh signal transduction we therefore need to identify which of the multiple cell biological processes downstream of Ptc primarily regulates Smo activation, and must clarify the connections between the events occurring at the level of Smo, the key signal transducer of the pathway. We have addressed these questions by combining a modelling approach with the direct visualisation of Smo phosphorylation status and the biophysical detection of Smo clustering. First, we simulated Smo activation in response to Hh with the help of a simplified, nondimensionalised equilibrium model, considering two scenarios corresponding to the alternative roles of Ptc in Smo regulation outlined above. Second, following a previously established strategy (Michel et al., 2011), we generated a fluorescence-based Smo activation reporter by inserting the conformation-sensitive core of the Inverse Pericam Ca2+ sensor (Nagai et al., 2001) into the cytoplasmic tail of Smo. Fluorescence of this reporter strictly reflects the phosphorylation dependent opening of the Smo tail, which can therefore be tracked with subcellular resolution both live and in fixed samples. We have used this reporter to test the alternative models and their underlying assumptions. Third, we have directly measured the oligomerisation state of fluorescently tagged Smo on the plasma membrane of cultured cells by dual-colour fluorescence cross-correlation spectroscopy (FCCS) (Weidemann et al., 2002; Bacia et al., 2006).
Here, we show that localisation of Smo to the plasma membrane is by itself sufficient to induce phosphorylation of the cytoplasmic tail, irrespective of the presence or absence of Ptc. In addition, we demonstrate that inactivation of Ptc by Hh controls Smo clustering independent of Smo tail phosphorylation. These results challenge models that place Smo phosphorylation at the apex of regulatory events. Our observations instead strongly support models of Hh pathway function whereby the subcellular localisation of Smo is the primary cell biological target of Ptc activity.
Modelling Smo regulation in response to Hh
Mechanistic dissection of Hh pathway activation is hampered by the large number of feedback events and regulatory inputs into the pathway. In addition, key steps within the pathway, for example, the inhibition of Smo by Ptc or the activation of stabilised Ci are not understood at the biochemical level, and have to be treated as black boxes in attempts to model the pathway. We wondered whether the Smo response to Hh could be modelled at a very abstract level while still allowing testable predictions about the behaviour of the system. We were particularly interested in any inferences that could be made about the role of Ptc from modelling, and therefore devised a simplified, formalised description of Smo behaviour in response to Hh. First, we treated the total Smo pool in the cells as four discrete populations that differ in their localisation and phosphorylation state (i.e. localised at the plasma membrane or on intracellular membranes, and a phosphorylated or nonphosphorylated cytoplasmic tail). Second, we defined these populations as being in pair-wise equilibrium with each other in exocytosis or endocytosis and phosphorylation or dephosphorylation (Fig. 1). This highly simplified treatment allowed us to focus on the position of these equilibria, subsuming all additional inputs on Smo trafficking mediated by Cos2 (Liu et al., 2007) or the nonvisual β-arrestin Kurtz (Molnar et al., 2011; Li et al., 2012) into the respective rate constants. By only considering the distribution of the Smo protein present in the cell between the four populations, we sidestepped for the moment the question of production and degradation rates. This can be justified in first approximation, because protein phosphorylation and endocytosis occur on shorter time-scales than protein synthesis or degradation. Third, we assumed that in all instances, phosphorylation has a positive feedback on Smo membrane localisation (Denef et al., 2000; Jia et al., 2003; Zhao et al., 2007), presumably through the suppression of Smo ubiquitylation and endocytosis (Li et al., 2012; Xia et al., 2012). Within this framework, we then simulated the Smo response to increasing Hh doses. Importantly, analytic treatment showed that the overall response is determined only by the ratios defining the equilibria and not the absolute values of the rate constants. As outlined above, we considered two distinct cell biological roles of Ptc: under the phosphorylation model, Ptc activity shifts the equilibria between the different Smo pools towards the nonphosphorylated forms both at the plasma membrane and within the cell (Fig. 1A). To match the observed intracellular accumulation of Smo in the absence of Hh, we had to additionally assume that trafficking of nonphosphorylated Smo is constitutively biased towards endocytosis.
Under the alternative endocytosis model (Fig. 1B), Ptc instead controls the position of the equilibrium between secretion and endocytosis of nonphosphorylated Smo. To recapitulate the observed ground state we had to additionally demand that the balance between kinase and phosphatase activities is biased towards phosphorylation at the membrane but towards dephosphorylation within the cell. Importantly, both models break down when these additional assumptions are omitted from the simulation (supplementary material Fig. S1) but capture the key features of Smo behaviour when they are incorporated: both models predict a shift from the nonphosphorylated, intracellular pool to the phosphorylated, plasma membrane pool when Ptc is gradually inactivated by increasing concentrations of Hh (Fig. 1A′–B″). Importantly, a key difference appears between the two models when Smo distribution is artificially biased towards the plasma membrane in the absence of Hh, corresponding to a pharmacological block of endocytosis using the dynamin inhibitor Dynasore (Macia et al., 2006). Under these conditions, the phosphorylation model predicts the accumulation of nonphosphorylated Smo at the plasma membrane: even though the plasma membrane pool becomes enlarged, Ptc should continue to bias the equilibrium towards the nonphosphorylated form of Smo, shifting to the phosphorylated form only when Hh is present (Fig. 2A–A″).
By contrast, the endocytosis model predicts that Smo retention at the plasma membrane is by itself sufficient to drive Smo towards the phosphorylated state: while the presumed activity of Ptc in promoting endocytosis is counteracted by the drug, the unchanged local bias towards phosphorylation at the plasma membrane is predicted to drive Smo phosphorylation even in the absence of Hh (Fig. 2B–B″). Simultaneously tracking the localisation and phosphorylation state of Smo in response to either Hh or endocytosis block would therefore allow discriminating between these models. This would, in turn, permit inferences about the cell biological process targeted by Ptc without requiring prior knowledge of the molecular mechanism. We therefore decided to generate a fluorescence-based sensor for Smo activation.
Smo-IP – a fluorescence-based sensor for Smo tail phosphorylation
We have previously shown that the conformation-sensitive cpYFP core of the Inverse Pericam (IP) Ca2+ sensor (Nagai et al., 2001) can be used to detect changes in protein interactions during the activation of signalling cascades, and have used this to selectively image active BMP receptors (Michel et al., 2011). We therefore adapted this strategy for the in vivo visualisation of Smo activation. To generate the Smo-IP reporter we replaced the bulk of the loop in the Drosophila Smo cytoplasmic tail between the Smo SAID domain (Zhao et al., 2007) and the distal, acidic patches (amino acids 757–915) with the IP cpYFP core. In the inactive state the closed conformation of the Smo cytoplasmic tail should keep the cpYFP core in a nonfluorescent conformation, whereas SAID phosphorylation should allow the IP core to relax into a fluorescent conformation (Fig. 3A). The Smo-IP construct was able to rescue amorphic smo alleles and therefore retains full signalling function (supplementary material Table S1).
Reporter fluorescence intensities under endogenous expression were too weak for practical use (supplementary material Fig. S2A). We therefore ubiquitously overexpressed Smo-IP from a tubulin promoter. Ptc is not expressed in the posterior compartment of the larval wing imaginal disc, where Smo is therefore constitutively active (Fig. 3B). In the anterior compartment, Hh pathway activation is determined by the Hh protein gradient (Torroja et al., 2005). Consistently, reporter fluorescence decreased with growing distance from the anteroposterior (AP) boundary (Fig. 3B,C). The Smo activity gradients had characteristic decay lengths of 10.8±1.8 µm (Fig. 3D and supplementary material Fig. S2B), which is consistent with previously reported ranges of the Hh gradient (Wartlick et al., 2011). However, in the wing disc Hh signalling also controls Smo protein stability. Similar to endogenous Smo, Smo-IP protein levels were therefore high in the posterior compartment, gradually decayed in front of the compartment boundary, and were low in the anterior compartment as a result of Ptc-mediated degradation (Denef et al., 2000; Li et al., 2012). Because only activated Smo is protected from degradation, reporter fluorescence in the anterior compartment necessarily closely followed protein levels (Fig. 3C). This made it impossible to unambiguously attribute the observed fluorescence to Smo phosphorylation and precluded validating reporter function in the disc by a ratiometric approach. In addition, although Hh signalling in the wing imaginal discs is quantitatively well understood (Nahmad and Stathopoulos, 2009), the disc epithelial cells are unsuitable for subcellular studies because of their pseudostratified arrangement and small diameter. To circumvent both problems, we instead turned to the larval salivary glands, whose large epithelial cells have previously been used for studies of Hh signalling (Zhu et al., 2003; Yavari et al., 2010).
Smo phosphorylation and localisation in the salivary gland
The larval salivary glands are situated adjacent to the fat body, which is a major site of Hh production and signalling (Pospisilik et al., 2010). Correspondingly, wild-type (WT) Smo-IP expressed in the salivary gland under 71B::Gal4 control was found largely at the plasma membrane and in its fluorescent state (Fig. 3E), indicating activation of the Hh pathway. This is at odds with a previous report which concluded that additional Hh expression is necessary for ptc::lacZ expression (Zhu et al., 2003). However, consistent with the presence of an endogenous Hh signal, co-overexpression of Ptc in the gland cells abolished reporter fluorescence and caused relocalisation of a large fraction of Smo from the cell surface onto internal membranes. (Fig. 3F). Both effects were reverted by overexpression of constitutively active PKA (Jia et al., 2004) (Fig. 3G), which also enhanced the phosphorylation and membrane localisation of WT Smo-IP (Fig. 3H). The apparent reduction in the level of intracellular Ptc is presumably caused by the slightly deeper imaging required as a result of the strong cell surface distortions induced by PKA. Thus, the Smo-IP reporter is switchable in a Ptc- and phosphorylation-sensitive manner. Importantly, stability of Smo protein in the glands appears to be less tightly linked to activation when compared with results in the wing disc. This allowed the validation of reporter function by a ratiometric approach, comparing reporter fluorescence and protein levels under different experimental conditions (Fig. 3I,J). The basal ratio of Smo-IP reporter fluorescence and anti-GFP immunostaining was reduced both at the plasma membrane and in the interior of the cell following co-overexpression of Ptc and increased in turn by addition of activated PKA. Coexpression of PKA also increased this ratio for WT Smo-IP (Fig. 3I and supplementary material Table S2A). This was mirrored by redistribution of the total Smo-IP pool from the membrane to the cell interior following Ptc overexpression and back to the plasma membrane after co-overexpression of PKA. The same membrane localisation of Smo-IP was achieved by expression of PKA in a WT Smo-IP background (Fig. 3J and supplementary material Table S2A). The Hh pathway in the glands is thus either endogenously only partially active or the equilibrium at saturating Hh levels is not near full Smo phosphorylation. Nevertheless, these experiments prove that the IP reporter cassette is switchable in the context of the Smo tail, and that its fluorescence reflects Hh pathway activation.
We confirmed this independently with the help of nonphosphorylatable and phosphomimetic versions of our reporter. We replaced the serine residues interspersed with the basic SAID patches (Jia et al., 2004; Zhang et al., 2004) with either aspartate (SmoSD-IP) or alanine (SmoSA-IP). These amino acid changes had been shown by FRET to force the Smo tail into an open or closed conformation, respectively, albeit without achieving subcellular resolution (Zhao et al., 2007). We expressed both wild-type and mutant Smo-IP constructs specifically in the dorsal compartment of the wing disc under ap::Gal4 control. The ventral compartment of each disc thus served as an internal control. Expression of WT Smo-IP reproduced the graded fluorescence in the anterior compartment also seen with ubiquitous expression. Overexpression of the reporter had little effect on the width of the expression domains of the Hh target genes collier (Fig. 4A) and ptc::lacZ (supplementary material Fig. S3A). By contrast, SmoSD-IP was fluorescent in the entire dorsal anterior compartment beyond the range of Hh protein (Fig. 4B). Similar to the equivalent phosphomimetic Smo versions lacking the reporter cassette (Jia et al., 2004; Zhang et al., 2004; Zhao et al., 2007), SmoSD-IP acted as a constitutive active protein driving expression of Hh target genes (Fig. 4B and supplementary material Fig. S3B). Conversely, the nonphosphorylatable SmoSA-IP reporter was nonfluorescent in both anterior and posterior compartments, even though protein levels were increased relative to the wild type (Fig. 4C). As expected (Jia et al., 2004; Zhang et al., 2004; Apionishev et al., 2005), SmoSA-IP strongly suppressed collier and ptc::lacZ expression (Fig. 4C and supplementary material Fig. S3C).
In the salivary glands, SmoSD-IP was strongly fluorescent (Fig. 4D and supplementary material Fig. S3D) and enriched at the plasma membrane (supplementary material Fig. S3E, Table S2A) regardless of Ptc overexpression (Fig. 4E). This confirms that the phosphomimetic mutations render Smo resistant to Ptc-mediated clearance from the cell surface (Zhao et al., 2007; Li et al., 2012; Xia et al., 2012). By contrast, SmoSA-IP expressed in the glands was only weakly fluorescent, indicating a nonphosphorylated, closed and inactive conformation. The remaining baseline fluorescence limits the signal to noise ratio of our reporter (Fig. 4F and supplementary material Fig. S3D).
Even under Ptc overexpression conditions, a considerable fraction of the reporter was still found outlining the plasma membrane, both for SmoSA-IP (Fig. 4F,G) and for WT Smo-IP (Fig. 3F). However, in all cases, the membrane fraction was reduced relative to WT Smo-IP under signalling conditions (Fig. 3E,I) or the phosphomimetic SmoSD-IP construct (Fig. 4D,E and supplementary material Fig. S3E). Together, these experiments show that, in the glands, the fluorescence of the conformation-sensitive IP cpYFP core is uncoupled from reporter protein levels and responds to the charge-dependent conformation of the Smo cytoplasmic tail. In addition, Smo phosphorylation itself cannot be required for the transport of Smo to the plasma membrane. Instead, the observations suggest a steady state trafficking equilibrium that is shifted towards internalisation for nonphosphorylated Smo. However, the observed behaviour of SmoSA-IP and SmoSD-IP is consistent with either of the two mathematical models (supplementary material Fig. S4), and is thus insufficient for discriminating between the alternatives. As argued above, experimentally testing the two models also requires a means of blocking Smo endocytosis.
Smo localisation in cultured cells
We first addressed Smo internalisation in cell culture experiments, where the localisation of Smo to the plasma membrane can be assayed unambiguously by immunostaining against the extracellular N-terminal domain performed under non-permeabilising conditions. S2R+ cells do not express Ci but contain endogenous Ptc and low amounts of Smo (Cherbas et al., 2011). In the absence of Hh, a C-terminally tagged Smo-GFP construct (Smo-GFP) transfected into these cells was therefore largely excluded from the cell surface (Fig. 5A,J and supplementary material Table S2B). Smo-GFP translocated to the plasma membrane following either stimulation by Hh (Fig. 5B,J) or treatment with Dynasore, a pharmacological inhibitor of dynamin-mediated endocytosis (Macia et al., 2006) (Fig. 5C,J). As expected (Jia et al., 2004), SmoSD-GFP was constitutively found at the plasma membrane (Fig. 5D–F,J), whereas the nonphosphorylatable SmoSA-GFP fusion protein remained at the detection limit at the plasma membrane of cells stimulated by Hh (Fig. 5G,H,J). However, significant amounts of SmoSA-GFP became trapped at the surface when endocytosis was inhibited by Dynasore (Fig. 5I,J).
This confirms earlier observations showing that both inhibition of dynamin-dependent endocytosis (Xia et al., 2012) and phosphomimetic mutations in the SAID-associated serines (Jia et al., 2004; Zhao et al., 2007) can enrich Smo at the plasma membrane. The observation that SmoSA can also be trapped at the membrane shows, first, that some exchange between the intracellular and plasma membrane bound pools must also occur for nonphosphorylated Smo, and second, that the forced membrane retention of WT Smo by Dynasore cannot be dependent on Smo phosphorylation.
Membrane localisation and phosphorylation of Smo
Finally, to test experimentally the two models of Smo activation, we combined the Smo reporter with inhibition of endocytosis. We treated S2R+ cells transfected with the Smo-IP reporter either with Hh to inactivate Ptc or with Dynasore to block dynamin-dependent endocytosis. As with the Smo-GFP fusion proteins (Fig. 5A–C), both treatments induced a significant increase in the levels of Smo protein detectable at the plasma membrane (Fig. 6A,B). As expected, stimulation with Hh caused increased phosphorylation of the Smo tail, which could be detected by Smo-IP reporter fluorescence. Importantly, treatment with Dynasore in the absence of ligand was equally sufficient to induce Smo phosphorylation (Fig. 6C,D).
To verify that this Hh-independent Smo activation is not merely an artefact of the pharmacological experiments in cultured cells, we turned to transgenic flies. Inactivation of Hh signalling in the salivary glands by Ptc overexpression suppressed Smo-IP fluorescence (Fig. 6E). In this background, endocytosis was inhibited by co-overexpression of a dominant-negative version of the Drosophila dynamin Shibire (UAS::shibireK44A) (Moline et al., 1999). Successful inhibition of endocytosis was reflected by Ptc accumulation at the membrane. Consistent with the cell culture results, this also induced reporter fluorescence, indicating Smo activation (Fig. 6F). Blocking Smo internalisation is therefore sufficient to induce Smo phosphorylation despite the presence of large amounts of Ptc both in vivo and in cultured cells. This challenges the traditional view that Smo membrane localisation is strictly a consequence of phosphorylation. We therefore also investigated the relationship between Smo phosphorylation and Smo clustering.
Smo clustering measured by FCCS
Phosphorylation in response to Hh has been shown by FRET microscopy to promote Smo clustering at the plasma membrane (Zhao et al., 2007). However, FRET-based techniques do not provide data on cluster size and do not possess great sensitivity. We instead used two-colour fluorescence cross-correlation spectroscopy (FCCS), which allowed us to quantitatively analyse Smo diffusion and oligomerisation under different experimental conditions (Bacia et al., 2006). This method is based on the statistical analysis of intensity fluctuations that arise when two spectrally distinct fluorophores diffuse through a microscopic detection volume (supplementary material Fig. S5A–B′). Whereas autocorrelation analysis of the individual fluorescence signals yields average dwell times and numbers of the observed particles, cross-correlation between the colour channels indicates co-diffusion of differently labelled molecules (Fig. 7A; supplementary material Fig. S5C). A useful readout is the ratio between auto- and cross-correlation amplitude, which is linked to various binding schemes and stoichiometries and can be used to measure the degree of binding (Weidemann et al., 2002; Weidemann et al., 2011).
To probe homotypic Smo interactions we generated C-terminally-tagged Smo-mRFP, SmoSD-mRFP and SmoSA-mRFP constructs analogous to the corresponding GFP fusion proteins and performed FCCS measurements in cells co-expressing green and red fluorescent Smo versions. Fluorescence was recorded at the membrane in the periphery of adherent S2R+ cells to maximise the contribution of plasma-membrane-resident proteins to the total signal (supplementary material Figs S5D, S6A–E). Co-transfection of independently membrane-anchored non-interacting GFP and RFP constructs (supplementary material Fig. S6A) fixed the baseline cross-correlation level, whereas a membrane-anchored GFP-mRFP fusion protein was used to define the maximum cross-correlation achievable under the experimental conditions (Fig. 7B and supplementary material Table S2C).
The phosphomimetics SmoSD-GFP and SmoSD-RFP exhibited strong cross-correlation, as expected for oligomerisation of constitutively active Smo at the plasma membrane (Zhao et al., 2007). Cross-correlation between the WT Smo constructs was also increased relative to the negative control, indicating a low level of Smo clustering in the absence of pathway activation. However, stimulation with Hh further increased WT Smo crosscorrelation, comparable to but weaker than the levels seen for the SmoSD constructs. This difference suggests that in contrast to the constitutively open phosphomimetic construct, a fraction of the total WT Smo pool exists as monomers or dimers even under Hh-induced signalling conditions, presumably in a nonphosphorylated state. Thus, the kinase and phosphatase equilibrium at the surface cannot be fully shifted towards phosphorylation and clustering. Steady-state levels of SmoSA at the surface were low in comparison to the corresponding WT and SmoSD constructs (supplementary material Fig. S6B–E), reflecting the preferential partitioning of nonphosphorylatable Smo to intracellular membranes (Fig. 4F). In the absence of ligand, SmoSA constructs showed clustering at the level of non-stimulated WT Smo. Importantly, treatment with Hh also induced significant oligomerisation of SmoSA that is comparable with the WT response (Fig. 7B and supplementary material Table S2C). As expected, blocking endocytosis with Dynasore, which was sufficient to trap Smo on the plasma membrane and induce phosphorylation (Fig. 5C), also increased Smo clustering (Fig. 7B and supplementary material Table S2C).
Finally, the maximum expected cross-correlation signal for homotypic dimerisation events is limited to roughly one third of the corresponding value of a dual-coloured fusion protein (Weidemann et al., 2002). The observed cross-correlation levels therefore indicate higher-order stoichiometries than dimers, but precise quantification of cluster sizes is difficult. Cross-correlation fractions suggest an average cluster size of around two Smo molecules in the absence of Hh, increasing to about five following Hh exposure (supplementary material Fig. S7A). However, these values are known to underestimate the true size of the oligomers, as the model function assumes ideal conditions that are typically not achieved under live-cell conditions (Weidemann et al., 2002). Smo cluster sizes in the range of tens of molecules are also consistent with the observed changes in diffusion times (Ramadurai et al., 2009), albeit with large uncertainty margins for both approaches (supplementary material Fig. S7B and Table S2C). In summary, as expected from the literature (Zhao et al., 2007), clustering of Smo is induced by the open, phosphorylated conformation. However, phosphorylation is per se not required for Smo oligomerisation, because Hh also induces clustering of SmoSA at the plasma membrane.
Previous observations have shown that phosphorylation of the cytoplasmic tail exerts positive feedback on Smo membrane accumulation (Jia et al., 2004; Zhao et al., 2007), presumably by regulating the Smo ubiquitylation state and subsequent endocytosis (Li et al., 2012; Xia et al., 2012). These studies have also provided evidence that interfering with Smo ubiquitylation or endocytosis leads to increased Smo membrane levels, Ci stabilisation and Hh target gene expression (Li et al., 2012; Xia et al., 2012). This was generally interpreted as affecting Smo protein stability by preventing internalisation and proteolytic degradation, thereby increasing the concentration of membrane resident, active Smo. These observations can thus be condensed into a simplified ‘phosphorylation’ model of Hh signal transduction, whereby Hh signalling first leads to Smo phosphorylation and translocation to the membrane. Smo phosphorylation then prevents clearance of the active Smo pool by counteracting ubiquitylation and endocytosis.
However, other studies have shown that modulation of the lipid environment can govern the subcellular localisation of Smo (Khaliullina et al., 2009; Yavari et al., 2010). Specifically, artificially increased PI4P levels can force Smo localisation to the membrane and activation of the Hh pathway in the absence of Hh. Intriguingly, inactivation of Ptc is sufficient to increase levels of this phospholipid (Yavari et al., 2010). These observations support an ‘endocytosis’ model of Hh pathway activation, whereby inactivation of Ptc primarily affects Smo redistribution to the plasma membrane, presumably by regulating the local lipid content of either the plasma membrane or Smo-containing endosomes.
We have developed a genetically encoded, fluorescence-based reporter for Smo phosphorylation. By simultaneously tracking Smo localisation and activation at the subcellular level we could demonstrate that enforced membrane localisation is sufficient to drive Smo phosphorylation, irrespective of the presence of Ptc. With the help of mathematical simulations, we show that this response is compatible with the endocytosis model but not the phosphorylation model. Importantly, the endocytosis model demands that the balance between the kinases and phosphatases acting on Smo differs between plasma membrane and internal membranes. Intriguingly, spatially differentiated regulation of enzyme activity at the subcellular level is a hallmark of the PKA system, at least as far as the membrane-associated AKAP-anchored type II isoforms are concerned (Wong and Scott, 2004). We therefore propose that Ptc controls activation of the Hh pathway by regulating access of the Smo substrate to the enzymes (kinases or phosphatases), thereby regulating its phosphorylation state, rather than by controlling the activity of the enzymes per se. To test this hypothesis, additional tools will have to be developed to directly measure PKA activity with subcellular resolution and independent of Smo. However, the assumption of asymmetric kinase to phosphatase equilibria required by our model helps to reconcile the two apparently contradicting mechanisms of Hh pathway activation. Although it is formally possible that Ptc regulates localisation and phosphorylation independently by separate downstream mechanisms, we would instead like to propose that Hh-induced translocation of Smo to the membrane by means of lipid modification (Khaliullina et al., 2009; Yavari et al., 2010) causes phosphorylation (Jia et al., 2004; Zhang et al., 2004; Apionishev et al., 2005) and opening (Zhao et al., 2007) of the Smo tail. This would, in turn, prevent Smo ubiquitylation and endocytosis (Li et al., 2012; Xia et al., 2012), leading to downstream signal transduction.
Nevertheless, our observations suggest that both active and inactive Smo are continuously shuttling between intracellular compartments and the plasma membrane. First, even the inactive SmoSA variant can readily be trapped at the cell surface by blocking endocytosis, although its detection at the membrane under normal conditions requires single-molecule sensitivity. Second, even though activation of Smo appears to occur at the membrane, the ratio of reporter activity to GFP level co-varies in the intracellular and plasma-membrane-associated fractions. Phosphorylation of the cytoplasmic tail is known to be sufficient for Smo clustering (Zhao et al., 2007). However, our FCCS experiments have shown Hh can also induce Smo clustering independent of phosphorylation. Thus, some unidentified activity of Ptc appears to directly inhibit Smo clustering. Future experiments must test whether this role is distinct from the regulation of Smo localisation, or whether these activities represent two aspects of the same molecular function. Finally, the higher quantitative resolution of FCCS compared with FRET approaches showed that SmoSD (where all monitored molecules are expected to be in the open conformation) shows significantly higher crosscorrelation than WT Smo exposed to Hh. The latter pool must therefore contain a significant fraction of monomeric Smo molecules, suggesting that the position of the equilibrium between kinase and phosphatase activities yields a considerable fraction of nonphosphorylated Smo even in the presence of Hh. Importantly, this does not affect the conclusions of our analytically robust modelling approach.
Smo is related to the GPCR family of signalling receptors, and it is therefore interesting to note that recently the formation of higher-order complexes has been recognised as a major mode of GPCR regulation. Consistently, the observed degree of cross-correlation of phosphorylated Smo cannot be explained by simple homodimerisation, but must involve the formation of higher-order oligomers (Worch et al., 2010; Weidemann et al., 2011). Intriguingly, while we were finishing this manuscript a study has shown, by an unrelated biochemical approach, that Hh can induce Smo oligomerisation within lipid rafts, and that this is required for signal transduction (Shi et al., 2013). It will therefore be interesting to test whether Smo clustering is related to downstream signalling through the G-protein cascade (Ayers and Thérond, 2010). We believe that our approach of combining the direct visualisation of the activation state of individual pathway components through fluorescent sensors with microscopy-based biophysical techniques shows great promise for uncovering the cell biological machinery underlying intercellular communication, both for the Hh pathway and other signalling cascades.
Materials and Methods
UAS::ptc (Martín et al., 2001), UAS::ptc1130x (Johnson et al., 2000), UAS::shibireK44A (Moline et al., 1999), ap::GAL4 (Calleja et al., 1996), 71B::GAL4 (Brand and Perrimon, 1993), dpp::LacZBS3.0 (Blackman et al., 1991), ptc::LacZ (Chen and Struhl, 1996), smo2 and smo3 (Nüsslein-Volhard et al., 1984) have all been described.
Transgenes and plasmid constructs
The SmoIP reporter was generated by replacing aa 757–915 of the Smo cytoplasmic tail with the cpYFP core of Inverse Pericam (Nagai et al., 2001) via fusion PCR. The SmoSA-IP and SmoSD-IP mutants were subsequently created by replacing the serines at Smo aa positions 667, 670, 673, 687, 690, 693, 740, 743 and 746 by fusion PCR with alanines and aspartates, respectively. Transgenic flies were generated by BestGene (Chino Hills, CA) or at the MPI-CBG, Dresden. To generate pUAS::SmoIP the reporter was inserted into pWRpA (gift from N. Brown, Gurdon Institute, Cambridge). pCaSpeR-tub::Smo-IP was generated by inserting smo-IP into pCaSpeR-tub. To create pUAST-SmoGFP and pUAST-SmoRFP, the eGFP and mRFP ORFs were fused to the C-terminus of Smo by fusion PCR and inserted into the pUAST vector. The SmoSA-RFP/GFP and SmoSD-RFP/GFP versions were derived analogous to the corresponding Smo-IP constructs. pUAST::memGFP, pUAST::memRFP and pUAST::memGFP-RFP were generated by fusing the respective ORFs to a Lyn palmytoylation/myrystoylation site (plasmid was a gift from G. Weidinger, Ulm) by fusion PCR.
Immunocytochemistry and microscopy
Established procedures for immunostaining of imaginal discs and testes (Michel et al., 2011; Michel et al., 2012) were followed also for salivary glands and cells. Primary antisera were then applied overnight at 4°C at the following dilutions: anti-GFP (Clontech 632460; 1:500), anti-β-galactosidase (Promega Z3781; 1:1000), anti-Col (1:100) (Dubois et al., 2007), anti-Sal (1:1000), (DSHB) anti-Smo (1:100) (Lum et al., 2003), anti-Ptc (1:100) (Capdevila et al., 1994), anti-En (1:100) (Patel et al., 1989) and anti-Ci (1:10) (Motzny and Holmgren, 1995). Secondary antisera (Santa Cruz) were used at 1:500 dilution. Confocal images were collected using Leica SP5/II or Zeiss LSM780 microscopes and processed with ImageJ/Fiji. For quantification, images were collected under nonsaturating conditions for each experimental setting. Membrane and intracellular areas were manually outlined in the anti-GFP channel on central sections through salivary gland cells intersecting the nucleus. After subtracting the nuclear signals in each channel as background, the average intensities (in a.u.) of the anti-GFP and reporter fluorescence channels were separately measured and their ratio averaged for 8–10 cells. To estimate the membrane-bound fraction, the average intensities and areas of the intracellular and membrane ROIs were multiplied, the membrane value divided by two to account for the contribution from neighbouring cells that cannot be resolved by light microscopy, and the ratios determined for the individual cells averaged. Assuming a roughly cuboidal geometry for the gland cells, this is a conservative estimate of the membrane-associated fraction, because the bottom and top surfaces are not taken into account. Observed differences were tested for significance by ANOVA followed by Tukey's HSD post-hoc test.
Insect cell culture
Drosophila S2R+ cells (a kind gift from Elisabeth Knust, MPI-CBG, Dresden) were cultured at 25°C, without CO2 in Schneider's Drosophila medium with L-Glutamine (Invitrogen or PAN BIOTECH) supplemented with 10% fetal bovine serum (FBS, Invitrogen) and transfected using the calcium phosphate method (Graham and van der Eb, 1973; Chen and Okayama, 1987).
Blocking endocytosis by Dynasore and ligand stimulation
Transiently transfected S2R+ cells were incubated with 25 µM Dynasore (Sigma-Aldrich) for up to 1 hour before imaging or fixation. Hh conditioned medium was generated by incubating S2R+ cells transfected with UAS::Hh and Actin5C::Gal4 plasmids for 6 days. Cells were stimulated by addition of conditioned medium 12 hours before fixation or imaging.
Sample preparation for FCCS
Cells were seeded into concanavalin-A-coated eight-well LabTek chambers (no. 1.5, 0.16–0.19 mm, Thermo Scientific) and allowed to adhere to the substrate for 1 hour. Before measurements, the medium was replaced with air buffer (20 mM HEPES, pH 7.4, 150 mM NaCl, 15 mM glucose, 20 mM trehalose, 0.15 mg/ml BSA, 5.4 mM KCl, 0.85 mM MgSO4, 0.75 mM CaCl2).
FCCS data acquisition
FCCS was performed at room temperature using a Zeiss LSM780 microsocope with a ConfoCor3 module and a Zeiss C-Apochromat 40×, N.A. 1.2 objective (Carl Zeiss). Fluorescence was recorded using avalanche photodiodes. Instrument settings were optimised before each session for maximum particle brightness using 25 nM solutions of Alexa Fluor 488 (Life Technologies) and CF568 (Biotium) dyes.
FCCS data analysis
Here, Veff denotes the effective focal detection volume, c the fluorophore concentration, τdiff,i and fi the dwell times and molar fractions of the two diffusion species, and τT and fT the lifetime and fraction of molecules in the dark state. The GFP autocorrelation and the cross-correlation curves were fitted with a two-component diffusion model (i = 2), whereas in the mRFP channel, one diffusion component (i = 1) was sufficient. Since blinking contributions of mRFP showed significant scatter between individual measurements (Hendrix et al., 2008), we fixed our experimental average value to 300 µseconds for evaluation of the particle numbers N.
For example, evenly expressed GFP and mRFP constructs (pg = 1/2) and a simple saturated dimerisation (n = 2) will produce CCg = 1/3. Uneven expression slightly increases this ratio but cannot exceed 1/2. Thus, measured CCg>1/2 for homotypic interactions indicate higher order stoichiometries (n>2). Differences between crosscorrelation levels were tested for significance by ANOVA followed by Tukey's HSD post-hoc test.
Mathematical modelling of Smo activation
We considered the following two non-dimensionalised models.
To visualise the general model behaviour we have chosen the following particular parameter values listed in Table 1. We confirmed independence of qualitative model behaviour from the particular choice of parameter by varying each parameter tenfold (data not shown).
NA, not applicable.
To simulate experimental manipulations, particular parameter values were set to zero. Thus, for SmoSA-IP: , for SmoSD-IP: and for Dynasore treatment: . Simulations were performed with the help of two independent modelling and simulation tools, Copasi (Hoops et al., 2006) and PottersWheel (Maiwald and Timmer, 2008). Results coincided and the diagrams show the steady state solutions as a function of Hh stimulation level. Our new endocytosis regulation model is available as SBML file for further analysis and use independent of specific simulation software.
We thank Nicholas H. Brown, Michele Crozatier, Christian Dahmann, Suzanne Eaton, Elisabeth Knust, Gilbert Weidinger, the Bloomington stock centre and the DSHB for providing flies and reagents, Raquel Pérez-Palencia for expert technical assistance and members of the Junior European Drosophila Investigators (JEDI) initiative for discussions.
A.K. and M.M. performed the Drosophila experiments; I.R. performed the cell culture and FCS experiments and analysed the FCS data together with T.W.; D.A. generated the proof of principle reporter construct; L.B. performed the mathematical modelling; C.B. initiated the project and wrote the manuscript.
The project was supported by the Center for Regenerative Therapies Dresden (CRTD); the Federal Ministry of Education and Research (BMBF) Consortium Mesenchymal Stem Cells; and the Deutsche Forschungsgemeinschaft [grant number BO 3270/2-1 to C.B.].