Ca²⁺ coordination controls sonic hedgehog structure and its Scube2-regulated release Free

Communication between cells requires dedicated signaling molecules. For the many ligands that are synthesized as membrane-tethered forms and therefore cannot diffuse, the range over which they can operate is limited to adjacent cells. Induced solubilization of some of these ligands enables additional paracrine and endocrine signaling communication between nonadjacent cells. Ectodomain shedding is one way to release the extracellular domain of membrane-associated molecules by proteolysis at or close to the cell membrane (at juxtamembrane sites). This process and the proteases involved, called sheddases, have therefore become recognized as important modulators of cell signaling. The mechanisms by which sheddase activation and substrate specificity are established and modulated are being increasingly investigated, and are the focus of this study.

Membrane-associated, dual-lipidated Sonic hedgehog (Shh) morphogens are one prominent family of signaling ligands that can be released by sheddases. Production of Shh begins with autocatalytic cleavage of a 45 kDa precursor molecule linked to the addition of a cholesteryl moiety to glycine 198 of the N-terminal Shh cleavage product (Porter et al., 1996) (Fig. 1A). This reaction is catalyzed by the C-terminal cholesterol transferase domain and tethers Shh to membranes of producing cells. Before its export to the cell surface, a palmitoyl group is attached to the N-terminus of the 20 kDa N-terminal signaling domain. This requires the expression of a separate gene product called Hh acyltransferase (Hhat) (Amanai and Jiang, 2001; Buglino and Resh, 2008; Chamoun et al., 2001; Lee and Treisman, 2001; Micchelli et al., 2002). Shh palmitoylation is special in that the palmitate is irreversibly attached via an amide bond to the α-amino group of the N-terminal cysteine, in contrast to reversible O-acylation, which targets the serine hydroxyl side chain in Wnt proteins (Takada et al., 2006), or S-acylation, which targets the thiol side chain in nearly all other palmitoylated proteins (Buglino and Resh, 2008; Pepinsky et al., 1998). Importantly, N-palmitoylation regulates Hh biofunction in most, but not all, biological systems. In invertebrates, mutation of the N-terminal palmitate acceptor cysteine to serine or alanine (C>A/S) or acyltransferase deficiency completely abolishes all Hh activity (Chamoun et al., 2001; Pepinsky et al., 1998). By contrast, ectopic overexpression of nonpalmitoylated Shh in vertebrate embryos can induce specific gain-of-function phenotypes, such as polydactyly, which, although less severe than phenotypes produced by Shh overexpression, indicate tissue-specific roles of palmitate in Shh biology (Chen et al., 2004; Kohtz et al., 2001; Lee et al., 2001; Williams et al., 1999). Consistent with this, loss of Hhat function in vertebrate embryos causes developmental malformations characteristic of only partially defective Shh signaling (Chen et al., 2004). Understanding this discrepancy between invertebrates and vertebrates, despite the otherwise conserved Hh reception and signaling, is key not only to deciphering the role of palmitate in Hh biology, but also to explaining its congruency with the various proposed models of Hh release and transport.

Shh release and transport have been suggested in unprocessed lipidated form, possibly aided by Scube2 [signal peptide, CUB domain, epidermal growth factor (EGF)-like protein 2] (Fig. 1A). Scube2 is a member of the you class mutants in zebrafish (Schauerte et al., 1998) and plays key cell nonautonomous roles in Shh release and signaling in vivo (Hollway et al., 2006; Johnson et al., 2012; Woods and Talbot, 2005). Notably, a Scube2 truncation mutant lacking the C-terminal cysteine-rich and CUB domains (ΔCUB) is inactive (van Eeden et al., 1996). CUB domains derive their name from the complement subcomponents C1r/C1s, sea urchin protein with EGF-like domains (UEGF), and bone morphogenetic protein 1 (BMP1), and contribute to protease activities in these proteins (Gaboriaud et al., 2011), possibly by inducing structural changes which boost turnover (Bourhis et al., 2013). In agreement with this established CUB function, Scube2 was also shown to strongly enhance proteolytic removal (shedding) of both terminal lipidated Shh peptides in a CUB-dependent manner (Jakobs et al., 2014, 2016), which solubilizes the morphogen in vitro (Damhofer et al., 2015; Dierker et al., 2009; Ohlig et al., 2011). Shh shedding itself requires expression of A Disintegrin and Metalloproteinase family members 10 (A10) and 17 (A17) (Damhofer et al., 2015; Dierker et al., 2009; Ohlig et al., 2011), two major sheddases known to cleave numerous substrate ectodomains from the cell surface. Importantly, because unprocessed palmitoylated peptides sterically inhibit Shh binding sites for the receptor Patched (Ptc), their proteolytic processing not only releases Shh but also unmasks these sites and thereby couples Shh solubilization with its bioactivation. Coupled Shh release and activation by N-terminal processing thus explains the importance of N-palmitoylation in two ways. First, N-palmitate facilitates shedding by positioning the N-terminal inhibitory peptide close to the cell surface (Baran et al., 2013; Scheller et al., 2011; Zhao et al., 2001). Second, with the continued membrane association of these N-terminal peptides, palmitate limits possible modes of morphogen solubilization to shedding and assures completion of this process. As a consequence, lack of acyltransferase activity in mutant cells or animals generates soluble inactive morphogens with (most) Ptc binding sites still blocked (Fig. 1A).

This work provides mechanistic insight into the regulation of this process. We show that Scube2 CUB domains are required for Shh processing and bioactivation, and that this requirement is overcome by the artificial increase of the length and accessibility of N- and C-terminal Shh peptides. This suggests that CUB domains regulate Shh cleavage site accessibility in the normal situation. Consistent with this, site-directed inactivation of Shh Ca²⁺ complexation also renders Scube2-regulated Shh shedding CUB independent. Thus, this work also reveals a new role of Ca²⁺ coordination in maintaining Shh in a resting, protease-resistant conformation, and shows that Scube2 converts this conformation into a relaxed, protease-accessible form. Finally, we show that Scube2 uncouples Shh processing from the otherwise strict requirement of membrane association of sheddase targets. This allows for the generation of truncated bioactive morphogens from their unlipidated inactive precursors and thereby explains the conundrum that nonpalmitoylated Shh shows variable bioactivity in vertebrates, whereas Hh from invertebrates that lack Scube orthologs does not. Therefore, our findings provide mechanistic insight into Scube2 domain functions and strongly support Scube2-enhanced Shh shedding over alternative modes of morphogen transport in unprocessed inactive form.

Low-level Shh shedding, called basal shedding, occurs in nonstimulated cells (Damhofer et al., 2015). It is dramatically increased by external signals converging on sheddases and their substrates, in a process called activated shedding. One Shh sheddase activator is Scube2 (Jakobs et al., 2014). To characterize its activator function, we first established optimal time points for Scube2-enhanced Shh solubilization from expressing Bosc23 cells. Relative amounts of cellular and soluble material were determined after 0, 6 and 16 h by SDS-PAGE and immunoblotting (Fig. 1B and Fig. S1A). We found that Scube2 concentration dependently (Fig. S1B) enhanced Shh release by ∼10-fold after 6 h; however, the background release and amounts of Shh released in the presence of Scube2 converged into more similar levels after 16 h (Fig. 1B). Therefore, throughout this work, conditioned media were collected after 6 h unless otherwise stated.

As shown in Fig. 1C, another important determinant of Shh release is its dual membrane association, because baseline shedding of Shh^C25S lacking the N-terminal palmitate acceptor amino acid was strongly enhanced (Fig. 1C and Fig. S1C). This was previously noted (Chamoun et al., 2001; Konitsiotis et al., 2015), and indicates that palmitoylated Shh N-termini control Shh release. We hypothesized that their regulated proteolytic removal is a prerequisite for subsequent C-terminal processing and Shh solubilization. Consistent with this, basal shedding releases Shh^C25S in N-terminally unprocessed form, as indicated by its unchanged electrophoretic mobility (Fig. 1C, compare the cellular material with corresponding media). By contrast, electrophoretic mobility of soluble (Scube2-released) Shh is increased as a consequence of additional N-terminal peptide processing (Fig. 1C). Notably, in the presence of Scube2, we also observed increased electrophoretic mobility of a fraction of Shh^C25S (Fig. 1C). Restored Shh^C25A bioactivity in the Shh-responsive cell line C3H10T1/2 (Nakamura et al., 1997) (Fig. S1D) is in line with the concept that unprocessed N-termini sterically inhibit Shh binding sites for the receptor Ptc and that their proteolytic removal unmasks these sites (Ohlig et al., 2011) (Fig. S2 provides a detailed description of this concept). Our observations thus suggest that Scube2 renders processing of N-terminal target peptides independent of their membrane-proximal positioning.

To test this possibility, we C-terminally inserted hemagglutinin (HA) peptides, resulting in Shh^HA and unlipidated Shh^C25A;HA with the C-terminal extended membrane anchor N¹⁹⁰SVAAKSG-YPYDVPDYA-G¹⁹⁸ (G¹⁹⁸ represents the cholesterol-modified glycine; underlined italicized letters represent the tag) (Fig. 2A, schematic) (Jakobs et al., 2014). We also used ^HAShh^HA carrying the additional N-terminal modification C²⁵GPGRGFG-YPYDVPDYA-KRRHPKK³⁹ [C²⁵ represents the palmitoylated cysteine; KRRHPKK is the previously determined N-terminal Cardin-Weintraub (CW) processing site] (Jakobs et al., 2014; Ohlig et al., 2012). Tagged and untagged proteins were expressed in Bosc23 cells, and relative amounts of cellular and soluble material were compared after 6 h in the absence of any Scube2 stimulation to determine baseline release (Fig. 2B–D). We also tested morphogen release by the A Disintegrin and Metalloproteinase activity enhancer ionomycin (IM) or methyl-β-cyclodextrin (MβCD). MβCD forms tilted stacks at the membrane surface to allow cholesterol to immerse into their hydrophobic interior (López et al., 2011, 2013). Importantly, this mechanism makes direct extraction of the Shh C-cholesterol impossible because the required position for the MβCD stack is occupied by the C-terminal Shh peptide. Instead, MβCD extraction of free membrane cholesterol changes membrane fluidity and the distribution of sheddases and their substrates, and thus increases the probability of their nonphysiological encounter (Murai et al., 2011).

As shown in Fig. 2B, we observed increased unstimulated (background) release of Shh^HA, Shh^C25A;HA and ^HAShh^HA over that of Shh and Shh^C25A/S, despite similar cellular expression [compare the cellular material (c) with the corresponding soluble proteins]. This observation demonstrates facilitated protease access to engineered terminal peptides in unstimulated cells. Stimulation by 2.5 µM IM for the same time period enhanced ^HAShh^HA processing and release, and moderately increased solubilization of the other forms. This confirms that A Disintegrin and Metalloproteases contribute to Shh solubilization (Damhofer et al., 2015; Dierker et al., 2009; Ohlig et al., 2011). Scube2 enhanced the proteolytic processing of Shh, and even more strongly enhanced the proteolytic processing of Shh^HA, Shh^C25A;HA and ^HAShh^HA, as indicated by the detection of truncated proteins in conditioned media (Fig. 2C; see Fig. S3 for a detailed explanation). Additional stimulation by 2.5 µM IM completed proteolytic processing of all forms, as again indicated by the 19 kDa product. By contrast, MβCD increased Shh and Shh^C25A/S N-terminal processing (Fig. 2D), but did not enhance terminal processing of Shh^HA, Shh^C25A;HA and ^HAShh^HA over background levels [compare bottom bands in lane 1 (Shh) with those in lanes 2–4]. Protein quantification revealed that Scube2 increased Shh and Shh^HA solubilization >10-fold, but MβCD increased only Shh solubilization (but not Shh^HA release) >20-fold (Fig. 2E). Notably, we found that all Scube2-released proteins were bioactive, demonstrating that N-terminal processing, and not N-palmitate per se, controls Shh biofunction (Fig. 2F).

We next examined how the multi-domain protein Scube2 specifies Shh processing at the cell surface. Like its homologs Scube1 and Scube3, Scube2 consists of a signal peptide for secretion followed by nine EGF domains linked by a spacer domain to a cysteine-rich/C-terminal CUB domain (Fig. 3A). The spacer domain has been shown to mediate HS binding and Scube recruitment to sites of Shh processing and release (Jakobs et al., 2016; Liao et al., 2016). However, the functions of the EGF domains in Shh release had remained unexplored. Therefore, we compared the Scube2 sheddase enhancer function for Shh substrates with that of a Scube2 variant that lacks all EGF domains (ΔEGF) by SDS-PAGE and immunoblotting. Polyclonal anti-Shh antibodies detected full-length and truncated Shh substrates, and anti-FLAG antibodies detected Scube2 and ΔEGF on the same (stripped) blot (Fig. 3B). After 6 h, Scube2 strongly enhanced the release of membrane-associated Shh and artificial variants Shh^HA, ^HAShh^HA, noncholesteroylated ShhN and Shh^C25A;HA over background levels [compare lanes 1 (empty vector; EV) and 2 (Scube)]. As shown earlier, processing is indicated by an electrophoretic size shift (compare the cellular material with corresponding media), which is incompatible with Shh extraction or membrane transport, which should leave its size unaffected. On average, Shh solubilization increased ∼8-fold in the presence of Scube2 compared with Shh amounts released in its absence (set to 100%). This activity strictly depended on one or more of the nine EGF domains because ΔEGF lacking these domains released much less Shh (Fig. 3B,C), despite its unimpaired expression. The same ΔEGF inactivity was observed for all other Shh variants tested (Fig. 3B,C). To rule out the possible explanation that ΔEGF was not located at the cell surface, we stained Scube2-transfected Bosc23 cells under nonpermeabilizing conditions with anti-FLAG antibodies and quantified the binding by fluorescence activated cell sorting (FACS). FACS confirmed comparable Scube2 and ΔEGF association with the Bosc23 cell surface (Fig. S4) (Jakobs et al., 2016). We next ruled out any possible indirect EGF domain function in Shh release by indirect EGF receptor (EGFR)-mediated signaling regulation of cell growth, cell cycle and apoptosis. To this end, we lysed Scube2- and ΔEGF-expressing Bosc23 cells and performed a sandwich immunoassay to detect EGF-mediated phosphorylation of 16 proteins predominantly belonging to the Akt signaling network (Fig. S5). No significant differences in the phosphorylation of any of the 16 proteins were determined after 10 min or overnight stimulation with Scube2/ΔEGF-conditioned medium, suggesting that Scube2 EGF domains regulate Shh shedding directly at the cell surface.

We used light microscopy to test this possibility. Bosc23 cells were grown in slide chambers and transfected, fixed and probed with primary antibodies directed against Shh and the established Shh sheddase A17 (Damhofer et al., 2015; Dierker et al., 2009; Ohlig et al., 2011). This was followed by incubation with two sets of secondary antibodies conjugated with specific oligonucleotides. Subsequent oligonucleotide ligation by a bridging probe in a proximity-dependent manner (40 nm being the upper limit of the interactions) was followed by rolling-circle amplification to visualize Shh/A17 interactions as fluorescent spots. The specificity of the reactions was confirmed by using only one primary antibody in conjunction with both secondary proximity ligation assay (PLA) probes (Fig. 4A,B), or both antibodies, on mock-transfected cells (Fig. 4C).

As shown in Fig. 4D, relatively few individual PLA signals were obtained from confocal scans of direct Shh and A17 interactions in the absence of Scube2. These interactions were highly variable between and within experiments, indicating that A17/Shh interactions were only transient or weak. By contrast, Scube2 increased direct interactions between Shh and A17 at the surface of Bosc23 cells (merged stacks from confocal scans are shown in Fig. 4E,E′). Co-transfection of a Scube2 variant lacking its C-terminal CUB domain (ΔCUB) also yielded significantly increased, yet variable, PLA signals (Fig. 4F,H). By contrast, as shown in Fig. 4G,H, we consistently observed only low levels of Shh/A17 colocalization in the presence of ΔEGF. We suggest that the most likely explanation for these findings is that Scube2 EGF domains colocalize sheddases with their substrate as a prerequisite for Shh release and signaling from the producing cell surface.

To determine the role of the cysteine-rich domain/CUB domain in Shh release, we performed the same protease-substrate-matching analysis by SDS-PAGE and immunoblotting as described earlier (Fig. 3B). We confirmed that, like ΔEGF, ΔCUB lacking the C-terminal cysteine-rich and CUB domains (Fig. 5A) was less effective in Shh release in vitro (Jakobs et al., 2014), consistent with its diminished upstream Shh activity regulation in vivo (Kawakami et al., 2005) (Fig. 5B,C). By contrast, Scube2-enhanced shedding of the artificial variants Shh^HA, ^HAShh^HA, ShhN and Shh^C25A;HA depended much less on the C-terminal CUB domain [Fig. 5B,C; compare lanes 2 (Scube2) and 3 (ΔCUB)]. This finding precludes the possibility that the CUB domain directly extracts Shh cholesterol from the membrane, because all cholesteroylated Shh variants were CUB independently released. Instead, Scube2-enhanced yet CUB-domain-independent processing of extended terminal peptides supports the idea that, after EGF domain-mediated sheddase/substrate coupling, CUB domains somehow facilitate protease access to otherwise inaccessible target peptides. The important question arising from this situation is how target accessibility of sheddase substrates is modulated by the Scube2 CUB domain.

Established mechanisms of shedding regulation at the protease level include intracellular protease trafficking, polarized secretion and specific activation of the protease, intracellular signals to induce post-translational modifications, structural changes of protease regulatory domains by redox regulation, and interaction with inhibitors and transmembrane regulators, such as integrins and tetraspanins (Hartmann et al., 2013; Hayashida et al., 2010). However, these mechanisms do not explain Scube2 CUB-domain specified Shh cleavage, which we hypothesized could be determined by structural features of the substrate (Hartmann et al., 2013; Stawikowska et al., 2013). Indeed, we and others noted that Shh harbors two Ca²⁺ ions that directly influence the surface surrounding their coordination site (Hitzenberger and Hofer, 2016; McLellan et al., 2008; Rebollido-Rios et al., 2014). As shown in Fig. 6A, the overlaid Shh crystal structures 3D1M [two Ca²⁺ ions bound (McLellan et al., 2008) and 3m1n (no Ca²⁺ bound) (Pepinsky et al., 2000)] supported increased flexibility of loop amino acids close to, or as part of, the Ca²⁺ coordination site (Ca²⁺ ions are labeled as blue spheres, maximum flexibility amino acids are red, and the most affected loop area is highlighted by a blue oval). Such changes in Shh ectodomain structure are facilitated by the absence of any cysteine bridges in the molecule (Hall et al., 1995). However, 3D1M was crystallized together with its ligand CDOFn3. This raised the possibility that the observed distance shift of loop amino acids was caused by the ligand and was independent of Ca²⁺ coordination. To rule out this possibility, we generated a Shh variant lacking the Ca²⁺-coordinating residues E90, E91 and E127 (Shh^ΔCa2+) to confirm the predicted change in the molecular surface in vitro. To this end, we first tested whether lack of Ca²⁺ coordination affected Shh multimerization and heparan sulfate (HS) association at the cell surface, both of which are required for proper signaling. Gel filtration chromatography failed to reveal significantly changed multimerization of Shh^ΔCa2+ (Fig. 6B, top). To test for the possibility that heparan sulfate proteoglycan (HSPG) binding of Shh^ΔCa2+ was impaired, we isolated HS from mouse embryos and coupled the material to HiTrap columns for subsequent fast protein liquid chromatography (Fig. 6B, bottom). Shh and Shh^ΔCa2+ were loaded onto the HS-coupled column, and proteins were eluted by increasing salt concentrations. Both proteins bound HS and eluted at 0.9 M NaCl from the column, revealing that HS binding was not significantly affected by Ca²⁺.

In contrast to the determination of such gross changes in molecular behavior, our structural comparison (Fig. 6A), together with the calculations of others (Hitzenberger and Hofer, 2016; Rebollido-Rios et al., 2014), predicted that the subtle distance shift of loop residues H133 and H134 may affect conformation-dependent Shh binding to the receptor Ptc (Bosanac et al., 2009), in turn affecting its function. Likewise, we predicted that the function-blocking activity of monoclonal, conformation-dependent 5E1 antibody would be affected. This antibody binds Shh residues H133/H134 and K87/R153/E176/K178 (Maun et al., 2010). To confirm these predictions, we again used the Shh-responsive (and therefore Ptc-expressing) cell line C3H10T1/2 (Nakamura et al., 1997). We first established that Shh-dependent differentiation of this cell line into alkaline phosphatase-producing osteoblasts was strongly reduced by 35 µg/ml 5E1 [Fig. 6C: Shh, 2.9±0.2 arbitrary units (au) (n=4); Shh+5E1, 0.35±0.02 au (n=4), P<0.0001; empty vector, 0.17±0.002 m au (n=8)]. As a control, we also determined that 5E1 inhibition was strongly reduced when targeted mutagenesis of Shh residue K87 implicated in its binding was performed (Maun et al., 2010), demonstrating that 5E1 inhibition depends on this residue [Fig. 6C: Shh^K87A, 2.87±0.24 au (n=3); Shh^K87A+5E1, 1.9±0.33 au (n=3)]. Confirming the predicted structural rearrangement of K87 in the absence of Ca²⁺ [Fig. 6A and Hitzenberger and Hofer (2016)], lack of Ca²⁺ coordination also decreased 5E1 binding to Shh^ΔCa2+ (Maun et al., 2010). For the same reason, Ptc binding of Shh^ΔCa2+ was also reduced [Fig. 6C: Shh^ΔCa2+, 1±0.06 au (n=8); Shh^ΔCa2++5E1, 0.53±0.02 au (n=4)]. Together, these results show that Ca²⁺ ions affect the surface topology of Shh in vitro, and that the structural rearrangement repositions residues H133, H134 and K87, as predicted from our in silico analysis.

Supporting our hypothesis that such a structure shift may also affect Shh processing at the cell surface, we observed that Scube2-dependent and Scube2-independent background release of Shh^ΔCa2+ was twofold increased over that of Shh, despite comparable cellular expression (Fig. S6A). This situation resembles increased background release of HA-tagged dual-lipidated or monolipidated Shh variants, as described earlier. Notably, Scube2 also increased Shh^ΔCa2+ proteolytic processing independent of its CUB domain [Fig. 7A, left: Shh+Scube2 was set to 100%; Shh, 12±2%, Shh+ΔCUB, 24±2%; mean±s.e.m., all n=12 (black columns). Shh^ΔCa2++Scube2 was set to 100%; Shh^ΔCa2+, 17±4%, Shh^ΔCa2++ΔCUB, 97±4%; mean±s.e.m., all n=6 (grey columns)]. By contrast, and consistent with our previous observation, ΔEGF was inactive [Fig. 7A, right: Shh+Scube2 was set to 100%, Shh, 12±1.5%; Shh+ΔEGF, 24±4%; mean±s.e.m., all n=7 (black columns). Shh^ΔCa2++Scube2 was set to 100%; Shh^ΔCa2+, 10±2%; Shh^ΔCa2++ΔEGF, 30±4%; mean±s.e.m., all n=10 (grey columns)]. Taken together, these results imply that Ca²⁺ renders terminal Shh peptides inaccessible and prevents unregulated morphogen processing (Fig. 7B, top). Scube2-enhanced shedding may then act on two levels. First, Scube2 binds to the HS chains of cell-surface Shh/HSPG complexes and, via its EGF domains, adjusts cell-surface sheddase active sites with the substrate cleavage sites (Fig. 7B, center). Proteolytic processing of otherwise protected terminal peptides may then involve a Ca²⁺-regulated switch of the Shh ectodomain structure. This would make Shh terminal peptides prone to proteolytic cleavage and release the activated morphogen from the cell surface. Importantly, analysis of the Shh cluster structure revealed close association of the N-terminal sheddase target peptide of one molecule in the cluster with the Ca²⁺ coordinating flexible loop of the adjacent protein in the cluster (Fig. 7C, the loop is shown in light blue). This situation indicates that Ca²⁺-regulated structural rearrangement of these loops, as shown in Fig. 6A, indirectly shifts N-terminal sheddase target sites into more accessible positions for subsequent cleavage. Our future work thus aims to reveal how CUB domain-dependent Shh release is linked to this process; preliminary sequence analysis suggests that the Scube2 CUB domain harbors a conserved Ca²⁺-binding site known to be required for CUB-dependent substrate recognition of complement proteases MASP1/2 and C1r/s (Fig. S7).

The mechanism of Shh signaling is only poorly understood, in a large part because all vertebrate and invertebrate Hh family members undergo N-terminal palmitoylation and C-terminal cholesteroylation during biosynthesis. This tethers the dual-lipidated morphogen to the plasma membrane. Because the energy required for the extraction of a single cholesterol molecule out of a diC16-PC/CHOL bilayer (0–40% cholesterol) requires 70–90 kJ mol⁻¹ (López et al., 2013), cell-surface-associated multi-lipidated Shh clusters would normally never accidentally flip out of the cell membrane, or get extracted without the use of energy. At first sight, the purpose of dual Shh lipidation therefore seems to be to prevent its solubilization.

Still, membrane-associated dual-lipidated Shh does signal to distant target cells. To explain this paradoxical situation, several hypotheses have been put forward that tie directly into this property. These hypotheses are largely based on the observed transfer of lipidated Hh-GFP fusion proteins on cellular extensions called cytonemes, or on secreted vesicles called exosomes (Bischoff et al., 2013; Gradilla et al., 2014; Sanders et al., 2013). The alternative situation that tight Shh membrane association merely serves to prevent unregulated Shh solubilization has been addressed by biochemical and structural studies suggesting proteolytic processing of both lipidated terminal peptides. This releases N- and C-terminally truncated, activated morphogens from producing cells (Dierker et al., 2009; Ohlig et al., 2011, 2012). Currently, the different concepts of Shh transport in lipidated or processed form are controversially discussed.

A first crucial criterion for the validity and usefulness of these hypotheses is their functional congruency. This criterion is fulfilled by the functional link between Scube2 CUB domain functions and Shh shedding, but not any mode of lipidated Shh transport, as supported by three observations. First, Shh solubilized by Scube2 lacks both lipidated peptides. Second, Shh target peptide modification renders Scube2-enhanced Shh processing CUB independent. Third, site-directed inactivation of Ca²⁺ coordination renders Scube2-enhanced Shh release CUB independent. Further support for Scube2 function as a sheddase enhancer comes from five published observations: First, Scube-type mutants (ty97, encoding Scube2 lacking its CUB domain) cause only mild phenotypes in zebrafish (van Eeden et al., 1996). Second, and likewise, mice made deficient in Scube2 expression show only minor Shh-related phenotypes (Lin et al., 2015). Third, reduced Shh signaling as a consequence of Scube2 knockdown in zebrafish is rescued by a moderate increase in Shh ligand expression (Johnson et al., 2012). Fourth, Scube orthologs are not expressed in Drosophila, despite the otherwise highly conserved signaling components. Finally, CUB-domains are established protease regulators in many biological settings (summarized in Jakobs et al., 2014).

A second criterion for the usefulness of the different concepts of Shh release and transport is the extent to which they can explain the variable functional relevance of N-palmitate for later signaling. Although N-palmitate is essential for fly Hh activity and often is essential for Shh activity (Chamoun et al., 2001; Chen et al., 2004; Dawber et al., 2005; Goetz et al., 2006; Kohtz et al., 2001; Lee et al., 2001), nonpalmitoylated Shh^C25S retains significant in vivo signaling activity in Ptc reporter mice and during vertebrate limb patterning, as well as in explant assays (Chen et al., 2004; Kohtz et al., 2001; Lee et al., 2001; Williams et al., 1999). In this work, we show that Scube2-modulated Shh shedding also explains this apparent paradox: Scube2-specified processing effectively targets nonpalmitoylated peptides and thereby activates otherwise inactive Shh^C25A/S to variable degrees. By contrast, sheddase regulation in flies that lack Scube orthologs is restricted to the general scheme that proteases require membrane association of target peptides to position themselves, and they cleave primarily with respect to a defined distance from the plasma membrane (Hooper et al., 1997). As a consequence, fly Hh^C85S is not cleaved at the unpalmitoylated N-terminus, resulting in biologically inactive morphogens. For the same reason, insufficient Scube2-expression in vertebrate tissues increases the relative amount of biologically inactive Shh^C25A/S.

How is Scube2-specified Shh processing achieved? The striking observation that distal relocation of the CW processing site and its replacement with a HA tag strongly impairs MβCD-induced unspecific shedding of palmitoylated and nonpalmitoylated Shh^HA, Shh^C25A;HA and ^HAShh^HA, but does not affect Scube2-assisted physiological shedding of the same proteins, suggests that Scube2 recognizes features of the substrate and ‘aligns’ physiological Shh cleavage sites with catalytic sheddase domains, irrespective of their juxtamembrane positioning. One domain type required for Scube2-assisted proteolysis is one or more of the nine Scube2 EGF repeats, because ΔEGF lacking these domains was inactive in all experimental settings tested in this study. Indeed, EGF domains are frequently found in sheddases and their substrates [L-selectin (CD62L, also known as SELL), TGFα, EGF, heparin binding EGF-like growth factor and Notch proteins] and are thought to mediate specific protein-protein interactions. For example, the L-selectin EGF domain contains a sheddase docking site required for induced specific processing at another site (Zhao et al., 2001), analogous to the membrane-distal region of angiotensin-converting enzyme (Sadhukhan et al., 1998). Likewise, BMP1 activity depends on its EGF and CUB domains (Wermter et al., 2007), as well as their HSPG-binding enhancers (Vadon-Le Goff et al., 2015). Because all Hhs lack EGF domains, one role of Scube2 enhancers may thus be to provide these domains externally to increase Shh target peptide encounters with their proteases.

Another important observation is that terminal extensions, as well as deletion of N- or C-terminal lipids or site-directed inactivation of Ca² coordination, all render Scube2-regulated Shh release CUB domain independent. How can this be explained by the different models of Shh release and transport? A general feature of sheddase recognition is that substrates require a susceptible membrane-proximal stalk of sufficient length between the membrane and the proximal extracellular globular domain to permit access of the protease to substrate, as demonstrated for Interleukin-6 (IL-6), L-selectin and T-cell immunoglobulin and mucin domain 3 (Tim-3, also known as Havcr2) (Baran et al., 2013; Möller-Hackbarth et al., 2013; Riethmueller et al., 2016; Zhao et al., 2001). Thus, membrane proteases cannot process domains if they are too close to the surface or even at the surface (in a protein with short stems). We note that this fact must be considered in the interpretation of models derived from permanent Hh-GFP membrane association, as this may have resulted, wholly or in part, from the elimination of sheddase target peptides from the cholesteroylated GFP-tag. Indeed, the cloning protocols used by Vincent et al. (2003) to generate a cholesteroylated GFP membrane marker, and those used by Torroja et al. (2004) to generate Hh-GFP, are similar, because both reduce the cholesteroylated stalk region to only two or three amino acids. The minimal required stalk length for protein shedding, however, is 10–12 amino acids (Baran et al., 2013; Möller-Hackbarth et al., 2013; Riethmueller et al., 2016; Scheller et al., 2011; Zhao et al., 2001), which is in agreement with the N-terminal Shh sequence Palm-C²⁵GPGRGFGKRRHPKK³⁹ (15 amino acids) and the C-terminal sequence E¹⁸⁹NSVAAKSGG¹⁹⁸-Chol (10 amino acids). However, because a large part of the N-terminal sequence closely associates with adjacent globular domains in the cluster in trans (Farshi et al., 2011), its inherent accessibility may be very low. This is consistent with the observation that Shh is not efficiently cleaved from unstimulated cells (this work and Damhofer et al., 2015). As expected from this situation, we found that impaired lipidation or increased length of terminal Shh peptides (as a consequence of HA insertion) relaxes the protective Shh domain orientation, facilitates sheddase access to its substrate, and enhances both baseline release and Scube2-induced shedding of mutated Shh variants. The observation that the Scube2 CUB domain is dispensable for the induced release of such ‘unprotected’ proteins, but is required for authentic dual-lipidated Shh, therefore, indicates that CUB domains enhance the accessibility of lipidated membrane-proximal Shh peptides. We note that such a mechanism must be highly specific because Xenopus laevis Scube2 does not enhance the release of dual-lipidated murine Shh (Jakobs et al., 2014). One possible way to increase target accessibility involves a change in the substrate to a conformation that is more susceptible to proteolysis, as described for the sheddase ligand Notch. Crystal structure analysis revealed that the membrane-proximal protease cleavage site of autoinhibited Notch isoforms is shielded by Ca²⁺-regulated intramolecular interactions that are removed by ligand-induced conformational changes, which, in turn, lead to substrate cleavage (Gordon et al., 2007).

We are the first to uncover an analogous mechanism for Ca²⁺-dependent autoinhibition of Shh release. We show that Shh at the cell surface attains a sheddase-resistant conformation, and that dissociation of Ca²⁺ increases baseline and induced Shh shedding independent of the Scube2 CUB domain. This indicates a novel Ca²⁺- and CUB domain-dependent mode of cell-surface sheddase regulation at the substrate level. We support this possibility by the following findings: first, Shh Ca²⁺ coordination sites are conserved and are solvent accessible, yet they are not surface exposed. In addition, the most solvent-accessible Ca²⁺ is only weakly bound. Two recent studies estimated its affinity at >100 mM (17–23 kJ/mol) (McLellan et al., 2008; Rebollido-Rios et al., 2014). This weak affinity explains the failure to observe this Ca²⁺ site in some crystal structures, as Ca²⁺ may have dissociated during purification (McLellan et al., 2008). It further suggests temporary vacancy of one or both Ca²⁺ sites at the average extracellular Ca²⁺ concentration of 1.3 mM (Hitzenberger and Hofer, 2016). Ca²⁺ occupancy, in turn, has a distinct influence on the properties of the protein surface surrounding the coordination site (Rebollido-Rios et al., 2014). As shown in this work, lack of Ca²⁺ destabilizes the surface because all but one Ca²⁺ coordinating residues are situated in a loop region, which becomes flexible when the acidic side chains are no longer restrained by the ions. A similar displacement or perturbation was observed in an unrelated study when only one (the more loosely bound) Ca²⁺ was removed (Hitzenberger and Hofer, 2016). We thus suggest that such changes in loop structure increase protease access to the interacting (autoinhibited) N-terminal Shh peptide, as has been reported for Ca²⁺ stabilization of thermolysin (Eijsink et al., 2011), a bacterial protease that is structurally related to Shh (Hall et al., 1995).

Our future studies will investigate how the Scube2 CUB domain affects Shh Ca²⁺ binding. These studies will be important for interpreting diseases linked to impaired Hh signaling, such as brachydactyly type A1 (BDA1). BDA1 is a congenital malformation characterized by apparent shortness or absence of the middle phalanges of all digits and is linked to missense substitutions of residues E95, D100 and E131 in the human Shh homolog Indian Hh (Ihh). These residues directly coordinate Ca²⁺ (Ihh E95 corresponds to mutated E91 in Shh^ΔCa2+; Ihh E131 corresponds to mutated E127) and their mutations act in a dominant gain of function comparable to the ectopic expression of Shh in developing digits (McLellan et al., 2008). It has been suggested that BDA1 is caused by disrupted interactions with receptors or increased signaling range. We note that the latter possibility is supported by increased unregulated (background) and regulated (Scube2-mediated) shedding of Ca²⁺-free morphogens from producing cells.

Shh constructs were generated from murine cDNA (NM_009170), and Hhat cDNA (NM_018194) was obtained from ImaGenes. Both cDNAs were cloned into pIRES (Clontech) for their coupled expression together with Hh acyltransferase from bicistronic mRNA in the same cells, resulting in authentic, dual-lipidated 20 kDa Shh. Shh^C25A (Hardy and Resh, 2012), ShhN^C25A and HA-tagged variants were generated by site-directed mutagenesis (Stratagene). Primer sequences can be provided upon request. Human Scube2 constructs were a kind gift from Ruey-Bing Yang (Academia Sinica, Taiwan) (Tsai et al., 2009).

Bosc23 cells, a HEK293 derivate [a kind gift from D. R. Robbins (University of Miami, USA) (Zeng et al., 2001)], were cultured in Dulbecco's modified Eagle's medium (DMEM, Lonza) supplemented with 10% fetal calf serum (FCS) and 100 µg/ml penicillin-streptomycin, and transfected with PolyFect (Qiagen). Transfected cells were cultured for 48 h, the medium was changed, and Shh was secreted into serum-free medium for 6 h or overnight. Harvested media were ultracentrifuged for 30 min at 125,000 g and proteins were trichloroacetic acid (TCA) precipitated. Where indicated, 500 µg/ml MβCD (Sigma-Aldrich) or 2.5 µM ionomycin ionophore (Biomol) were added to serum-free media to increase A10/A17 enhanced shedding. All proteins were analyzed by 15% SDS-PAGE, followed by western blotting using polyvinylidene difluoride membranes. Blotted proteins were detected by anti-HA antibodies (1:10.000, mouse IgG; Sigma-Aldrich), polyclonal anti-FLAG antibodies (1:5000, rabbit IgG; Sigma-Aldrich), anti-Shh antibodies (1:1000, goat IgG; R&D Systems), or polyclonal anti-CW antibodies directed against the CW sheddase target sequence (1:5000, rabbit IgG; Cell Signaling Technology). Incubation with peroxidase-conjugated donkey-anti-goat/rabbit/mouse IgG (1:10.000, Dianova) was followed by chemiluminescent detection. Signals were quantified by using ImageJ. Adobe Photoshop was used to convert greyscale blots into merged RGB pictures for improved visualization and quantification of N- and C-terminal peptide processing.

C3H10T1/2 reporter cells (Nakamura et al., 1997) were grown in DMEM containing 10% FCS and 100 µg/ml penicillin-streptomycin. At 24 h after seeding, serum-free Shh-conditioned media were diluted 1:1 with DMEM containing 20% FCS and 100 µg/ml antibiotics, and applied to C3H10 T1/2 cells. Cells were lysed 5–6 days after induction (20 mM HEPES, 150 mM NaCl, 0.5% Triton X-100 pH 7.4) and osteoblast-specific alkaline phosphatase activity was measured at 405 nm after addition of 120 mM p-nitrophenylphosphate (Sigma-Aldrich) in 0.1 M glycine buffer (pH 9.5). 5E1 antibody was purchased from Development Studies Hybridoma Bank.

HS columns were generated as follows: 1 g (wet weight) C57/B16 mouse embryos (E12-E18) were homogenized and digested in 320 mM NaCl, 100 mM sodium acetate (pH 5.5) containing 1 mg/ml pronase, and 1 mg/ml proteinase K for 72 h at 40°C. Fresh enzymes were added every 12 h. The digested samples were centrifuged, filtered and diluted 1:3 in water, and 2.5 ml aliquots were applied to DEAE Sephacel columns. Eluted glycosaminoglycans were lyophilized, diluted and quantified by the carbazole reaction. The material was then coupled to NHS-activated sepharose columns. The columns were tested by using recombinant HS-binding protein fibroblast growth factor 2 and 8, vascular endothelial growth factor and semaphorin 3F as positive controls. Fast protein liquid chromatography was conducted on an Äkta protein purifier (GE Healthcare). Samples were applied to the columns in the absence of salt, and bound material was eluted by using a linear 0 to 1.5 M NaCl gradient in 0.1 M phosphate buffer (pH 7.0). Eluted fractions were TCA precipitated and analyzed by SDS-PAGE as described earlier. Signals were quantified using ImageJ. Gel filtration analysis was performed with a Superdex200 10/300 GL column (GE Healthcare) equilibrated with PBS at 4°C.

Scube2-transfected Bosc23 cells were nonenzymatically removed from the culture dish by using Versene (PAA) and suspended in PBS containing 5% FCS in a total volume of 0.5 ml. Cells were washed and treated with anti-FLAG antibody (1:500 dilution) and 7-amino actinomycin (7AAD, Cayman Chemical) for 1 h, and with fluorescein isothiocyanate-conjugated goat-anti-rabbit secondary antibody (1:200 dilution, Dianova) for 30 min, on ice. FACS analysis was performed on a BD Accuri C6 flow cytometer (BD Biosciences). A total of 50,000 cells were counted. Histograms were created using FlowJo single cell analysis software.

Transfected cell lines were fixed in 4% paraformaldehyde under nonpermeabilizing conditions and subjected to Duolink in situ fluorescence detection (Sigma-Aldrich) according to the manufacturer's instructions. Briefly, slides were blocked, incubated with primary antibodies directed against A17 (anti-A17 antibodies, mouse monoclonal IgG; R&D Systems) and Shh (anti-Shh antibodies, goat IgG; R&D Systems), washed, and incubated with secondary antibodies conjugated to oligonucleotides (PLA probes, Sigma-Aldrich). Circularization and ligation of the oligonucleotides was followed by an amplification step with nucleotides and fluorescent oligonucleotides. Negative controls always included transfected cell lines expressing both target proteins. These cells were incubated with each single primary antibody and both PLA probes. Slides were mounted with Duolink in situ mounting medium and evaluated with an LSM 700 confocal microscope (Zeiss). Z-stack micrographs taken with ×40/63 objectives were obtained. Representative results are shown from experiments repeated at least twice. Integrated morphometric analysis was performed using MetaMorph Microscopy Automation and Image Analysis Software (Molecular Devices).

Bosc23 cells were incubated for 10 min or overnight with serum-free DMEM, Scube2-conditioned serum-free DMEM, or serum-free DMEM conditioned with ΔEGF. Cells were lysed, the lysate was centrifuged, and phosphorylation of 16 cytosolic EGFR target proteins was determined using a PathScan Akt Signaling Antibody Array Kit, according to the manufacturer's instructions (Cell Signaling Technology). Briefly, cell lysates were incubated on glass slides coupled with target-specific capture antibodies. Biotinylated capture antibodies detect cellular proteins only when phosphorylated at functionally relevant positions. Bound detection antibodies were detected by chemoluminescence and quantified with Biodoc 1D (Vilber Lourmat).

The crystal structure of human Shh [Protein Data Bank (PDB): 3m1n] (Pepinsky et al., 2000) was displayed by using the PyMOL Molecular Graphics System, Version 1.3.

Sequence analysis was conducted on the CFSSP secondary structure prediction server (http://www.biogem.org/tool/chou-fasman/). All statistical analysis was performed in GraphPad Prism by using the Student's t-test (two-tailed, unpaired, confidence interval 95%). Data are presented as mean±s.e.m. PLA data were analyzed by one-way Anova (Kruskal–Wallis test) and ranked by Dunn's post hoc analysis based on nonparametric data distribution in GraphPad Prism.

We thank Ruey-Bing Yang (Academia Sinica, Taiwan) for human Scube2 constructs, and S. Kupich and R. Schulz for excellent technical and organizational assistance.