Hh signal transduction has been the focus of intense research over the past decade due to the central role it plays in development and its emerging biomedical relevance in areas ranging from regenerative medicine to oncology [2,30]. Our goal in these studies was to isolate and characterize small-molecule modulators of Hh signaling in order to understand better the regulation of pathway activation and to generate potential therapeutics. Our work shows firstly that it is possible to identify potent small molecule agonists of Hh signaling, secondly that these compounds can mimic the effects of recombinant Hh protein in multiple assays used to define the properties of Hh signaling, thirdly that these compounds act by binding directly to Smo, and finally that two Ptc-independent inhibitors of Hh signaling compete for this binding to Smo, strongly suggesting they too act directly on Smo.
Models of Smo-ligand interaction
To interpret the results of the competition binding studies, we assume that the mutation in Smoact, like those in constitutively activate mutants of other GPCRs , indirectly influences ligand binding by creating a change in the normal equilibrium between the different conformations of Smo. Thus, the mutation would not directly influence the binding pocket for either ligand. A simple two-state model (Figure 8a) predicts that the agonist (Ag, green square) and the antagonist (Ant, pink circle) compete for the same site on Smo to activate or inactivate Hh-pathway signaling. It also suggests that antagonists should bind Smoact with a lower affinity than they bind Smowt, while the agonist should bind with a higher affinity, as it prefers the active conformation. Such a model cannot accommodate our observations with the gain-of-function Smo mutant. Thus we introduce a ternary complex model (Figure 8b), used traditionally to describe the behavior of GPCRs in binding studies with agonist and antagonists , as well as constitutively active receptors . The ternary complex model for Hh signaling suggests that there are two independent binding sites on Smowt, one specific for the agonist and another specific for antagonists. Binding at either site would decrease the affinity for interactions at the other site (allosteric binding with high negative cooperativity). The agonist-bound form represents the normal activated state, while the antagonist-bound form is considered the inactive conformation. There are also other conformations that would not be bound, or would be transiently bound, by both ligands. A signaling pathway coupler, or effector (in blue), is proposed to bind the activated state of Smowt so as to generate a complex competent to initiate Hh signaling. Throughout the discussion the term 'coupler/effector' is used to describe an unknown molecule that binds activated Smo in such a way as to trigger signal transduction. The model further suggests that the Smoact protein resides in a stable conformation in the absence of agonist that is capable of forming an active coupler/effector complex resistant to antagonist, but not agonist, binding.
Specifically, our data suggest that the agonist binds and stabilizes (or induces) an active signaling state of Smo while the antagonists bind and stabilize (or induce) an inactive form. Furthermore, the gain-of-function Smo mutation renders the protein less sensitive to the inhibitors, presumably because the amino-acid substitution directly stabilizes or induces an active conformation. On the basis of a simple two-state model, one might predict an increased affinity of the agonist for the mutant form, but in our studies binding of the agonist, unlike the antagonist, is not affected by the activating mutation, suggesting that a more complex model requiring two binding sites and perhaps multiple active conformations is needed to account for the observations. Thus, we propose a variation of the classic 'ternary complex model' (Figure 8
), a decades-old paradigm that has provided the foundation for describing ligand induced conformational changes of GPCRs
Briefly, if this model is applied to Hh signaling, it proposes that, firstly, agonist and antagonists act at independent sites to select active and inactive conformations of Smo; secondly, that Smo engages an undefined coupler/effector, when it is in its signaling state; and finally, that the gain-of-function mutant form of Smo, Smoact, adopts an abnormal conformation that resembles the coupler/effector-bound signaling state of Smo with low affinity for antagonists but normal affinity for the agonist.
Activating Hh signaling through the GPCR-like Smo receptor
As a receptor class, GPCRs are considered excellent drug targets because they are often regulated through interactions with small natural ligands [32,33]. Specifically, studies of classic GPCRs, such as the β-adrenergic receptors, show that in the absence of endogenous ligand (agonists) these receptors exist in multiple interconvertible conformations that are predominately inactive . Upon exposure to their natural ligands, however, the active receptor forms are preferentially stabilized, allowing them to readily engage G-protein couplers and to create signaling-competent complexes. Multiple compound classes have been isolated on the basis of their ability to compete for the binding of β-adrenergic receptors by their natural ligands. These competitors can mimic the natural ligand activity (agonists) or interfere with it (antagonists).
In addition to the binding sites for natural or endogenous agonists (orthosteric sites), many GPCRs have also been found to have allosteric sites . These sites can bind natural ligands, as in the case of Zn ions and heparin for the dopamine and neurokinin receptors, respectively, or bind synthetic drugs such gallamine, in the case of the muscarinic receptors . Binding of small molecules to these allosteric sites can modulate activity of a receptor without directly mimicking or competing out the interaction of ligands to the orthosteric sites. In summary, GPCRs have an array of potential regulatory binding sites, or potential drug targets.
How does Smo compare with other GPCRs with regard to the properties described above? Although there is clear structural homology between Smo and other GPCRs, endogenous ligands have yet to be discovered. Early models of Hh signaling proposed a Hh-regulated Ptc-Smo complex that directly controlled the conformation of Smo, making endogenous ligands unnecessary. But recent studies argue against this stoichiometric model [5,36], indicating that perhaps natural ligands should be considered. Furthermore, regarding G-protein-effector coupling for Smo, the results are also equivocal. Although no compelling data have been presented that directly link classic G-protein activation  to the canonical Hh pathway involving Ci/Gli stimulation , recent studies show that under certain conditions Smo can engage G-protein subunits in a Ptc-dependent manner  and that G-protein-mediated cAMP modulation may underlie certain effects of Hh on neuronal tissue .
Finally, with respect to pharmacological properties, our studies indicate that Smo behaves like a classic GPCR in many regards and that the models used to describe this large family of receptors can be applied to Hh signaling (Figure 8). Two relatively novel concepts for Hh signaling are raised by these GPCR models: firstly, the importance of considering the active Smo-coupler/effector complexes when modeling pathway regulation, and secondly the potential for endogenous ligands in the regulation of Smo activity.
GPCR models for Smo and potential mechanisms of Ptc function
A recent study in Drosophila suggested that Hh stimulates the pathway via Ptc degradation and, as a result, Smo is stabilized through extensive phosphorylation at the plasma membrane where it initiates signaling . This led to speculation that a Ptc-regulated phosphatase may control the subcellular distribution and stability of Smo . Although suggestive, the studies in Drosophila did not establish that phosphorylation, translocation or stabilization of Smo is required for pathway stimulation. It is plausible that some or all of these effects on Smo result from a feedback inhibition mechanism that targets the activated Smo receptor. Interestingly, while we did observe stabilization of Smo by Hh in our mammalian cell experiments (Figure 5), we did not detect Smo phosphorylation or translocation to the plasma membrane (data not shown). Perhaps the cellular (or species) context dictates the degree to which the active conformations of Smo are associated with such changes. Thus, in considering models of Ptc function based on Hh-stimulated effects it is important to consider whether the form of Smo that is being observed is the active (coupler/effector bound) signaling state, or perhaps a downregulated and inactive form.
Several potential points for regulation by Ptc during the formation of a Smo-coupler/effector complex are apparent in a ternary complex model. The ideas proposed for the Drosophila system, in which Ptc may affect the levels of Smo or its subcellular localization, are easily accommodated. Ptc-induced instability of Smo protein would indirectly reduce the concentration of an active Smo-coupler/effector complex. Furthermore, limiting access of Smo to its effector through targeted vesicle trafficking would prevent a signaling-competent complex from forming. Alternatively, Ptc could more directly maintain Smo in an inactive state. On the basis of our studies it is tempting to speculate that native small molecules with properties similar to our agonist and antagonists act directly on Smo in a Ptc-dependent manner. These putative endogenous Smo modulators could represent orthosteric or allosteric ligands.
The simplest model would have Ptc acting catalytically to dock a natural antagonist directly onto (or remove an agonist from) Smo. A more complex model would involve Ptc restricting the distribution of Smo such that it is forced into compartments containing the natural antagonists or lacking the natural agonists. Finally, Ptc could control the distribution of the endogenous small-molecule modulators themselves. Ptc shares sequence homology with molecules associated with vesicle trafficking and transporter activity, namely SCAP and NPC1 [36,40-42]; if it also shares the activities of these molecules, as suggested by a recent study , then the possibility of the existence of endogenous Smo ligands that are docked via Ptc should be explored.
Therapeutic potential of a Hh-pathway agonist
Various studies in mammals have shown that Hh genes are expressed in discrete areas of the adult organism and may function in the normal maintenance of mature organ systems [43-46]. In addition, the regenerative healing of vascular and skeletal tissues following acute injuries appears to be aided by re-activating the Hh-signaling cascade [47,48]. Taken together, these observations suggest that the Hh pathway may represent a point of intervention for treating certain degenerative disorders. Two recent studies in models of Parkinson's disease and peripheral nerve damage support this claim, by demonstrating that pathway activation with a Hh-protein ligand has therapeutic value [49,50]. On the basis of our current understanding of these models and the specific mechanism of action of the Hh agonists, we predict that an agonist-derivative with low toxicity and favorable pharmacokinetics would replicate these positive results. As a drug, a Hh agonist would represent an attractive alternative to an expensive Hh-protein therapeutic. Beyond the economics, for disorders of the central nervous system a small molecule with the potential to cross the blood-brain barrier would eliminate the need for injections directly into the brain, the current delivery mode for central nervous system protein therapies.
Materials and methods
Chemical libraries and medicinal chemistry
The compound libraries used in our screens were purchased from a number of commercial vendors and were primarily generated by combinatorial chemistry approaches. The Hh-agonist class was isolated from a library synthesized by Oxford Asymmetry International, now EvotecOAI. The derivatization of this compound class utilized standard procedures, the details of which will be published elsewhere.
Cultured cell line assays
TM3 and C3H10T1/2 cells (ATCC; Manassas, USA) were maintained according to the instructions of ATCC. Stable Hh-signaling reporter cell lines were established by G418 selection following transfection with a luciferase reporter plasmid  containing the neomycin-resistance gene. Hh signaling was monitored by plating cells at 70% confluence in growth medium. After 24 hours the cells were changed to 0.5% serum-containing medium, and Hh protein or compounds were added; 24 hours later the cells were either monitored for luciferase activity using the Luc-lite assay kit (Packard Instrument Company, Meriden, USA) or harvested for RNA isolation using an RNA isolation kit (Qiagen; Valencia, USA). RNA was subjected to quantitative RT-PCR analysis (Taqman; Applied Biosystems, Foster City, USA) utilizing Gli1, Ptc1 and GAPDH primers and probes. Assays were run on a Prism 7700 instrument (ABI; Applied Biosystems).
Recombinant Hh protein
The Hh protein used in the studies described here was bacterially overexpressed amino-terminal human Shh modified at its amino-terminal cysteine by an octyl maleimide moiety . This lipophilic Shh form showed comparable potency to native Shh in the cell-based reporter assay (data not shown).
Retroviral cell lines
Mouse Smo and Ptc1 genes were introduced by a retroviral approach utilizing the pLPCX vector (Clontech; Palo Alto, USA) to limit the copy number per cell. Stable HA-Smo and Ptc-GFP lines were established by puromycin selection following infection of TM3 cells with the respective retroviruses. TM3 cells expressing both epitope-tagged Ptc and Smo were derived by infecting first with an HA-Smo construct and subsequently with a Ptc-GFP construct. The levels of Ptc were relatively low in these lines and a standard immunoprecipitation procedure followed by western blotting was required to detect the Ptc-GFP protein. The HA-Smo protein was highly expressed and was easily detected in western blots of whole cell extracts. The HA tag was sub-cloned into the Smo gene so that it would reside immediately after the Smo signal sequence. The GFP tag was inserted before the stop codon of the Ptc open reading frame.
In utero Hh-signaling assays
Generation of Ptc1lacZ, Shh and Smo mutant mice has been described previously [25,26]. Ptc1lacZ/+ mice were kindly provided by Matthew Scott. Shh+/- and Smo+/- mice were kindly provided by Andrew McMahon. Agonist solution was prepared in fine suspension in 0.5% methylcellulose/0.2% Tween 80 at 1.5 mg/ml. Compound was administered by oral gavage to pregnant mice once a day for two days at 100 μl per 10 g body weight. Embryos were collected 24 hours later. Whole-mount in situ hybridization with Ptc1 probe and X-gal staining for whole-mount β-galactosidase detection were performed as described . For histology, embryos stained with X-gal were post-fixed in 4% paraformaldehyde, wax-embedded, and 20 μm sections were prepared.
Primary cerebellar cultures
Cerebellar neurons were dissected out of postnatal (one week) rat brains, and placed into primary cell culture. Briefly, cells were placed in 96-well plates at a density of approximately 150,000 cells per well in basal medium of Eagle (Gibco; Carlsbad, USA) supplemented with 26 mM KCl, 2 mM glutamine and 10% calf serum. Treatment agents were added once, on the first day of culture (0 DIV). Cells were left in culture until 2 DIV, when [3H]-thymidine was added for 5 hours. Cells were then lysed, and the incorporation of [3H]-thymidine was determined by scintillation counting.
Neural plate explant assay
Intermediate regions of the open neural tube (i-explants) were dissected from stage 10-11 chick embryos and embedded in collagen gel . Explants were cultured in Ham-F12 supplemented with 3 g/l D-glucose, Mito Serum Extender (Collaborative Research; Bedford, USA), penicillin/streptomycin (Gibco), 2 mM L-glutamine (Gibco) and Hh-Ag 1.3 (0.1, 1, 10, 20, 200 and 1000 nM prepared as 1000X stocks in DMSO; n = 6 explants). As a control, some explants were cultured with vehicle alone or with octylated Hh-N recombinant protein. Cultures were fixed after 22 hours, stained with mouse monoclonal antibodies against Pax7, MNR2 or rabbit polyclonal antibodies against Nkx2.2 and the number of immunoreactive cells per explant counted.
Whole cell/immunocomplex binding assay
Cultured cells - 70% confluent 293T cells in 6-well plates - were either left untransfected or transfected using Fugene6 with a pCDNA3.1 construct containing HA-tagged Smo or v5-epitope tagged β2AR (Invitrogen; Carlsbad, USA). After 48 hours cells were switched from 10% fetal bovine serum containing DME media to 0.5% FBS containing media supplemented with either 5 nM [3H]-Hh-Ag 1.5 alone or 5 nM [3H]-agonist plus 5 μM of various competitors (see Results). After 2 hours of incubation at 37°C, cells were washed one time with PBS and subsequently lysed in 0.5 ml of lysis/wash buffer containing 1% NP40 in Tris-buffered saline plus an EDTA-free protease inhibitor cocktail (Roche; Indianapolis, USA) for 20 minutes at 4°C. Cell extracts were spun at 14,000 rpm in a microcentrifuge and supernatants were incubated for 40-60 minutes with either anti-HA beads (Roche) or anti-v5 antibody (Invitrogen) and Protein A beads (Pierce; Rockford, USA) to form immunocomplexes. Immunobeads were then spun down and washed three times with 0.5 ml lysis/wash buffer per wash. The washed pellets were then resuspended in SDS sample buffer and combined with scintillation fluid. Counts per minute (cpm) for each sample were then determined in a scintillation counter (Packard topcount).
Membrane binding assays
Membranes were prepared as follows. Briefly, approximately 108 cells were transfected with pcDNA 3.1 constructs (Invitrogen) bearing either murine Smo (wild-type or W539L mutant), GFP, rat β2AR (Invitrogen), or murine Ptc cDNAs using Fugene 6 (Roche). After 48 hours cells were harvested by scraping in PBS, centrifuged at 1,000 × g for 10 minutes, and gently resuspended in around 10 ml of a 50 mM Tris pH 7.5, 250 mM sucrose buffer containing an EDTA-free protease inhibitor cocktail (Roche). This cell suspension was then placed in a nitrogen cavitation device (Parr Instrument Co, Moline, USA) and exposed to nitrogen gas (230 psi) for 10 minutes. Lysed cells were released from the device and centrifuged at 20,000 rpm in an SS34 rotor for 20 minutes at 4°C. Supernatants were discarded and the pellets were resuspended in 10% sucrose, 50 mM Tris pH 7.5, 5 mM MgCl, 1 mM EDTA solution using three 10-second pulses with a Polytron (Brinkman; Westbury, USA) at a power setting of 12. Using these membranes, filtration binding assays were performed according to previously described protocols . To reduce nonspecific binding, 96-well filtration plates (fiberglass FB filters; Millipore, Bedford, USA) were pre-coated as suggested by the manufacturer with 0.5% polyethyleneimine + 0.1% BSA and then washed four times with 0.1% BSA.
For association and dissociation studies, membranes (1.5 μg total protein) were incubated in polypropylene tubes with 2 nM [3H]-Hh-Ag 1.5 in the presence or absence of 2 μM competitor in binding buffer (50 mM Tris pH 7.5, 5 mM MgCl, 1 mM EDTA, 0.1 % bovine serum albumin) plus EDTA-free protease inhibitor cocktail (Roche) in a final volume of 250 μl for 1-26 hours at 37°C. For saturation and competition binding analysis, membranes (1.5 μg total protein) were incubated on the plates with various concentrations of the [3H]-Hh-Ag 1.5 (plus and minus competitors) in binding buffer plus EDTA-free protease inhibitor cocktail (Roche) at a final volume of 1 ml for approximately 45 hours at 37°C to allow binding to reach apparent equilibrium. Binding reaction mixtures (0.2 ml for association/dissociation studies and 0.75 ml in saturation and competition experiments) were then transferred to the pre-coated 96-well filtration plates (Millipore fiberglass FB filters), filtered and washed over a vacuum manifold with six 300 μl per well washes of binding buffer supplemented with 2% hydoxypropyl cyclodextrin (HPCD; Sigma; ST Louis, USA) + 0.1% BSA to decrease non-specific binding. Identical results were obtained if incubations were done in borosilicate glass or siliconized plastic tubes. Centrifugation assays were also performed that replicate the filtration assay results (data not shown). Additionally, these experiments showed that the extent of ligand depletion was less than 10% in these studies. Binding-kinetics experiments were performed similarly to the saturation and competition studies. All binding data were evaluated using a nonlinear regression analysis program (Prism; GraphPad; San Diego, USA). Ki values were calculated using the Cheng-Prusoff correction equation , where Ki = IC50/1+ [L]/Kd, and Kd for Hh-Ag 1.5 was determined to be 0.37 nM by the saturation analysis.
Note added in proof
Related results demonstrating the action of cyclopamine on Smo have been reported by Beachy and colleagues [51,52].
We thank Douglas Melton, Andrew McMahon, Thomas Jessell and Phillip Beachy for comments on the manuscript. Shh-, Smo- and Ptc-transgenic lines were graciously provided by the labs of Andrew McMahon and Matthew Scott. We thank James Chen and Phillip Beachy for providing us with KAAD-cyclopamine and for sharing their observations that cyclopamine and its derivatives bind directly to Smoothened and that our agonists compete this interaction. Superior medicinal chemistry support as well as screening libraries were provided by Larry Kruse, Andy Boyd, Steven Price and the team at Evotec/Oxford Asymmetry International.