Isolation of Hh agonists by high-throughput screening
To identify small-molecule agonists of Hh signaling, we established a mammalian-cell-based assay. After testing several cell lines for Hh-dependent induction of the target genes Ptc1 and Gli1, we identified C3H10T1/2 and TM3 cells as optimal responders. We then introduced into each line a plasmid containing a luciferase reporter downstream of multimerized Gli binding sites and a minimal promoter . An isolated stable clone of the 10T1/2 cell transfectants (referred to as clone S12) gave a 10-20-fold up-regulation of luciferase activity (Figure 1a) when stimulated with Hh protein  for 24 hours. Using this assay system, we screened 140,000 synthetic compounds at a concentrations of 2-5 μM and isolated several putative agonists. One of these molecules - Hh-Ag 1.1 (Figure 1a,b) - was studied further. Hh-Ag 1.1 exhibited half-maximal stimulation (EC50) at around 3 μM, and an activation maximum (Amax) of approximately 35% compared to the Hh protein control (Figure 1a). In the presence of sub-threshold signaling levels of Hh protein (0.3 nM), the EC50 of Hh-Ag 1.1 was reduced to around 0.4 μM and the Amax approached 70% (Figure 1a).
We next tested whether expression of endogenous Hh-responsive genes was stimulated by the agonist. Using quantitative PCR, Hh-Ag 1.1 was shown clearly to elevate the expression of Gli1 and Ptc1 in a dose-dependent manner (Figure 1c).
Chemical modifications increase potency
In an effort to improve the potency of Hh-Ag 1.1, over 300 derivatives were synthesized and tested in the cell-based reporter assay. The relative potencies of the most active derivatives - 1.2, 1.3, 1.4 and 1.5 - are shown in Figure 1d. The most potent, Hh-Ag 1.5, had an EC50 of approximately 1 nM. Thus, potency was increased over 1000-fold by chemical modification. The structures of compounds 1.2 and 1.3 are shown in Figure 1e. Hh-Ag 1.2 was the most stable derivative in vivo and in vitro (data not shown) and was used for most cell-based assays. Hh-Ag 1.3 showed lower toxicity in embryonic tissue cultures (data not shown) and was used for the neural plate explant assays described below. These experiments suggest that the agonist may have many of the properties of the Hh ligand. To specifically test this, we used two established in vitro assay systems that detect the effects of Hh on primary neuronal precursors.
In vitro assay of neuronal precursors
Proliferation activity of the agonist
It has recently been shown that primary neonatal cerebellar granule neuron (CGN) precursors proliferate in response to Hh stimulation . To determine whether the Hh agonist could elicit this response, we monitored [3H]-thymidine incorporation of cultured rat CGN precursors treated with Hh protein, Hh-Ag 1.1, Hh-Ag 1.2, or vehicle (DMSO). The original active molecule, Hh-Ag 1.1, stimulated thymidine incorporation at 5 μM, but not at 1.75 μM (Figure 1f). The extent of proliferation was around 50% of that seen with a high dose of Hh protein (50 nM). Hh-Ag 1.2 stimulated proliferation at 300 nM and 100 nM to levels comparable to those seen with Hh protein (Figure 1f). These data demonstrate that the agonists can elicit a biological response in primary cells similar to that produced by Hh protein.
Morphogenic activity of the agonist
Neural progenitors within the intermediate region of the chick neural plate (Figure 2a) respond to increasing concentrations of Hh protein by adopting specific fates. The identity of these cells can be assessed by their distinct expression patterns of a set of transcription factors . Three of these transcription factors - Pax7, MNR2 and Nkx2.2 - whose expression is differentially sensitive to increasing concentrations of Hh protein were assayed in response to varying concentrations of the agonist (Hh-Ag 1.3). The dorsal spinal cord marker Pax7 is normally repressed by low concentrations of Hh . Pax7 expression was extinguished by 1-10 nM agonist (Figure 2b,c,d,e,f). Higher concentrations of agonist (10-200 nM) induced expression of the motor neuron progenitor marker MNR2 (Figure 2b,g,h,i,j), and yet higher concentrations (20 nM-1 μM) induced the most ventral interneuron progenitor marker Nkx2.2 (Figure 2b,k,l,m,n). This dose-dependent profile of expression closely resembles the response achieved by increasing concentrations of Hh protein [22-24], demonstrating that the Hh agonist mimics the concentration-dependent inductive activity of Hh on neural precursors.
Activity of the agonist in vivo
To explore the site of action of the Hh agonist within the Hh pathway, we developed an in vivo assay for the agonist that would allow us to test its activity in Shh- and Smo-mutant mouse embryos in utero. First, we compared the expression of Ptc1 in vehicle- and agonist-treated Ptc1lacZ/+ mouse embryos . The Ptc1lacZ/+ mouse expresses β-galactosidase under control of Ptc1-regulatory elements and thus reports Hh-pathway activity in mouse tissues. Hh-Ag 1.2 (Figure 1e) was chosen for study on the basis of its relatively low toxicity, long serum half-life and ability to cross the placenta (data not shown). Hh-Ag 1.2 was delivered by oral gavage to pregnant mice at 7.5 and 8.5 days post coitum (7.5 and 8.5 dpc). Embryos were collected at embryonic day (E) 9.5 and analyzed by staining with the β-galactosidase chromogenic substrate X-gal. In vehicle-treated embryos, Ptc1 expression was confined primarily to the ventral neural tube (Figure 3a,c). In embryos treated with Hh-Ag 1.2, however, expression of Ptc1lacZ was greatly extended dorsally in the neural tube and throughout the adjacent mesoderm (Figure 3b,d). These embryos also displayed open rostral neural tubes, similar to those of Ptc1-/- embryos. These experiments demonstrate that the agonist compound effectively activates Hh signaling in vivo following oral administration.
Agonist site of action in vivo
Having established an in utero assay for Hh signaling, we next investigated whether the agonist could rescue aspects of Shh- or Smo-mutant phenotypes, by monitoring lacZ expression in Smo-/-Ptc1lacZ/+ embryos  and Ptc1 mRNA levels in Shh-/- embryos.
Pregnant mice from Shh+/- and Smo+/- intercrosses were treated by oral gavage with vehicle or agonist (15 mg/kg) at 6.5 and 7.5 dpc. Embryos were collected at 6-8 somite stages (E8.5) when the midline defects are first detectable in both Shh-/- and Smo-/- embryos, but prior to any general retardation of growth and development [26,27]. In both Shh+/- and Smo+/-Ptc1lacZ/+ embryos, Ptc1 was detected in ventral neural tube, somites and lateral plate mesoderm (Figure 3e,i,m). Treatment with the agonist dramatically enhanced and expanded the expression of Ptc1 in these heterozygous embryos (Figure 3f,j,n). This was consistent with what we have observed in wild-type embryos (Figure 3a,b,c,d). It is worth noting that the agonist-treated embryos exhibited overgrowth of the headfolds and hindbrain, reminiscent of Ptc1-/- embryos (compare Figure 3e,m with 3f,n).
Shh-/- and Smo-/- embryos at this stage (6-8 somites) started to show fused ventral lips of the cephalic folds, and a single continuous optic vesicle, indicating lack of a clearly defined midline (red arrow, Figure 3g, and data not shown). As expected, Ptc1 expression was not detected in the ventral neural tube of the vehicle-treated Shh-/- embryos (arrowhead, Figure 3g), whereas expression was seen in lateral plate mesoderm and weakly in somites (Figure 3g,k). This is most likely due to Ihh signaling in these tissues . Both Shh and Ihh signaling were dependent on Smo, however, because Ptc1 expression could not be detected in Smo-/- embryos (Figure 3o).
Following agonist treatment, we observed that the neural tube and somite expression of Ptc1 in Shh-/- embryos was greater than vehicle-treated wild-type levels (compare Figure 3h,l with 3e,i). The midline defects in Shh-/- embryos were at least partly rescued by agonist treatment (compare Figure 3g and 3h; red arrows). Like Shh+/- embryos, Shh-/- embryos had overgrown headfolds after administration of the Hh agonist (Figure 3f and h). In contrast, agonist treatment had no detectable effect on either morphology or Ptc1 expression in Smo-/- embryos (compare Figure 3o and p). In summary, these studies demonstrate that agonist activity in vivo does not depend upon Shh, but that Smo is absolutely required.
Mechanism of action
Chemical epistasis studies
We sought to determine the level at which the agonist acts in the Hh pathway, in cultured cell assays. To begin addressing this question, we used the Hh reporter cell line to conduct competition experiments between the Hh agonist and known Hh- signaling antagonists that block the pathway at different levels (Figure 4a). These include: a Hh-protein-blocking antibody, 5E1 ; a natural product derivative, cyclopamine [9,10] that has recently been shown to act downstream of Ptc, perhaps at the level of Smo ; a recently identified synthetic small-molecule inhibitor, Cur61414, which has inhibitory properties similar to cyclopamine ; and forskolin, an adenylate cyclase/protein kinase A activator that is thought to block Hh signaling by stimulating degradation of members of the Gli family of transcriptional activators .
The Hh-blocking antibody 5E1 had no effect on pathway activation by the agonist (Figure 4b), while forskolin (Figure 4c), cyclopamine (Figure 4d) and Cur61414 (Figure 4e), were all inhibitory. The lack of inhibition by 5E1 eliminates the possibility that the small molecule agonist activates signaling indirectly via stimulation of Hh expression. Furthermore, this supports the data showing that the agonist can activate signaling in Shh-/- embryos (Figure 3) and suggests that the agonist function is not only downstream of the Hh protein but also independent of the endogenous Hh-signaling modulators, Tout veloux and HIP, that act via the Hh ligand . The competition experiment with forskolin showed identical inhibition curves for Hh protein and the agonist, strongly suggesting that the action of the small molecule is upstream of the protein-kinase-A-sensitive step in the pathway. In contrast, the competition experiments with cyclopamine (Figure 4d) and Cur61414 (Figure 4e) showed that Hh protein and the agonist differ in their sensitivity to these antagonists. Specifically, the agonist appears somewhat resistant to the inhibitory effect of cyclopamine and Cur61414. Identical results were seen using the slightly less active cyclopamine-related natural compound jervine, and the more potent synthetic derivative of cyclopamine, KAAD-cyclopamine (data not shown). These results argue that the agonist activates the pathway downstream of the Hh-Ptc interaction while cyclopamine, Cur61414 and the agonist may act at a similar level in the Hh-signaling cascade.
Regulation of Ptc and Smo by Hh protein and Hh agonist
Recent work in Drosophila tissue culture has shown that endogenous Ptc and Smo proteins are differentially affected by the addition of Hh to the growth medium . Ptc was destabilized, while Smo accumulated following post-translational modification. To test whether similar phenomena occur in mammalian cells with Hh protein and agonist, we generated stable cell lines expressing two epitope-tagged proteins, Ptc coupled to green fluorescent protein, Ptc-GFP, and Smo coupled to a fragment of influenza hemagglutinin, HA-Smo. Figure 5a shows an immunoprecipitation (anti-GFP) plus protein blot (anti-Ptc) analysis of extracts from these cells treated for 4, 8 and 24 hours with vehicle, 25 nM Hh protein or 0.2 μM Hh agonist (see Figure 1e; Hh-Ag 1.2). This experiment shows that Ptc-GFP appears to be destabilized by Hh protein but not by the agonist. Similar results were seen at higher doses of agonist (up to 2 μM) and in several independent lines (data not shown). These data further support the idea that Hh protein and the agonist act in distinct ways to stimulate the pathway.
Figure 5b shows an immunoblot (anti-HA) of total extracts from HA-Smo-expressing cells treated for 2, 5, 8 and 20 hours with vehicle, 35 nM Hh protein or 0.5 μM Hh agonist. In contrast to the results with Ptc-GFP, incubation of cells with both Hh protein and the small-molecule agonist resulted in the apparent accumulation of HA-Smo protein after 5 hours of incubation. To test whether the accumulation of HA-Smo in response to Hh protein or the agonist required protein synthesis, a similar study was performed in the presence of cycloheximide (Figure 5c). Under these conditions, HA-Smo accumulation was detectable 5 hours after addition of either Hh protein or the agonist (Figure 5c); this result argues that the effect of Hh protein and the agonist on HA-Smo levels does not require new protein synthesis. Finally, with increasing concentrations of Hh protein and the agonist there is a clear dose-dependent increase of HA-Smo levels (Figure 5d). These effects on epitope-tagged Smo protein were observed in multiple lines (data not shown). Taken together, these data suggest that Hh protein and the agonist share the ability to stabilize Smo, but only Hh protein can destabilize Ptc. Yet the agonist is fully capable of activating the full signaling pathway.
Testing Smo as the molecular target
Binding in whole cells
Our biochemistry experiments (above) show that the agonist modulates Smo levels, and thus may activate Hh signaling by directly binding Smo. To explore this possibility we tested whether a tritiated form of the agonist analog Hh-Ag 1.5 could form a complex with Smo, when Smo is transiently overexpressed in 293T cells. Figure 6a shows immunoprecipitable counts of extracts from cells incubated at 37°C for 2 hours with 5 nM [3H]-Hh-Ag 1.5 either in the absence (columns 1-3) or presence of competitors (columns 4-9).
Immunocomplexes from untransfected control and β-adrenergic-receptor transfected cells did not contain significant counts (Figure 6a, columns 1, 2). Immunocomplexes derived from cells expressing Smo (Figure 6a, column 3) resulted in the recovery of approximately 40,000 of the 800,000 added counts, however. To test the specificity of this apparent Hh-Ag/Smo complex, cells were incubated with 5 μM (1000-fold molar excess) of unlabeled Hh Ag 1.5 or an unlabeled, signaling-inactive but structurally similar compound, an Hh-Ag 1.1 derivative that has a two-carbon linker in place of the cyclohexane ring (Figure 6a; Hh-Ag 1.5, column 4; Hh-Ag control, column 5). The addition of the unlabeled Hh-Ag 1.5, but not the inactive Hh-Ag 1.1 derivative agonist control, resulted in the complete absence of counts in the immunocomplex. These results suggest that a stable, specific interaction can form between Smo and the Hh agonist.
It has been shown that the Hh-pathway antagonists cyclopamine and Cur61414 block signaling in a Ptc-independent manner [11,17] and therefore may act directly on Smo. Having established a binding assay for a small-molecule agonist binding to Smo-expressing cells, we next tested whether the Hh antagonists could selectively compete out binding of [3H]-Hh-Ag 1.5. To perform theses studies, Smo-overexpressing 293T cells were incubated for 2 hours at 37°C with 5 nM [3H]-Hh-Ag 1.5 in the presence of either KAAD-cyclopamine at 5 μM (Figure 6a, column 6), the related but inactive plant compound tomatadine at 5 μM (Figure 6a; Antag control 1, column 7), Cur61414 at 5 μM (Figure 6a, column 8), or a related but inactive Cur61414 derivative (Figure 6a; Antag control 2, column 9) at 5 μM. These data show that the Hh-signaling inhibitors, but not structurally related inactive compounds, can significantly compete with the binding of the Hh agonist to Smo-expressing cells. This supports the model that all of these small-molecule modulators of Hh signaling are direct ligands of Smo.
We next asked whether a derivative of the Hh agonist carrying a photoactivatable crosslinker could be coupled directly to Smo, to facilitate further biochemical characterization of the binding site. To perform this experiment we synthesized a tritiated diazirine derivative of Hh-Ag 1.2 with an EC50 in the cell-based assay of 35 nM (data not shown). We incubated this compound at 0.5 μM with HA-Smo- or control, GFP-transfected 293T cells and subsequently ultraviolet-irradiated them to initiate crosslinking. Fractionation by SDS-polyacrylamide gel electrophoresis and autoradiography of the resulting immunocomplexes from these cells showed crosslinking exclusively to HA-Smo, but with an efficiency of less than 1% (data not shown). This result demonstrates that a Hh-agonist derivative can be covalently crosslinked to Smo in living cells. More efficient crosslinkers are required to extend these studies, however.
Cell-free membrane-binding assays
To test whether the Hh agonist could interact with Smo in vitro, we transiently overexpressed murine Smo, murine Ptc, rat β2-adrenergic receptor and GFP in 293T cells, harvested membranes and performed a filtration membrane-binding assay in a 96-well plate with [3H]-Hh-Ag 1.5 added at 2 nM. Figure 6b shows a bar graph of the bound counts from these binding assays (murine Smo, column 1; GFP, column 2; βAR, column 3; murine Ptc1, column 4; and a no-membrane plate control, column 5). The no-membrane control (column 5) was included to show the degree of non-specific binding to the filter-plate apparatus. The Smo-containing membranes (column 1) are the only samples that exhibit significant binding above that seen in the absence of membranes.
To assess the specificity of binding, we repeated the experiment in the presence and absence of a 1000-fold molar excess of unlabeled agonist (2 μM). The addition of 'cold' compound completely competed out these counts (Figure 6c, column 1 compared to column 2). To control for this observation, we added an unlabeled, inactive Hh agonist to the binding assay at 2 μM (a 1000-fold molar excess). This compound was unable to compete out the binding of [3H]-Hh-Ag 1.5 to the Smo-containing membranes (Figure 6c, column 3). These results argue that Smo and the agonist form a specific complex in vitro, as predicted by the whole cell/immunocomplex binding assay (Figure 6a).
Having established an in vitro binding assay for the Hh agonist to Smo, we next tested whether the Ptc-independent Hh antagonists could selectively compete with the interaction. Binding was assayed in the presence of KAAD-cyclopamine at 10 μM (Figure 6c, column 4), tomatadine at 10 μM (Figure 6c; Antag control 1, column 5), Cur61414 at 10 μM (Figure 6c, column 6), or the inactive Cur61414 derivative (Figure 6c; Antag control 2, column 7) at 10 μM. These data show that the Hh-signaling inhibitors, but not structurally related inactive compounds, can significantly compete with the binding of the Hh agonist to Smo membranes.
Kinetics, saturation and competition binding analysis
Next, we sought to generate association, dissociation and saturation-binding curves, in order to derive affinity constants for the interaction of the Hh agonist and Smo. To control for nonspecific binding we used either Cur61414 or Hh-Ag 1.5 as unlabeled competitors. Similar results were generated if control membranes (from cells transfected with GFP, βAR, or Ptc) were used to define the non-specific level (data not shown).
First, we performed a kinetic analysis to establish the reversibility of the binding reaction and the approximate incubation time required for equilibrium binding studies. Association assays were performed at 37°C by combining 2 nM [3H]-Hh-Ag 1.5 with Smo-containing membranes for various times prior to harvesting and counting. Dissociation studies were initiated by adding 2 μM unlabeled Hh-Ag 1.5 after 2 hours of association. Samples were then incubated for 1-26 hours prior to harvesting and counting. Figure 7a shows the association and dissociation phases of agonist binding to Smo-containing membranes. Using the Prism GraphPad software, these data were fit to one-phase exponential association and decay curves, respectively, and gave an association t½ of approximately 1 hour and a dissociation t½ of approximately 10 hours. These results demonstrate that the binding of the agonist to Smo is reversible and that equilibrium binding will require binding reaction times of approximately 50 hours (five times the t½ of dissociation).
Next, we performed a saturation binding experiment. To establish total, nonspecific, and specific binding curves (Figure 7b), we added a range of [3H]-Hh-Ag 1.5 concentrations (0.01-3 nM) in the presence or absence of unlabeled Hh-Ag-1.5 at 2 μM. Identical results were seen if Cur61414 at 10 μM was used as the competitor (data not shown). On the basis of the binding kinetics, incubations were carried out for approximately 45 hours at 37°C, to allow for equilibrium to be reached. Using the Prism GraphPad software to perform non-linear regression analysis and curve fitting, we concluded that the data best fit a simple one-site binding model with a predicted Kd of 0.37 nM for Hh-Ag 1.5. This Kd is in general agreement with the EC50 values observed in the cell-based assay (0.37 nM as compared to 1 nM).
To further validate our binding results, we performed a competition assay using several agonist derivatives across a range of concentrations (0.01 nM to 1 μM). Figure 7c shows the competition curves for Hh-Ag 1.5, Hh-Ag 1.3, Hh-Ag 1.2, Hh-Ag 1.1, and the signaling-inactive Hh-Ag 1.1 derivative described above. With the exception of the inactive derivative, these compounds all compete out the binding of [3H]-Hh-Ag 1.5 (0.4 nM) to the Smo-containing membranes. These data are best fit to a single-binding-site competition model that predicts the following Ki values: Hh-Ag 1.5, 0.52 nM; Hh-Ag 1.3, 8.4 nM; Hh-Ag 1.2, 22 nM; and Hh-Ag 1.1, 96 nM. These Ki values are in general agreement with the agonist EC50 values in cell culture for these compounds, with the exception that the Hh-Ag 1.1 compound is not as potent in signaling assays (EC50 2 μM) as its Ki (96 nM) might predict. This suggests an uncoupling of binding and signaling for certain agonists. Although binding affinities and signaling efficacy can correspond for certain ligand/receptor complexes, exceptions often arise  because binding affinity does not necessarily measure the ability of a compound to induce an active receptor conformation. As a control for these binding studies, identical competition experiments were performed with membranes from cells transfected with GFP, or with the β-adrenergic receptor. No specific binding or apparent competition was seen under these conditions (data not shown).
We next compared binding of KAAD-cyclopamine, Cur61414 and Hh-Ag 1.5 to a constitutively active mutant of Smo (Smoact) or to wild-type Smo (Smowt). The two Hh-signaling antagonists, KAAD-cyclopamine and Cur61414, have shown decreased potency on Smoact-expressing cells, leading to the speculation that they may bind this mutant form of Smo less well than the wild-type form [11,17]. We sought to determine whether the Hh agonist binds Smoact with a higher affinity, an observation seen with certain ligands and constitutively active mutants of GPCRs . To perform this experiment, we isolated membranes from cells transfected with a cDNA construct encoding a tryptophan-to-leucine mutation at residue 539 (W539L) of murine Smo. This oncogenic mutation has been found in human basal cell carcinoma  and the correspondingly mutated protein is capable of ligand-independent activation of the Hh pathway in cell-culture assays . A kinetic and saturation binding assay with Smoact-containing membranes showed that this mutant protein binds the Hh agonist with an affinity identical to that of Smowt (data not shown).
Using Smoact- and Smowt-containing membranes, we then performed competition binding studies by adding increasing concentrations of unlabeled Hh-Ag 1.5, KAAD-cyclopamine or Cur61414 in the presence of [3H]-Hh-Ag 1.5 (0.4 nM). These binding curves (Figure 7d) can be fit to a single-site competition model. Although the Ki for Hh-Ag 1.5 on Smoact-containing membranes was essentially identical to that observed for Smowt-containing membranes (approximately 0.5 nM), the Ki values of the Hh antagonists were seven-fold higher on the Smoact- compared to the Smowt-containing membranes for both KAAD-cyclopamine (38.3 nM versus 5.8 nM) and Cur61414 (309 nM versus 44 nM). These results strongly support the model, initially hypothesized for cyclopamine , that the reduced potency observed for Hh antagonists on Smoact-expressing cells is directly due to a reduced affinity of the antagonist for the mutated Smo protein. The agonist, on the other hand, would be predicted to bind to a site on Smo that is not affected by this gain-of-function Smoact mutation.