Structural basis of Smoothened regulation by its extracellular domains

Abstract

Developmental signals of the Hedgehog (Hh) and Wnt families are transduced across the membrane by Frizzledclass G-protein-coupled receptors (GPCRs) composed of both a heptahelical transmembrane domain (TMD) and an extracellular cysteine-rich domain (CRD). How the large extracellular domains of GPCRs regulate signalling by the TMD is unknown. We present crystal structures of the Hh signal transducer and oncoprotein Smoothened, a GPCR that contains two distinct ligand-binding sites: one in its TMD and one in the CRD. The CRD is stacked a top the TMD, separated by an intervening wedge-like linker domain. Structure-guided mutations show that the interface between the CRD, linker domain and TMD stabilizes the inactive state of Smoothened. Unexpectedly, we find a cholesterol molecule bound to Smoothened in the CRD binding site. Mutations predicted to prevent cholesterol binding impair the ability of Smoothened to transmit native Hh signals. Binding of a clinically used antagonist, vismodegib, to the TMD induces a conformational change that is propagated to the CRD, resulting in loss of cholesterol from the CRD-linker domain-TMD interface. Our results clarify the structural mechanism by which the activity of a GPCR is controlled by ligand-regulated interactions between its extracellular and transmembrane domains.

Extended Data Figure 1. Sequence alignment of SMO orthologues

Numbering corresponds to that of human SMO. Secondary structure assignments are displayed above the alignment and colour-coded as in Fig. 1. Black arrows and numbers (fX.50) below alignment show class F Ballesteros-Weinstein nomenclature for GPCR helices. Residues interacting with cholesterol are highlighted in red. Disulphide bridges are highlighted in yellow and numbered. N-linked glycosylation sites are depicted by a hexagon. Position of the Val329Phe point mutation is highlighted in purple.

Extended Data Figure 2. Characterization of the SMO Val329Phe mutation

a, Superposition of SMOΔC structure (blue) with the SMO-SANT-1 complex structure, which lacks the CRD (green, PDB 4N4W ), showing the TMD ligand-binding pocket as a yellow surface. Inset shows Val329, mutated to Phe in our structure. b, SEC analysis of fluorescently-labelled SMOΔC showing difference in expression levels of wild-type and Val329Phe variant (main protein peak ~20 min). c, 20(S)-OHC beads can bind both mouse WT SMO and Val333Phe (mouse Val333 corresponds to human Val329). Immunoblots, using an anti-SMO antibody directed against the ICD, were used to measure SMO captured on 20(S)-OHC beads. Adding 50 μM free 20(S)-OHC as a competitor reduced binding. d, Purified human SMOΔC (the crystallization construct) binds to 20(S)-OHC beads. e, Smo^−/− mouse fibroblasts stably expressing SMO-WT or SMO-Val333Phe were exposed to SHH, SAG or 20(S)-OHC. Levels of endogenous Gli1 mRNA (mean +/− SD, n=4), measured by qRT-PCR, were used as a metric of Hh pathway activity since Gli1 is a direct Hh target gene. Asterisks indicate statistical significance for the difference in Gli1 mRNA levels between identically treated SMO-WT and SMO-Val333Phe cells. f, Immunoblot shows SMO and GLI1 protein levels in these stable cell lines, with p38 as loading control.

Extended Data Figure 3. Crystallization, structure solution and oligomeric state of SMOΔC

a, SMOΔC crystal packing. Asymmetric unit consists of two antiparallel SMOΔC chains. Chain A coloured as in Fig. 1 with BRIL-fusion in yellow; Chain B in grey. LCP crystal packing with alternating hydrophobic and hydrophilic layers perpendicular to the c axis. Molecules coloured as for Chain A. b, Pearson correlation coefficient (CC) analysis used to relate data quality with model quality. A CC_work and CC_free smaller than CC* indicates that the model does not account for all of the signal in the data (and is therefore not overfit). c-e, SigmaA-weighted 2Fo-Fc electron density maps of final refinement at 1.0 σ contour level. c, Val329Phe mutation; d, extra density within TMD ligand-binding pocket (Fo-Fc maps shown at contour level of +3σ (green) and −3σ (red)) (N.B. This density could not be confidently assigned, likely because of low occupancy within the crystal.); e, ‘connector’ region linking CRD and LD, with Asn188 and linked N-acetyl glycosamine moiety. f, SEC-MALS analysis of amphipol-solubilised SMOΔC. Molar masses (M_W, black lines) and 280 nm absorption (grey line) plotted against elution time. M_W derived from protein-conjugate analysis indicated in parentheses. For clarity, graphs of M_W are only shown around main absorption peak. Theoretical M_W of SMOΔC based on sequence is 71 kDa with the extra mass observed in MALS (~13 kDa) likely due to three N-linked glycosylation sites. Expected mass of protein-bound amphipol (A8-35) was previously determined to be 40-75 kDa, in agreement with our data. This analysis suggest that SMOΔC is a monomer under our purification conditions.

Extended Data Figure 4. Interfaces of the SMO CRD

Detailed interactions of the CRD with the connector region (a) and with the LD-TMD segment (b). The number of interactions are indicated in the top panel and coloured as indicated in the key box. For non-bonded contacts, the width of the striped line is proportional to the number of atomic contacts. Residue colouring is according to amino acid: blue, positive (H,K,R); red, negative (D,E); green, neutral (S,T,N,Q); grey, aliphatic (A,V,L,I,M); purple, aromatic (F,Y,W); orange, proline (P) or glycine (G); yellow, cysteine (C). The figure is adopted from the PDBSUM server.

Extended Data Figure 5. Comparison of SMOΔC with previously determined SMO TMD structures

Superposition of the SMOΔC structure with SMO TMD structures. Structural alignment was performed using the 7TM bundle as template (not including the LD or TMD helix 8). SMOΔC (red), SMO TMD complexed with cyclopamine (light orange, PDB 4O9R, r.m.s.d. 0.598 Å for 243 equivalent Cα positions), antaXV (light blue, PDB 4QIM, r.m.s.d. 0.515 Å for 233 equivalent Cα positions), SANT1 (pale cyan, PDB 4N4W, r.m.s.d. 0.483 Å for 240 equivalent Cα positions), LY2940680 (pale green, PDB 4JKV, r.m.s.d. 0.493 Å for 230 equivalent Cα positions), SAG1.5 (pale yellow, PDB 4QIN, r.m.s.d. 0.623 Å for 262 equivalent Cα positions). The box shows a close-up view of the LD region revealing a structural rearrangement in the SMOΔC structure compared to the previously determined SMO TMD structures lacking the native extracellular domain.

Extended Data Figure 6. Cholesterol stabilises SMO. a-e. Molecular dynamics simulations of SMO in a lipid bilayer

a, SMO embedded in a lipid bilayer with the CRD in orange, the 7-pass transmembrane region excluding intra- and extracellular loops (7TM) in blue and cholesterol in cyan. b-d, Relative r.m.s. fluctuations of the Cα atoms over the course of 5 × 100 ns of atomistic molecular dynamics simulation in the presence and absence of the cholesterol. The structures in b and c are shown as putty representations coloured from high conformational stability (i.e low r.m.s. fluctuations; blue/thin) to low stability (i.e high r.m.s. fluctuations; red/thick). e, Secondary structure DSSP matrices for each of the simulations. The asterisks in b, c and e all mark the helix spanning from residues 155-160, that is destabilised in the absence of bound cholesterol. f-g, Thermostability of purified SMOΔC. See supplementary discussion for details. f, Compiled peak heights from thermostability SEC analysis of purified SMOΔC after treatment with different MBCD concentrations. g, Example of raw SEC data used for the analysis in f. Samples were incubated at 35°C for 1 h prior to loading onto the SEC column.

Extended Data Figure 7. Effect of domain interface mutations on expression levels and 20(S)-OHC binding

a, Protein levels of SMO and also of PTCH1 and GLI1 (each of which is encoded by a direct Hh target gene) measured by immunoblot from Smo−/− mouse fibroblasts stably expressing one of five SMO variants: wild-type SMO (WT); SMO lacking the entire CRD (ΔCRD); SMO with two mutations (Pro120Ser or Ile160Asn/Glu162Thr) that introduce glycosylation sites in the LD-CRD interface; and, SMO lacking a conserved disulphide bond (Cys197Ser/Cys217Ser, marked “6” in Fig. 4c) in the LD. Elevated levels of GLI1 and PTCH1 reflect high constitutive signalling activity of each mutant. NS: a non-specific band detected by anti-PTCH1 antibody; SUFU: loading control. Different patterns seen in SMO panel is caused by different numbers of N-linked glycosylation sites. b, Gli1 mRNA levels were used to assess Hh signalling activity in Smo−/− cells stably expressing the indicated mouse SMO variants. D477R and M2 (Trp539Leu) are two previously described mutations in the TMD that increase constitutive signalling. c, Oxysterol-binding capacity of each SMO variant was determined (right blot) by its ability to bind to 20(S)-OHC beads in the absence or presence of 50 μM free 20(S)-OHC. Inputs for each binding reaction are shown on the left.

Extended Data Fig. 8. SAXS analysis of SMOΔC

a, Overlay of size-exclusion chromatograms monitored at 280 nm (A₂₈₀) collected during SAXS measurements for apo-SMOΔC (red), (+) 20(S)-OHC SMOΔC (blue), amphipol (green) and BSA standard (black). Amphipol and BSA were injected at 10 mg/ml. Insert shows curves normalized to peak height. BSA was used as a reference with a radius-of-hydration of 3.7 nm. Absorbance of the free amphipol is negligible and elutes ~5 minutes after the amphipol-stabilised SMOΔC samples. b, Dimensionless Kratky plot of apo and (+) 20(S)-OHC-loaded SMOΔC SAXS data. Cross-hairs denote the Guinier–Kratky point (√3, 1.1), the peak position for an ideal, globular particle. The slower decay of the transformed scattering intensities for (+) 20(S)-OHC (blue) indicate a comparably less spherical particle.

Extended Data Fig. 9. Crystal structure of the SMOΔC-vismodegib complex and structural analysis of mutations found in vismodegib-resistant cancers

a, Chemical structure of vismodegib. b, Close-up view of vismodegib binding site. Colour-coding follows Fig. 5b. Composite omit map calculated with PHENIX at 1.0σ shown as magenta chicken-wire. c, Mapping of residues that are mutated in vismodegib-resistant tumours (yellow highlights). Brackets indicate mutant residues. d-f, Close-up views of selected interactions. Native residues in blue and mutated residues in yellow. Arrows indicate position of potential clashes that could disrupt vismodegib binding. d, Gln477/Asp473 hot spot. The Gln477Glu mutation leads to a loss of the potential hydrogen bond of the glutamine sidechain to the chloride of the vismodegib chlorophenyl-methylsulfone moiety. The Asp473His mutation potentially destabilises the hydrogen-bonding network around Arg400 that coordinates the vismodegib chlorophenyl-methylsulfone moiety. e, The imidazole ring of His231 is in hydrogen-bonding distance to two carbonyl main-chain atoms of residues Ser385 and Val386 located on a loop coordinating the interaction of Asp384 to vismodegib's amide linker. f, Trp281 forms a key hydrophobic interaction with the vismodegib pyrimidine ring that is deeply buried in the SMO helical bundle core. Mutation to cysteine would significantly destabilise this interaction while mutation of nearby Val321 to the bulkier methionine would likely result in a rearrangement of the Trp281 side chain. g-h, SMOΔC captured on cholesterol beads in the presence of increasing concentrations of free vismodegib or 20(S)-OHC (h).

Figure 1. Structure of human SMO

a, Two views of the overall structure showing extracellular and transmembrane domains of human SMO in cartoon representation with the CRD in orange, LD in pink, TMD in blue. The inactivating point mutation Val329Phe is depicted in red, cholesterol in cyan, nine numbered disulphide bridges in black, and two N-linked glycans (NAG) as yellow sticks. A schematic of SMO is shown above (SP: signal peptide, BRIL: position of the BRIL fusion protein inserted between TMD helices 5 and 6). b, The ‘connector’ region between CRD and LD highlighted as sticks in atomic colouring, with the CRD shown as a solvent accessible surface and the LD and part of TMD ECL3 loop as cartoons. c, Interface between CRD, LD and TMD shown in cartoon representation with ECL3-NAG and cholesterol as yellow and cyan sticks, respectively.

Figure 2. The cholesterol binding site

a-b, Close-up of cholesterol with interacting residues as sticks. Initial 2Fo-Fc map at 1.0 σ before inclusion of cholesterol shown as chicken-wire. Colour-coding follows Fig.1. Inset shows the potential hydrogen-bonding network coordinating the cholesterol 3β-hydroxyl group. Interatomic distances (Å) are shown in black. c, Solvent accessible surface colour-coded by hydrophobicity (red: hydrophobic, white: hydrophilic). d, Sequence conservation (based on 55 vertebrate SMO sequences) mapped onto SMOΔC (black: conserved, white: not conserved). e, Superposition of human (orange, this study) and fly (purple, PDB 2MAH) SMO CRD structures. The fly CRD region occupying the cholesterol binding site is highlighted.

Figure 3. The cholesterol-binding site regulates SMO signalling activity

a, Purified SMOΔC captured on beads coupled to increasing concentrations of cholesterol in the presence or absence of free 20(S)-OHC. b, SMOΔC captured on cholesterol beads in the presence of increasing concentrations of free 20(S)-OHC or c, in the presence of a synthetic epimer 20(R)-OHC. d, Protein levels of SMO variants stably expressed in Smo^−/− mouse fibroblasts. e, Binding of SMO variants to 20(S)-OHC-coupled beads. Asp99Ala and Tyr134Phe are predicted to disrupt hydrogen bonding with cholesterol (Fig.2a). f, Levels of Gli1 mRNA (mean +/− SD, n=4) were used as a metric for Hh signalling activity in cell lines shown in d after stimulation with SHH, SAG or 20(S)-OHC. Statistical significance is denoted for the difference in Gli1 mRNA levels between cells expressing SMO-WT compared to each mutant SMO protein.

Figure 4. SMO activity is regulated by the stability of its extracellular domain

a-b, MD simulations of SMO. Cα r.m.s.d. versus time for CRD (a) and 7TM region (b) with (blues) and without (reds) cholesterol. c-d, Mutations in the extracellular region increase constitutive signalling activity of SMO. c, CRD-LD interface with mutated residues highlighted. Corresponding mouse residues are in brackets. d, Gli1 mRNA levels (mean +/− SD, n=4) after treatment with agonists (SHH, SAG, 20(S)-yne) or antagonists (cyclopamine, SANT-1). Gli1 levels in untreated cells reflect constitutive activity of SMO. Numbers above bars indicate Gli1 fold-increase compared to untreated cells. Asterisks denote statistical significance for comparison with WT SMO. e, Solution-state SAXS profiles of untreated (grey) and 20(S)-OHC-treated (red) SMOΔC. Lines are fits derived from the indirect Fourier transform of e. f, Paired-distance, P(r), distribution functions, normalised to their respective I(0) values, with maximum particle sizes (d_max) indicated.

Figure 5. Structure of SMO in complex with the antagonist Vismodegib

a, Overall structure showing full extracellular and transmembrane domains of human SMO (cartoon) in complex with vismodegib (green). Colour-coding follows Fig.1. b, Close-up of the vismodegib-binding site. The asterisk denotes a residue mutated in vismodegib-resistant tumours. c-e Comparison of apo-SMO (red) and vismo-SMO (blue). c, Superposition of the two SMO structures based on the TMD. Arrows indicate domain rotations. Cholesterol (red) and vismodegib (blue) shown as spheres. d, Close-up of the TMD/LD/CRD interface. Phe484 contacts the vismodegib methylsulfone group. e, Close-up of sterol-binding site. Cholesterol, Arg161 and NAG shown as sticks. The partly-disordered ECL3 loop of vismo-SMO is depicted as dotted line.