The Notch pathway is a fundamental signalling system in most multicellular animals. We have determined the X-ray crystal structure of the extracellular domain of the Notch ligand delta-like ligand-1 (Dll-1). The structure incorporates the N-terminal C2 domain, receptor-binding DSL domain and the first six (of eight) EGF (epidermal growth factor)-like repeats, which form a highly extended conformation, confirmed by analytical ultracentrifugation. Comparison of our structure with a fragment of Jagged1 ligand allows us to dissect the similarities and differences between the ligand families. Differences in the C2 domains of Dll-1 and Jagged1 suggest their lipid-binding properties are likely to differ. A conserved hydrophobic patch on the surface of both Dll-1 and Jagged1 provides a likely receptor-interaction site that is common to both ligands. We also explore the binding affinity of Dll-1 for a fragment of Notch1 using different techniques. Apparent binding affinities vary when different techniques are used, explaining discrepancies in the literature. Using analytical ultracentrifugation, we perform for the first time binding analyses where both receptor and ligand are in solution, which confirms a Kd of 10 μM for this interaction.

INTRODUCTION

The field of Notch signalling was born with the observation of the Notched wing phenotype in Drosophila, as characterized by John S. Dexter in 1914 [1]. The sequencing and molecular analysis of the gene responsible for the wing phenotype was performed in the 1980s and as a consequence the Notch receptor was identified [2,3]. The Notch pathway is an evolutionarily conserved signalling system found in most multi-cellular animals. It regulates many aspects of development and tissue homoeostasis, playing a vital role in the regulation of cell fate by maintaining a balance between cell proliferation, differentiation and apoptosis [4]. Both the Notch receptors and the ligands are single-pass type I transmembrane proteins and as such direct cell–cell contact is required for signal transmission. This direct contact allows precise control of cell differentiation and patterning, which is required for tissue formation. Given the fundamental role of Notch signalling, it is perhaps not surprising that mutations in the Notch signalling pathway are associated with a number of diseases, including several types of cancer [5].

Notch signalling is unique in that the cell-surface receptor that recognizes ligand also functions as part of the transcriptional complex that elicits the biological response. Binding of ligand to the extracellular domain of the Notch receptor initiates a series of proteolytic events, which ultimately result in release of the intracellular domain of the receptor. A poorly-understood conformational change in the membrane-proximal ‘negative-regulatory region’ of the receptor [6] is triggered by ligand binding, providing access for ADAM (a disintegrin and metalloproteinase) family metalloproteases, which in turn produces a substrate for the membrane-bound γ-secretase complex [7,8]. After cleavage by γ-secretase, the intracellular domain is released from the membrane and translocates to the nucleus where it associates with co-transcription factors RBP–JK (recombination signal binding protein for immunoglobulin kappa J region) [9] and MAML (mastermind-like) [10] to initiate the transcription of Notch target genes.

In mammals, there are four Notch receptors (Notch1–4) and five canonical ‘DSL’ (Delta-Serrate-Lag) family ligands: delta-like ligands (Dll) 1, 3 and 4 and Jagged 1 and 2. Both ligand and receptor are made up of a series of epidermal growth factor (EGF)-like repeats, increasing in number from Dlls to Jagged ligands to receptor (which has over 30 tandem EGF repeats). A recent structure of an N-terminal fragment of the Notch ligand Jagged1 revealed the presence of a calcium-dependent C2 phospholipid-binding domain at the N-terminus of all Notch ligands [11]. However, after decades of research on Notch signalling, the molecular details of the receptor–ligand interaction still elude us. Atomic resolution information on even small portions of receptor or ligand have been scarce [1114]. In the present study, we report the X-ray crystal structure of the ectodomain of the Notch-ligand Dll-1 and investigate the binding affinity of a fragment of the Notch1 receptor for Dll-1 using ELISA and analytical ultracentrifugation techniques.

EXPERIMENTAL

Protein expression and purification

The extracellular domain of human Dll-1 (residues 1–545) was cloned into the pApex vector to generate a C-terminal Fc–FLAG (fragment crystallisable region of human IgG1), fusion protein. A TEV (tobacco etch virus) cleavage site was incorporated before the Fc to allow removal of the Fc–FLAG tag. Freestyle™ 293-F cells (Life Technologies) were grown to a density of 2×106/ml in FreeStyle 293 expression medium and transiently transfected with DNA and polyethyleneimine (PEI) at a 3:1 PEI–DNA ratio (1 mg DNA per litre). Cells were grown for 7 days after transfection and supplemented with Glutamax (Life Technologies), 0.2 mM butyric acid (Sigma–Aldrich) and 5 g/l lupin (Solabia) 1 and 4 days after transfection. Secreted recombinant protein was purified using Protein G resin (GE Life Sciences). Where required, TEV cleavage was performed in the presence of 2 mM 2-mercaptoethanol. The protein was subsequently concentrated and applied to a Superdex 200 size exclusion column (GE Life Sciences) equilibrated in TBS, pH 7.5.

Human Notch1-1–526 [N11–13 (Notch1 EGF repeats 1–13)] was cloned into pApex to generate an Fc–FLAG fusion, then sub-cloned into the pEE14.4 vector (Lonza) and transfected into CHO (Chinese Hampster Ovary) cells. A stable cell line was generated in the presence of the glutamine synthase inhibitor L-methionine sulfoximine (Sigma–Aldrich). Cells were grown in glutamine-free DME (Dulbecco's Modified Eagle's) (Lonza) supplemented with 10% dialysed fetal calf serum, GS-supplement (Sigma) and 50 μM L-methionine sulfoximine. Cells were grown in roller-bottles for 7 days prior to harvesting conditioned medium. Protein secreted into the medium was purified using M2 resin (Sigma), followed by size exclusion chromatography using an S200 column (GE Life Sciences). Typical protein yields after M2-purification were 0.25–0.5 mg/l.

Crystallization and structure elucidation of Dll-1

Crystallization was accomplished by hanging-drop vapour diffusion at 20°C, using protein at 4 mg/ml and 1:1 ratio of protein and reservoir solution. Crystals were obtained from 0.4 M potassium thiocyanate, 16% PEG3350, 0.1 M bis-tris propane, pH 7.5, 5 mM CaCl2 and grew to ~70 μm in 30–60 days, after initial phase separation. Crystals were flash frozen in liquid nitrogen, using reservoir solution supplemented with 20% glycerol as cryoprotectant.

Diffraction data were collected at 100 K on beamline MX2 at the Australian Synchrotron. Data were processed using XDS and scaled using XSCALE. Data cut-off was based on CC1/2, which was 0.2 in the outer shell. A molecular replacement solution was found using PHASER [27], using the C2 domain of Jagged1, from PDB 4CBZ, as a model [11]. Iterative rounds of restrained refinement and manual model building were performed using PHENIX [28] and COOT [29] respectively. Due to the low resolution of the data and limited secondary structure, Ramachandran restraints were used throughout the refinement.

ELISA assays

TEV-cleaved N11–13 or Dll-1 was diluted to 10 μg/ml in sodium carbonate buffer, pH 9.6 and use to coat wells of a 96-well plate by overnight incubation at 4°C. Protein was discarded and wells washed four times with wash buffer (20 mM Tris/HCl, pH 7.5, 150 mM NaCl, 0.05% Tween 20, 0.5% BSA, 5 mM CaCl2) followed by blocking with wash buffer for 1 h at 37°C. The buffer was discarded and Fc-tagged protein (N11–13 or Dll-1) added (6 μM to 0.1 nM) for 2 h at 37°C. Wells were washed again four times in wash buffer. Horseradish peroxidase (HRP)-coupled protein G (Life Technologies) was used to detect bound Fc-fusion protein, incubating at 37°C for 1 h. Wells were washed four times with wash buffer and once with PBS before the addition of Super Signal West Fempto Chemiluminescent substrate (Pierce). Luminescence was recorded using a BMG Labtech Lumistar Optima.

Analytical ultracentrifugation

Sedimentation velocity and sedimentation equilibrium experiments were performed using a XL-I analytical ultracentrifuge (Beckman Coulter) equipped with UV/Vis scanning optics. Buffer reference (TBS, 5 mM CaCl2, pH 7.5) and sample solutions were loaded into 12-mm double-sector cells with quartz windows and the cells were mounted in an An-60Ti 4-hole rotor or An-50Ti 8-hole rotor. Buffer density (1.0055 g/ml) and buffer viscosity (1.022 cp) were calculated using SEDNTERP [30].

Sedimentation velocity: Dll-1 and N11–13 (5.5 μM, 380 μl) were each centrifuged alone at 201600 g and 20°C. Radial absorbance data were collected at 280 nm in continuous mode every 6 min. A titration series of Dll-1 and N11–13 (350 μl) was produced holding the concentration of N11–13 constant (4 μM) and altering the concentration of Dll-1 (1, 4, 16, 24, 32 μM). These samples were centrifuged at 201600 g and 20°C. Radial absorbance data were collected in continuous mode every 4 min at 280 nm for samples containing 1, 4 and 16 μM Dll-1 and at 290 nm for samples containing 24 and 32 μM Dll-1. Sedimentation velocity data were fitted to a continuous sedimentation coefficient distribution [c(s)] model using SEDFIT[31]. c(s) Distributions for mixtures containing 24 and 32 μM Dll-1 were normalized to absorbance at 280 nm using a correction factor of 2.0. Axial ratios for Dll-1 and N11–13 were calculated with SEDFIT using frictional ratios (f/f0) estimated from the fit to sedimentation velocity data. Theoretical sedimentation coefficients were calculated from the Dll-1 crystal structure co-ordinates using HYDROPRO [32] under experimental conditions described above.

Sedimentation equilibrium: Dll-1 and N11–13 (5.0 μM, 120 μl) were each centrifuged alone at 3513, 10297 and 29111 g and 20°C until sedimentation equilibrium had been reached. Radial absorbance data at equilibrium were obtained at 280 nm in step mode with a radial increment of 0.001 cm. Data obtained at all rotor speeds for each protein were fitted globally to a single species model with mass conservation using the program SEDPHAT [33], to determine accurate, apparent molecular masses for the proteins in isolation.

A solution containing equimolar concentrations of Dll-1 and N11–13 was serially diluted (10.0–1.2 μM, 120 μl). Samples were centrifuged at 3513, 6532 and 15805 g at 20°C until sedimentation equilibrium had been reached. Radial absorbance data at equilibrium were obtained at 250 nm in step mode with a radial increment of 0.001 cm. Data obtained at all rotor speeds and protein concentrations were fitted globally using the ‘AB hetero-association’ model, implemented in SEDPHAT, incorporating mass conservation, fixing the apparent molecular masses determined for each protein and constraining the Dll-1–N11–13 molar ratio as identical for all samples. The molar absorption coefficients at 250 nm of Dll-1 (42272 M−1 cm−1) and N11–13 (36914 M−1 cm−1) were determined from the calculated absorption coefficients at 280 nm using absorbance data collected with the XLI analytical ultracentrifuge at both wavelengths and these were fixed in the analysis. The Dll-1–N11–13 molar ratio and total loading concentration in each cell were fitted in separate fitting steps before the global fit to the data was minimized. All fits were carried out assuming an operational partial specific volume for the two proteins of 0.73, corresponding to that used for apparent mass calculations. Confidence intervals for the Dll-1–N11–13 complex dissociation constant were estimated using the error surface projection method within SEDPHAT.

RESULTS AND DISCUSSION

X-ray crystal structure of Dll-1 extracellular domain

We have determined the structure of the entire extracellular domain of the human Notch ligand Dll-1. The ectodomain of Dll-1 was expressed as an Fc fusion protein in 293F cells. A three-step purification strategy of protein G affinity chromatography, TEV cleavage and size exclusion chromatography yielded protein that was >95% pure by SDS/PAGE analysis. Crystals of the ectodomain were obtained which belonged to the space group P212121, with two copies of Dll-1 in the asymmetric unit (Table 1). The structure was solved using the C2 domain of Jagged1 (PDB ID 4CBZ) as a model for molecular replacement. The structure was refined including data to 3.2 Å (1 Å=0.1 nm). A representative section of the electron density is shown in Supplementary Figure S1. Whereas the entire ectodomain was present in the crystals, as confirmed by SDS/PAGE analysis of the crystals, not all of Dll-1 was visible in the X-ray structure. In one copy of Dll-1, residues 21–442 were visible, comprising the C2 domain, DSL domain and EGF-like repeats 1–6 (i.e. missing the last two EGF-like repeats); in the second copy, only the C2, DSL and first two EGF domains were visible (residues 22–291). The two copies of Dll-1 in the asymmetric unit overlay with an RMSD of 2.3 Å. They form what is essentially a symmetrical dimer interface mediated by the C2 and DSL domains (Supplementary Figure S2A).

Table 1
Data collection and refinement statistics
Data collection  
Wavelength (Å) 0.9537 
Resolution range (Å) 67.4–3.2 (3.314–3.2) 
Space group P 21 21 21 
Cell dimensions  
 a, b, c (Å) 108.15, 118.87, 134.81 
 α, β, γ (°) 90, 90, 90 
Total reflections 212113 (21408) 
Unique reflections 29271 (2872) 
Multiplicity 7.2 (7.5) 
Completeness (%) 99.9 (99.9) 
I/sigma(I) 6.27 (0.80) 
Wilson B-factor 85.71 
R-merge 0.432 (3.096) 
CC1/2 0.988 (0.266) 
CC* 0.997 (0.648) 
Refinement  
Resolution range (Å) 67.4–3.2 (3.314–3.2) 
Reflections used in refinement 29267 (2871) 
Reflections used for R-free 1997 (196) 
R-work 0.2522 (0.3644) 
R-free 0.2791 (0.3648) 
 CC-work 0.837 (0.509) 
 CC-free 0.806 (0.527) 
Number of non-hydrogen atoms 4993 
 Protein 4983 
 Carbohydrate 10 
Protein residues 654 
RMSD (bonds) 0.003 
RMSS (angles) 0.93 
Ramachandran favoured (%) 92 
Ramachandran allowed (%) 6.8 
Ramachandran outliers (%) 0.79 
Average B-factor 87.75 
Data collection  
Wavelength (Å) 0.9537 
Resolution range (Å) 67.4–3.2 (3.314–3.2) 
Space group P 21 21 21 
Cell dimensions  
 a, b, c (Å) 108.15, 118.87, 134.81 
 α, β, γ (°) 90, 90, 90 
Total reflections 212113 (21408) 
Unique reflections 29271 (2872) 
Multiplicity 7.2 (7.5) 
Completeness (%) 99.9 (99.9) 
I/sigma(I) 6.27 (0.80) 
Wilson B-factor 85.71 
R-merge 0.432 (3.096) 
CC1/2 0.988 (0.266) 
CC* 0.997 (0.648) 
Refinement  
Resolution range (Å) 67.4–3.2 (3.314–3.2) 
Reflections used in refinement 29267 (2871) 
Reflections used for R-free 1997 (196) 
R-work 0.2522 (0.3644) 
R-free 0.2791 (0.3648) 
 CC-work 0.837 (0.509) 
 CC-free 0.806 (0.527) 
Number of non-hydrogen atoms 4993 
 Protein 4983 
 Carbohydrate 10 
Protein residues 654 
RMSD (bonds) 0.003 
RMSS (angles) 0.93 
Ramachandran favoured (%) 92 
Ramachandran allowed (%) 6.8 
Ramachandran outliers (%) 0.79 
Average B-factor 87.75 

Statistics for the highest-resolution shell are shown in parentheses.

Criteria used for data cut-off was based on CC1/2, as suggested by Karplus and Diederichts [34]

The structure reveals Dll-1 has an extended conformation, consisting of a linear arrangement of the N-terminal C2 domain, the ligand-binding DSL domain and the first four EGF-like repeats. The junction between EGF repeats 4 and 5 mediates a turn of approximately 90°, with EGF repeats 5 and 6 again in a linear arrangement [Figure 1A(i)]. The final two repeats (7 and 8) are not visible. The longest axis measures approximately 195 Å. To the best of our knowledge, this structure represents the longest sequence of tandem EGF-like repeats in the Protein Data Bank. The C2, DSL and first two EGF repeats of Dll-1 and Jagged1 overlay with an RMSD of 1.2 (Supplementary Figure S3) suggesting these domains adopt a rigid, linear arrangement in solution. EGFs 3 and 4 both adopt a classical EGF repeat fold and the tandem repeats are essentially linear, almost identical in conformation to EGFs 12 and 13 of Notch1, despite lacking the Ca2+ chelated by these EGF repeats in Notch1 (RMSD 1.2 Å) [12]. This implies the interface between these two domains also has limited flexibility. The sharp bend between EGFs 4 and 5 is not represented between other tandem EGFs in the PDB, however as this region mediates a crystal contact (Supplementary Figure S2B), it is hard to determine whether the bend is a feature of the structure in solution or simply represents one of a range of possible orientations between adjacent EGF repeats. EGF repeats 5 and 6 again adopt a linear conformation, but slightly more bent than EGFs 3 and 4.

Structural analyses of Dll-1

Figure 1
Structural analyses of Dll-1

(A; i) Overall structure of Dll-1. C2 domain (red), DSL (green) and six EGF-like repeats (alternating beige and blue); (ii) residues conserved across all human Notch ligands (C2 domain excluded): 100% identity in dark blue, 100% homology in light blue. (B) DSL and EGF repeats 1–3 of Dll-1 (left) and Jagged1 (right). A large hydrophobic patch exists on EGF repeat 2 (yellow), on the same side of the ligand as key residues of the DSL domain (blue).

Figure 1
Structural analyses of Dll-1

(A; i) Overall structure of Dll-1. C2 domain (red), DSL (green) and six EGF-like repeats (alternating beige and blue); (ii) residues conserved across all human Notch ligands (C2 domain excluded): 100% identity in dark blue, 100% homology in light blue. (B) DSL and EGF repeats 1–3 of Dll-1 (left) and Jagged1 (right). A large hydrophobic patch exists on EGF repeat 2 (yellow), on the same side of the ligand as key residues of the DSL domain (blue).

Residues Thr303 and Ser342 appear to be O-glycosylated in our structure. Thr321 and Ser411 also appear to be at least partially glycosylated. The equivalent position to Thr303 is also seen to be O-glycosylated in the structure of Jagged1 (Thr311). No N-linked glycosylation is observed in the Dll-1 crystals.

Our 3D structure allows the conservation of surface residues across all ligands to be mapped in detail. Mapping the primary sequence conservation across the human Notch ligands Dll-1, Dll-4, Jagged1 and Jagged2 on to the surface of Dll-1 EGF repeats reveals several regions of conservation [Figure 1A(ii)] (Note that Dll-3 is excluded is it does not have the capacity to activate Notch signalling [15]). The first highly conserved region represents the well-characterized DSL domain. Residues in this region of Jagged have been shown to be essential for receptor binding and/or activation [12]. The DSL domains of Jagged1 and Dll-1 overlay with an RMSD of 0.7 Å and the B-factors are the lowest in this region. Together, these observations imply that this is a key region for ligand–receptor recognition and that this region may be pre-ordered for receptor binding. A second region of high homology is present at the junction between EGF repeats 2 and 3 [Figure 1A(ii)]. Residues His265, Trp280, Asp288, Asn289, Leu290, Tyr291, Tyr312 are completely conserved (Supplementary Figure S4.). This is on the opposite face of the protein to the DSL domain and appears to influence the relative orientation of EGFs 2 and 3. EGF repeat 3 is only visible in one copy of Dll-1 and a similar phenomenon was observed for Jagged1 fragment structures, the three structures of the C2–DSL–EGFs1–3 fragment contained two copies of the molecule, but in two crystal forms (PDB IDs 4CCO, 4CBZ [11]), density was only observed for the third EGF repeat of one of the two copies. This implies that the relative orientations of EGF repeats 2 and 3 can adopt multiple conformations and the sequence conservation of the linker region would suggest this will be the case for all ligands.

Although it is a site of crystal contact and must therefore be viewed with some caution, there are also four conserved residues at the EGF4–EGF5 interface where the distinct bend in the structure is observed. The side chain of Asn339 (EGF4) has clear electron density that places it within hydrogen bonding distance of Gly387 (EGF5) and Phe358 (EGF4) forms van der Waals contacts with Tyr389 (EGF5). Asn339 and Gly387 are completely conserved across the ligand family and equivalent homologous aromatic residues are observed conserving the hydrophobic interaction between Phe358 and Tyr389. This implies that this bend may also be conserved across the family of Notch ligands.

Surface-exposed hydrophobic residues were also mapped, to identify possible protein interaction sites. The residues surrounding conserved Trp272 (Val247, Ile258, Tyr260, Pro261, Leu264, Pro271, Leu283, Phe284) form a continuous hydrophobic patch on the surface of Dll-1 (Figure 1B), which covers the entire length of EGF 2. Comparison with the Jagged1 structure reveals a large hydrophobic patch is present in the same region of Jagged1 (Figure 1B), implying a conserved interaction surface. Intriguingly, this is on the same face as the key DSL residues. EGFs 1 and 2 of ligand have been shown to be important for efficient binding of ligand to receptor [16]; this may provide an explanation for the importance of EGF repeat 2 in engaging receptor. The fact that Dll-3 is divergent in this region (Supplementary Figure S4) may contribute to its inability to activate receptor [15].

C2 domain and calcium binding

The recent structure of the N-terminus of Jagged1 revealed that the conserved MNNL domain (module at the N-terminus of notch ligands) is in fact a membrane-recognition motif, known as a C2 domain [11]. C2 domains are globular membrane-recognition domains that are found only in eukaryotes. They are composed of two four-stranded β-sheets that form a compact β-sandwich, with highly variable surface loops connecting the strands [17]. One subfamily of these domains [the protein kinase C (PKC) sub-family] binds Ca2+ and the structure of the N-terminus of Jagged1 revealed a canonical Ca2+-binding site in the C2 domain [11]. Our structure shows that Dll-1 also contains a C2 domain, but we do not observe Ca2+ in the Dll-1 structure despite crystallizing in the presence of 5 mM Ca2+. Indeed, the Ca2+-binding aspartate residues present in Jagged1 are substituted for a histidine (His128) and arginine (Arg57) in Dll-1 (Figure 2A). Sequence alignments imply that this is a common difference between Dll and Jagged ligands, with Ca2+-binding residues seemingly conserved in Jaggeds and consistently absent in Dlls (Figure 2D; Supplementary Figure S3). The majority of C2 domains do not bind calcium but do bind lipid [17], therefore this does not preclude the C2 domain of Dll-1 from being a membrane-recognition motif. However it is hard to reconcile with the data showing that the C2 domains of both Dll-1 and Jagged1 are stabilized by calcium and that liposome binding appears to be enhanced for both proteins in the presence of calcium [11].

C2 domain comparisons

Figure 2
C2 domain comparisons

(AC) Comparison of Dll-1 (bottom, green) and Jagged1 (top, beige) C2 domains. (A) Key features of the C2 domain; (B) comparison with PKCα C2 domain (blue) bound to PtdIns(4,5)P2; (C) conserved loop, H-bonds shown as black dashed lines. Additional H-bonds only present in Dll-1 are shown in green. (D) Partial sequence alignment, C2 domain. 100% Identity, black highlight; 100% homology grey highlight; Jagged Ca2+-binding aspartate, green highlight; conserved lysine (arginine in non-bilaterals) red highlight; basic residues in ‘basic loop’ region, blue highlight. For a more detailed sequence alignment (C2, DSL and EGFs 1–2), see Supplementary Figure S4.

Figure 2
C2 domain comparisons

(AC) Comparison of Dll-1 (bottom, green) and Jagged1 (top, beige) C2 domains. (A) Key features of the C2 domain; (B) comparison with PKCα C2 domain (blue) bound to PtdIns(4,5)P2; (C) conserved loop, H-bonds shown as black dashed lines. Additional H-bonds only present in Dll-1 are shown in green. (D) Partial sequence alignment, C2 domain. 100% Identity, black highlight; 100% homology grey highlight; Jagged Ca2+-binding aspartate, green highlight; conserved lysine (arginine in non-bilaterals) red highlight; basic residues in ‘basic loop’ region, blue highlight. For a more detailed sequence alignment (C2, DSL and EGFs 1–2), see Supplementary Figure S4.

The end of the C2 domain that binds Ca2+ in the PKC-type C2 domains often displays a basic patch in other C2 domain families, which is implicated in lipid binding [17]. There is a basic patch present in Jagged1 (Figure 2A, ‘basic loop’). This region is not entirely ordered in our Dll-1 structure, however it can be seen from sequence alignments that the four lysine/arginine residues in Jagged-1 are not conserved in Dll-1 or indeed in Dlls in other species (Figure 2D; Supplementary Figure S4). This lack of conservation of the calcium-binding site and adjacent basic residues between Jagged1 and Dll-1 (and by homology Dll-4) suggest there may be differences in membrane interactions between the ligand families.

A second membrane-interaction site is typically observed on the concave face of C2 domains. The curvature comes about in part due to the presence of conserved hydrophobic residues on the surface of β2, 5, 6 [17]. These are Phe59, Lys127, Lys143 in Dll-1 and Tyr74, Trp139, Ile151 in Jagged1 (Figure 2B). This region has been shown to bind phosphatidylinositol-4,5-bisphosphate [PtdIns(4,5)P2] in the C2 domain of PKCα, therefore we superimposed the C2 domains of Jagged1 and Dll-1 on to that of the PKCα–PtdIns(4,5)P2 structure (PDB ID 3GPE [18]). The phosphate-binding tetrad of PKCα is completely conserved in Jagged1 (Tyr74, Trp139, Lys76, Glu137), with Tyr74 and Lys76 H-bonding to phosphate 5 of PI-4,5-P2, which is stacked above Trp139. This implies Jagged1 has the capacity to bind phosphate, however the lysine residues on β3 of PKCα that contact additional phosphate groups in PtdIns(4,5)P2 are not conserved. Given that we would expect this C2 domain to bind to the outer leaflet of the membrane and phosphatidyl inositols are usual present on the inner leaflet, this is perhaps not surprising. In Dll-1, the equivalent Tyr74 in Jagged1 is substituted for phenylalanine (Phe59), losing the capacity to H-bond to phosphate and Trp139 is substituted for leucine, implying that Dll-1 may not bind phospholipids in the same fashion. As the lipid selectivity of the C2 domain can determine the cellular location of a C2 domain-containing protein [19], these differences in both the Ca2+-binding and the β-groove suggest that the ligands may show preference for different cell types and/or locations on a cell based on membrane composition.

In addition to these relatively well-characterized features of C2 domains, there are two other features common to both Dll-1 and Jagged1 structures and, by sequence homology, to other Notch ligands. Adjacent to the concave surface described above, there is a charged [K/R]-E-[K/R] motif in Jagged1 and Dll-1, which is well conserved throughout the Notch ligand family (Figure 2A; Supplementary Figure S4). This is adjacent to a structured loop on concave face of the C2 domain that is shared by Dll-1 and Jagged-1 and is not a described feature of other C2 domain families, suggesting it may be a feature unique to Notch ligands. The loop is oriented by a central conserved lysine at the end of β2 (Lys65 in Dll-1, Lys80 in Jagged1) present in all Notch ligands (Figure 2D; Supplementary Figure S4). A conserved network of hydrogen bonds (Figure 2C) maintains the structure of the loop, which is also tied to the surface by a disulfide bond (Dll-1 C63–C77). In Dll-1, the adjacent R-E-R motif also interacts with the loop through Arg61, whereas in Jagged1 the equivalent Lys76 is presumably too short to reach the loop. Whereas the function of this loop cannot be inferred by comparison with other C2 domains, the high degree of conservation implies a functional importance and it is a region of the protein that must be the subject of further investigations. It is conceivable that it is a protein–protein interface that allows interaction with other components of the Notch activation machinery.

Dll-1–Notch1 binding affinities

Despite numerous attempts to determine Notch-ligand binding affinities, it is not clear whether there are intrinsic differences in affinity among the various Notch ligands or whether ligand–receptor preferences are based largely on expression patterns. Results of binding experiments vary enormously depending on how the assay is performed. Andrawes et al. [20] have performed perhaps the most biologically-relevant study on large, recombinant protein fragments in combination with cellular activity assays and found that there is a 10-fold difference in affinity between Dll-1 and Dll-4, with the affinity of Notch1 EGF repeats 6–15 for Dll-1 determined to be ~3.5 μM by biolayer inferometry and ~12 μM by surface plasmon resonance. We therefore used our recombinant protein to measure the Dll-1 binding affinity for Notch1.

We prepared recombinant Notch1 residues 1–526 (EGF repeats 1–13, N11–13), incorporating the ligand-binding site (EGFs 11–13), as a TEV-cleavable Fc fusion protein and performed ELISAs in two alternative formats: either N11–13 was plated and a concentration series of Dll-1–Fc was presented in solution; or Dll-1 was plated and N11–13–Fc was presented in solution at different concentrations. In both cases, bound Fc-fusion protein was detected using HRP–protein G. The interaction was calcium-dependent, as expected [21]. The value obtained for Kd varied depending on how the experiment was performed. Plating Dll-1, with N11–13–Fc in solution, led to an apparent Kd of 0.163±0.019 μM, whereas the reverse experiment, with Dll-1–Fc in solution, gave an apparent Kd of 0.011±0.002 μM.

Given the discrepancy between these two systems and the large difference in affinity as compared with the recently published data of Andrawes et al. [20], we opted to study the complex by an alternative method. To the best of our knowledge, the Notch–ligand interaction has never been studied directly when both components are in solution. We chose to use analytical ultracentrifugation to determine the affinity of the Notch–ligand complex in solution. Initial sedimentation velocity experiments were conducted on separate samples of Dll-1 and N11–13 alone in order to determine their sedimentation behaviour and detect any self-association. The resulting continuous c(s) distributions show single, narrow, symmetrical peaks with modal sedimentation coefficients of approximately 3.0 S and 3.2 S for Dll-1 and N11–13 respectively [Figure 3B(i)]. A lack of concentration-dependent changes in sedimentation behaviour was confirmed for Dll-1 in separate experiments (Supplementary Figure S5). The best fit to the raw sedimentation data (Supplementary Figure S6) provided f/f0 for both Dll-1 and N11–13 of 1.84. These parameters suggest significant deviation from globular structure for both proteins, consistent with our crystal structure of Dll-1 and the predicted N11–13 structure as an extended arrangement of EGF repeats. As the extent and composition of glycosylation of both Dll-1 and N11–13 are unknown, partial specific volumes for the proteins cannot be calculated and therefore molecular masses for each protein cannot be accurately derived from sedimentation data. To confirm that the c(s) distributions indicate monomeric forms of the proteins the theoretical sedimentation coefficient for a Dll-1 monomer was calculated from the crystal structure co-ordinates assuming no glycosylation. The calculated sedimentation coefficient of 3.11 S agrees very well with our experimentally observed value, indicating that the two proteins exist as monomers in solution with very similar sedimentation coefficients.

Dll-1–Notch 1–13 binding analyses

Figure 3
Dll-1–Notch 1–13 binding analyses

(A) ELISAs: (i) plated N11–13, Dll-1–Fc in solution; (ii) plated Dll-1, N11–13–Fc in solution. Graphs show duplicate data points from an individual experiment and are representative of all experiments. Kd values are average of three independent experiments, each performed in duplicate. Curve fitting assumed 1:1 association. (B) Sedimentation velocity analysis of Dll-1 and N11–13. (i) Continuous c(s) distributions for Dll-1 and N11–13, calculated from best fits to sedimentation data shown in Supplementary Figure S6. (ii) c(s) distributions for 4 μM notch in the presence of increasing concentrations of Dll-1, calculated from best fits to sedimentation data shown in Supplementary Figure S7. (C) Sedimentation equilibrium analysis of the Dll-1–Notch complex. Sedimentation equilibrium distributions were determined for solutions with initial loading concentrations of 10 μM (top), 3.5 μM (middle) and 1.2 μM (bottom). Radial absorbance data at sedimentation equilibrium are shown at three rotor speeds overlaid with the global best fit to a 1:1 hetero-association model (solid lines).

Figure 3
Dll-1–Notch 1–13 binding analyses

(A) ELISAs: (i) plated N11–13, Dll-1–Fc in solution; (ii) plated Dll-1, N11–13–Fc in solution. Graphs show duplicate data points from an individual experiment and are representative of all experiments. Kd values are average of three independent experiments, each performed in duplicate. Curve fitting assumed 1:1 association. (B) Sedimentation velocity analysis of Dll-1 and N11–13. (i) Continuous c(s) distributions for Dll-1 and N11–13, calculated from best fits to sedimentation data shown in Supplementary Figure S6. (ii) c(s) distributions for 4 μM notch in the presence of increasing concentrations of Dll-1, calculated from best fits to sedimentation data shown in Supplementary Figure S7. (C) Sedimentation equilibrium analysis of the Dll-1–Notch complex. Sedimentation equilibrium distributions were determined for solutions with initial loading concentrations of 10 μM (top), 3.5 μM (middle) and 1.2 μM (bottom). Radial absorbance data at sedimentation equilibrium are shown at three rotor speeds overlaid with the global best fit to a 1:1 hetero-association model (solid lines).

In order to detect complex formation between Dll-1 and N11–13, a titration series comprising a constant concentration of 4 μM N11–13 in the presence of increasing concentrations of Dll-1 between 1 and 32 μM was subjected to sedimentation velocity analysis (Supplementary Figure S7). c(s) distributions resulting from these experiments show a prominent peak centred around approximately 3.1 S, corresponding to the additive signal from monomeric Dll-1 and N11–13 [Figure 3B(ii)]. As the concentration of Dll-1 was increased a second peak centred at approximately 4.3 S appeared and increased in magnitude, indicating complex formation between the two proteins. Qualitative analysis of the signal above 4 S in these c(s) distributions suggested that the dissociation constant for the complex was in the low micromolar range.

To facilitate the analysis of the dissociation constant of the Dll-1–N11-13 complex by sedimentation equilibrium, apparent molecular masses for the isolated proteins were determined experimentally by sedimentation equilibrium and by assuming a partial specific volume for each protein of 0.73 (Supplementary Figure S8). Apparent molecular masses using this scale for Dll-1 (67755 Da) and N11–13 (69450 Da) were then fixed for the analysis of the complex and all fits were conducted using an operational partial specific volume of 0.73. Sedimentation equilibrium radial distribution data obtained from three equimolar mixtures of Dll-1 and N11–13 at three rotor speeds were fitted globally (Figure 3C) yielding a Kd of 10.4 μM (95% confidence interval: 9.4–11.2 μM). The analysis provides an excellent fit to data at all centrifugal fields and protein concentrations coupled with a narrow confidence interval, indicating high accuracy in the Kd estimate.

We determined binding affinity of N11–13 and Dll-1 using three different methods, each of which yielded an apparent Kd that differed by an order of magnitude. Avidity effects are likely to be contributing to the differences here, as the species in solution for the ELISAs is an Fc-fusion and therefore dimeric. The measurements determined by analytical ultracentrifugation used monomeric proteins in solution and are therefore perhaps more accurate. Certainly, understanding the true affinity in solution is a critical part of understanding the complex molecular mechanisms of receptor–ligand engagement at the cell surface. However, since clustering of ligand appears to be important for Notch activation [22,23] avidity effects may be in play in a cellular setting and measuring the affinity of the monomeric proteins in solution may not be representative of the true biological affinity. For example, the extracellular domain of the EGF receptor binding to ligand has a Kd of 500 nM when measured in solution, compared with 20 pM and 1 nM (two binding modes) on the cell membrane [2426]. In addition, whereas combined mutagenesis and binding-affinity data certainly provide crucial information about receptor interactions and ligand preferences, they do not take into account potential lipid-mediated participation of the C2 domain and effect on the efficiency of the interaction due to the membrane composition of the receptor-bearing cell. Even if the C2 domain does not participate directly in the interaction, avidity effects due to anchoring of the ligand to the target cell by the correct ligand type will presumably increase efficiency of signalling.

In summary, our structure, in combination with the smaller N-terminal fragment of Jagged1 [11], provides insight into the conserved architecture of the Notch ligands and a platform for interrogation of the role of key conserved regions of the ligands. While this manuscript was being revised, the structure of an N-terminal fragment of Dll-4 in complex with EGF repeats 11–13 of Notch1 was published [35]. Comparison of our structure with that of receptor-bound Dll-4 will provide additional insight into the key details of Notch-ligand interactions. Further structures of larger fragments of Notch in complex with different ligands will provide the detailed understanding required to allow specific targeting of the Notch-ligand interface for therapeutic gain. Our current efforts lie in this direction.

AUTHOR CONTRIBUTION

Nadia Kershaw conceived and designed the study and acquired data. Michael Griffin designed the study and acquired data. Nicole Church and Cindy Luo acquired data. Antony Burgess and Timothy Adams were involved in design/conception. All authors contributed to writing or critically revising the manuscript.

We thank CSIRO collaborative crystallisation centre (C3) for initial crystallization trials and scientists at the MX beamlines at the Australian Synchrotron. The authors declare that they have no conflict of interest.

FUNDING

This work was supported by the National Health and Medical Research Council of Australia (NHMRC) programme [grant number 487922]; the Ludwig Institute for Cancer Research; The Victorian State Government Operational Infrastructure Support Grant; the NHMRC Independent Research Institutes Infrastructure Support Scheme; the C.R. Roper Fellowship to M.D.W.G.; and the Australian Research Council Future Fellowship [grant number FT140100544 (to M.D.W.G.)].

Abbreviations

     
  • c(s)

    sedimentation coefficient

  •  
  • Dll-1

    delta-like ligand 1

  •  
  • EGF

    epidermal growth factor

  •  
  • f/f0

    frictional ratios

  •  
  • HRP

    horseradish peroxidase

  •  
  • Notch11–13

    Notch1 EGF repeats 1–13

  •  
  • PEI

    polyethyleneimine

  •  
  • PKC

    protein kinase C

  •  
  • PtdIns(4

    phosphatidylinositol-4,5-bisphosphate

  •  
  • TEV

    tobacco etch virus

References

References
1
Dexter
J. S.
The analysis of a case of continuous variation in drosophila by a study of its linkage relations
Am. Naturalist
1914
, vol. 
48
 (pg. 
712
-
758
)
2
Kidd
S.
Kelley
M. R.
Young
M. W.
Sequence of the notch locus of Drosophila melanogaster: relationship of the encoded protein to mammalian clotting and growth factors
Mol. Cell. Biol.
1986
, vol. 
6
 (pg. 
3094
-
3108
)
[PubMed]
3
Wharton
K. A.
Johansen
K. M.
Xu
T.
Artavanis-Tsakonas
S.
Nucleotide sequence from the neurogenic locus notch implies a gene product that shares homology with proteins containing EGF-like repeats
Cell
1985
, vol. 
43
 (pg. 
567
-
581
)
[PubMed]
4
Artavanis-Tsakonas
S.
Rand
M. D.
Lake
R. J.
Notch signaling: cell fate control and signal integration in development
Science
1999
, vol. 
284
 (pg. 
770
-
776
)
[PubMed]
5
Ntziachristos
P.
Lim
J. S.
Sage
J.
Aifantis
I.
From fly wings to targeted cancer therapies: a centennial for notch signaling
Cancer Cell
2014
, vol. 
25
 (pg. 
318
-
334
)
[PubMed]
6
Gordon
W. R.
Vardar-Ulu
D.
Histen
G.
Sanchez-Irizarry
C.
Aster
J. C.
Blacklow
S. C.
Structural basis for autoinhibition of Notch
Nat. Struct. Mol. Biol.
2007
, vol. 
14
 (pg. 
295
-
300
)
[PubMed]
7
Brou
C.
Logeat
F.
Gupta
N.
Bessia
C.
LeBail
O.
Doedens
J. R.
Cumano
A.
Roux
P.
Black
R. A.
Israel
A.
A novel proteolytic cleavage involved in Notch signaling: the role of the disintegrin-metalloprotease TACE
Mol. Cell
2000
, vol. 
5
 (pg. 
207
-
216
)
[PubMed]
8
Mumm
J. S.
Schroeter
E. H.
Saxena
M. T.
Griesemer
A.
Tian
X.
Pan
D. J.
Ray
W. J.
Kopan
R.
A ligand-induced extracellular cleavage regulates gamma-secretase-like proteolytic activation of Notch1
Mol. Cell
2000
, vol. 
5
 (pg. 
197
-
206
)
[PubMed]
9
Schroeter
E. H.
Kisslinger
J. A.
Kopan
R.
Notch-1 signalling requires ligand-induced proteolytic release of intracellular domain
Nature
1998
, vol. 
393
 (pg. 
382
-
386
)
[PubMed]
10
Wu
L.
Sun
T.
Kobayashi
K.
Gao
P.
Griffin
J. D.
Identification of a family of mastermind-like transcriptional coactivators for mammalian notch receptors
Mol. Cell. Biol.
2002
, vol. 
22
 (pg. 
7688
-
7700
)
[PubMed]
11
Chillakuri
C. R.
Sheppard
D.
Ilagan
M. X.
Holt
L. R.
Abbott
F.
Liang
S.
Kopan
R.
Handford
P. A.
Lea
S. M.
Structural analysis uncovers lipid-binding properties of Notch ligands
Cell Rep.
2013
, vol. 
5
 (pg. 
861
-
867
)
[PubMed]
12
Cordle
J.
Johnson
S.
Tay
J. Z.
Roversi
P.
Wilkin
M. B.
de Madrid
B. H.
Shimizu
H.
Jensen
S.
Whiteman
P.
Jin
B.
Redfield
C.
, et al. 
A conserved face of the Jagged/Serrate DSL domain is involved in Notch trans-activation and cis-inhibition
Nat. Struct. Mol. Biol.
2008
, vol. 
15
 (pg. 
849
-
857
)
[PubMed]
13
Hambleton
S.
Valeyev
N. V.
Muranyi
A.
Knott
V.
Werner
J. M.
McMichael
A. J.
Handford
P. A.
Downing
A. K.
Structural and functional properties of the human notch-1 ligand binding region
Structure
2004
, vol. 
12
 (pg. 
2173
-
2183
)
[PubMed]
14
Taylor
P.
Takeuchi
H.
Sheppard
D.
Chillakuri
C.
Lea
S. M.
Haltiwanger
R. S.
Handford
P. A.
Fringe-mediated extension of O-linked fucose in the ligand-binding region of Notch1 increases binding to mammalian Notch ligands
Proc. Natl. Acad. Sci. U.S.A.
2014
, vol. 
111
 (pg. 
7290
-
7295
)
[PubMed]
15
Ladi
E.
Nichols
J. T.
Ge
W.
Miyamoto
A.
Yao
C.
Yang
L. T.
Boulter
J.
Sun
Y. E.
Kintner
C.
Weinmaster
G.
The divergent DSL ligand Dll3 does not activate Notch signaling but cell autonomously attenuates signaling induced by other DSL ligands
J. Cell. Biol.
2005
, vol. 
170
 (pg. 
983
-
992
)
[PubMed]
16
Shimizu
K.
Chiba
S.
Kumano
K.
Hosoya
N.
Takahashi
T.
Kanda
Y.
Hamada
Y.
Yazaki
Y.
Hirai
H.
Mouse jagged1 physically interacts with notch2 and other notch receptors. Assessment by quantitative methods
J. Biol. Chem.
1999
, vol. 
274
 (pg. 
32961
-
32969
)
[PubMed]
17
Zhang
D.
Aravind
L.
Identification of novel families and classification of the C2 domain superfamily elucidate the origin and evolution of membrane targeting activities in eukaryotes
Gene
2010
, vol. 
469
 (pg. 
18
-
30
)
[PubMed]
18
Guerrero-Valero
M.
Ferrer-Orta
C.
Querol-Audi
J.
Marin-Vicente
C.
Fita
I.
Gomez-Fernandez
J. C.
Verdaguer
N.
Corbalan-Garcia
S.
Structural and mechanistic insights into the association of PKCalpha-C2 domain to PtdIns(4,5)P2
Proc. Natl. Acad. Sci. U.S.A.
2009
, vol. 
106
 (pg. 
6603
-
6607
)
[PubMed]
19
Cho
W.
Stahelin
R. V.
Membrane binding and subcellular targeting of C2 domains
Biochim. Biophys. Acta
2006
, vol. 
1761
 (pg. 
838
-
849
)
[PubMed]
20
Andrawes
M. B.
Xu
X.
Liu
H.
Ficarro
S. B.
Marto
J. A.
Aster
J. C.
Blacklow
S. C.
Intrinsic selectivity of Notch 1 for delta-like 4 over delta-like 1
J. Biol. Chem.
2013
, vol. 
288
 (pg. 
25477
-
25489
)
[PubMed]
21
Cordle
J.
Redfieldz
C.
Stacey
M.
van der Merwe
P. A.
Willis
A. C.
Champion
B. R.
Hambleton
S.
Handford
P. A.
Localization of the delta-like-1-binding site in human Notch-1 and its modulation by calcium affinity
J. Biol. Chem.
2008
, vol. 
283
 (pg. 
11785
-
11793
)
[PubMed]
22
Morrison
S. J.
Perez
S. E.
Qiao
Z.
Verdi
J. M.
Hicks
C.
Weinmaster
G.
Anderson
D. J.
Transient Notch activation initiates an irreversible switch from neurogenesis to gliogenesis by neural crest stem cells
Cell
2000
, vol. 
101
 (pg. 
499
-
510
)
[PubMed]
23
Varnum-Finney
B.
Wu
L.
Yu
M.
Brashem-Stein
C.
Staats
S.
Flowers
D.
Griffin
J. D.
Bernstein
I. D.
Immobilization of Notch ligand, delta-1, is required for induction of notch signaling
J. Cell. Sci.
2000
, vol. 
113
 
Pt 23
(pg. 
4313
-
4318
)
[PubMed]
24
Domagala
T.
Konstantopoulos
N.
Smyth
F.
Jorissen
R. N.
Fabri
L.
Geleick
D.
Lax
I.
Schlessinger
J.
Sawyer
W.
Howlett
G. J.
, et al. 
Stoichiometry, kinetic and binding analysis of the interaction between epidermal growth factor (EGF) and the extracellular domain of the EGF receptor
Growth Factors
2000
, vol. 
18
 (pg. 
11
-
29
)
[PubMed]
25
Elleman
T. C.
Domagala
T.
McKern
N. M.
Nerrie
M.
Lonnqvist
B.
Adams
T. E.
Lewis
J.
Lovrecz
G. O.
Hoyne
P. A.
Richards
K. M.
, et al. 
Identification of a determinant of epidermal growth factor receptor ligand-binding specificity using a truncated, high-affinity form of the ectodomain
Biochemistry
2001
, vol. 
40
 (pg. 
8930
-
8939
)
[PubMed]
26
Ferguson
K. M.
Active and inactive conformations of the epidermal growth factor receptor
Biochem. Soc. Trans.
2004
, vol. 
32
 (pg. 
742
-
745
)
[PubMed]
27
McCoy
A. J.
Grosse-Kunstleve
R. W.
Adams
P. D.
Winn
M. D.
Storoni
L. C.
Read
R. J.
Phaser crystallographic software
J. Appl. Crystallogr.
2007
, vol. 
40
 (pg. 
658
-
674
)
[PubMed]
28
Adams
P. D.
Afonine
P. V.
Bunkoczi
G.
Chen
V. B.
Davis
I. W.
Echols
N.
Headd
J. J.
Hung
L. W.
Kapral
G. J.
Grosse-Kunstleve
R. W.
, et al. 
PHENIX: a comprehensive Python-based system for macromolecular structure solution
Acta Crystallogr. D.
2010
, vol. 
66
 (pg. 
213
-
221
)
29
Emsley
P.
Cowtan
K.
Coot: model-building tools for molecular graphics
Acta Crystallogr. D.
2004
, vol. 
60
 (pg. 
2126
-
2132
)
30
Laue
T. M.
Shah
B. D.
Ridgeway
T. M.
Pelletier
S. L.
Analytical Ultracentrifugation in Biochemistry and Polymer Science
1992
Cambridge
Royal Society of Chemistry
31
Schuck
P.
Size-distribution analysis of macromolecules by sedimentation velocity ultracentrifugation and lamm equation modeling
Biophys. J.
2000
, vol. 
78
 (pg. 
1606
-
1619
)
[PubMed]
32
Ortega
A.
Amoros
D.
Garcia de la Torre
J.
Prediction of hydrodynamic and other solution properties of rigid proteins from atomic- and residue-level models
Biophys. J.
2011
, vol. 
101
 (pg. 
892
-
898
)
[PubMed]
33
Vistica
J.
Dam
J.
Balbo
A.
Yikilmaz
E.
Mariuzza
R. A.
Rouault
T. A.
Schuck
P.
Sedimentation equilibrium analysis of protein interactions with global implicit mass conservation constraints and systematic noise decomposition
Anal. Biochem.
2004
, vol. 
326
 (pg. 
234
-
256
)
[PubMed]
34
Karplus
P. A.
Diederichs
K.
Linking crystallographic model and data quality
Science
2012
, vol. 
336
 (pg. 
1030
-
1033
)
[PubMed]
35
Luca
V.C.
Jude
K.M.
Pierce
N.W.
Nachury
M.V.
Fischer
S.
Garcia
K.C.
Structural biology. Structural basis for Notch1 engagement of Delta-like 4
Science
2015
, vol. 
347
 (pg. 
847
-
853
)

Author notes

The atomic coordinates and structure factors have been deposited in the PDB with the following accession number: PDB ID 4XBM.

Supplementary data