NELL2-Robo3 complex structure reveals mechanisms of receptor activation for axon guidance

Abstract

Axon pathfinding is critical for nervous system development, and it is orchestrated by molecular cues that activate receptors on the axonal growth cone. Robo family receptors bind Slit guidance cues to mediate axon repulsion. In mammals, the divergent family member Robo3 does not bind Slits, but instead signals axon repulsion from its own ligand, NELL2. Conversely, canonical Robos do not mediate NELL2 signaling. Here, we present the structures of NELL-Robo3 complexes, identifying a mode of ligand engagement for Robos that is orthogonal to Slit binding. We elucidate the structural basis for differential binding between NELL and Robo family members and show that NELL2 repulsive activity is a function of its Robo3 affinity and is enhanced by ligand trimerization. Our results reveal a mechanism of oligomerization-induced Robo activation for axon guidance and shed light on Robo family member ligand binding specificity, conformational variability, divergent modes of signaling, and evolution.

Fig. 1. Robo3.1 is necessary and sufficient for axon repulsion from NELL2.

a Domain structure of NELLs, Robos, and Slits, drawn true to scale for human NELL2, Robo3, and Slit1. CC with numbers indicate conserved cytoplasmic motifs. SP signal peptide. TM transmembrane helix. CK C-terminal cysteine knot. b Transverse spinal cord sections from E9.75 mouse embryos were labeled with Hoechst stain and antibodies against Robo3 and NELL2. Robo3-expressing precrossing commissural axons grow around areas of NELL2 expression in the ventral horn. Dotted line indicates spinal cord border. Fluorescence intensity profile (bottom) was generated by line scan along the dashed arrow. c DIC images of a commissural axon turning away from a gradient of NELL2 in a Dunn chamber (0 and 2 h). d Quantification of commissural axon turning angles for increasing concentrations of NELL2 (n = 3 independent experiments for all conditions). e, f DIC images of Robo3^−/− neurons in a NELL2 gradient. Robo3^−/− axons do not turn in response to NELL2 (e), unless rescued with Robo3.1 (f). g Quantification of turning angles in response to NELL2 for commissural (labeled C) axons (wild type or Robo3^−/− with or without Robo3.1 or Robo3.2 rescue) or ipsilateral (labeled I) axons (n = 4 for wild-type commissural and ipsilateral axons, n = 3 for all other conditions). Scale bar, 10 μm (b, c, e, f). Error bars indicate SEM.

Fig. 2. NELL2 EGF2–3 bind Robo3 FN1 to mediate complex formation.

a, b Domains mediating NELL2–Robo3 interactions were mapped using a COS-7-based AP-fusion protein binding assay. NELL2 EGF2 and EGF3 together are necessary and sufficient for binding to Robo3 (a). Robo3 FN1 is necessary and sufficient for NELL2 binding (b). c Size-exclusion chromatography of the hNELL2 EGF1–6 and hRobo3 FN1–3 complex. hNELL2 EGF1–6, hRobo3 FN1–3, and a molar 1:1 mixed complex samples were injected in a Superdex 200 Increase 10/300 column, and the elution profile was recorded by following absorbance at 280 nm with a pathlength of 0.2 cm. Dark blue: hNELL2 EGF1–6 + hRobo3 FN1–3 complex sample; Cyan: hRobo3 FN1–3; and Red: hNELL2 EGF1–6. AU: Absorbance units. SPR sensorgrams (d) and equilibrium response fitting to a Langmuir 1:1 binding model (e) for biotinylated hRobo3 FN1–3 as ligand on an NLC/neutravidin chip with hNELL2 EGF1–6 as the analyte (mobile phase) collected on a ProteOn XPR36. Legend on the right in (d) refers to concentration of the analyte injected on the SPR chip. Black curve in (e) is the Langmuir model fit to response values measured at equilibrium. ± refers to standard error of the fit. Scale bar, 100 μm, (a, b).

Fig. 3. Crystal structure of the NELL2–Robo3 complex.

a The structure of hRobo3 (green) bound to hNELL2 (cyan). The three calcium ions are depicted as balls; the two glycan moieties and the side chains they are linked to are shown as sticks. b Interactions of the EGF2 domain with Robo3 are dominated by polar contacts. c Interactions of the EGF3 domain with Robo3 are exclusively hydrophobic. d Binding isotherms for biotinylated wild type and mutant hNELL2 EGF1–6 measured against S2 cells expressing hRobo3 FN1–3. y-axis represents mean fluorescence intensity (MFI) from fluorescent streptavidin. e Binding isotherms for biotinylated wild type and mutant hRobo3 FN1–3 measured against S2 cells expressing hNELL2 EGF1–6. Apparent K_D values for S2 cell staining experiments are listed in Supplementary Fig. 3a.

Fig. 4. Axon repulsive activity of NELL2 correlates with its affinity towards Robo3.

a, b Binding isotherms (a) for SPR experiments testing the interaction of mRobo3 FN1–3 with mNELL2 EGF1–6 WT and mutants. Original sensorgrams are in Supplementary Fig. 3d. SPR responses are fit with 1:1 Langmuir isotherm model to calculate dissociation constants (K_D) (b). The ± errors represent standard error of the fit. DIC images of commissural axons exposed to mutant forms of NELL2 with Robo3-binding affinities of 1.7 µM (c) and ≥100 µM (d) (0 and 2 h). Commissural axons respond to NELL2 mutants with 1.7 µM binding affinity, but not with ≥100 μM affinity. (e) Quantification of commissural axon turning angles in response to wild-type NELL2 or NELL2 containing point mutations (n = 3 for WT NELL2, R452A, and L498A/F500A; n = 4 for R448A/Y450A; n = 5 for V509A). K_D values are dissociation constants measured for mNELL2 mutants via SPR (b). Scale bar, 10 µm (c, d).

Fig. 5. Axon repulsive activities and interactions of NELL family members with Robo3 are conserved to varying degrees.

a Crystal structure of hRobo3 FN1 bound to hNELL1 EGF1–3 (purple and yellow), overlaid with the hRobo3-hNELL2 structure (cyan and green). b Partial alignment of NELL sequences from Xenopus laevis (x), mouse (m), human (h), and the arthropod Aedes aegypti (a). Green squares indicate NELL residues at the Robo3-binding interface, which are highly conserved, except in aNELL. Blue positions are identical to human NELL1 or NELL2, while light blue represents conservative substitutions. Sequence numbering above the alignment is for hNELL2. c Quantification of commissural axon turning angles for increasing concentrations of NELL1 (n = 3 for all conditions). d Quantification of commissural axon turning angles in response to 50 ng/ml NELL1/2 chimeric molecules (n = 3 for all conditions). Error bars represent SEM.

Fig. 6. Robo1 interacts weakly with NELL2.

a Partial alignment of Robo FN1 sequences. Green squares indicate Robo3 residues at the NELL2-binding interface. Blue positions are identical in Robo1 and Robo2, while light blue represents conservative substitutions. b Surface plasmon resonance sensorgrams for the mRobo1 ectodomain-mNELL2 interaction. Expected maximal response is ~1,100 RU. c ECIA results for hRobo1 and hNELL2 show that FN1 and the EGF1–3 are necessary for the interaction. See Supplementary Fig. 6a for expression levels of each construct used. d ECIA results for Robo1 binding to NELL2 mutants designed at the Robo3-interaction interface. Comparison with Robo3 binding to NELL2 mutants in Supplementary Fig. 3c shows that Robo1 binding has the same binding surface on NELL2 with similar energetics of binding. Expression levels of ECIA constructs are measured with western blotting in Supplementary Fig. 6b. e Surface plasmon resonance sensorgrams for the mRobo1 FN1–3-mNELL2 EGF1–6 interaction. f Langmuir binding isotherm for (e). Dashed red line indicate the dissociation constant, K_D. Calculated R_max is 143 R.U. The fit to the isotherm is reliable given a theoretical R_max of 218 R.U. assuming a 100% active chip surface. g NELL2-AP-binding assay with cells expressing Robo3.1, Robo3^ΔFN1, or Robo3^Robo1-FN1. h Quantification of turning angles in response to NELL2 for Robo3^−/− commissural axons rescued with wild-type Robo3.1, Robo3^ΔFN1, or Robo3^Robo1-FN1 (n = 3 for all conditions). Scale bar, 100 μm (g). Error bars indicate SEM.

Fig. 7. NELL2 trimers signal through multimerization of Robo3 monomers.

Sedimentation velocity (SV) AUC (a) and SEC (b) show that mRobo1 ectodomain is mostly dimeric (red curves), while mRobo3 ectodomain is a monomer (blue curves). Single-concentration SV runs (a) were performed with 2.5 µM Robo1 and 1.8 µM Robo3. See Supplementary Fig. 7 for a series of concentrations, which shows that mRobo1 ectodomain is in an equilibrium of monomer to dimer to oligomer, unlike the stably monomeric mRobo3. c Bead models calculated from SAXS data indicate a fully extended shape for hRobo3 ECD. Guinier plot for SAXS data is in Supplementary Fig. 8a. Pair distance distribution (P(r)) analysis is in agreement with the extended shape of Robo3. d Bead models calculated from SAXS data for hNELL2 indicate a trimer more compact than Robo3 ECD (c). Guinier plot analysis and P(r) plot are in Supplementary Fig. 9a. e DIC images of commissural axons exposed to wild-type NELL2 or NELL2^ΔCC (0 and 2 h). f Quantification of axon turning angles in response to NELL2 and NELL2^ΔCC (n = 3 for all conditions). NELL2^ΔCC repels axons to a lesser degree than wild-type NELL2. g A model for conformational changes in mammalian Robos. Open Robo3 prefers to bind trimeric NELLs, while multiple conformational models exist for Robo1 and Robo2^,, which are in “closed” states for NELL binding. The conformational flexibility is created by the IG4-IG5, IG5-FN1, and FN1-FN2 linkers. The lightly shaded area in the NELL–Robo complex model was structurally characterized in this study (shown on the left side of the model). “x3” labels indicate trimerization via the CC domain of NELLs. Scale bar, 10 μm (e).