Operational redundancy in axon guidance through the multifunctional receptor Robo3 and its ligand NELL2

See allHide authors and affiliations

Science  20 Nov 2015:
Vol. 350, Issue 6263, pp. 961-965
DOI: 10.1126/science.aad2615

No going back

The mammalian spinal cord coordinates neuronal systems across the body. Axons that cross the spinal cord midline during development first need permission to cross and then instruction not to keep crossing back and forth. Jaworski et al. studied the axonal guidance receptor ROBO3 and found a ligand NELL2 in mice that appears to help in this process.

Science, this issue p. 961


Axon pathfinding is orchestrated by numerous guidance cues, including Slits and their Robo receptors, but it remains unclear how information from multiple cues is integrated or filtered. Robo3, a Robo family member, allows commissural axons to reach and cross the spinal cord midline by antagonizing Robo1/2–mediated repulsion from midline-expressed Slits and potentiating deleted in colorectal cancer (DCC)–mediated midline attraction to Netrin-1, but without binding either Slits or Netrins. We identified a secreted Robo3 ligand, neural epidermal growth factor-like-like 2 (NELL2), which repels mouse commissural axons through Robo3 and helps steer them to the midline. These findings identify NELL2 as an axon guidance cue and establish Robo3 as a multifunctional regulator of pathfinding that simultaneously mediates NELL2 repulsion, inhibits Slit repulsion, and facilitates Netrin attraction to achieve a common guidance purpose.

Nervous system development requires that axons of neurons navigate to their correct targets. Pathfinding is directed by molecular cues sensed by receptors in the axonal growth cone (1). We studied axon guidance in the mouse spinal cord. Commissural neurons in dorsal spinal cord send axons across the floor plate at the ventral midline in response to multiple cues (2). Netrin-1 and other signals expressed by floor plate attract these axons (3). After midline crossing, floor plate–derived repellants, including Slit proteins, expel them and prevent recrossing (4, 5). Commissural axons are insensitive to the repellants before crossing and insensitive to the attractants after crossing, indicating a switch in cue responsiveness (5, 6).

Slits signal through Robo family receptors. All three Slit genes are expressed by the floor plate, and commissural neurons express Robo1, Robo2, and Robo3 (7). Slits bind Robo1 and Robo2 with high affinity to mediate midline repulsion in post-crossing axons (4, 5, 8, 9), but Robo3 does not bind Slit proteins (1013). Robo3 exists in two splice isoforms with different intracellular domains: Robo3.1 is expressed on precrossing commissural axons and allows midline crossing by inhibiting Slit repulsion, whereas Robo3.2 is expressed postcrossing and prevents midline recrossing (12, 14). In mice lacking Robo3, all commissural axons fail to cross the midline and instead project through the ipsilateral ventral horn (12). Midline crossing is partially, but incompletely, rescued in mice lacking Robo1, Robo2, and Robo3 (8). Thus, Robo3 allows midline crossing partly by inhibiting Slit signaling through Robo1/2, but also through additional mechanisms. Indeed, Robo3 potentiates attraction by Netrin-1, mediated by the Netrin receptor DCC (13). However, Robo3 binds neither Netrin-1 nor Slits directly, suggesting that a Robo3 ligand might further contribute to midline guidance.

We screened an extracellular protein microarray (15, 16) with the Robo3 ectodomain fused to Fc (Robo3-ECD-Fc) and found that a fusion of NELL2 and Fc (NELL2-Fc) binds Robo3-ECD-Fc (fig. S1A). NELL2 is a secreted glycoprotein containing a laminin G–like domain, six epidermal growth factor (EGF)–like domains, and five von Willebrand factor (vWF) C domains (Fig. 1C) and is a human ortholog of chick Nel (neural EGF–like) (17, 18). Nel can inhibit retinal ganglion cell axon growth and induce growth cone collapse (19). We confirmed the interaction between NELL2 and Robo3 by means of surface plasmon resonance (SPR) (Fig. 1A). Robo3-ECD-Fc also binds human embryonic kidney (HEK) 293T cells expressing a NELL2–enhanced green fluorescent protein (EGFP) fusion construct tethered to cell surfaces by a glycosylphosphatidylinositol (GPI) anchor (NELL2-EGFP-GPI) in a flow cytometry assay (FACS) (fig. S1, F to J) with high affinity [median effective concentration (EC50) of 10.9 ± 1.6 nM] (Fig. 1B). To determine whether NELL2 binds Robo3 expressed on cells, we transfected COS-7 cells with expression constructs for Robo3.1, Robo3.2, and as controls, DCC, Robo1, or Robo2 and tested for binding of alkaline phosphatase (AP) fusions of Netrin-1, Slit2 N-terminal fragment (Slit2-N), or NELL2. NELL2-AP binds cells that express Robo3.1 or Robo3.2 but not any of the other axon guidance receptors (Fig. 1E and fig. S1B); binding of Netrin-1-AP to DCC and of Slit2-N-AP to both Robo1 and Robo2 confirmed that these receptors are expressed on the cell surface. Robo3.1 does not bind either Netrin-1 or Slit2-N (fig. S1, B and C) (1013). Thus, Robo3 and NELL2 bind to each other specifically and with high affinity.

Fig. 1 NELL2 binds Robo3.

(A) NELL2-Robo3 binding was studied by means of SPR. SPR sensograms show binding response of NELL2 or negative controls at indicated concentrations to immobilized Robo3-ECD-Fc. (B) NELL2-Robo3 binding was studied by means of FACS. The binding curve shows that EC50 binding occurs at a Robo3-ECD concentration of 10.9 ± 1.6 nM. (C and D) Schematics of the (C) NELL2 and (D) Robo3 domain structures. (E) NELL2 domains mediating Robo3 binding were mapped by using a COS-7–based binding assay. NELL21-638-AP and NELL2397-638-AP specifically bind Robo3.1-expressing cells, but NELL21-397-AP does not. TM, transmembrane domain. (F) COS-7–based Robo3 domain mapping. Cells expressing Robo3.1delta(73−535) specifically bind NELL2-AP, but not Netrin-1-AP. (G) NELL2-AP binding was studied on commissural neuron explants from E11.5 Robo3+/− and Robo3−/− embryos (grayscale). NELL2-AP binds cell bodies and axons (arrowheads) of Robo3+/− commissural neurons [which express Robo3.1 (inset)], but not Robo3−/− axons. Scale bars, (E) and (F) 100 μm; (G) 20 μm; inset, 40 μm. Error bars indicate SEM.

We next delineated which regions of Robo3 and NELL2 mediate their interaction. NELL21-638-AP, which lacks the C-terminal vWF domains, binds to Robo3.1-expressing cells, whereas NELL21-397-AP, which also lacks the six EGF-like domains, does not (Fig. 1, C and E). Furthermore, NELL2397-638-AP, which only contains the EGF-like domains, binds to Robo3.1–expressing cells (Fig. 1, C and E). We also found that NELL2-AP binds to COS-7 cells that express Robo3.1delta(73−535), which lacks its five N-terminal immunoglobulin (Ig)–like domains (Fig. 1, D and F), and that in an SPR assay, NELL2-Fc fails to bind Robo3-Ig-Fc, which only contains the Ig-like domains of Robo3 (fig. S1D) but does bind Robo3-FNIII-FLAG, which only contains the three fibronectin type III (FNIII) domains of Robo3 (fig. S1E). As shown with FACS, Robo3-FNIII-Fc binds NELL2-EGFP-GPI–expressing cells (fig. S1K). Thus, the NELL2 EGF-like domains mediate binding to the Robo3 FNIII domains.

Because Robo3 is expressed by commissural neurons (12), we studied NELL2-AP binding to these neurons in vitro. NELL2-AP binds to commissural axons from Robo3 heterozygous embryos but less to axons from Robo3-null embryos (relative axonal binding, 100.0 ± 8.4% for Robo3+/−, 26.4 ± 4.0% for Robo3−/−; P = 0.0044, n = 3 embryos each) (Fig. 1G). Thus, Robo3 is a NELL2 receptor on commissural axons.

We also tested binding of the NELL2 relative NELL1 (18) to Robo3. NELL1-AP binds to COS-7 cells that express Robo3.1 (fig. S1L), and Robo3-FNIII-AP binds to cells that express either NELL1 or NELL2 (fig. S1M). Thus, Robo3 can bind both NELL1 and NELL2 through its FNIII domains.

We next examined NELL1 and NELL2 expression in spinal cord at the time of commissural axon growth to the midline [embryonic day (E) 9.5 to E12.5] by means of in situ hybridization. At E9.5, NELL2 is expressed in the presumptive motor column in the ventral horn (Fig. 2A). At E10.5, additional sites of expression include dorsal root ganglia, the dorsal mantle layer of the spinal cord, and a triangular population of cells close to the ventricle (Fig. 2B). We stained adjacent E10.5 spinal cord sections with antibodies against Robo3.1 and L1 to label pre- and postcrossing commissural axons, respectively (Fig. 2C). Pre-crossing axons project along the lateral edge of a NELL2-free corridor between NELL2-expressing cells in the ventral horn and the ventricular zone (Fig. 2D). NELL2 expression in cells forming this ventral corridor persists at E11.5 (Fig. 2E) and E12.5 (Fig. 2F). We stained adjacent E12.5 sections with an antibody against HB9 to label motor neurons (Fig. 2G) and found that sites of highest NELL2 expression in the ventral horn represent motor neurons (Fig. 2H). In situ hybridization for NELL1 mRNA did not produce detectable signal in E9.5–E12.5 spinal cord sections (fig. S2). Thus, NELL2 is the predominant NELL family member expressed in mouse spinal cord during commissural axon guidance, and areas of NELL2 expression demarcate regions that these axons avoid en route to the midline.

Fig. 2 NELL2 is expressed in the developing spinal cord.

(A, B, E, and F) Transverse spinal cord sections from (A) E9.5, (B) E10.5, (E) E11.5, and (F) E12.5 mouse embryos were used for radioactive in situ hybridization to detect NELL2 mRNA. NELL2 exhibits strong expression in the developing motor column (arrowheads) at all ages examined [the arrow in (A) indicates a labeling artifact from folding of the tissue section]. From E10.5 through E12.5, NELL2 is also expressed in cells surrounding the central canal (asterisk). (C and D) Labeling of E10.5 brachial spinal cord sections with (C) Robo3.1 and L1 antibodies was combined with (D) NELL2 in situ hybridization on adjacent sections. Robo3.1–positive commissural axons do not project through areas of high NELL2 expression. (G and H) Labeling of E12.5 and brachial spinal cord sections with (G) L1 and HB9 antibodies was combined with (H) NELL2 in situ hybridization on adjacent sections. In the ventral horn, areas of high NELL2 expression correspond to the positions of HB9-positive motor neurons. Scale bar, 200 μm.

The interaction of NELL2 with Robo3 and its expression during commissural axon growth to the midline suggest that NELL2 might regulate commissural axon guidance. We tested this idea by exposing commissural axons to NELL2 in vitro. Dorsal spinal cord explants from E11.5 mouse embryos were cultured with Netrin-1, which promotes radially symmetric growth of commissural axon fascicles (20). When confronted with aggregates of COS-7 cells that express either NELL2 and red fluorescent protein (RFP) or RFP alone (Fig. 3, A and B, and fig. S3, A, D, and E), commissural axons grow away from NELL2-expressing cells (Fig. 3B) but not control cells (Fig. 3A), an effect that is statistically significant (Fig. 3E and fig. S3F). Overall axon growth is similar with NELL2-expressing and control cells (normalized total axon length, 100 ± 11.9% for control, 103 ± 13.8% for NELL2; P = 0.7022, n = 5 experiments). Growing axons also turn away from NELL2-expressing cells (ratio of axons turning toward versus away, 0.928 ± 0.046 for control, 0.505 ± 0.012 for NELL2; P = 0.001, n = 5 experiments; quantification method is provided in the supplementary materials) (Fig. 3C). With COS-7 cells that express NELL1 (fig. S3, B and C), we found no statistically significant effect on commissural axon repulsion (fig. S3, G to I). Thus, NELL2 repels commissural axons in vitro, but NELL1 does not.

Fig. 3 NELL2 repels precrossing commissural axons through Robo3.

(A and B) Wild-type E11.5 mouse dorsal spinal cord explants were cocultured with COS-7 cells that express (A) RFP or (B) RFP and NELL2 and stained for class III β−tubulin (TuJ1) to visualize axons (two examples each). NELL2 expression causes axon turning (arrowheads) away from cell aggregates. (E) Commissural axon growth in the proximal quadrant relative to COS-7 aggregates was quantified. Proximal growth is significantly lower with NELL2-expressing cells than with control cells (P = 0.0009). In (C) to (E), dorsal spinal cord explants from E11.5 Robo3 mutant embryos were cocultured with COS-7 cells that express (C) RFP or (D) RFP and NELL2 and labeled with an antibody to TuJ1, and (E) axon growth relative to COS-7 cells was quantified. Proximal growth relative to NELL2-expressing cells is significantly higher in Robo3−/− explants (n = 3 independent experiments) than in explants from control (Robo3+/− and wild-type) littermates (P = 0.0123, n = 3 independent experiments). Scale bar, 200 μm. Error bars indicate SEM.

We next examined whether Robo3 is required for NELL2-mediated repulsion. When dorsal spinal cord explants from Robo3 heterozygous and wild-type embryos were cocultured with cells that express RFP or both NELL2 and RFP, we observed repulsion of commissural axons from NELL2-expressing but not control cells (Fig. 3E). However, commissural axons from Robo3 mutant explants are not significantly repelled by either control (Fig. 3, C and E) or NELL2-expressing cells (Fig. 3, D and E), indicating that Robo3 is required for NELL2-mediated axon repulsion.

To determine whether Robo1 and Robo2 are recruited by Robo3 to mediate NELL2 repulsion, we studied the response of commissural axons from Robo1−/−;Robo2−/− embryos and wild-type littermates to NELL2. Robo1 and Robo2 are not required for NELL2 repulsion (fig. S3J), arguing against cross-talk between Robo3 and Robo1/2 in this pathway.

The switch in Robo3 isoforms upon midline crossing (14) might affect the response of commissural axons to NELL2. We used a postcrossing explant assay (5) to study the response of axons that have crossed the floor plate (fig. S3K). In postcrossing explants cocultured with control cells (fig. S3L) or NELL2-expressing cells (fig. S3M), axonal growth is comparable (fig. S3N). Thus, precrossing commissural axons, which express Robo3.1, are repelled by NELL2, but post-crossing axons, which express Robo3.2, are not, even though both isoforms bind NELL2.

To assess the in vivo function of NELL2, we generated mice deficient in NELL2 (fig. S4A). The abundance of NELL2 mRNA is dramatically reduced in the mutant, as determined with quantitative reverse transcription polymerase chain reaction (RT-PCR) (fig. S4B) and in situ hybridization (fig. S4, C and D), suggesting that the mutant NELL2 transcript is subject to nonsense-mediated mRNA decay and that the allele is null or near null.

On the basis of the expression of NELL2 by motor neurons and its Robo3-dependent repulsive activity for commissural axons, we reasoned that invasion of the motor column in Robo3 mutant mice (12) might be caused not just by reduced midline Netrin attraction and increased midline Slit repulsion—the two known effects of loss of Robo3—but also at least partly by loss of NELL2-mediated repulsion. To compare the Robo3−/− phenotype to a possible NELL2−/− phenotype, we first quantified commissural axon invasion of the ventral horn in Robo3 mutants. We stained spinal cord sections for transient axonal glycoprotein–1 (TAG-1) and HB9 to label precrossing commissural axons and motor neurons, respectively. In wild-type (Fig. 4, A and B) and Robo3 heterozygous animals (Fig. 4C and fig. S4E), the majority of commissural axons avoid the region occupied by motor neurons, whereas in Robo3−/− embryos, many more project through the ventral-most portion of the motor column (Fig. 4D and fig. S4F). Invasion of the motor column in Robo3 mutants was about sixfold greater than in wild-type or Robo3+/− mice (Fig. 4J).

Fig. 4 Commissural axons aberrantly enter the spinal cord ventral horn in NELL2−/−;Robo3+/− mice.

(A to I) Transverse E11.5 mouse spinal cord sections were stained for TAG-1 and HB9. Shown are (A) double stainings and [(B) to (I)] inverted grayscale images of TAG-1 alone. (G) to (I) show high magnifications of the left ventral horn. In [(A) and (B)] wild-type and (C) Robo3+/− embryos, commissural axons project to the midline and avoid the ventral horn occupied by motor neurons (dotted outline). In (D) Robo3−/− embryos, axons fail to cross the midline (asterisk) and project through the ventral horn (arrowheads). In (E) NELL2−/− embryos, axons cross the midline and avoid the ventral horn. In [(F) to (I)] NELL2−/−;Robo3+/− embryos, numerous axons (arrowheads) invade the ventral horn. (G) is a higher-magnification view of a region of (F). (J) Ventral horn invasion was quantified and is significantly higher in Robo3−/− embryos (n = 5 embryos) than in wild-type (P < 0.0001, n = 10 embryos), Robo3+/− (P < 0.0001, n = 3 embryos), NELL2−/− (P < 0.0001, n = 4 embryos), and NELL2−/−;Robo3+/− embryos (P < 0.0001, n = 7 embryos). Axon growth in the ventral horn of NELL2−/−;Robo3+/− embryos is significantly higher than in wild-type (P < 0.0001), Robo3+/− (P = 0.001), and NELL2−/− embryos (P = 0.0002). Robo3+/− and NELL2−/− embryos are indistinguishable from wild type (P = 0.8837 and P = 0.6645, respectively). Scale bars, (A) to (F) 100 μm; (G) to (I) 30 μm. Error bars indicate SEM.

We did not observe an increase in axon growth through the motor column in NELL2−/− (Fig. 4, E and J, and fig. S4G) or NELL1−/−;NELL2−/− mice (fig. S4, I to K). However, because Robo3-dependent mechanisms, including Slit silencing and Netrin potentiation, might be sufficient to promote guidance to the midline and avoidance of the motor column even in absence of NELL2, we generated mice that lacked not only both copies of NELL2 but also one copy of Robo3 and analyzed motor column invasion in this genetically sensitized background. In NELL2−/−;Robo3+/− embryos, there is a significant (approximately threefold) increase in commissural axon growth through the motor column (Fig. 4, F to J, and fig. S4H), an effect confirmed by using Robo3 itself as marker for these axons (fig. S4, L and M). Thus, in collaboration with other cues, NELL2 helps prevent commissural axons from entering the motor column, supporting a role for NELL2 as a repulsive guidance cue in vivo. The function of the low level of NELL2 expression in E10.5 dorsal spinal cord (Fig. 2B) remains to be determined.

Our results identify NELL2 as a repulsive guidance cue that contributes to commissural axon guidance to the midline. NELL2 acts directly through Robo3, which also serves indirectly as both a negative regulator of Slit signaling (12) and a positive regulator of Netrin signaling (13). Commissural axons invade the domain normally defined by NELL2 expression only in mice that completely lack NELL2 and are also partially deficient in the receptor Robo3 (while remaining excluded in mice that either lack NELL2 or partially lack Robo3 in isolation), implying that the NELL2 signaling pathway collaborates with others to ensure avoidance of the motor column. Our results also show that Robo3.1 serves as an integrative hub: Its three diverse actions in response to three different cues—mediating NELL2 repulsion from the motor column, potentiating midline Netrin-1 attraction, and antagonizing midline Slit repulsion—act simultaneously, are mutually reinforcing, and serve the common purpose of steering commissural axons toward and across the midline. This multiplicity of mechanisms likely helps ensure high-fidelity steering of axons to their targets.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

Table S1

References (2127)

References and Notes

  1. Acknowledgments: We thank the members of the Tessier-Lavigne laboratory for discussion and suggestions. We are grateful to C. Culiat for sharing NELL1 mutant mice, J. Ernst for help with purification of recombinant proteins, S. Warming for advice on the assembly of the NELL2 targeting vector by recombineering, M. Roose-Girma for help with generation of NELL2 mutant mice, and M. Yaylaoglu for providing probe templates for in situ hybridization. The NELL2 conditional and full knockout mice as well as cDNA constructs described here are available from A.J. under a material transfer agreement with Genentech. We also thank N. Velarde, D. Collado, K. Sono, and Y. Zhou for technical assistance. This work was supported by Genentech, The Rockefeller University, and Brown University. The supplementary materials contain additional data. Roche provided part of the support for the study but does not stand to benefit from it because no patents were filed.
View Abstract

Navigate This Article