Presenilin-dependent Receptor Processing is Required for Axon Guidance

Columbus mutants exhibit multiple errors in motor axon guidance

To identify genes involved in motor axon navigation we conducted a forward genetic screen in GFP reporter mice mutagenized with ENU (Lewcock et al., 2007). The embryonic MN-specific transgenic reporter ISL^MN:GFP-F was crossed with heterozygous ENU mutants and the offspring were intercrossed to generate homozygous mutants. We identified a mutant we called Columbus that displays a severe defect in ventral root formation (Figure S1A–D) (Lewcock et al., 2007). Normally motor axons preferentially grow through the anterior half of the somite, whereas Columbus motor axons exhibited no preference for the anterior- or posterior-somite, leading to a loss of segmentally organized ventral roots. Transverse sections of Columbus mutants also revealed that a subset of MNs had extended axons into the FP rather than out the ventral roots (Figure 1A, 1B, 1D and 1E). To characterize the midline axon growth defect in more detail we imaged spinal cords using an open book preparation at E13 and found MN misprojections at all levels of the spinal cord (Figures 1G, 1H, 1J, 1K and data not shown). Some motor axons crossed the midline and projected to the contralateral side while others stalled in the FP and formed bundles (Figure S1E and S1F). Likewise, midline motor axon guidance defects were observed with other MN reporters such as Hb9:GFP (Figure S1G and S1H).

Presenilin-1 (PS1) is mutated in Columbus

SNP markers for the mutagenized DBA mouse strain were used to map the Columbus mutation to a 16.7 Mb segment on chromosome 12. The ventral root segmentation and midline-crossing motor axon phenotypes did not segregate during the outcrosses. Next genomic DNA for candidate genes was sequenced. We identified a T/A base conversion in intron 11 of the mouse presenilin-1 gene (PS1), which has 12 exons and 11 introns in the genome (Figure 1M, S1I and S1J). To examine how this mutation affects PS1 expression, we first performed RT-PCR on PS1 mRNA from Columbus mutants using primers that flank the 11^th intron. We found that the point mutation shifted the PCR product from 164 bp to 401 bp in Columbus mutants (Figure 1N). Sequencing Columbus mutant PS1 transcripts showed that the T/A base conversion disrupted the normal splice site for the 12^th exon of PS1. The mutation unmasked a cryptic splice acceptor within intron 11, that resulted in a 237 bp insertion which introduces 2 premature stop codons (Figures 1M). Western-blot analysis revealed that PS1 protein was undetectable in Columbus mutants using both anti-N and -C terminus antibodies. These findings demonstrate that the Columbus mutation alters PS1 splicing, which severely disrupts PS1 protein expression (Figure 1O).

To confirm that the motor axon guidance defects observed in Columbus mutants were due to PS1 rather than another mutation we crossed the ISL^MN:GFP-F MN reporter into PS1 knockout mice (Shen et al., 1997). Embryos with a targeted disruption of the PS1 gene displayed a similar combination of pathfinding errors to those observed in the Columbus mutant, including failure to form discrete ventral roots and midline-crossing of motor axons (Figures 1C, 1F, 1I, 1L; S1K–P, and data not shown). Next we analyzed the distribution of PS1 protein in mouse embryos using immunochemistry and found it was expressed at high levels by MNs and interneurons in the spinal cord as well as peripheral tissues (Figure S2A–C). In contrast, progenitor cells in the ventricular zone expressed much lower levels of PS1. Within MNs PS1 was detected as the cells became post-mitotically born and began axonogenesis, and both cell bodies and axons were labeled.

PS1 function in cell fate specification

Based on the expression of PS1, it remained unclear whether it was required for spinal neuron differentiation, motor axon guidance, and/or the proper development of peripheral tissues. Notch-delta signaling is required for both spinal cord neurogenesis and somite development, and cleavage of the Notch receptor by γ-secretase is required to generate the notch intracellular domain involved in gene regulation (Selkoe and Kopan, 2003). To determine which tissues require PS1 activity we crossed floxed PS1 mice with Nestin-Cre transgenic mice to generate a neural cell-specific PS1 conditional knockout (PS1 cKO) mouse. In PS1 cKO embryos the segmentally repeated ventral roots developed normally (data not shown), whereas the inappropriate midline-crossing motor axon tracts still formed (Figure S2D and S2E). These data suggest that non-neuronal expression of PS1 is required for proper ventral root formation, likely because PS1 is necessary for anteroposterior somite patterning (Shen et al., 1997). In contrast, the midline motor axon-crossing defect in PS1 mutants arises from defects within the neural tube.

Next we examined progenitor cell growth and neural differentiation in developing spinal cords from Columbus mutants using immunofluorescent staining. We found that loss of PS1 function did not change progenitor cell patterning, MN specification, or MN subtype diversification (Figure S2F–K), presumably because PS2 can partially compensate for PS1 inactivation in maintaining Notch signaling (Donoviel et al., 1999). Taken together, our results indicate that the midline motor axon growth defect is due to a neural tube-intrinsic function of PS1 that is unrelated to MN fate specification.

Motor axons grow toward the floor plate in PS1 mutants

Since the inappropriate midline-growth of motor axons was intrinsic to the neural tube we tested whether the FP and/or MNs behaved abnormally in PS1 mutants. The FP and GFP-labeled MNs were dissected from ISL^MN:GFP-F E12 embryos. Tissue was taken separately from either PS1 knockout mutants or controls (a mix of heterozygous and wild type embryos) and cocultured in 3D Collagen/Matrigel matrices for 24 hours. A semi-quantitative assessment of axon outgrowth was performed by counting GFP-positive neurite numbers on the side facing towards (proximal) and away (distal) from the FP (Figure 2A).

In the presence of control FP explants, MNs from wild-type embryos extended more axons from the distal side of the explant, indicating the FP either represses or repels motor axon growth (Figure 2C and 2G). Likewise, control MNs failed to extend axons toward FP explants dissected from PS1 mutant embryos (Figure 2E and 2G). On the other hand, MNs dissected from PS1 mutant embryos extended numerous axons toward FP explants regardless of whether the FP was dissected from control or PS1 deficient embryos (Figure 2D, 2F and 2G). Thus PS1 inactivation in FP cells does not alter motor axon growth compared to controls, whereas the loss of PS1 function in neural tissue leads to aberrant growth of MNs in explant cultures. These findings indicate that PS1-deficient MNs either lose responsiveness to a chemorepellent from the FP or acquire responsiveness to a chemoattractant.

PS1 deficient motor neurons are attracted to Netrin-1

To examine whether PS1 mutant MNs displayed abnormal responsiveness to guidance signals we cocultured GFP-labeled MN explants from ISL^MN:GFP-F embryos with Cos cell aggregates that had been transfected with cDNAs encoding known guidance signals. Since semaphorins are well-established repellents in the midline, we tested whether MN responsiveness to Sema3A was altered in PS1 mutants. We found that both control and PS1 deficient MNs were repelled by Cos cell aggregates expressing Sema3A after 24 hours in culture (Figure 2H–L). Neither wild type nor PS1 deficient spinal MNs from E12 embryos were repelled by Slit2-secreting cell clusters (Figure 2L), suggesting that midline-growth of PS1 mutant MNs is not due to loss of Slit responsiveness.

To assess whether MN attraction to guidance signals was altered we shortened our coculture assay to 15–18 hours to minimize axonal outgrowth in the absence of growth-promoting signals (Figure 2B). We first tested Shh and Netrin-1, which are both known chemoattractants expressed by the FP (Charron et al., 2003; Kennedy et al., 1994). Neither control nor PS1 deficient MNs were attracted to Cos cells secreting Shh (Figure 2O and 2P). Likewise, Netrin-1 failed to promote axonal outgrowth from control MNs (Figure 2Q). Interestingly, there was a marked increase in axonal outgrowth from PS1 mutant explants when cocultured with Netrin-1 secreting Cos cells (Figure 2R). To exclude the possibility that Netrin-1 acted indirectly by altering the Cos cell aggregates, we tested whether purified recombinant Netrin-1 was active. We found that MN explants from control embryos were unresponsive, whereas explants from PS1 mutants extended axons in a dose-dependent manner to bath-applied recombinant Netrin-1 (Figure 2T). These results suggest that PS1 mutant MNs acquire abnormal responsiveness to Netrin-1.

Inhibition of Netrin-1 - DCC signaling rescues the midline-motor axon phenotype

To test if abnormal Netrin-responsiveness caused motor axon guidance defects, we crossed PS1 knockout mice to Netrin-1 hypomorphic mutants to generate PS1/Netrin-1 double mutant embryos (Serafini et al., 1996). We found that the inappropriate growth of motor axons within the FP was significantly reduced by disruption of Netrin-1 expression (Figure 3D–I). Next, we tested whether DCC was required for PS1-deficient MN chemoattraction to the FP. The axon outgrowth from PS1 mutant MNs toward FP explants was inhibited by function blocking antibodies to DCC (Figure 3A–C). Furthermore, we found that MNs in PS1/DCC double knockout embryos avoided the FP (Figure 3J–O). These data suggest that MNs lacking PS1 become attracted to Netrin-1 due to abnormal signaling from the DCC receptor, which leads to inappropriate motor axon growth into the FP.

Inhibition of γ-secretase activity confers responsiveness to Netrin

PS1 encodes the catalytic component of the multi-subunit γ-secretase, but it was unclear if protease activity was required for regulating MN axon growth. We added γ-secretase antagonist L-685458 to wild-type explant cultures and assayed motor axon responses to recombinant or Cos cell produced Netrin-1. Netrin-mediated motor axon growth was stimulated when L-685458 was added to the medium (Figure 4A, 4B and data not shown), indicating that the catalytic activity of PS1 is normally required to prevent MNs from responding to Netrin-1. Furthermore, the finding that MNs become responsive to Netrin-1 following the acute inhibition of PS1 activity with L-685458 suggests that the axon guidance defects in PS1 mutants are not indirectly due to developmental defects that arise from the chronic loss of γ-secretase activity throughout embryonic development.

Next we tested whether isolated MNs lacking γ-secretase activity turn in response to Netrin-1. We established a Dunn chamber assay to create a Netrin-1 gradient (Yam et al., 2009). In this assay, the outer annular well was filled with media containing Netrin-1 and the inner well was filled with control media (Figure 4D). A gradient was formed over the annular bridge through diffusion of the Netrin from the outer to the inner well that was stable for many hours (Yam et al., 2009) (data not shown). ISL^MN:GFP-positive (wild type) chick MNs were dissociated and cultured on a coverslip, which was then inverted over the Dunn chamber (Figure 4D and 4E). In the control condition, the direction of motor axonal growth remained unchanged (Figure 4F). In contrast, when γ-secretase inhibitor L-685458 was added axons turned toward the source of Netrin-1 (Figure 4G). To quantify the extent of turning, we measured the initial angle (α), defined as the angle between the initial orientation of the axon and the gradient, and the angle turned (β), defined as the angle between the initial and final trajectories of the axon (positive for turns up the gradient; negative for turns down the gradient) (Yam et al., 2009) (Figure 4H). Scatter plots of these angles showed that for the control condition, no net turning occurred. However, in the presence of γ-secretase inhibitor, there was a significant bias toward positive turning angles (Figure 4I and 4J). Together, these data indicate that γ-secretase activity is required cell autonomously for newly generated MNs to prevent the Netrin receptor expressed by MNs from responding to Netrin-1 as a chemoattractant.

DCC stubs promote axon growth

Next we explored how γ-secretase was involved in regulating MN responsiveness to Netrin. Previous studies have found that DCC is the target of protease cleavage (Taniguchi et al., 2003). First, DCC is cleaved by metalloprotease/s that lead to “shedding” of the ectodomain segment, generating a membrane-tethered DCC “stub”. This DCC stub is the substrate of γ-secretase, which releases the intracellular domain (ICD) from the membrane (Figure 5A) (Taniguchi et al., 2003). Although the full-length DCC receptor is typically viewed as the primary Netrin-signaling component, it has been shown that the DCC stub and DCC-ICD also have signaling properties (Gitai et al., 2003; Parent et al., 2005; Taniguchi et al., 2003). We found that the levels of DCC stubs and DCC-ICD fragments were very low in wild type neurons, whereas DCC stubs accumulated to high levels in PS1 mutants (Figure 5B, 5C and data not shown). As expected, γ-secretase inhibition also caused the DCC stub to accumulate (see below). To examine if cleavage of DCC played a role in Netrin signaling in MNs we performed collagen assays with wild type MN explants and Netrin-1-producing cells in the presence or absence of metalloprotease and/or γ-secretase inhibitors. Application of metalloprotease inhibitor GM6001 failed to stimulate motor axon outgrowth, while the γ-secretase inhibitor L-685458 caused motor axons to become responsive to Netrin-1 (Figures 5D–F and 5H). Interestingly, the addition of GM6001 to the culture blocked the ability of L-685458 to trigger MN chemoattraction to Netrin-1, indicating that metalloprotease-mediated cleavage is a prerequisite for acquiring Netrin-responsiveness (Figures 5G and 5H). Thus, full-length DCC is unlikely to be sufficient to cause MN chemoattraction. Taken together, these data suggest that DCC in MNs is normally the target of a sequential protease pathway.

Blocking γ-secretase activity should have two effects on DCC processing: (1) cause an accumulation of membrane-tethered DCC stubs, and (2) reduce the generation of DCC-ICD fragments (Figure 5A and 5B). To test whether one or both of these DCC fragments influenced motor axon growth, we made two constructs, one containing the intracellular domain of DCC (DCC-ICD) and the other containing a myristoylated form of the intracellular domain of DCC (Myr-DCC-ICD) to mimic membrane-tethered DCC stubs. First, Myr-DCC-ICD was coelectroporated into HH stage 14 chick embryo neural tubes with a MN-specific reporter Hb9:DsRed. After 2 days incubation, the spinal cords were flat mounted and imaged. We found that the Myr-DCC-ICD construct mimicking DCC stubs induced DsRed-positive motor axons to cross the midline and project contralaterally like MNs in PS1 mouse mutants (Figure 5I and 5J). In contrast, expression of: (1) DCC full-length receptor, (2) the DCC extracellular domain, and (3) the DCC-ICD each failed to induce motor axon growth toward the FP (Figure 5K–M).

Next we tested whether a loss of DCC-ICD signaling caused MNs lacking γ-secretase activity to become responsive to Netrin-1. MN explants were dissected from Hb9:DsRed electroporated chick embryos and cocultured with Cos cells expressing Netrin-1. Wild type DsRed-positive MNs were not attracted to Netrin-1, whereas explants treated with the γ-secretase inhibitor L-685458 became Netrin-1 responsive (Figure 5N, 5P and 5R). This finding suggests that Netrin receptor regulation by γ-secretase is evolutionarily conserved in chick. Likewise, MNs were attracted to Netrin-1 when electroporated with Myr-DCC-ICD (DCC stub mimic) (Figure 5O and 5R). Electroporation of the DCC-ICD failed to prevent L-685458 treated MNs from responding to Netrin-1 (Figure 5Q and 5R), suggesting that inappropriate Netrin attraction is not due to an inability to generate the intracellular domain fragments in PS1 mutants. Rather, these data indicated that the accumulation of DCC stubs in PS1 mutants cause newly generated MNs to become responsive to Netrin-1 in the FP.

Inhibition of Slit/Robo signaling switches motor neurons to a Netrin-1 responsive state

Since MNs express high levels of DCC (Figure S3A, S3B and data not shown) (Keino-Masu et al., 1996), we wondered what normally prevents these neurons from growing to the Netrin-positive FP. Growth cone turning assays suggest that commissural neurons silence Netrin signaling when their axons encounter the Slit-expressing FP, which activates the Robo receptor and leads to Robo-DCC intereactions that prevent Netrin attraction (Figure 6A) (Stein and Tessier-Lavigne, 2001). Interestingly, MNs coexpress both Slit and Robo (Figure S3C–F) (Brose et al., 1999), leading to the possibility of a “self-silencing” mechanism that inhibits Netrin responsiveness (Figure 6A). To test this hypothesis we cocultured chick DsRed-positive MN explants with Netrin-1 expressing Cos cell aggregates. Wild type chick MNs are normally not attracted to Netrin, but electroporation of a dominant-negative form of Robo1 (Robo1-ectodomain+transmembrane domain) into the explants induced motor axon growth toward the Netrin-1 source (Figure 6C and 6F). A similar effect was observed when Robo1-Fc was bath-applied to the explants to sequester Slits (Figure 6D and 6F). Next we generated a gain-of-function variant of Robo1 by combining a myristoylation signal to the intracellular domain of Robo1 (Myr-Robo1), which has been demonstrated to constitutively interact with the DCC cytoplasmic domain (Stein and Tessier-Lavigne, 2001). Electroporation of Myr-Robo1 into chick MNs markedly inhibited the attraction to Netrin-1 caused by Robo1-Fc (Figure 6E and 6F). A similar effect was observed in mouse explants in which bath-applied Robo1-Fc induced motor axon growth toward Netrin-1 (Figure 6G and 6H). To confirm these observations in vivo, we crossed Robo1/2 double mutants to MN transgenic reporter Hb9:GFP and found a subset of MNs extended axons into the FP (Figure 6J and 6K). These findings suggest MNs normally use a self-silencing mechanism based on Slit/Robo coexpression to prevent DCC from causing attraction to Netrin produced by the FP.

DCC stubs are immune to Robo-mediated silencing

Next we examined why the Slit/Robo silencing of Netrin-chemoattraction was ineffective in PS1 mutants. Since DCC stubs accumulate in PS1 deficient embryos, first we tested whether DCC stubs dominantly bind to Robo and release full-length DCC for signaling attraction to Netrin. We performed coimmunoprecipitation assays to examine the interactions between DCC and Robo1 in the absence or presence of DCC stubs. Full-length DCC was coexpressed with myc epitope-tagged Robo1 (Robo1-Myc). γ-secretase inhibitor was added into the cell culture bath to induce the accumulation of DCC stubs. When DCC was immunoprecipitated with an antibody to the extracellular domain of DCC (N-terminus), similar amounts of Robo1 coimmunoprecipitated with DCC, regardless of whether or not DCC stubs were present (Figure 7A). Likewise, using anti-Myc antibody to immunoprecipitate Robo1, we found that coimmunoprecipitated full-length DCC pull down was similar in the presence or absence of DCC stubs (Figure 7A). Interestingly, Robo1 immunoprecipitation did not pull down DCC stubs, suggesting these fragments of DCC are not capable of high affinity interactions with Robo1 (Figure 7A and 7B). We coprecipitated DCC stubs with full-length DCC, however, suggesting Robo1 is excluded from full-length DCC complexes containing the DCC stub (Figure 7B). Taken together, these findings indicate that the receptor complexes containing DCC stubs are insensitive to silencing because they do not interact with Robo.

Ligand binding induces the accumulation of DCC stubs

Since DCC stubs possess potent activity, we tested whether the generation and/or lifespan of this Netrin receptor fragment was regulated. After screening a variety of growth factors and neurotrophins (data not shown), we discovered that Netrin itself induced the accumulation of DCC stubs in a dose-dependent manner (Figure S4B, S4C and data not shown). A slight increase in DCC-ICD was also detected when Netrin was present, but the overall levels of full-length DCC were not substantially changed under these conditions (Figure S4B and S4C). Because DCC is sequentially cleaved by proteases, accumulation of the DCC stub intermediate could occur through one or more mechanisms including: enhancement of the metalloprotease cleavage to generate the stub and/or inhibition of γ-secretase activity involved in clearing the stub (Figure S4A). We first isolated the metalloprotease activity in this process by inhibiting γ-secretase with L-685458. Under this condition we found DCC stub levels increased upon Netrin stimulation (Figure S4E). Next, we examined γ-secretase cleavage by comparing the ratio of DCC stub (substrate) to DCC-ICD (product) and found that DCC-ICD:DCC stub levels declined in the presence of Netrin (Figure S4D). These findings indicate that Netrin regulates the accumulation of DCC stub likely through influencing the sequential processing of DCC mediated by metalloprotease and γ-secretase, thereby fine tuning DCC signaling (Figure S4F).

Commissural neurons require PS1

Although commissural neurons are initially attracted to the midline source of Netrin-1, they use Slit/Robo silencing to switch off their responsiveness in order to cross the midline. To test whether PS1 function is required for this switch, we first performed TAG-1 immunostaining of E12.5 spinal cord sections from control and PS1 mutant embryos. TAG-1-positive commissural axons project ventrally toward the FP in both control and PS1 mutant embryos, however instead of forming a tightly bundled commissure at the FP, the axons appeared disorganized and defasciculated in the mutants (Figure 7C). In addition, the ventral commissure was thicker and the TAG-1 signal in the ventral funiculus was largely absent, suggesting that commissural axons had failed to exit the FP (Figure 7C). DiI labeling of dorsal commissural neurons revealed that commissural axons either stall in the FP of PS1 mutants or inappropriately re-cross the midline (Figure S4G–L). These findings suggest that PS1 is required to switch the guidance properties of commissural neurons once they encounter the FP. In summary, PS1 plays a critical role in establishing neuronal connectivity in both efferent-motor and afferent-sensory pathways.