Transmembrane semaphorin5B is proteolytically processed into a repulsive neural guidance cue


Address correspondence and reprint requests to Timothy P. O’Connor, Ph.D., 2350 Health Sciences Mall, Vancouver, BC V6T 1Z3, Canada.


Developing neuronal growth cones respond to a number of post-transcriptionally modified guidance cues to establish functional neural networks. The Semaphorin family has well-established roles as both secreted and transmembrane guidance cues. Here, we describe the first evidence that a transmembrane Semaphorin, Semaphorin 5B (Sema5B), is proteolytically processed from its transmembrane form and can function as a soluble growth cone collapsing guidance cue. Over-expression of A Disintegrin and Metalloprotease (ADAM)-17, results in an enhanced release of the Sema5B ectodomain, while removal of a predicted ADAM-17 cleavage site prevents its release. In contrast, knockdown of ADAM-17 does not significantly reduce Sema5B release, indicating there are additional unknown compensating proteases. This modulation of the transmembrane Sema5B to a diffusible cue represents a sophisticated method to regulate neuronal guidance in vivo.

Abbreviations used

A Disintegrin and Metalloprotease


amyloid precursor protein


chondroitin sulfate proteoglycan


Dulbecco’s Modified Eagle’s Medium


dorsal root ganglion


extracellular matrix


fetal bovine serum


heparin sulfate proteoglycan


membrane-type-1 matrix metalloproteinase


phosphate-buffered saline






trichloroacetic acid


tissue inhibitor of metalloproteases


thrombospondin repeat

The establishment of neural networks requires the precise guidance of neuronal growth cones to their correct targets. Growth cones are guided by a wide variety of precisely regulated signaling ligands, which mediate both axon growth and guidance (O’Donnell et al. 2009). Guidance cues are regulated by a number of post-transcriptional modifications including alternative splicing (Holmberg et al. 2000), proteolytic processing (Adams et al. 1997; Galko and Tessier-Lavigne 2000; Hattori et al. 2000), formation of multimeric complexes (Davis et al. 1994; Chen et al. 1997; Takahashi et al. 1999; Castellani et al. 2000), and interactions with the extracellular matrix (ECM) (Brose et al. 1999). Many guidance cues are secreted (for example class 3 Semaphorins, Netrins, and Slits) and then either diffuse from their source of production or associate with the ECM (Brose et al. 1999; Suto et al. 2005). These ligands may establish gradients of activity based on their diffusion properties or can be spatially patterned through their interaction with cell-associated binding proteins or the ECM (Brose et al. 1999; Hiramoto et al. 2000). Other guidance cues are membrane-associated or transmembrane and thus provide local cell-associated signals (Suto et al. 2005). There are, however, several instances in which transmembrane or membrane-associated guidance cues undergo protease-mediated proteolysis such that they are released and affect neurite outgrowth and guidance (Galko and Tessier-Lavigne 2000; Schimmelpfeng et al. 2001). Growth cone responsiveness is therefore dependent on the combined effects of a multitude of processed guidance cues and the context in which they are encountered during navigation.

The Semaphorin family consists of both transmembrane and secreted guidance molecules, some of which require proteolytic processing to regulate their function (Adams et al. 1997; Basile et al. 2007). Semaphorins contain an extracellular plexin/semaphorin/integrin and a Semaphorin (Sema) domain, the latter of which is also found in plexins and certain receptor tyrosine kinases (Gherardi et al. 2004). The eight classes of Semaphorin are subdivided according to the presence of other domains including Ig-like, C-terminal basic domains, and thrombospondin repeats (TSRs). Of the five vertebrate Semaphorins (classes 3–7), only one (class 3) is secreted. However, Sema4D is released from certain squamous cell carcinoma cell lines as a result of membrane-type-1 matrix metalloproteinase (MT1-MMP) cleavage at the cell surface (Basile et al. 2007). This release of a membrane-associated Semaphorin is not likely unique to class 4 Semaphorins, as recent evidence suggests that Sema5B is proteolytically processed(Lett et al. 2009; O’Connor et al. 2009).

Several lines of evidence suggest that Sema5B is a repellent guidance cue in the developing brain. First, ectopically expressed Sema5B has been shown to repel cortical axons in vitro (Lett et al. 2009). Furthermore, knockdown of Sema5B in embryonic brain slices resulted in aberrant cortical fiber extension into the ventricular zone (Lett et al. 2009). Whether these effects result from cell-associated or diffusible Sema5B was unclear. However, HA-Sema5B-transfected HEK293-conditioned media has been shown to stimulate synaptic elimination in hippocampal neuron cultures (O’Connor et al. 2009). In addition, multiple Sema5B bands were detected on western blots of hippocampal cell lysates, suggesting that it is processed in vivo (O’Connor et al. 2009). Processing has never been described for Semaphorin5A (Sema5A), suggesting this may be a regulatory function which distinguishes Sema5B from Sema5A. The major aim of this work then was to determine whether Sema5B was released via a proteolytic mechanism, followed by characterization of the cleavage product and the protease responsible for cleavage.

Two families of metalloproteases are largely responsible for most of the known ligand and receptor cleavage events which effect growth cone steering; MMPs, and A Disintegrin and Metalloproteases (ADAMs) (McFarlane 2003). Both ADAMs and MMPs are expressed in the developing CNS, where they are associated with neurogenesis, myelination, cell migration, and axon guidance (reviewed in (Yong et al. 2001). Metalloproteases often mediate ectodomain shedding of target proteins including transmembrane Sema4D in squamous cell carcinoma, and GPI-linked Sema7A in activated platelets (Galko and Tessier-Lavigne 2000; Hattori et al. 2000; Fong et al. 2011). Secreted Sema3C is also cleaved by a metalloprotease (ADAMTS1), releasing it from the ECM associated with breast cancer cells to promote cell migration (Esselens et al. 2010). Thus, there is evidence that metalloprotease-mediated cleavage is a widespread mode of regulating Semaphorin activity, however, this has yet to be shown for transmembrane Semaphorins in the CNS. It is therefore especially relevant to detect and characterize Semaphorin cleavage events, such that both processed and unprocessed forms can be examined for specific activities. This type of regulation will be important when considering the broader roles of the numerous Semphorins in nervous system development.

Here, we describe the ectodomain shedding of the guidance cue Sema5B. Sema5B cleavage was observed in HEK293, C6, and Neuro-2A cells lines. Release was enhanced by phorbol myristate acetate (PMA) treatment and by co-expression with ADAM-17, suggesting that it was likely a protease that mediates Sema5B cleavage. Cleavage was not significantly reduced by siRNA knockdown of ADAM-17, suggesting there are compensating proteases present, likely also members of the ADAM family. The processed form of Sema5B was found to be more repulsive than a non-cleaved mutant, suggesting proteolytic processing may be a means of producing a more potent molecule. These data indicate that Sema5B processing could be an important step in regulating its activities in axon guidance and synapse elimination.



GM6001 was purchased from Calbiochem (La Jolla, CA, USA), MG-132 was from Sigma (St. Louis, Mo, USA), PMA was purchased from Sigma and used at 1 μM in all assays indicated. PNGaseF was from Sigma. Tissue Inhibitor of Metalloproteases (TIMP)-1 and TIMP-2 recombinant proteins were a kind gift from Dr. Chris Overall (University of British Columbia) and their activities were confirmed using active site titration as described by Willenbrock et al. (Willenbrock et al. 1993). siRNAs used for knockdowns were L-003453-00-0005 ON-TARGETplus SMARTpool Human ADAM17 (6868) and D-001810-10-05, ON-TARGETplus Non-targeting Pool from Dharmacon (Lafayette, CO, USA).

Cell lines, constructs, and transfections

HEK293 cells were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM)/F12 (Sigma) supplemented with 10% fetal bovine serum (FBS), penicillin/streptomycin (Pen/Str), 1 mM Glutamax (Invitrogen, Carlsbad, CA, USA). Stable cell lines expressing the full-length chick HA-C5B in pDisplay have been described previously (O’Connor et al. 2009). The secretable HA-Sema was constructed similarly. HA-C5B-Myc, HA-C5B6xHis, ADAM-17-Flag pcDNA3.1, and all deletion constructs were designed and fabricated using restriction-free cloning protocols outlined at using iProof polymerase (Bio-rad Laboratories, Hercules, CA, USA) (Bond and Naus 2012). All HA-C5B mutations were generated using the HA-C5B pDisplay construct as a template (O’Connor et al. 2009). ADAM-17 pcDNA3.1 was used as a template for ADAM-17-Flag {originally Addgene plasmid 19141, [previously described by Lemieux et al. (2007)]}. Control plasmids used were either pEGFP-N1 (Clontech, Mountain View, CA, USA) or pDisplay (Invitrogen). Cells were transfected using Polyethylenimine, Linear (MW 25 000) (PEI) (Polysciences Inc., Warrington, PA, USA) as described by Durocher et al. (Durocher et al. 2002), Lipofectamine2000 or RNAiMax (both Invitrogen).

Generation of concentrated conditioned media, lysates, and immunoblotting

Serum-free conditioned media were collected after 6–8 h, spun briefly to pellet debris, then concentrated using trichloroacetic acid (TCA) precipitation, or, for PNGaseF and outgrowth assays, with AMICON 30MWCO centrifugal filters (Millipore, Billerica, MA, USA). PNGaseF treatments were conducted according to manufacturer’s instructions (Sigma). Lysates were made in radio-immunoprecipitation assay buffer, sonicated briefly, and loaded equally following bicinchoninic acid assay (Pierce, Rockford, IL, USA). For MG-132 proteosome inhibition assays, 24-h post transfection, transiently transfected HEK293 cells were incubated with 1 μM PMA and 10 μM MG-132 for 6 h before cell lysis. siRNA transfections were assayed at 24, 48, and 72 h with no observable difference, so media for replicate experiments was collected on three separate occasions following a 24 h incubation in serum-free with 1 μM PMA media (added at 24 h post transfection). Except where otherwise indicated, proteins were separated on 8% sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) gels and transferred to polyvinylidene fluoride membrane. Polyvinylidene fluoride membranes were stained with GelCode Blue (Pierce) according to manufacturer’s instructions. Antibodies used were mouse anti-HA clone HA7 (1 : 10 000; Sigma),mouse anti-Myc clone 9E10 (1 : 3 of unconcentrated supernatant; American Type Culture Collection (Evan et al. 1985), mouse anti-Flag M2 (1 : 1000; Stratagene, Santa Clara, CA, USA), mouse anti-gamma tubulin (1 : 10 000; Sigma). Horseradish peroxidise-conjugated goat anti-mouse (1 : 10 000; Jackson ImmunoResearch, West Grove, PA, USA), rabbit anti-ADAM-17 (1 : 1000, Abcam, Cambridge, MA, USA). The rabbit anti-Sema5B antibody was described in O’Connor et al. (O’Connor et al. 2009). Relative densities for siRNA immunoblots were done using the Gel Analyzer plugin from ImageJ.

Membrane preparations

Chick brains were dissected and homogenized in 10% sucrose, 50 mM TrisHCl pH 7.5, 1 mM EDTA, then spun at 35 000 g for 10 min. The supernatant was discarded and replaced with 50 mM TrisHCl pH 7.5, 1 mM EDTA and incubated on ice for 30 min. The sample was then centrifuged at 35 000 g for 20 min, the supernatant removed, and the pellet resuspended in a minimal volume of 50 mM TrisHCl pH7.5. To analyze proteins released from membrane preparations, the membrane pellet was thoroughly homogenized through a 26 g needle, then aliquoted into several tubes. Across several time points, the membranes were pelleted at 13 000 g and the supernatant was collected for western blot analysis.

On-cell western blots using Li-Cor odyssey

Transiently transfected HEK293 cells were seeded onto 20 μg/mL laminin, 10 μg/mL poly d Lysine-coated 96-well plates at ∼104 cells per well 24 h post transfection. Following an overnight incubation to allow the cells to fully adhere, cells were rinsed with phosphate-buffered saline (PBS) then fixed for 20 min in 4% Paraformaldehyde, 10% sucrose. Cells were washed 2x then blocked for 1.5 h with Odyssey Blocking Buffer (Li-Cor Biosciences, Lincoln, NE, USA). Primary antibody incubations were conducted overnight at 4°C with rabbit anti-HA in Odyssey Blocking Buffer. Following 3x 10-min washes in PBS, cells were permeabilized by incubation for 15 min in PBS + 0.1% TritonX-100 (PBST). A normalizing mouse anti-β-actin antibody (Sigma) was then added in Odyssey Blocking Buffer at 1 : 1000 for 1 h at 22°C. Cells were washed 3x10 min with PBS then incubated for 1 h in goat anti-rabbit IRDye 680 and goat anti-mouse IRDye 800 (Li-Cor Biosciences). Cells were washed again, then integrated intensities were measured using the Li-Cor Odyssey System. Surface HA-staining was normalized to cell number using the β-actin staining.

Collapse assays

White leghorn chicken eggs were obtained from the University of Alberta and incubated at 38°C. Embryos were terminated at E8 by decapitation, and dorsal root ganglions (DRGs) were dissected then dissociated for 20 min with 0.25% trypsin. Cells were gently triturated, then plated onto 40 μg/mL laminin-coated coverslips in a 12-well plate. Cells were cultured overnight in growth media (DMEM/F-12 supplemented with Pen/Str, 1 mM Glutamax, 0.5% FBS, 0.01 xB27 (Invitrogen) and 50 ng/mL neural growth factor 7S (Invitrogen). Cells transiently transfected in 150 mm plates were switched into serum-free media for 24 h, then media was collected and concentrated down to ∼200x using an AMICON 30MWCO column. HA-C5B6xHis medias were then exchanged into 20 mM TrisHCl pH7.5 and applied to a Ni2+ column (Qiagen, Valencia, CA, USA). The column was washed five times with 20 mM TrisHCl pH7.5, then bound protein was eluted with 250 mM imidazole in 20 mM TrisHCl pH7.5. Imidazole was removed by > 5 exchanges into PBS using an AMICON 100MWCO column. The sample was concentrated down to a concentration relative to 200x the original media. Relative purity was determined on a Gelcode Blue stained membrane, and percent recovery and relative concentration to HA-C5B media was determined by immunoblotting for HA. To assay collapse, concentrated media/purified protein was added to the cell media at a final concentration of ∼10x for 30 min at 37°C. Cells were fixed with 4% Paraformaldehyde, 10% sucrose for 15 min, washed 2x with PBS, then blocked for 1 h in Blocking Buffer (0.3% TritonX-100, 0.1 M Glycine, 7% bovine serum albumin in PBS).Growth cones were stained using Phalloidin Alexa Fluor®488 (Molecular Probes, Carlsbad, CA, USA). Coverslips were mounted with Prolong Gold (Invitrogen). A growth cone was considered collapsed if it completely lacked lamellopodia and had fewer than four filopodia. Assays were done in triplicate counting a minimum of 100 growth cones per experiment, then average percent collapse was compared using a Student’s t-test.

Avoidance and overlay assays

For avoidance and overlay assays, HEK293 cells were transiently transfected and the cells were then replated at low or high density onto Poly d Lysine + Laminin-coated coverslips. For the avoidance assay, E8 chick DRGs were chopped into explants and added to the low-density HEK293 cultures. For the overlay assay, E8 chick DRGs were dissociated with Accustase (Innovative Cell Technologies, San Diego, CA, USA) and added to HEK293 cells that had reached confluency. Cultures were grown overnight in DMEM/F-12 supplemented with Pen/Str, 1 mM Glutamax, 0.5% FBS, 0.01 xB27, and 50 ng/mL neural growth factor 7S. For overlay assays, axons from three coverslips from three separate experiments totaling at least 100 axons per experiment were measured using ImageJ. Average lengths were normalized to controls and compared using a Student’s t-test. Avoidance assays were measured by defining an average cell area, then counting the number of axons crossing through an island per unit cell area. Should one or less axons cross per cell area unit, the island was considered avoided. Two separate experiments were conducted, each with a Fisher test p < 0.00001 and with > 50 islands counted per experiment. Data presented represents the average of both experiments as compared by a Student’s t-test.


Sema5B is proteolytically released from the cell membrane

Following the observation that N-terminally tagged chick Sema5B (HA-C5B) could be detected in transfected HEK293 cell media, it was suggested that the Sema5B ectodomain was released as a result of proteolytic processing (O’Connor et al. 2009). Previous work showed multiple Sema5B bands were detected in western blots of hippocampal cell lysates, supporting this hypothesis (O’Connor et al. 2009).To further characterize Sema5B cleavage in vitro, a C-terminal Myc tag was incorporated into the previously described HA-Sema5B pDisplay construct (HA-C5B-Myc) (O’Connor et al. 2009). When transfected into HEK293 cells, HA-C5B-Myc was detected in cell lysates at ∼160 kDa with both anti-HA and anti-Myc antibodies, whereas a similar, but slightly smaller band was detected with anti-HA, but not anti-Myc, in the media (Fig. 1a).The ∼55 kDa band detected with both HA and Myc antibodies was likely non-specific because of its presence in both untransfected and transfected cells. We observed release of the 160 kDa Sema5B band in additional cell lines, including Neuro2A and C6 cells (Figure S1).

Figure 1.

 Sema5B is Proteolytically Processed. (a) Concentrated media and lysates from HA-C5B-Myc and pDisplay (control)-transfected HEK 293 cells were analyzed by western blot. When probed with anti-HA, bands of approximately 160 kDa were detected in the HA-C5B-Myc lysate and media. When probed with an antibody against the C-terminal Myc tag, the same 160 kDa band was detected only in the lysate. This suggested the media HA-tagged band was separated, likely by proteolytic processing, from the C-terminal Myc-tag. Note also the minor size difference between the lysate and media bands. γ-Tubulin and Gelcode Blue staining were used as loading controls for lysates and media respectively. (b) Media from HA-C5B and HA-Sema-expressing HEK293 cells was concentrated using Amicon MWCO 30 filter devices, then N-deglycosylated with PNGaseF. This shifted the detected bands downwards by approximately 30 kDa on 6% SDS-PAGE immunoblots for HA. This level of glycosylation supports the predicted amount of glycosylation that is expected to occur in the Sema domain (see Figure S2) (Julenius et al. 2005; Julenius 2007, K. Julenius, unpublished data). (c) HA-C5B-Myc, HA-C5B, and pEGFP-N1 (control) were transiently transfected into HEK293 cells. A period of 24 h post transfection, the media was changed for serum-free + 1 μM PMA ± 10 μM MG312 for 2 h. Lysates were then collected and immunoblotted for anti-Myc with γ-tubulin as a loading control. A ∼16–17 kDa band was detected only in the HA-C5B-Myc-transfected cells indicating a C-terminal Myc-tagged fragment. This fragment was detected in higher abundance following MG-312 treatment, suggesting that it is normally degraded by the proteosome. (d) Left- Membranes from E8 chick cortex and spinal cord were assayed for Sema5B using an anti-Sema5B antibody. A 160 kDa band approximating the HEK293 Sema5B band was detected along with several other possible proteolytic fragments, degradation products or non-specific bands. Right-Supernatants from cortex membranes were collected to assay for the release of Sema5B. An increase in released Sema5B was detected over time, suggesting it was being cleaved from cortical membranes similar to Sema5B in HEK293, Neuro2A and C6 cells. The 65 kDa band may also be a Sema5B cleavage product, but there was no supporting evidence from HEK293, Neuro2A, or C6 cells.

The predicted molecular weight of chick Sema5B is only approximately 120 kDa, with the ectodomain accounting for 106 kDa. Both these sizes are significantly smaller than what we detected by western blot, suggesting that both the full-length and released bands were glycosylated or otherwise post-translationally modified. Glycosylation prediction software predicted a high degree of N-glycosylation in the Sema domain, as well as O-fucosylation and C-mannosylation in the TSRs (Figure S2) (Julenius 2007; Julenius et al. 2005; K. Julenius, unpublished data). Removal of N-Glycosylation from HA-C5B-conditioned media using PNGaseF resulted in a size reduction of ∼30 kDa, indicating that most of the size discrepancy was because of N-glycosylation (Fig. 1b). The vast majority of glycosylation appeared to be in the Sema domain as confirmed by a similar deglycosylation experiment conducted with an HA-Sema peptide construct (a secreted polypeptide that contained only the Sema domain) (Fig. 1b). It should be noted that the HA-C5B deglycosylation western in Fig. 1b was run using a 6% instead of 8% SDS-PAGE gel, resulting in a size discrepancy between this and other westerns described.

To further show that Sema5B was proteolytically released into the media, we confirmed the presence of a small C-terminal Sema5B fragment that was left behind in the cell lysates (Fig. 1c). To isolate this fragment, HA-C5B-Myc, HA-C5B (negative control with no C-Terminal Myc tag) and an empty vector control were transfected into HEK293 cells, then immunoblotted for Myc. Because of the possibility of C-terminal fragment degradation following cleavage, a second set of identical transfections were further treated with the proteasome inhibitor MG312 prior to cell lysis. As shown in Fig. 1c, a band was detected at ∼16–17 kDa in the HA-C5B-Myc lysates only, reflecting a C-Terminal Myc-tagged fragment. Furthermore, this band was considerably stronger in the MG312-treated lysates, suggesting that this product is normally degraded by the proteasome after Sema5B processing.

Assessing whether Sema5B is also cleaved and released from neurons proved difficult because of the lack of availability of specific and sensitive Sema5B antibodies. Cultured dissociated neurons were assayed for Sema5B release into the media, but no signal was detected and only weak signal was detected in lysates (data not shown). In attempt to address this issue, several classes of neurons were transfected or electroporated with the HA-C5B-Myc construct, however, the efficiency was such that only a weak 160 kDa band could be detected in lysates after immunoprecipitation (data not shown). As a final attempt to concentrate Sema5B in high enough quantities to assay with the rabbit-anti-Sema5B antibody we have used previously (O’Connor et al. 2009), membranes were isolated from E8 chick spinal cord and cortex, then assayed by western blot. We have previously shown that this antibody can detect the 160 kDa HA-C5B media fragment. A strong signal was consistently detected at 160 kDa in both E8 cortex and spinal cord, along with some smaller products which may be degradation products, alternate proteolytic products or non-specific peptides(Fig. 1d).

To assay whether Sema5B was being processed, cortical membranes were collected, and the supernatant was assayed for Sema5B release over time. We found that Sema5B was shed from the membrane preparations, along with a 65 kDa fragment (Fig. 1d). The 65 kDa fragment may be a true Sema5B cleavage product, however, as we had no corroborating evidence in transfected cells and were unable to isolate sufficient quantities for mass spectrometry identification, it was not further characterized. Based on these results, and the above data showing Sema5B processing in three non-related cell lines (including a neuroblastoma line) (Figure S1), all evidence suggests that Sema5B is proteolytically processed in vivo. However, because of the difficulties of detecting Sema5B in media from cultured neurons, this release is likely constitutively low and subject to spatial, temporal, and activity-dependent induction.

Sema5B cleavage can be prevented through deletion of amino acids 962–964

Following confirmation that Sema5B was indeed proteolytically processed, we examined whether this processing was a requirement for Sema5B’s function as a neural repellent. To this end, a cleavage-deficient mutant was generated using site-directed mutagenesis. A target region was selected between amino acids 955–968, which was proximal to the transmembrane domain, prior to the first TSR, and highly conserved from humans through to zebrafish (Fig. 2a). Cleavage within this region would generate a C-terminal fragment between 15 and 16 kDa, which closely approximates the mass estimated in Fig. 1c.

Figure 2.

 Generation of a non-cleaved Sema5B construct. (a) The region just N-Terminal to the transmembrane region was targeted for site-directed mutagenesis to generate a non-cleaved Sema5B. A region between amino acids 955–968 was targeted because of the high degree of conservation across species relative to surrounding areas and its location correlating with the proposed C-terminal fragment size. Deletion ranges are indicated. (b) Deletion constructs were transiently transfected into HEK293 cells, then lysates and media were assayed 24 h post transfection. Despite relatively equal expression levels (bottom), media showed a slight decrease in shedding with the Δ958–961 mutant, and almost complete inhibition with the Δ962–964 deletion. (c) Surface accessibility assays were done on the HA-C5BΔ962–964 construct by using non-permeabilizing immunolabeling for HA, followed by permeabilization and labeling for β-actin to normalize for minor differences in plating density. Labeling was detected using IRDye-conjugated antibodies and the Li-Cor Odyssey System. Quantification of normalized integrated intensities indicated that HA-C5BΔ962–964 was successfully trafficked to the cell surface with a slight but significant increase relative to HA-C5B-Myc. This possibly reflected additional retention of Sema5B at the cell surface in the absence of cleavage.* indicates p < 0.05 relative to HA-C5B-Myc.

A series of three to four amino acid deletion constructs was generated between amino acids 955–968 to identify any cleavage-deficient mutants. When expressed in HEK293 cells, all constructs were found to have relatively equal expression levels in cell lysates, however, shedding was reduced slightly in the Δ958–961 deletion and almost completely inhibited with the Δ962–964 deletion (Fig. 2b). To establish whether this reduction in shedding was because of inhibition of cleavage or reduced trafficking to the cell surface, transiently transfected HEK293 cells were assayed using an On-Cell Western using the Li-Cor Odyssey system. HA-C5B-Myc, and HA-C5BΔ962-964-transfected cells were surface labeled for HA, then permeabilized to label β-actin as a normalizing control. When analyzed, the HA-C5BΔ962–964 construct was indeed found to be successfully trafficked to the cell surface, and was also found to a have a slight but significant increase in surface expression relative to HA-C5B-Myc (Fig. 2c). These data suggest that HA-C5BΔ962–964 is trafficked to the cell surface, and supporting its function as a cleavage-deficient mutant, is retained at the cell surface to a greater degree than wild-type Sema5B. Thus, the Δ962–964 construct was deemed appropriate to use as a cleavage-deficient mutant in neurite outgrowth assays.

Released Sema5B collapses neurite growth cones from E8 chick dorsal root ganglia

Previous data have shown that Sema5B released from a recombinant C-terminal deletion construct (HA-C5BΔC) can collapse hippocampal neurite growth cones and stimulate synapse elimination when applied as concentrated media (O’Connor et al. 2009). There is therefore precedent for a released Sema5B molecule having a repulsive function. However, this Sema5B fragment was not characterized to any extent, thus to more precisely test whether a cleaved Sema5B fragment could repel neurons, a secretable C-terminally His-tagged (HA-C5B6xHis) construct was generated and purified. This construct was truncated at amino acid 963 to generate a soluble Sema5B fragment of the exact same size as determined by our mutation analysis (Fig. 2). When E8 DRG neurons were treated with the purified HA-C5B6xHis construct, an average of 44% of growth cones were collapsed relative to a 15% average in controls (p < 0.05) (Fig. 3a and b). Similar observations were made when concentrated, conditioned media from HA-C5B cells was used in parallel collapse assays (Fig. 3b). These results indicate that the shed Sema5B ectodomain was indeed repulsive toward DRG neurons.

Figure 3.

 Cleaved Sema5B is repulsive toward dorsal root ganglion growth cones. (a) Not collapsed (left) and collapsed (right) growth cones are shown with anti-Tuj (green) labeling the microtubules and Phalloidin (Red) labeling the actin filaments in growth cones. Growth cones with no lamellopodia and four or less lamellopodia were considered collapsed. (b) Quantification of E8 chick DRG neuron growth cone collapse following application of concentrated conditioned media (GFP, HA-C5B) or the purified HA-C5B6xHis. HA-C5B and HA-C5B 6xHis had very similar activities, collapsing 44% of growth cones compared with the 15% seen in controls (p < 0.05). There was no significant difference between the HA-C5B-Myc-concentrated media and purified HA-C5B 6xHis suggesting there was no other concentrated media product inducing collapse (p = 0.92).* indicates p < 0.05 relative to control.

Proteolytic processing increases DRG neurite repulsion in response to Sema5B

To determine the necessity of cleavage for Sema5B activity, the cleavage-deficient HA-C5BΔ962–964 mutant was used for both cell island avoidance assays and overlay assays with E8 DRG axons. For avoidance assays, HA-C5B-Myc and HA-C5BΔ962–964 cells were plated at low density with E8 DRG explants. To normalize for differing cell island areas, an average single cell size was determined and then used to approximate the number of cells per cell island. Avoidance was then defined as islands which had one or less contacting axons per number of cells in the island. For example, should the area of an island have been determined to include three cell-areas, then if three or less axons crossed over the island it was said to be avoided. When quantified, approximately 80% of the HA-C5B-Myc and 70% of HA-C5BΔ962–964 islands were avoided, while significantly fewer control islands were avoided (p > 0.05) (Fig. 4a). The difference between HA-C5BΔ962–964 and HA-C5B-Myc avoidance was also significant, with axons showing less avoidance of the HA-C5BΔ962–964 islands (p < 0.05). Qualitatively, it appeared as though the pockets of axons formed around HA-C5B-Myc cells gave a larger berth than for HA-C5BΔ962–964, likely reflecting a zone of diffusion and therefore the larger area of avoidance (Fig. 4b).

Figure 4.

 Non-Cleaved Sema5B is less repulsive than wild-type Sema5B. (a) Quantification of the DRG axon avoidance assay. E8 DRG explants were grown in the presence of low confluence HA-C5B-Myc, HA-C5B-Δ962–964 or pEGFP-N1-transfected cell islands. An island was determined to be avoided should an area representative of a single cell contact one or less axons (see results section). HA-C5B-Myc and HAC5BΔ962–964 cell islands were significantly avoided by DRG axons with averages of 80% of HA-C5B-Myc and 70% of HA-C5B-Δ962–964 islands being avoided (p < 0.05). Control islands were avoided an average of only 6% of the time. The difference between HA-C5B-Δ962–964 and HA-C5B-Myc avoidance was also significant (p < 0.05). (b) Representative images of cell island avoidance assay. Tuj-1 labeled axons are labeled in red and HA is labeled in green. To better illustrate the size of the avoidance ‘pocket’, the cell islands were outlined with dashes and overlayed onto the Tuj-1 staining. Avoidance pockets are apparent, with a larger berth being given to the cleaved wild-type HA-C5B-Myc relative to the cleavage-deficient mutant. pEGFP-N1-expressing cells showed relatively little avoidance. (c) Quantification of DRG neuron overlay assay. Dissociated E8 chick neurons were cultured on top of a confluent layer of transiently transfected HEK293 cells. HA-C5B-Myc-expressing cells induced an average axon length that was 48% of controls, while HA-C5BΔ962–964 was an average length of 57% of controls (three experiments, minimum of 75 axons/experiment, > 100 axons per experiment, p < 0.05). The shorter relative axon lengths of HA-C5B-Myc compared with HA-C5BΔ962–964 cocultures was also significant with p < 0.05.* indicates p < 0.05.

To further examine the repulsive activity of the cleavage-deficient mutant, dissociated E8 chick DRG axons were grown on a confluent monolayer of HEK293 cells transfected with HA-C5B-Myc, HA-C5BΔ962–964 or green fluorescent protein (GFP) (control) cells. Axon lengths were measured after 16 h then normalized to axon lengths from GFP control assays. As shown in Fig. 4c, both Sema5B constructs led to reduction in axon length, with HA-C5B-Myc reduced to 48% and HA-C5BΔ962–964 reduced to 57% of control axons lengths (p < 0.05) (Fig 4c). The shorter relative axon lengths of HA-C5B-Myc compared to HA-C5BΔ962–964 cocultures was also significant (p < 0.05). This is in agreement with the cell island avoidance assay, which indicated that full-length Sema5B is more repulsive than the cleavage-deficient construct. The difference in both cases, although significant, was only approximately 10–15%, indicating that while cleavage does produce a more repulsive molecule, the membrane-anchored Sema5B is nonetheless a potent repulsive protein.

Sema5B cleavage is induced by co-expression of ADAM-17

To address the mechanism through which Sema5B is released from the cell surface, stably transfected HA-C5B cells were grown in the presence of a panel of protease inhibitors including cysteine, serine, and metalloprotease inhibitors. Of these, the metalloprotease inhibitor GM6001 was found to inhibit HA-C5B release in a dose-dependent manner between 5 and 50 μM (Fig. 5a). To further narrow down the possible metalloproteases, recombinant TIMP-1 and 2 were assayed for their ability to inhibit HA-C5B shedding. As seen in Fig. 5b, neither TIMP-1 nor TIMP-2 were found to have any effect on the cleavage of Sema5B. This result strongly indicated that the metalloprotease in question was not a MMP as TIMPs-1 and -2 together inhibit all known MMPs (Visse and Nagase 2003).

Figure 5.

 Sema5B cleavage results from metalloprotease activity. (a) The broad-spectrum metalloprotease inhibitor GM6001 was used to treat transiently transfected HA-C5B cells with a dosage range of 5–50 μM over 6 h. A dose-dependent decrease in release was observed, indicating cleavage was metalloprotease dependent. Gelcode Blue staining was used as a general loading control. (b) Recombinant TIMPs -1, and -2 were used to narrow down the class of metalloprotease responsible for Sema5B cleavage (in the presence of 1 μM PMA). TIMPs -1 and -2 had no effect on Sema5B release between 1 and 100 nM. Gelcode Blue was again used as a loading control.

The protein kinase C agonist PMA has often been shown to up-regulate ectodomain shedding in cell culture systems [reviewed in (Hayashida et al. 2010)]. Enhanced HA-C5B-Myc release was observed within an hour of 1 μM PMA treatment (Fig. 6a). PMA is a known activator of several candidate proteases, particularly the TIMP-1 and -2 insensitive metalloprotease ADAM-17 (Lohi et al. 1996; Peschon et al. 1998). ADAM-17 is fairly promiscuous in terms of its substrate recognition sites, however, Caescu et al. (Caescu et al. 2009) have identified ADAM-17 sequence preferences. They reported that ADAM-17 prefers a sequence of P5-P4-P3-P2-P1-X-P1′-P2′-P3′-P4′, where X is the cleavage site and a hydrophobic residue (most often Proline) is observed at position P5 or P3, with small residues at positions P3, P2, and P1, an aliphatic hydrophobic residue at P1′, and a small residue at position P3′. These sites correlate well with our identified cleavage site when defining P5 as the hydrophobic Proline positioned at site 957, followed by the small hydrophobic residues Isoleucine, Leucine, and Proline as P3, P4, and P1, Alanine at 962 as P1′, then Serine at 964 as P3′. These results provide further support for ADAM-17 to function as a candidate protease of Sema5B.

Figure 6.

 Sema5B cleavage can be induced by A Disintegrin and Metalloprotease (ADAM)-17, but is subject to cleavage by other unidentified proteases. (a) HA-C5B stably transfected cells were treated with the protein kinase C agonist PMA over a time course of 1–2 h, then the media were TCA concentrated and immunoblotted for HA. No release was observed from control-treated cells after this short time period, while HA-C5B was detected strongly in the media of PMA-treated cells. (b) ADAM-17-Flag and a control vector (pEGFP-N1) were transfected into stably transfected HA-C5B HEK293 cells to determine whether ADAM-17 could induce Sema5B cleavage. Lysates and media were collected and media was TCA precipitated. While lysate expression levels remained comparable (lanes shown are from same membrane and exposure), the media showed a significant increase in release of Sema5B with ADAM-17 –Flag transfection. The second, possibly less glycosylated or alternative cleavage product at 100 kDa was consistently detected in this assay. ADAM-17-Flag expression was confirmed with anti-FLAG, while loading was confirmed with γ-tubulin (lysates) and Gelcode Blue (media). All blots are from the same gel, however, black lines were used to indicate where lanes were repositioned for clarity purposes. (c) ADAM-17 was knocked down using siRNA in HA-C5B-Myc transiently transfected HEK293 cells. Knockdown was confirmed using an anti-ADAM-17 antibody, and loading was controlled for using γ-tubulin. When the media was analyzed, only a slight decrease in HA-Sema5B release was seen consistently over three experiments.

To determine whether addition of ADAM-17 could induce higher levels of Sema5B cleavage, an ADAM-17-Flag construct was over-expressed in HA-C5B-transfected HEK293 cells. A large increase in Sema5B shedding was seen in all cases (reproduced on > three occasions). Also, a slightly smaller band was occasionally seen in highly concentrated medias (Fig. 6b). This secondary product may be a second cleavage product or reflect a lesser glycosylated cleavage product. There was no concomitant decrease of full-length Sema5B in the lysates (Fig. 6b). This may suggest that the relative amount of full-length Sema5B held at the membrane is much higher than is shed, such that an increase in shedding does not visibly affect the surface levels of Sema5B. Alternatively, Sema5B-transfected cells may be able to modulate Sema5B expression in response to increased cleavage. Nonetheless, these data together suggest that Sema5B cleavage is induced by ADAM-17.

To further examine whether ADAM-17 is the only candidate protease responsible for Sema5B cleavage, we knocked down its expression with siRNA in HEK293 cells expressing HA-C5B-Myc. Knockdown was confirmed by immunoblot, and medias were assayed for reduction in HA-C5B-Myc shedding (Fig. 6c, top). We observed a minor but not significant reduction in Sema5B cleavage, suggesting that alternative metalloproteases may be able to compensate for the loss of ADAM-17.


Prior to this work, transmembrane Semaphorins were thought to act primarily as membrane-associated, short-range guidance molecules which signaled through transmembrane or membrane-associated receptors such as plexins, neuropilins, and integrins (Gherardi et al. 2004). Previous work has shown that Sema4D is released from certain squamous cell carcinoma cell lines by MT-MMP1, resulting in a soluble product which can act as a chemoattractant in vitro and promote angiogenesis in vivo (Basile et al. 2007). This is the first evidence, however, of a transmembrane Semaphorin that is proteolytically processed into a potential diffusible neuronal guidance cue. Indeed, growth cone assays show that Sema5B can collapse DRG growth cones when not membrane associated, a result which is consistent with our previous work using dissociated hippocampal cultures and cortical explants (Lett et al. 2009; O’Connor et al. 2009). Furthermore, both overlay and avoidance assays suggest that a processed Sema5B may be more potent than a non-cleavable full-length form. Nonetheless, the non-cleaved form of Sema5B is also a potent repellant guidance cue. This suggests a simple mechanism for Sema5B expressing cells to modulate their effective repulsiveness with the expression of specific proteases such as ADAM-17.

It remains unknown whether Sema5B is additionally modified following its release from the membrane. Sema5A can function as an attractive or repulsive guidance cue depending on the presence of heparan sulfate proteoglycans (HSPGs) or chondroitin sulfate proteoglycans (CSPGs), respectively (Kantor et al. 2004). Similarly, based on glycosaminoglycan microarrays, Sema5B has the ability to interact with both CSPGs and HSPGs (Shipp and Hsieh-Wilson 2007), therefore it is tempting to speculate that processed Sema5B may be retained and functionally modulated by the extracellular matrix (ECM) via TSR-mediated proteoglycan interactions. Thus, Sema5B could potentially be a bifunctional guidance cue, depending on whether the released molecule interacts with HSPG or CSPGs.

Our results indicate that cleavage of Sema5B is largely metalloprotease dependent, and can be induced by the protease ADAM-17. Over-expression of ADAM-17 potently enhanced Sema5B cleavage, however, reduction of ADAM-17 using siRNA had little effect on Sema5B release. Although at first, these results appear to conflict, it is important to consider that protease networks are complex, and redundancies often exist such that removal of one component does not necessarily prevent cleavage of the target protein. This is especially the case with the ADAM family of proteases. ADAM-8, ADAM-9, ADAM-10, and ADAM-17 can all cleave and shed amyloid precursor protein (APP); however, it is still unclear which functions in a physiological setting (Asai et al. 2003). ADAM-9, -10, and -17 knockout mice phenotypes and tissue expression patterns indicate that there is functional redundancy in place, with the potential for tissue-specific ADAM combinations regulating APP release (Hartmann et al. 2002). Meanwhile, ADAMs -10 and -17 both mediate cleavage of PrPc protein at the 110/111–112 peptide bond to produce the non-pathogenic N1 breakdown product (Chen et al. 1995; Vincent et al. 2001). Interestingly, while recognizing the same site, ADAM-10 is responsible for constitutive cleavage, while ADAM-17 mediates regulated cleavage (Vincent et al. 2001). In fact, ADAM-10 and ADAM-17 share many substrates including APP (Hartmann et al. 2002; Asai et al. 2003), PrPc (Vincent et al. 2001), L1 (Maretzky et al. 2005), and Notch (Brou et al. 2000; Hartmann et al. 2002). Furthermore, while there are preferences at certain positions, ADAMs-10 and -17 often share identical consensus sequences (Esch et al. 1990; Mullberg et al. 1994; Chen et al. 1995; Brou et al. 2000; Mumm et al. 2000; Hinkle et al. 2004; Caescu et al. 2009; Kummer et al. 2009). ADAMs-10 and-17 further have some overlap in tissue expression with Sema5B. Specifically in hippocampal neurons (Karkkainen et al. 2000; Skovronsky et al. 2001; O’Connor et al. 2009; Restituito et al. 2011), and the ventricular zone and DRGs of embryonic chick spinal cord (Lin et al. 2010). Also, ADAM-10 (and not ADAM-17) is known to be expressed in the cortical subventricular zone in a manner similar to Sema5B (Yang et al. 2005; Lett et al. 2009; Demars et al. 2011). ADAM-9 shares many of these expression patterns as well (Weskamp et al. 2002; Lin et al. 2010). Thus, it is highly plausible that Sema5B is the target of multiple proteases in addition to ADAM-17 whose activities are likely spatially, temporally, and contextually regulated.

One such scenario may exist in the hippocampus, where ADAM-10 and -17 have been shown to localize throughout hippocampal neurons and are particularly concentrated both pre- and post-synaptically (Restituito et al. 2011). Their activation leads to attenuation of synaptic transmission and loss of synaptic proteins, partly as a result of N-Cadherin cleavage (Restituito et al. 2011). Treatment with the Sema5B cleavage fragment and over-expression of Sema5B in hippocampal neurons results in decreased synapse number only when Sema5B was expressed on the post-synaptic side, consistent with a role in synapse elimination (O’Connor et al. 2009). Hippocampal neurons also express Plexin A3, which has been suggested to be a component of the Sema5B receptor (Liu et al. 2005; Matsuoka et al. 2011). One could hypothesize that activity-dependent ADAM-17/10 cleavage of Sema5B could regulate synapse disassembly by facilitating diffusion of Sema5B to target pre-synaptic Plexin-A3 receptors on neighboring cells, while restricting its range through a combination of limited cleavage (thereby only affecting immediately adjacent receptors) and/or restricting diffusion (while concentrating cleavage product) through association with local ECM components.

Having the Sema5B ectodomain act in a paracrine and autocrine manner is an intriguing possibility. In addition to the scenario described above in the hippocampus, ADAM-expressing neurons that are non-Sema5B responsive could proteolytically remove Sema5B, thereby creating a path through which Sema5B-responsive follower axons could navigate. Alternatively, there are a number of known Sema5B-responsive neurons that also express Sema5B. Thus, one way to control neuronal outgrowth and steering could be by the precise ADAM-mediated release of Sema5B and consequent autocrine signaling to modulate growth cone navigation (Serini et al. 2003; Casazza et al. 2007). This would provide a unique and complex method of controlling neuronal growth. At this point, both overlay assays and published results in the hippocampus suggest that the ectodomain minimally acts in a paracrine manner, with further experiments being necessary to further explore autocrine signaling possibilities.

The receptor proteins Plexin-A1 and -A3 have recently been shown to be necessary in part for the Sema5A- and 5B-mediated inhibition of retinal ganglion cell neurite growth (Matsuoka et al. 2011). In addition, Sema5A, and not Sema5B, can bind to and elicit a functional response to several Plexin-B3-expressing cells (Artigiani et al. 2004; Pan et al. 2009; Li and Lee 2010). This suggests that Sema5A and Sema5B, while similar and possibly semiredundant, can trigger different cellular responses and raises the questions as to why and how these two otherwise homologous molecules differ. The Sema5A ectodomain has been shown to have a very limited function when expressed as a secreted molecule unless oligomerized using an Fc fusion domain and antibody (Artigiani et al. 2004). In contrast, we have shown that Sema5B is functional and more potent in its processed, released form. Cleavage may therefore be one of the unique features conferring additional functionality to Sema5B which is not present in Sema5A.

At this time, it is unclear how Sema5B cleavage may be actively regulated. ADAM member activity can be induced by numerous stimuli including growth factors, serum factors, intracellular Ca2+, mechanical or osmotic stress, and protein kinase C activation (Reiss and Saftig 2009). Furthermore, the TIMP proteins are endogenous inhibitors of ADAM proteins, particularly TIMP-3, which is ECM associated and can inhibit both ADAM-17 and ADAM10 (Amour et al. 1998, 2000), thus providing an additional level of Sema5B regulation. For example, TIMP-3 (which increases throughout CNS development) could inhibit Sema5B diffusion throughout the nervous system, restricting Sema5B function to a membrane-associated repulsive cue. Further exploration using in vivo studies will be necessary to probe deeper into the intricacies of Sema5B proteolytic regulation.


We thank the laboratory of Dr. Chris Overall from the University of British Columbia for providing the TIMP-1 and -2 recombinant proteins. This work was supported by the Canadian Institutes for Health Research (grant# MOP13246) and the Natural Sciences and Engineering Research Council of Canada (grant# RGPIN171387-09). We have no conflicts of interest to report.

Authorship credit

Kristen Browne conducted all experiments, generated all figures, and wrote the manuscript. Wenyan Wang generated the original HA-C5B pDisplay constructs and conducted numerous preliminary experiments for this project. Qian Qian Liu also conducted preliminary experiments. Matthew Piva provided the imaging, measurements, and data analysis for both the overlay and avoidance assays. Senior author Timothy O’Connor supervised the experiments and edited the manuscript.