Molecular mechanisms of proteoglycan-mediated semaphorin signaling in axon guidance

Significance

The development of the nervous system relies on precise axon guidance, orchestrated by cues like semaphorins. These cues not only engage their cognate receptors but also interact with extracellular matrix components like proteoglycans, whose role remains poorly understood. Our study reveals that secreted semaphorins bind specifically to proteoglycan glycosaminoglycan chains via multiple sites, with an essential high-affinity site located at the C-terminal tail. Deleting this site disrupts semaphorin activity in guiding olfactory receptor neuron axons in vivo. We propose that secreted semaphorins attach to cell surfaces via proteoglycan interactions, aiding their recognition by axon receptors.

Abstract

The precise assembly of a functional nervous system relies on axon guidance cues. Beyond engaging their cognate receptors and initiating signaling cascades that modulate cytoskeletal dynamics, guidance cues also bind components of the extracellular matrix, notably proteoglycans, yet the role and mechanisms of these interactions remain poorly understood. We found that Drosophila secreted semaphorins bind specifically to glycosaminoglycan (GAG) chains of proteoglycans, showing a preference based on the degree of sulfation. Structural analysis of Sema2b unveiled multiple GAG-binding sites positioned outside canonical plexin-binding site, with the highest affinity binding site located at the C-terminal tail, characterized by a lysine-rich helical arrangement that appears to be conserved across secreted semaphorins. In vivo studies revealed a crucial role of the Sema2b C-terminal tail in specifying the trajectory of olfactory receptor neurons. We propose that secreted semaphorins tether to the cell surface through interactions with GAG chains of proteoglycans, facilitating their presentation to cognate receptors on passing axons.

The wiring of the nervous system relies on the precise guidance of axons to reach their targets. Central to this process are axon guidance cues, a diverse group of signaling proteins that provide spatiotemporal instructions to growing axons, allowing them to navigate through the complex environment. Over the past decades, four main families of axon guidance cues have been identified: semaphorins, netrins, slits, and ephrins. These attractive and repulsive cues bind to their cognate receptors and coreceptors and regulate signaling pathways that govern axonal pathfinding by controlling cytoskeletal dynamics (reviewed in ref. 1). Intriguingly, a limited repertoire of guidance cues orchestrates the wiring of a vast number of neuronal connections. However, the mechanisms underlying the functional versatility and complexity of these cues are just beginning to be elucidated. One plausible mechanism involves the modulation of guidance cue signaling by components of the extracellular matrix, particularly proteoglycans.

Proteoglycans are composed of core proteins and covalently attached linear polysaccharides known as glycosaminoglycan (GAG) chains. While there is a limited number of core proteins, the GAG chains undergo a myriad of modifications such as sulfation, epimerization, or acetylation, giving them unique binding affinities and specificities. Also, the composition and modifications of GAG chains vary significantly during development and aging (reviewed in ref. 2). The vast diversity of GAG chains might add another layer of complexity to the intricate network shaped by the combinatorial expression of axon guidance cues and their cognate receptors.

The pivotal role of proteoglycans in axon guidance was initially observed in cultured cockroach embryos. The introduction of exogenous heparin or heparan sulfate led to pronounced pathfinding defects in pioneering axons within limb buds, while enzymatic degradation of heparan sulfate using heparinase yielded similar axonal perturbations (3). The role of heparan sulfate in axon navigation was also demonstrated in the development of the vertebrate visual system (4, 5). These in vitro findings have been further supported by a number of genetic studies. Mice with CNS-specific disruption of the Ext1 gene, responsible for the biosynthesis of heparan sulfate, exhibited notable anatomical and axonal guidance defects similar to those in Slit1/Slit2 double-knockout mice (6). Similarly, studies on mice lacking specific heparan sulfate sulfotransferases, Hst2st or Hs6st1, showed aberrant axon navigation at key brain regions such as the optic chiasm and corpus callosum (7). In Drosophila, the loss of proteoglycan Syndecan led to impaired midline crossing by longitudinal axons, resembling the phenotype of Slit and Robo mutants (8, 9). Remarkably, the midline crossing defects caused by the loss of Syndecan could be rescued by overexpression of another proteoglycan, Dally-like protein (Dlp), suggesting the functional redundancy of proteoglycan core proteins (8). Additionally, the loss of secreted proteoglycan Perlecan resulted in motor axon guidance defects reminiscent of those observed in Sema1a and PlexA mutants. However, these motor axon defects could not be rescued by overexpressing Syndecan or Dally-like proteoglycans, indicating distinct functional roles for these two proteoglycans (10).

In Caenorhabditis elegans, an elegant study examining mutations in each individual heparan sulfate biosynthetic enzyme unveiled that distinct axons relied on different combinations of enzymes for proper guidance. This observation led to the proposal of the heparan sulfate code hypothesis, suggesting that the modification pattern of GAG chains provides a specific code for axon guidance (11). This hypothesis has since been expanded beyond heparan sulfate proteoglycans, encompassing other nonheparan sulfate variants, and has been termed the “sugar code” for axon guidance (12).

A growing body of evidence in recent years highlights the intricate interplay between semaphorins and proteoglycans. Semaphorins, the largest family among axon guidance cues, encompass secreted proteins (classes 2, 3, and V), single-membrane spanning proteins (classes 1, 4, 5, and 6), and GPI-anchored proteins (class 7) (13). Studies on Sema3A unveiled its association with the extracellular matrix and enhanced growth cone collapsing activity upon heparin treatment (14). Furthermore, Sema3A has been shown to associate with the densely organized extracellular matrix structures termed perineuronal nets via interaction with chondroitin sulfate proteoglycans (15). The subsequent biochemical study revealed that the binding of Sema3A to chondroitin sulfates is specific to their sulfation pattern (16). Moreover, Sema5A has been demonstrated to bind GAG chains of proteoglycans during the formation of the fasciculus retroflexus, a diencephalic axon tract associated with limbic function. Intriguingly, Sema5A exhibited a dual role, exerting both attractive and repulsive effects contingent upon GAG chains. Specifically, heparan sulfate chains mediated Sema5A attraction, while chondroitin sulfate chains mediated repulsion (17).

Although the binding of semaphorins to proteoglycan GAG chains has been explored in vitro and in vivo, the mechanistic view and structural data remain elusive. In this study, we delve into the intricate binding properties of secreted Drosophila semaphorins to the GAG chains of proteoglycans. Our findings highlight that secreted semaphorins, particularly Sema2a and Sema2b, exhibit selective binding to GAG chains within proteoglycans, displaying distinct affinities based on sulfation degrees. Structural insights into Sema2b elucidate multiple GAG-binding sites, notably a high-affinity site at the C-terminal tail, emphasized by a conserved lysine-rich helical arrangement across secreted semaphorins. Moreover, our experiments suggest that these semaphorins preferentially associate with the more sulfated clusters of heparan sulfate on cell surfaces, establishing an interaction preference within the GAG chains. Last, in vivo experiments with Sema2b variants underscore the pivotal role of the Sema2b C-terminal tail in guiding olfactory receptor neuron (ORN) axons.

Results

Secreted Semaphorins Bind GAG Chains of Proteoglycans on the Cell Surface.

We investigated the GAG-binding properties of secreted semaphorins using surface plasmon resonance (SPR) binding experiments. We produced Sema2a and Sema2b in HEK293S cells and analyzed binding to heparin, heparan sulfate, and chondroitin sulfate immobilized on the chip. We found that both Sema2a and Sema2b exhibit comparable micromolar and submicromolar binding affinities to GAG chains (Fig. 1 A and B). Notably, we observed the strongest binding between Sema2a and Sema2b with heparin, the most sulfated GAG. The binding affinity between Sema2a and Sema2b with heparan sulfate, the second-most sulfated GAG, was substantially weaker. The binding of Sema2a and Sema2b to the less sulfated GAG, chondroitin sulfate, was the weakest. This suggests a correlation between the degree of sulfation of GAG chains and the binding affinities of Sema2a and Sema2b, underscoring the significance of electrostatic interactions between the negatively charged sulfate groups of GAG chains and the positively charged residues in Sema2a and Sema2b. Although we were not able to determine precisely the association (k_on) and dissociation (k_off) constants, sensorgrams of the semaphorin–GAG interactions indicate a relatively fast association and a relatively slow dissociation rate, suggesting that the semaphorin–GAG complex is relatively stable over time (SI Appendix, Fig. S1).

Fig. 1.

Secreted semaphorins bind GAG chains of proteoglycans on the cell surface. (A and B) Binding isotherms showing the interaction of Sema2a (A) or Sema2b (B) with immobilized heparin (orange), heparan sulfate (blue), and chondroitin sulfate (black). The apparent K_D values are displayed at the top. Error bars represent SD of two or three biological replicates. (C) Binding isotherms showing the interaction of Sema2b with immobilized Dlp^core (blue) and Dlp^ΔGPI (orange). The apparent K_D values are displayed at the top. Error bars represent SD of two biological replicates. (D) Representative confocal microscopy images of CHO-K1 or CHO-PgsA745 cells demonstrating the binding of Sema2a^EmGFP on the cell surface of CHO-K1 and the colocalization of Sema2a^EmGFP with the anti-NS antibody, which recognizes more sulfated NS clusters within the heparan sulfate chains. (Scale bar, 10 μm.) (E) Quantitative colocalization analysis of Sema2a with more sulfated NS clusters of heparan sulfates (NS-HS). The plot shows Pearson’s correlation coefficient (PCC) values in black for the colocalization of Sema2a^EmGFP with the anti-NS antibody. Manders’ colocalization coefficient values are shown in red for Sema2a (M_Sema2a) and in green for NS-HS (M_NS-HS). Each point represents a single-cell measurement. N = 55 cells from two independent experiments. Statistical analysis was performed using the Wilcoxon test. (F) Representative confocal microscopy images of CHO-K1 or CHO-PgsA745 cells demonstrating the lower level of colocalization between Sema2a^EmGFP and anti-NA antibody, which recognizes the NA clusters within the heparan sulfate chains. (Scale bar, 10 μm.) (G) Quantitative colocalization analysis of Sema2a with less sulfated NA clusters of heparan sulfates (NA-HS). The plot shows PCC values in black for the colocalization of Sema2a^EmGFP with the anti-NA antibody. Manders’ colocalization coefficient values are shown in red for Sema2a (M_Sema2a) and in green for NA-HS (M_NA-HS). Each point represents a single-cell measurement. N = 71 cells from two independent experiments. Statistical analysis was performed using the Wilcoxon test.

To examine whether secreted semaphorins bind GAG chains of proteoglycans, we performed SPR binding experiments between Sema2b and the proteoglycan, Dally-like protein (Dlp), which has been recently shown to be crucial in the extracellular distribution of Wnt (18). We purified a secreted form of Dlp that lacks the glycophosphatidylinositol (GPI) membrane anchor (Dlp^ΔGPI) and immobilized it on the surface of the SPR chip. We obtained the best fit for the SPR data using a two-site binding model, indicating that Dlp binds Sema2b via a high-affinity and a lower-affinity binding site (Fig. 1C). Furthermore, we have observed no binding between Sema2b and Dlp core protein lacking the GPI anchor and heparan sulfate chains (Dlp^core), which we used as a control (Fig. 1C). These binding experiments demonstrate the ability of secreted semaphorins to specifically interact with the GAG chains of proteoglycans.

To further explore the binding of secreted semaphorins to endogenous membrane-associated GAG chains on the cell surface, we performed a cell-based binding assay. A previous study indicated that sulfation within a single heparan sulfate chain is not uniformly spread but typically comprises two distinct clusters, the NA cluster rich in GlcA and GlcNAc, and the more sulfated NS cluster containing IdoA and GlcNS derivates (19). To investigate whether the binding site for semaphorins is formed by the NA or NS clusters, we carried out a colocalization analysis by using two types of antibodies against heparan sulfate, which recognize either the NS or NA clusters. For microscopy, we used CHO-K1 cells which express high levels of heparan sulfate proteoglycans on the cell surface (20, 21). As a control, we used CHO-PgsA745 cells, which do not produce GAG chains due to a defect in xylosyltransferase (20, 21). We observed that both antibodies exhibited similar punctate staining patterns on the surface of CHO-K1 cells (Fig. 1 D and F). To quantify the degree of 3D colocalization between Sema2a and the NA or NS clusters, we assessed their correlation and co-occurrence by calculating Pearson’s correlation coefficient (PCC) and Manders’ colocalization coefficients (M). The colocalization analysis with Sema2a fused with EmGFP (Sema2a^EmGFP) revealed that Sema2a^EmGFP specifically associates with the cell surface of CHO-K1 and colocalizes predominantly with the anti-NS, rather than the anti-NA antibody (Fig. 1 E–G and SI Appendix, Fig. S2A). We also confirmed these findings using STED microscopy, which provides better lateral resolution than confocal microscopy, typically below 30 nm. The quantitative analysis performed with STED microscopy corroborates the findings from confocal microscopy (SI Appendix, Fig. S2 B–E), although the PCC and Manders’ colocalization coefficients obtained using STED microscopy are slightly lower due to the smaller effective point spread function of STED microscopy. Taken together, these findings suggest that the secreted semaphorins can indeed bind endogenous GAG chains on the cell surface, with a preference for the sulfo-rich NS clusters.

Sema2b Contains Multiple Binding Sites for GAGs.

To identify the GAG-binding sites on secreted semaphorins, we attempted to crystallize Sema2a and Sema2b in complex with sucrose octasulfate (SOS), a synthetic heparin-mimic. Unfortunately, the crystallization trials using full-length constructs failed; however, we successfully crystallized SOS with truncated Sema2b, in which a 66-residue long C-terminal tail was omitted from the construct (Sema2b^ΔC). We determined the crystal structure of Sema2b^ΔC in complex with SOS to a resolution of 2.79 Å (Fig. 2A and SI Appendix, Table S1). The overall architecture of Sema2b^ΔC closely resembles the previously reported unliganded Sema2b crystal structure (22), with a Cα rmsd of 0.73 Å over 612 Cα equivalent residues, indicating no significant conformational changes upon SOS binding (SI Appendix, Fig. S3A). Each chain of Sema2b^ΔC interacts with two SOS molecules, which bound to positively charged patches on the surface of Sema2b^ΔC (SI Appendix, Fig. S3B). Intriguingly, both GAG-binding sites are out of the canonical binding site for Plexin B receptors (23), suggesting that GAG binding does not directly affect semaphorin–plexin interaction. Additionally, the positions of both GAG-binding sites in Sema2b^ΔC are different from the binding site of the coreceptor neuropilin in the previously reported Sema3A-PlexinA4-Nrp1 ternary complex (24) and also from the cis binding site of the PlexA in the previously reported Sema1b–PlexA complex (25).

Fig. 2.

GAG-binding sites of Sema2b. (A) Ribbon representation of Sema2b^ΔC in complex with SOS, the sema domain is shown in orange, the PSI domain in blue, the Ig-like domain in red, and SOS is shown in stick representation. A 66-residue long C-terminal tail, which was omitted from the construct used for crystallization, is represented by a dashed gray line. The C-terminal central helix predicted by AlphaFold is shown in green. Schematic domain organization of Sema2b is shown below the ribbon representation, highlighting the locations of SOS binding sites with yellow stars. The high-affinity binding site is indicated by a violet star. (B and C) Close-up view of the SOS binding site I (B) and the SOS binding site II (C). SOS is overlaid with the composite omit 2mF_o-DF_c map displayed in gray at a contour level of 1.5 σ. (D and E) Binding isotherms showing the interaction of Sema2b^R559A (D) or Sema2b^ΔC (E) with immobilized heparin (orange), heparan sulfate (blue), and chondroitin sulfate (black). The apparent K_D values are displayed at the top. (F and G) Ribbon representation of the AlphaFold model for the basic C-terminal tail, with arginines and lysines shown in stick representation. (G) is rotated 90° around the y axis relative to (F).

The first SOS binding site is formed by the PSI domain and the neighboring sema domain with a total buried surface area of 735 Ų. The interaction interface primarily involves hydrogen bonds and electrostatic interactions, predominantly formed by K61, H63, R535, and R559 (Fig. 2B and SI Appendix, Fig. S4A). The second SOS binding site is localized to the Ig-like domain, burying a total surface area of 531 Ų (Fig. 2C and SI Appendix, Fig. S4B). The most extensive interactions with SOS in the second binding site are made by residues R652 and K585.

To confirm the existence of the first SOS binding site as observed in the crystal structure, we aimed to substitute the key interaction residues and analyze their impact on GAG binding. Unfortunately, the substitution of positively charged residues with aspartic acid led to protein misfolding. However, we did observe correct folding and expression levels similar to wild-type (WT) Sema2b when we substituted arginine 559 with alanine (R559A). Subsequent SPR binding experiments with the Sema2b^R559A mutant revealed about a 30-fold decrease in binding affinities to heparin and heparan sulfate, with no detectable K_D for chondroitin sulfate (Fig. 2D). We could not employ the same approach to confirm or reject the existence of the second SOS binding site, as substitutions of arginine 552 and 585 with aspartic acid or alanine also resulted in protein misfolding. Therefore, the evidence for the second SOS binding site remains inconclusive.

We further focused on the GAG-binding properties of the C-terminal tail of Sema2b, which was omitted from the Sema2b^ΔC construct used for crystallization. The SPR binding experiments with Sema2b^ΔC indicated a 232-fold decrease in binding affinity to heparin, a 128-fold decrease to heparan sulfate, and more than a 16-fold decrease to chondroitin sulfate when compared to WT Sema2b (Fig. 2E). This observation suggests that the C-terminal tail serves as an additional binding site for GAGs. Alphafold (26) prediction revealed that the C-terminal tail of Sema2b is predominantly disordered with one central helix. Although the disordered region shows a relatively low per-residue confidence score, the central helix is predicted confidently with pLDDT > 70, suggesting correct backbone prediction (SI Appendix, Fig. S4C). The C-terminal tail contains one arginine and nine lysine residues, with seven of these lysines located in the central helix. The lysine side chains are arranged in a single plane along the central helix (Fig. 2 F and G), which could facilitate binding to negatively charged sulfate groups distributed along the linear GAG chains. Interestingly, the central helix with its positively charged residues in the C-terminal tail appears to be conserved and can be observed in the AlphaFold predictions across all members of the human class 3 semaphorins (SI Appendix, Fig. S5), the orthologs to class 2 semaphorins in Drosophila. Our observation is further supported by previous studies, which have demonstrated helical motifs in the C-terminal tail of Sema3A and Sema3F using circular dichroism analyses (27, 28).

In summary, the crystal structure of Sema2b^ΔC, along with the SPR binding experiments, demonstrates that Sema2b possesses multiple binding sites for GAGs. The high-affinity binding site is located at the C-terminal tail, while the low-affinity binding site is jointly occupied by the PSI domain and the adjacent sema domain. The existence of an additional binding site, the second SOS binding site, within the Ig-like domain could not be confirmed or refuted using the SPR binding experiments. Importantly, all GAG-binding sites are distinct from the canonical Plexin B binding site, indicating that GAG binding on Sema2b does not interfere with Plexin B binding.

GAG Chains Establish Concentration Gradients of Secreted Semaphorins on Cell Surfaces In Vitro.

We further investigated whether GAG chains of proteoglycans play a role in establishing concentration gradients of secreted semaphorins via the mechanism of restricted diffusion. This model has been previously reported for a number of morphogens, such as Wg, Hh, and Dpp (reviewed in ref. 29). We hypothesized that the interplay between secreted semaphorins and GAG chains influences their distribution, restricting their diffusion and concentrating them along cell membranes, particularly closer to their source of production.

To explore whether GAG chains of proteoglycans mediate the formation of a concentration gradient of secreted semaphorins on the cell surface, we conducted experiments in microfluidic channels (Fig. 3A). CHO-K1 cells, which we previously showed to bind the GAG chains of proteoglycans, were seeded on the microfluidic channel to form a confluent cell monolayer. Following thorough washing of the cell monolayer, we introduced Sema2a^EmGFP into one of the reservoirs, allowing it to spread along the channel for 30 min. After incubation, we washed the cells and measured fluorescence intensity along the cell surface using a confocal microscope. Imaging revealed a linear gradient distribution of Sema2a^EmGFP along the channel (Fig. 3B), with the highest intensity proximal to the point of injection. To characterize the Sema2a distribution, we conducted linear regression analysis to determine the line of best fit, yielding slope values. For the Sema2a^EmGFP distribution, we identified a linear decrease in intensity with an average slope of −3.69 ± 0.04 mm⁻¹ (Fig. 3C).

Fig. 3.

Role of GAG chains in establishing semaphorin gradients on cell surfaces in vitro. (A) Schematic representation of the semaphorin gradient experiment within a microfluidic channel. Fluorescent semaphorins were introduced into a reservoir and allowed to diffuse over a cell monolayer in the channel. After washing, a confocal microscope was used to measure the resulting concentration gradient of semaphorins along the cell monolayer across the channel. (B) Representative confocal images of the cell monolayer along the channel, demonstrating the gradient distribution of semaphorins with the highest intensity proximal to the point of injection (Left). (Scale bar, 1 mm.) (C–F) A plot of normalized fluorescent intensity line traces (orange) along the channel for Sema2a^EmGFP (C), Alexa Fluor 488-conjugated Sema2b (D), Sema2b^R559A (E), and Sema2b^ΔC (F). Additionally, the plot includes a band indicating the 95% CI (gray). The following P values for slope comparison were determined using one-way ANOVA: Alexa-Sema2b versus Alexa-Sema2b^ΔC, ***P = 0.0004; Alexa-Sema2b versus Alexa-Sema2b^R559A, *P = 0.0334; and Alexa-Sema2b^ΔC versus Alexa-Sema2b^R559A, ***P = 0.0002.

To investigate the involvement of GAG chains in gradient formation, we repeated the experiment using Sema2b and its mutant variants, each exhibiting distinct affinities to GAGs. Instead of fusion with EmGFP, we conjugated Sema2b, Sema2b^ΔC, and Sema2b^R559A with the fluorescent dye, Alexa Fluor 488. Sema2b^R559A showed a similar gradient distribution to that observed for WT Sema2b (Fig. 3 D and E), with an average slope of −1.75 ± 0.04 and −1.59 ± 0.04 mm⁻¹, respectively. However, the gradient distribution of Sema2b^ΔC demonstrated a significantly less steep (with an average slope of −0.84 ± 0.03 mm⁻¹) (Fig. 3F). Additionally, variations in the Y-intercept were also observed for the Sema2b variants (Fig. 3 D–F), likely caused by varying degrees of Alexa Fluor labeling or alterations in GAG occupancy by the Sema2b variants. Taken together, these findings suggest the involvement of GAG chains, specifically their interaction with the C-terminal tail of Sema2b, in the formation of the observed semaphorin gradient.

Sema2b with Disrupted High-Affinity GAG Binding Loses Its Activity in Guiding ORN Axons In Vivo.

Are there any in vivo functional effects of the binding between GAGs and semaphorins? To answer this question, we tested the functions of different Sema2b GAG-binding sites in the ORN. During the development of the fly olfactory circuit, pioneering antennal ORN axons take either ventromedial or dorsolateral trajectory to circumnavigate the antennal lobe. Sema2b is highly expressed in ORN axons that take the ventromedial trajectory and is largely absent from those that take the dorsolateral trajectory and instruct the trajectory choice (Fig. 4E) (30). As reported in the previous study, we observed that ORN axons, labeled by pan-ORN pebbled-GAL4 (31) driver, segregated roughly equally into dorsolateral and ventromedial bundles at 24 h after puparium formation (APF) in WT controls (Fig. 4 A and F).

Fig. 4.

In vivo analysis of GAG-binding variants of Sema2b in axon trajectory choice of ORN. (A) At 24 h APF, pebbled+ ORN axons choose either a ventromedial or dorsolateral trajectory to circumnavigate the antennal lobe. All ORN axons are labeled by pebbled-GAL4 driving UAS-mCD8-GFP (green). Neuropil staining by the N-cadherin (NCad) antibody indicates the antennal lobe (blue). D, dorsal; L, lateral. Midline is to the left. (Scale bar, 10 μm for all images of the antennal lobe.) (B–D) Overexpression of WT Sema2b (B), and Sema2b^R559A (D) in pebbled+ ORNs causes more axons to choose the ventromedial trajectory comparing to control. Overexpression of Sema2b^ΔC (C) in pebbled+ ORNs does not cause this bias. (E) Summary of ORN axon trajectory choice at 24 h APF with different Sema2b levels. (Left) In WT flies, Sema2b+ axons (dark magenta) and Sema2b– axons (light magenta) choose the ventromedial and dorsolateral trajectory, respectively. The antennal lobe is shown in blue. (Right) Sema2b overexpression in pebbled+ ORNs biases most ORN axons to take the ventromedial trajectory. (F) Quantification of dorsolateral/ventromedial axon bundle ratio for different Sema2b variants. Boxes indicate geometric mean and 25 to 75% range. Geometric means/sample sizes are as follows: Control 0.84/12; ORN > Sema2b 0.02/18; ORN > Sema2b^ΔC 1.04/15; ORN > Sema2b^R559A 0.01/20. ***P < 0.001; n.s., not significant, Kruskal–Wallis test with Bonferroni’s multiple comparison.

To examine the in vivo functions of the Sema2b variants with different GAG-binding properties, we constructed transgenic flies that integrate UAS-Sema2b-HA, UAS-Sema2b^ΔC-HA, or UAS-Sema2b^R559A -HA transgenes at the same chromosome locus. Pan-ORN overexpression of WT Sema2b redirected nearly all dorsolateral ORN axons to take the ventromedial trajectory, with a mean dorsolateral/ventromedial ratio around 0.02 (Fig. 4 B and F). This phenotype is similar to but stronger than previously reported (30), likely because the UAS-Sema2b flies used in this study expressed Sema2b at a higher level. By contrast, when UAS-Sema2b^ΔC was overexpressed in ORN axons, we did not observe ORN axons biased toward the ventromedial trajectory. Instead, ORN axons took both dorsolateral and ventromedial trajectories about equally (mean dorsolateral/ventromedial ratio = 1.04) (Fig. 4 C and F). To rule out the possibility that Sema2b^ΔC is not properly expressed, we stained for the HA-tag and found Sema2b^ΔC was expressed in a comparable level as WT Sema2b (SI Appendix, Fig. S6). These results suggest the deletion of Sema2b’s C-terminal tail disrupts Sema2b’s functions in ORN trajectory specification. In contrast, when Sema2b^R559A was overexpressed in pebbled+ ORN axons, we observed a strong bias of ORN axons toward ventromedial trajectory (mean dorsolateral/ventromedial ratio = 0.01) (Fig. 4 D and F), similar to the WT Sema2b overexpression group. Taken together, the in vivo experiments support that the Sema-2b’s high-affinity binding site at the C-terminal tail is crucial for its function in ORN trajectory specification, while the low-affinity binding site, formed by the PSI domain and the adjacent sema domain, is not as essential. However, we cannot exclude the possibility that the C-terminal deletion disrupted a critical function of Sema2b that is independent of GAG binding.

Discussion

Proteoglycans play crucial roles in cell signaling by acting as key components of the extracellular matrix and modulating various signaling pathways. We found that proteoglycans bind secreted semaphorins and act as repositories of these cues. We propose the semaphorin bridge model, in which secreted semaphorins are tethered to the cell surface through interactions with GAG chains of proteoglycans and presented to their cognate receptors on passing axons. According to this model, semaphorins bind to GAG chains on one cell while simultaneously binding to their cognate receptors on another cell, effectively bridging the two cell surfaces. A similar mechanism has been previously suggested for leukocyte adhesion, where GAG-bound chemokines on endothelial cells are presented to chemokine receptors on passing leukocytes (32).

Two prerequisites are essential for the semaphorin bridge model: 1) a relatively high stability of the semaphorin–GAG complex and 2) the ability of secreted semaphorins to simultaneously bind GAG chains and their cognate plexin receptors.

The first prerequisite is supported by our cellular and biophysical assays. We have demonstrated that Sema2a and Sema2b can directly bind to the heparan sulfate chains of Dlp proteoglycan. The binding affinities between GAG chains and these semaphorins fall within the micromolar and submicromolar range. Notably, the highest binding was observed with heparin, with apparent K_D values of 0.4 μm for Sema2a and 0.5 μm for Sema2b. Our previously determined binding affinities between Plexin B and Sema2a (K_D of 25 nM) or Sema2b (K_D of 3 nM) (22) suggest that the binding of Sema2a or Sema2b to their Plexin B receptor is stronger than their binding to GAG chains. Furthermore, our SPR experiments revealed relatively slow dissociation rates of the semaphorin–GAG complexes, suggesting that the complexes are relatively stable over time. Previous studies have reported that a single heparan sulfate chain is not uniformly spread but rather consists of two distinct clusters: the less sulfated NA cluster and the more sulfated NS cluster (19). Our colocalization experiments suggest a pronounced preference of secreted semaphorins for binding specifically to the sulfated-rich NS clusters of heparan sulfate.

The second prerequisite for the semaphorin bridge model, the ability to simultaneously bind GAG chains and plexin receptors, is supported by our structural analyses, which revealed multiple GAG-binding sites on Sema2b. Importantly, all of these binding sites are located outside the canonical Plexin B binding site (23), indicating that Sema2b can simultaneously accommodate both receptors and GAG chains without any apparent steric clashes. Furthermore, our binding assays imply that the interaction with GAG chains is primarily mediated by the Sema2b terminal tail with its positively charged residues. Our analyses suggest that this binding mode is conserved among all secreted human class 3 semaphorins, which is also supported by previous structural studies between the C-terminal tail of human Sema3A and GAG chains (27). We also demonstrated that disrupting this binding site resulted in the loss of Sema2b’s ability to guide ORN axons in vivo. In contrast, disrupting the low-affinity binding site did not significantly affect ORN axon guidance in vivo; however, it may play a role in other aspects of semaphorin biology beyond ORN axon navigation.

We propose that a central role of GAG chains in the semaphorin bridge model is to concentrate secreted semaphorins in precise anatomical positions near their sites of production and protect them from diffusing away. As a result, the secreted semaphorins can act as short-range guidance cues mediating a haptotactic response rather than the typical chemotactic response observed for cues in solution (Fig. 5A). This substantially expands a long-range guidance paradigm by which diffusible cues travel long distances to guide distant axons, an idea that Ramon y Cajal proposed over a century ago (33).

Fig. 5.

Semaphorin bridge model. (A) Proposed role of GAG chains in the semaphorin–plexin signaling: GAG chains play a central role by concentrating secreted semaphorins near their production sites, preventing diffusion and facilitating binding to their cognate plexin receptors. This mechanism enables secreted semaphorins to function effectively as short-range guidance cues. (B) Potential GAG chain’s role in semaphorin stability: GAG-bound semaphorins are protected from cleavage by proteases like furin as the GAG chains can occlude the furin cleavage site on semaphorins. (C) A model of concentration gradient formation through restricted diffusion. Secreted semaphorins diffuse away from their source along the cell layer and are restricted by interactions with GAG chains.

The short-range guidance by secreted semaphorins is supported by several previous studies. In particular, pioneering research on chicken Sema3A indicated that Sema3A is rather directly attached to chick brain membranes and not released into solution (34, 35). Also, recent in vivo study on the development of embryonic ventral nerve cord and ORN demonstrated that Sema2b guides rather locally at the membrane, and secretion is not obligatory for Sema2b function (30, 36). Indeed, in this system, we have shown here that C-terminal deletion of Sema2b with disrupted high-affinity GAG binding did not exhibit any biological activity in guiding ORN axon trajectory choice in an overexpression assay (Fig. 4), consistent with GAG binding playing a crucial role in axon guidance in vivo. However, this experiment by itself does not rule out the possibility that the C-terminal deletion disrupted an essential function of Sema2b unrelated to GAG binding.

Interestingly, the bridge model may not be limited to semaphorins alone. Secreted netrins, another group of guidance cues originally proposed to act over long distances (37), may also be subject to similar mechanisms. Indeed, an increasing body of evidence suggests that netrins do not travel far from their site of production (38–40); instead, they remain membrane-bound (41), guiding cells through haptotaxis (42, 43).

We speculate that GAG chains also play a role as stabilizing agents on the cell surface (Fig. 5B). In particular, several vertebrate class 3 semaphorins possess one or multiple furin cleavage sites and previous studies have demonstrated their susceptibility to cleavage by furin-like endoproteases, resulting in a significant impact on their function (44–46). We found that, in some, but not all, human class 3 semaphorins, the furin cleavage site is positioned precisely at the AlphaFold-predicted helical motifs of the C-terminal tail (SI Appendix, Fig. S5), suggesting that the positively charged furin cleavage site (RX(K/R)R) may also be involved in GAG binding. Thus, when semaphorins are bound to GAGs, they may be shielded from cleavage by furin as the GAG chains occlude the furin cleavage site. This idea is supported by recent work demonstrating that membrane-bound Sema3A remains uncleaved, unlike its soluble counterpart, which is secreted into the media during Sema3A large-scale production in HEK cells (47). Thus, the presence of GAG chains on the cell surface might help protect the semaphorins from furin-mediated cleavage, thereby stabilizing them in their intact, active form.

The last, but certainly not least, crucial role of GAGs in the semaphorin binding might be their capacity to shape concentration gradients (Fig. 5C). Concentration gradients of axon guidance cues have been identified for various secreted guidance cues, including Sema2a (48), Sema2b (48), Sema3A (49), or netrins (37, 50), as well as for transmembrane cues such as Sema1a (51) or Ephrin-A (52, 53). The presence of gradients of axon guidance cues plays a fundamental role in axonal pathfinding by providing spatial information to growing axons. How is the concentration gradient of guidance cues established? Gradient formation for transmembrane cues seems to be under transcriptional regulation. For instance, in the chick tectum, a continuous gradient of transmembrane ephrin-A has been shown to be transcriptionally regulated via the gradient expression of the transcription factor Engrailed2 (54). This, in turn, is controlled by FGF (55), a family of GAG-binding proteins (56). However, the precise molecular mechanisms underlying the establishment of concentration gradients for secreted semaphorins remain poorly understood. In this study, we have demonstrated the establishment of a concentration gradient of Sema2a and Sema2b in vitro on cell monolayers. To explain the observed in vitro gradient formation, we propose a theoretical restricted diffusion mechanism, where secreted semaphorins diffuse away from their source along the cell layer but are restricted by interactions with GAG chains. As a result, the highest concentration of secreted semaphorins is found near the producing cells and their neighboring cells, while the concentration gradually decreases as semaphorins disperse further away. Notably, similar gradient formation models based on the restricted diffusion have been suggested for various morphogens, such as Wg, Hh, and Dpp (reviewed in ref. 29).

The semaphorin bridge model sheds light on the role and involvement of proteoglycans in semaphorin signaling. While supported by numerous experiments, further rigorous investigations, particularly involving mammalian semaphorins, will be required to broaden its scope and reliability. Given that secreted semaphorins serve as inhibitors, impeding axon regrowth following spinal cord injuries, the model could have important implications in axon regeneration and repair mechanisms.

Material and Methods

Protein Production.

Constructs encoding Drosophila melanogaster Sema2a; Sema2b; Sema2b^ΔC; Dlp^ΔGPI and Dlp^core (residues 26D–724V; 34D–737V; 35Y–679Q; 53P–723G; and 53P–723G S625A, S629A, S631A, S643A, S686A, respectively) were cloned into the pHR-CMV-TetO2-IRES-EmGFP vector in-frame with a C-terminal 3C-Avi-His6 tag (57). A construct encoding Sema2a^EmGFP was built by placing EmGFP at the C terminus of Sema2a. For protein production, stable cell lines were generated using lentiviral transduction. Apart from Sema2a^EmGFP, all semaphorins were produced in HEK293S cells and secreted into the FreeStyle 293 expression medium (ThermoFisher). Sema2a^EmGFP, Dlp^ΔGPI, and Dlp^core were produced in Expi293F cells and secreted into Expi293 expression medium (ThermoFisher). Suspension cells were grown at 37 °C, 110 rpm, and 8% CO₂.

For purification, the conditioned medium was collected and subjected to immobilized metal-affinity chromatography using a HisTrap FF column (Cytiva) and further purified using size-exclusion chromatography with a Superdex 200 16/60 column (Cytiva). Additionally, conditioned media based on the Expi293 expression medium were dialyzed before loading onto the HisTrap FF column. Site-directed mutagenesis was carried out using overlap-extension PCR, and the resulting PCR products were cloned into the pHR-CMV-TetO2 vector as described above. The mutant proteins were secreted at levels similar to those of the WT proteins.

Fluorescent Labeling of Semaphorins Using Alexa Fluor 488 Succinimidyl Ester.

Sema2b, Sema2b^R558A, and Sema2b^ΔC were labeled with Alexa Fluor 488 succinimidyl ester (ThermoFisher) using a 500-fold excess of the dye in PBS. The initial concentration of semaphorins was 7.5 μm. The labeling reaction took place over 2 h at room temperature. Any unbound dye was removed via thorough dialysis. The protein concentration was assessed using formulas provided in the manufacturer’s manual.

SPR Equilibrium Binding Experiments.

For SPR experiments, heparin, heparan sulfate, and chondroitin sulfate (Toronto Research Chemicals) were biotinylated following the previously described method (58) and immobilized onto SA Biacore sensor chips (Cytiva). The Dlp constructs were biotinylated in vivo in HEK293T cells through cotransfection with pDisplay-BirA-ER (59) (pHR-CMV-TetO2:pDisplay ratio was 3:1). The cells were supplemented with 100 μm D-biotin and maintained at 37 °C and 5% CO₂. Three days posttransfection, the conditioned medium was collected and dialyzed against 15 mM Tris (pH 7.5), 150 mM NaCl, and 0.05% (v/v) Tween 20. The biotinylated Dlp constructs were then immobilized onto SA Biacore sensor chips (Cytiva). SPR experiments were performed using a Biacore S200 instrument (Cytiva) in a buffer containing 15 mM Tris (pH 7.5), 150 mM NaCl, and 0.05% (v/v) Tween 20 at 25 °C. The signal from the experimental flow cells was corrected by subtraction of the nearest blank injection and the reference signal from a blank flow cell. Surface regeneration was performed twice per run using a buffer containing 0.1 M Tris (pH 8.0), 1.0 M NaCl, and 1% CHAPS. All data were analyzed with Biacore Insight Evaluation software (Cytiva). The SPR experiments were performed with two or three biological replicates.

Protein Crystallization.

Prior to crystallization, Sema2b^ΔC was supplemented with 10 mM SOS obtained from Toronto Research Chemicals. Crystals were grown using the sitting drop vapor diffusion method at 20 °C. The Sema2b^ΔC in complex with SOS crystallized in 0.1 M MES (pH 6.0) and 5% (w/v) PEG 6000. Crystals were cryoprotected by soaking in a reservoir solution supplemented with 25% (v/v) glycerol and then flash-cooled in liquid nitrogen.

Data Collection and Structure Determination.

Diffraction data were collected at 100 K using the I03 beamline at Diamond Light Source. The data were indexed, integrated, and scaled using automated autoPROC software (60). Anisotropy correction was performed using the automated STARANISO as a part of autoProc software. The anisotropy correction yielded an ellipsoidal resolution boundary with limits of 2.73, 2.73, and 4.96 Å along the a*, b*, and c* axes, respectively. The structure of Sema2b^ΔC in complex with SOS was determined by molecular replacement in Phaser (61) with the Sema2b^ΔC structure (22) (PDB 6QP8) as a search model. This initial molecular replacement solution was further completed by several cycles of manual rebuilding in COOT (62) and refinement in Phenix (63). Refinement statistics are given in SI Appendix, Table S1. Electrostatics potentials were generated using APBS (64), structural alignment was performed using PDBeFold (65), a composite omit map was calculated in Phenix (66), and schematic diagrams of protein–ligand interactions were produced using LigPlot (67). Figures were created using PyMOL (Schrodinger, LLC) and CorelDraw (Corel Corporation).

Immunofluorescence.

Adherent Chinese hamster ovary (CHO) cells, including wild-type (K1) and xylosyltransferase-deficient mutant (PgsA-745) cells, were seeded on 18 mm high-precision coverslips at a density of 15,000 or 25,000 cells per coverslip in DMEM supplemented with 2% FBS. After 24 h of growth, the medium was aspirated, and the cells were incubated with Sema2a^EmGFP in DMEM for 1 h at 37 °C followed by 30 min on ice. Subsequently, the cells were washed with DMEM and fixed with 4% paraformaldehyde in PBS for 20 min. After five 5-min washes with PBS, the cells were blocked with 4% BSA in PBS for 1 h. Next, the cells were incubated with 10 µg/mL anti-GFP antibody (ab290, RPIgG, Abcam) and either anti-NA heparan sulfate antibody, dilution 1:50 (a gift from Prof Masanori Taira and Dr Yusuke Mii, The University of Tokyo) or 1 µg/mL anti-NS heparan sulfate antibody (370-255-5, AMSBIO, MMIgM, F58-10E4) in PBS with 0.3% BSA for 1 h. Following extensive washes, the cells were incubated with 3.3 µg/mL goat anti-rabbit IgG antibody conjugated with Abberior STAR 635P and 5 µg/mL goat anti-mouse IgM antibody conjugated with Alexa Fluor 594 in PBS with 0.3% BSA for 1 h. After five 5-min washes with PBS, the cells were incubated with 1 µg/mL DAPI in PBS for 10 min, extensively washed with PBS and Milli-Q water, air-dried, and mounted in VectaShield.

Confocal Microscopy.

The images were acquired using the Nikon CSU-W1 spinning disk confocal microscope equipped with the NIS-Elements AR 5.41.01 software and a CFI Plan Apo VC (100×, NA 1.40) immersion oil objective. The z-scans were analyzed, and the 3D colocalization was estimated using the Imaris 9.8.2 software. The statistical significance of colocalization was determined using the Wilcoxon test in the Prism 9 software package (GraphPad).

STED Microscopy.

Images were acquired using an Abberior STED superresolution microscope, equipped with a 40 MHz pulsed STED 775 nm laser, and controlled by Imspector 16.3 software. A Nikon CFI Plan Apo Lambda 60x oil immersion objective (NA 1.40) was employed. Imaging was focused on the most spread part of the cell membrane. Images were deconvolved using Huygens Professional 19.10 software and colocalization analysis was performed using the Coloc2 plugin in ImageJ. Statistical significance of the colocalization was assessed using the Wilcoxon test in GraphPad Prism 10.1.2 software.

Visualizing Semaphorin Gradients In Vitro.

Adherent Chinese hamster ovary cells (CHO-K1) were seeded into µ-Slide VI^0.4 chambers (Ibidi) at a density of 80,000 cells per channel using phenol red-free DMEM supplemented with 10% FBS. After a 24-h incubation period, the medium was aspirated, and the cells were washed twice with phenol red-free DMEM. Subsequently, the cells were treated with either 100 µg of Sema2a^EmGFP or Alexa Fluor 488-conjugated Sema2b proteins in phenol red-free DMEM for 30 min at 37 °C followed by five washes with phenol red-free DMEM. The final semaphorin concentration used in the microfluidic channel (4 μm) was approximately 10 times higher than the apparent K_D values calculated in our SPR experiments. Following the 30-min incubation, the cells were washed five times with phenol red-free DMEM.

For gradient imaging, large-scale images of each channel (6 × 34 fields, with a 1% overlap) were acquired and stitched together using the spinning disk confocal microscope Nikon CSU-W1 equipped with NIS-Elements AR 5.41.01 software. Throughout imaging, the temperature of the media and the µ-Slide VI^0.4 chamber was maintained at 37 °C, while the CO₂ concentration surrounding the chamber was sustained at 5%. Imaging was performed using the CFI Plan Apo VC objective (20×, NA 0.75). To analyze the semaphorin gradient, images were processed using NIS-Elements AR 5.41.01 software and Visual Studio Code. We developed a Python script for Visual Studio Code for semiautomated analysis of the concentration gradient on the cell surface. Initially, background subtraction was performed using the rolling ball correction algorithm (the rolling ball radius was 40.62 μm), followed by automatic thresholding using the Otsu algorithm. Subsequently, the injection origin was designated as a single-pixel point, and the image was divided into radial zones at 10 μm intervals. Mean fluorescent intensity was determined for each zone and then normalized to the maximal determined value in the image. Data analysis involved simple linear regression (normalized fluorescent intensity versus distance from the injection origin) using GraphPad Prism 10.1.0. The experiments were conducted in duplicate.

Fly Genetics and Fly Brain Immunostaining.

Constructs encoding Sema2b, Sema2b^ΔC, and Sema2b^R559A were cloned into the pUAST-attB vector in-frame with a C-terminal 3xHA tag. All the UAS constructs were integrated into the 86Fb landing site. Transgenic flies carrying different Sema2b variants were crossed to the pebbled-GAL4, UAS-mCD8-GFP flies. Progenies were raised at 29 °C and their brains were dissected and stained at 24 h APF as previously reported (68). For primary antibodies, we used chicken antibody to GFP (1:1,000; Aves Labs) and rat antibody to N-cadherin (1:40, N-Ex #8; 1:40; Developmental Studies Hybridoma Bank (DSHB)). Secondary antibodies raised in donkey against chicken and rat antisera were used, conjugated to Alexa 488 or 647 (Jackson ImmunoResearch). Confocal images were collected with a Zeiss LSM 780 and processed with ImageJ.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Structure factors and coordinates have been deposited in the Protein Data Bank with identification number 8RMJ (69).