Regulation of ECM degradation and axon guidance by growth cone invadosomes

During development, growth cones guide neurites to their proper synaptic partners by functioning as highly specialized molecular sensors. Receptors on the surface of growth cones detect the molecular constitution of the surrounding extracellular matrix (ECM) and the secreted protein gradients in the environment (Kolodkin and Tessier-Lavigne, 2011). Receptor engagement elicits adhesion and intracellular signaling cascades that affect growth cone motility (Lowery and Van Vactor, 2009). Integrated cell signals converge onto the actomyosin and microtubule (MT) cytoskeletons, which power the force-generating machinery responsible for growth cone motility and axon guidance (Dent et al., 2011).

Growth cones contain many structural projections that rearrange in response to guidance cues. Growth cones are also compartmentalized into specialized regions where the cytoskeleton is differentially organized. Within the peripheral domain, F-actin is organized into bundles and a dense meshwork array that generate protrusive filopodia and lamellipodia. Towards the rear of the growth cone is the central domain where stable, bundled MTs splay from the axon shaft. Lastly, the transition zone exists at the interface between the peripheral and central domains (P- and C-domains, respectively), and is a region of strong actomyosin contraction and F-actin depolymerization (Medeiros et al., 2006).

In two-dimensional environments, growth cones use filopodia to help direct neurite extension upon planar substrata. However, growth cones in vivo must penetrate and transverse tissues, and detect secreted guidance cues from sources deep within the embryo. To direct guidance through three-dimensional (3D) environments, growth cones may form protrusions on their apical and basal surfaces, reminiscent of podosomes and invadopodia of immune and metastatic cancer cells, respectively (Linder et al., 2011; Murphy and Courtneidge, 2011). Podosomes and invadopodia, collectively referred to as invadosomes, are filopodia-like protrusions that are associated with ECM adhesion and remodeling. Although growth cones are known to secrete matrix metalloproteinases (MMPs) during axon guidance (McFarlane, 2003; Yong et al., 2001), they have not been shown to form invadosomes.

In this study, we show that growth cones form stable, F-actin-rich puncta that are reminiscent of invadosomes of invasive cells (Murphy and Courtneidge, 2011; Murphy et al., 2011). F-actin foci form primarily within the C-domain of growth cones and colocalize with numerous invadosomal markers, such as cortactin (Cttn), active Src and Tks5. Growth cones of varied neuron types across species form invadosomes on many different substrata, as well as in the intact spinal cord, suggesting that they are a fundamental subcellular specialization of developing neurons. Growth cone invadosomes are more stable than point contact (PC) adhesions (Gomez et al., 1996; Renaudin et al., 1999), yet colocalize with several actin polymerization and bundling proteins. Consistently, growth cone invadosomes are sites of rapid F-actin turnover with concentrated free F-actin barbed ends, as well as monomeric actin. Super resolution 3D microscopy shows that the F-actin within invadosomes is oriented as a vertical column that spans the C-domain in the z-plane and that F-actin can extend as orthogonal 3D-protrusions from the C-domain of growth cones. Moreover, in 3D collagen gels and in vivo, similar projections occur from the apical and basal surfaces of growth cones. Using fluorescent gelatin, we show that growth cone invadosomes target proteolytic activity in a Tks5-dependent manner. Lastly, inhibiting invadosome targeting of MMP activity with dominant-negative Tks5 prevents proper extension of motoneuron axons into the periphery. From these findings, we have identified novel structural components of growth cones that are analogous to the invadosomes of invasive cells and are crucial in guiding axons throughout tissues of the developing nervous system.

When we imaged actin in fixed growth cones by using total internal reflection fluorescence (TIRF) or confocal microscopy, we observed prominent F-actin foci within the C-domain (Fig. 1), which was predominantly devoid of bundled F-actin. To begin to understand the role of F-actin foci in growth cones, we set out to determine whether foci formation depends on the culture substratum. For this, we cultured neurons isolated from embryonic Xenopus spinal cord on ECM proteins, cell adhesion molecules and non-biological substrata. We found that foci formed within the C-domain of growth cones when cultured on all of the tested substrata (Fig. 1A), including bare glass (supplementary material Fig. S1A), suggesting that these specializations do not require specific receptor engagement. To determine whether F-actin foci formed in the growth cones of diverse neuron types and across species, we cultured a variety of primary neuron types, as well as human neurons derived from induced pluripotent stem cells (iPSCs). F-actin foci formed in growth cones of all cell types tested, including human iPSC-derived neurons (Fig. 1B; supplementary material Fig. S1B-F). Although the average number (Fig. 1C) and size of F-actin foci were similar to PC adhesions (Robles and Gomez, 2006), their location in the C-domain was distinct.

To visualize F-actin in live growth cones, we expressed either β-actin fused to green fluorescent protein (GFP-β-actin) or the F-actin binding probe Utrophin-CH conjugated to mCherry (mCh-UtrCH) in neurons. High-resolution live-cell imaging showed that both GFP-β-actin and mCh-UtrCH concentrated in F-actin bundles at the growth cone leading edge, as expected, but also in prominent actin foci within the C-domain (Fig. 1E; supplementary material Fig. S2A). It is noteworthy that GFP-tagged γ-actin also targeted to actin foci when expressed in growth cones (supplementary material Fig. S2B). Time-lapse imaging showed that actin foci were relatively long-lived, with an average lifetime of 7.76±0.32 min (Fig. 1E; supplementary material Movies 1, 2), which is notably longer than that of PC adhesions (Robles and Gomez, 2006; Santiago-Medina et al., 2013). In extreme cases, actin foci were observed to form at the leading edge and remained stable until they disassembled at the rear of the growth cone (supplementary material Fig. S2C). Although the majority of foci were stable, some were motile, resembling actin comets (supplementary material Fig. S2D).

Although the growth cone morphologies observed in vitro appeared similar to those seen in vivo, it is possible that foci are only a tissue culture artifact. Therefore, to determine whether F-actin foci also formed within growth cones in vivo, we labeled F-actin in commissural interneurons and motoneurons with mCh-UtrCH or GFP-UtrCH, and then fixed embryos for whole-mount spinal cord imaging (see Materials and Methods). By using confocal microscopy, we observed prominent F-actin foci within motoneuron growth cones on the ventral fascicle (Fig. 1F) and commissural interneurons at the ventral midline (Fig. 1G). We conclude that growth cones of distinct neuron populations and of different species can assemble F-actin-rich foci under a variety of environmental conditions.

Growth cones are known to form transient PC adhesions, predominantly at the leading edge and within filopodia (Robles and Gomez, 2006) but not typically within the C-domain, where foci form and remain stable (Fig. 1). Therefore, we hypothesized that foci are not classical PC adhesions but that they might still recruit proteins implicated in integrin-dependent adhesion. To test whether F-actin foci function as adhesion sites, we immunolabeled growth cones that had been cultured on laminin for β1 integrin. We found that β1 integrin receptors targeted to F-actin foci, as well as to PC adhesions and filopodia tips (Fig. 2A). Because F-actin foci contain integrin receptors, we tested for other adhesion proteins. Paxillin (PXN) and focal adhesion kinase (FAK) both serve essential roles in PC dynamics (Myers and Gomez, 2011; Woo and Gomez, 2006), so we tested whether these adhesion proteins colocalized with F-actin foci in fixed growth cones. Immunolabeling for PXN that was phosphorylated at residue Y118 (pY118-PXN) or FAK that was phosphorylated at residue Y397 (pY397-FAK), together with F-actin, showed that adhesion proteins colocalized with a minority of F-actin foci within the C-domain (Fig. 2B-D). Next, we used two-color live-cell imaging of mCh-UtrCH with GFP-tagged PXN (PXN-GFP) to assess the dynamic assembly of F-actin foci and PC adhesions (supplementary material Movie 3). PXN-containing PC adhesions assembled at the leading edge of growth cones (Fig. 2E), which were typically devoid of F-actin foci (Fig. 2F,G). However, a majority of foci (65.7%) did emerge from pre-existing PXN PCs within the C-domain. This finding suggests that a subset of PXN-containing PC adhesions could mature into longer-lived F-actin foci that persist within the C-domain as growth cones migrate forward. By contrast, F-actin foci did not colocalize with exocytic or endocytic vesicles (supplementary material Figs S3, S4). These results suggest that F-actin foci might function as previously unidentified adhesion sites, because they form in close proximity to the substratum, remain stable and colocalize with integrins and adhesion proteins.

With live-cell imaging, F-actin foci appeared as stable aggregates of F-actin, yet the dynamics of actin polymerization at foci were unclear. Moreover, some F-actin foci resembled actin comets, suggesting that rapid actin polymerization occurs at these structures. Because the process of actin polymerization requires the addition of monomeric actin (G-actin) to the barbed ends of F-actin, we tested whether actin monomers concentrated at foci. Similar to other sites of rapid actin polymerization in growth cones (Lee et al., 2013; Marsick et al., 2010), we found monomeric G-actin at F-actin foci (Fig. 3A; supplementary material Fig. S4). To determine whether foci contained uncapped barbed actin filaments for G-actin addition during actin polymerization, we used the F-actin barbed-end binding probe tetramethylrhodamine kabiramide-C (TMR-KabC) (Petchprayoon et al., 2005; Tanaka et al., 2003). TMR-KabC is a small, cell-permeable molecule that can bind to the barbed ends of F-actin and track the retrograde flow of actin filaments (Keren et al., 2008; Santiago-Medina et al., 2013,, 2011). Consistent with actin polymerization at F-actin foci, TMR-KabC rapidly labeled foci in live growth cones (Fig. 3B). Note that KabC also strongly labeled F-actin filaments undergoing retrograde flow at the leading edge, which was distinct from the static labeling of actin at foci (Fig. 3C). Finally, to directly measure the rate of actin polymerization within foci, we bleached or photo-activated fluorescent protein-conjugated actin in live growth cones (supplementary material Movie 4). Either GFP-β-actin or photoactivatable-GFP (PA-GFP)-γ-actin together with mCh-UtrCH were bleached or photoactivated within foci, and the recovery rate and decay were measured, respectively (Fig. 3D-G). In fluorescence recovery after photobleaching (FRAP) experiments, the turnover rate of GFP-β-actin at foci was significantly faster compared with that of regions immediately adjacent to foci (t_1/2=6.56±0.01 s versus 8.53±0.02 s; P<0.004; Fig. 3D,E), but slower than the recovery of actin within the growth cone veil (t_1/2=5.55±0.02 s; P<0.0001). Similar rates were obtained in fluorescence decay after photoactivation (FDAP) experiments using photoactivatable-GFP at foci (t_1/2=7.11±0.01 s; Fig. 3F,G). Taken together, these data demonstrate that F-actin is highly dynamic within stable foci.

In comparison with the F-actin distribution of other cell types, growth cone F-actin foci exhibited a striking similarity to the invadosomes of immune and invasive cells. Invadosomes are long-lived actin-based protrusions located on the basal surface of cells that appear as punctate accumulations of F-actin in vitro (Linder et al., 2011; Murphy and Courtneidge, 2011). Invadosomes regulate cell migration by adhering to and remodeling the underlying ECM. As sites of cell adhesion, invadosomes contain integrin receptors, PXN and FAK (Chan et al., 2009). In order to begin to test the hypothesis that F-actin foci are invadosome-like protrusions of neuronal growth cones, we immunolabeled growth cones for several key invadosome markers. Cttn is a key modulator of F-actin polymerization that is necessary for invadosome formation (Kirkbride et al., 2011; MacGrath and Koleske, 2012). By using immunocytochemistry, we found that Cttn was present at the leading edge of growth cones (Fig. 4A), as has been reported previously (Decourt et al., 2009; Kurklinsky et al., 2011). However, we also found robust colocalization of Cttn with F-actin foci in the C-domain of growth cones (Fig. 4A; supplementary material Fig. S4). To confirm this distribution, we co-expressed GFP-tagged Cttn (GFP-Cttn) with mCH-UtrCH in live cells (supplementary material Movie 5). Along with Cttn, other actin regulatory proteins that have been previously implicated in invadosome formation (Murphy and Courtneidge, 2011; Philippar et al., 2008) targeted to F-actin foci, including the nucleating proteins Arp2/3 and N-WASP, the cross-linking protein α-actinin and the anti-capping factor Mena (Fig. 4B-E; supplementary material Fig. S4). Taken together, these results suggest that F-actin foci in the C-domain of growth cones represent invadosomes, which might regulate axon guidance through tissues.

One classic marker and specific modulator of invadosome formation is the adaptor protein Tks5, which is phosphorylated by Src (Courtneidge, 2012). Phosphorylated Tks5 targets through its N-terminal PX domain to phosphatidylinositol containing lipid rafts (Murphy and Courtneidge, 2011). Activated Tks5 recruits additional proteins that are involved in invadosome formation, such as the actin regulators described above. To begin to understand the roles that both Tks5 and Src play in growth cone invadosomes, we examined their expression and localization. Tks5 and Tks4 were expressed in developing Xenopus spinal cord tissues during stages of robust axon outgrowth (Fig. 5A,B). In addition, Tks5 and active Src (phosphorylated on residue Y418, pY418-Src) localized to F-actin foci, providing strong evidence that these structures represent growth cone invadosomes (Fig. 5C; supplementary material Fig. S4). Finally, in order to visualize Tks5 dynamics in live growth cones, we expressed human Tks5, tagged with GFP (Tks5-GFP), along with mCh-UtrCH in spinal neurons. As expected, Tks5 colocalized with F-actin foci (Fig. 5E), but surprisingly these foci appeared to be more mobile compared to those foci labeled with only F-actin probes. Because Tks5 recruits several actin regulatory proteins, it is possible that overexpression of Tks5 stimulates actin polymerization, leading to mobile F-actin foci.

Active Src regulates invadosome formation in non-neuronal cells through phosphorylation of both Tks5 and Cttn (Boateng and Huttenlocher, 2012). As active Src targets to growth cone invadosomes, we tested Src inhibitors on growth cone invadosome formation. As predicted, pharmacological inhibition of Src with PP2 and SU6656 resulted in fewer and less-stable invadosomes, assessed in fixed and live growth cones (Fig. 5F,G). Tks5 function is also regulated by recruitment to phosphatidylinositol 4,5-bisphosphate (PIP₂)-containing lipid rafts via its PX domain, which might be generated downstream of phosphatidylinositol (3,4,5)-trisphosphate (PIP₃) because inhibition of phosphoinositide 3-kinase (PI3K) with LY294002 blocks invadopodia formation by Src-transformed NIH-3T3 cells (Yamaguchi and Oikawa, 2010). Consistent with this, growth cones that were treated with LY294002 exhibited reduced actin foci (Fig. 5F). More generally, to test whether intact lipid rafts were necessary for growth cone invadosome formation and stability, we treated GFP-β-actin-expressing growth cones with methyl-β-cyclodextrin (MβCD). MβCD is a cholesterol-sequestering agent that has been previously shown to disrupt lipid rafts in growth cones (Guirland et al., 2004). Acute treatment of growth cones with MβCD led to a modest acceleration in the rate of neurite outgrowth and a decrease in growth cone size (Fig. 5H). Interestingly, growth cones that were treated with MβCD failed to form new stable invadosomes (Fig. 5I-J). Taken together, these results demonstrate that Src tyrosine kinases and intact lipid microdomains are required for the proper formation and maintenance of growth cone invadosomes, and provide further support for the notion that F-actin foci in growth cones are analogous to the invadosomes of invasive cells.

Invadosomes are often restricted to the basal surface, below the nucleus of non-neuronal cells in two-dimensional (2D) culture, whereas in 3D environments, the polymerization of actin can generate basal protrusions containing bundled F-actin (Linder et al., 2011). To accurately resolve the cytoskeletal structure of growth cone invadosomes and to visualize their 3D projections, we used structured illumination microscopy (SIM), which can achieve a lateral resolution of 130 nm and an axial resolution of 250 nm (Toomre and Bewersdorf, 2010). Neurons cultured from embryonic Xenopus spinal cord, as well as iPSC-derived human forebrain neurons, were fixed and labeled for F-actin in 2D culture. Consistent with improved resolution by using SIM, we observed F-actin bundles with a fine interdigitating cross-linked actin meshwork within the peripheral veil (Fig. 6A-C). When viewed orthogonally (Fig. 6D), F-actin foci often appeared as longitudinal columns of F-actin that extended from the apical membrane of the growth cone.

Invadosomes of invasive cells often contain MTs for structural stability and for the delivery of cargo (Murphy and Courtneidge, 2011). To assess whether growth cone invadosomes contained MTs, we immunolabeled for β-tubulin together with F-actin and imaged with SIM. MTs were found alongside bundled F-actin within apical invadosomal protrusions (Fig. 6E,F). It is unclear why some invadosomes formed protrusions and contained MTs yet others remain restricted to the basal and apical surfaces of growth cones, but this is likely to depend on the 3D environment and the maturation state of the invadosome (Linder et al., 2011; Murphy and Courtneidge, 2011).

As invadosomes are stabilized by their 3D environment in vivo, we wanted to visualize the 3D morphology of growth cones with super resolution in collagen gel and in vivo. Xenopus explant cultures in collagen-I gels extended axons that spread in three dimensions, as detected by using low magnification confocal microscopy (supplementary material Fig. S5), and can be labeled with standard immunocytochemistry protocols for high-resolution SIM imaging. Similar to on 2D substrata, growth cones in 3D collagen that were then labeled for F-actin and Cttn exhibited punctate distributions when viewed as flattened 2D projections (Fig. 6G). However, when viewed orthogonally and as 3D rotations, F-actin and Cttn puncta often appeared as 3D protrusions that extended from the central domain of growth cones (Fig. 6H; supplementary material Movie 6). In addition, we visualized the 3D morphology of peripheral Rohon-Beard neuron growth cones in the skin by using SIM. To specifically label neurons in vivo, we immunolabeled whole-mount stage-25 Xenopus embryos using antibodies against NCAM and Cttn, and cleared the embryos with Murray's clearing solution. SIM imaging of RB neurons within the skin revealed extensive invadosome-like protrusions that extended orthogonally toward the skin from the C-domain of growth cones (Fig. 6I-K). Although growth cones in vivo remained largely planar with many peripheral filopodia directed along the axis of growth, discrete orthogonal protrusions could extend up to 10 µm perpendicular to the axis of extension (Fig. 6J,K; supplementary material Movie 7).

A defining characteristic of invadosomes is their ability to degrade the local matrix, allowing cells to remodel and penetrate tissues (Linder et al., 2011). Cell surface and secreted proteases, such as the MMPs, accumulate and are released at invadosomes (Poincloux et al., 2009). To determine whether growth cone invadosomes contain MMPs, we first screened for the transcripts of known invadosomal MMP by using PCR analyses of pure spinal cord extracts. We found that MMP2, MMP9 and MMP14 were expressed in developing Xenopus spinal tissue (Fig. 7A), as well as several ADAM proteins (not shown). MMP14 (also known as MT1-MMP in mammals) expression in developing neurons was confirmed by immunoblotting, which revealed bands at approximately 130 kDa (not shown), 60 kDa and 30 kDa, which is likely to represent an MMP14 complex, an active form and a proteolytic fragment, respectively (Fig. 7B). To determine whether MMPs are active in the embryonic spinal cord, we performed gelatin zymography on spinal cell extracts; these showed degradation at molecular masses corresponding to the MMP9 propeptide, MMP9, the MMP2 propeptide and MMP2 (Fig. 7C). Next, to directly examine whether MMPs were present at invadosomes, we immunolabeled growth cones for MMP14 together with F-actin. Although MMP14 was present in a punctate distribution throughout growth cones, we did observe colocalization between MMP14 and F-actin foci (Fig. 7D-F). Moreover, using SIM imaging, we found that ADAM17 concentrated to the tips of invadosomal protrusions viewed in three-dimensions (Fig. 7G,H; supplementary material Movie 8).

Finally, in order to assess whether there is proteolytic activity at growth cone invadosomes, we performed fluorescent gelatin degradation assays. Xenopus spinal neurons were cultured on two variants of fluorescent gelatin that either show a loss or a gain of fluorescence upon proteolytic degradation by MMPs (Artym et al., 2009; Sloane et al., 2006). Both assays showed that growth cones exhibited a high level of proteolytic activity over regions of ECM contact, with most intense degradation occurring near F-actin foci (Fig. 7I-K; supplementary material Fig. S6). Importantly, ECM degradation was attenuated in growth cones cultured in the presence of GM6001, a non-specific MMP inhibitor, as well as in growth cones that had been treated with the invadosome disrupting agents PP2, SU6656 and MβCD (Fig. 7L). Finally, growth cones and neural crest cells expressing a dominant-negative variant of Tks5, ΔPX-Tks5 (Fig. 7L; supplementary material Fig. S7), also showed reduced ECM degradation. This specific manipulation allowed actin foci to form, but foci were highly mobile (supplementary material Movie 9), possibly due to the inability of mutant Tks5 to dock with PIP₂-containing microdomains (Abram et al., 2003; Oikawa et al., 2012; Seals et al., 2005). Taken together, these results suggest that MMPs are present and active within the developing spinal cord, and target to growth cone invadosomes where they remodel the local ECM.

We hypothesize that growth cone invadosomes promote axon extension across tissue barriers, as reported previously for both normal and metastatic migratory cells (Linder et al., 2011; Murphy and Courtneidge, 2011). One location that we expect invadosomes will function is at sites where motoneurons exit the spinal cord and extend into the peripheral axial myotomal tissues. To begin to test this notion, we expressed GFP-tagged ΔPX-Tks5 or GFP by using blastomere injection at the eight-cell stage, thereby targeting dominant-negative Tks5 or control GFP to the ventral spinal cord. Embryos were allowed to develop to the neural tube stage, when the pioneering motoneuron axons are exiting the spinal cord into the periphery. Note that, although other cell types were labeled using the CMV promoter, we compared embryos that had relatively sparse labeling in order to both limit potential non-cell autonomous effects and facilitate visualization of motoneurons. Fixed and methanol-dehydrated embryos were immunolabeled for GFP plus acetylated tubulin or synaptotagmin 2 (using the Znp-1 antibody), cleared with Murray's solution and imaged intact by using confocal microscopy. Three-dimensional reconstruction of motoneuron axons on the ventral fascicle near exit points revealed that motoneuron growth cones often extended peripheral protrusions from their MT-containing central domain (Fig. 8A-C). These invadosome-like protrusions were particularly evident in wild-type motoneurons labeled with the Znp-1 antibody, which labels the most distal extent of developing axons (Fig. 8D,E). Motoneuron axon terminals exhibited numerous fine protrusions that extended laterally into the periphery (Fig. 8F; supplementary material Movie 10), some of which matured into peripheral axons (Fig. 8D). Importantly, invadosome function appeared to be necessary for proper motoneuron development, as expressing dominant-negative ΔPX-Tks5-GFP in motoneurons reduced the extension of axons into the peripheral myotomal tissue (Fig. 8G-I). Although not quantified, axon extension within the spinal cord appeared to be qualitatively normal, and dominant-negative ΔPX-Tks5-GFP did not inhibit axon extension in vitro (not shown), suggesting a specific effect on peripheral tissue invasion. Taken together, these data suggest that motoneuron growth cones in the spinal cord use invadosomes to promote axon extension into the peripheral myotome.

Many studies have identified F-actin-rich protrusions from the basal surface of invasive cells as sites of membrane attachment and degradation of the ECM, but analogous structures have not been reported in developing neuronal growth cones. In this study, we show for the first time that growth cones form 3D protrusions, which share many characteristics of invasive cell invadosomes. We found that distinct F-actin foci formed within the C-domain of growth cones from a variety of neuron types and species in vitro, as well as in the spinal cord and skin of intact Xenopus embryos. Growth cone invadosomes contained β1 integrin receptors and recruited PXN and FAK transiently, yet they were distinct from leading edge PC adhesions. Invadosomes appeared to be sites of F-actin polymerization as they contained F-actin barbed ends, as well as actin monomers, and underwent rapid actin turnover, as determined by using FRAP and FDAP. Consistent with this notion, the actin regulatory proteins Arp3, N-WASP, α-actinin and Mena localized to invadosomes. Importantly, the classic invadosome markers Cttn, pY418-Src and Tks5 robustly concentrated at growth cone invadosomes. Pharmacological manipulations suggested that Src and lipid rafts, known Tks5-targeting signals, were both necessary for invadosome formation in growth cones. Viewing growth cones by using super-resolution microscopy showed that F-actin within invadosomes oriented as axial columns, perpendicular to the direction of axon extension. Moreover, a subset of axial F-actin bundles, which often contained MTs, penetrated the apical cell surface of growth cones to form protrusions in 2D and 3D culture, as well as in vivo. These growth cone invadosomes targeted MMPs and locally remodeled the ECM through proteolytic activity, which required Tks5 function. Finally, dominant interference of Tks5 function in vivo inhibited the exit of motoneurons from the spinal cord into the periphery. Taken together, these findings demonstrate that growth cones form protrusions analogous to the invadosomes of invasive cells, which control axon guidance through 3D environments of the developing nervous system.

Invadopodia and podosomes are collectively referred to as invadosomes, although they have some notable differences in structure and function (Linder et al., 2011; Murphy and Courtneidge, 2011). Three common functions ascribed to invadosomes are cell adhesion, matrix degradation and mechanosensing (Alexander et al., 2008; Collin et al., 2008; Juin et al., 2013). Our evidence suggested that neuronal growth cones form invadosomes that share common functions to both invadopodia and podosomes (supplementary material Fig. S8). For example, growth cone invadosomes appeared to be adhesive structures, because stable F-actin foci localized to areas that were in close contact to the substratum and colocalized with adhesion-related proteins, such as integrin receptors, PXN and FAK (Fig. 2). However, although invadosomes are likely to be adhesive contacts, they differ from PCs, which typically assemble within filopodia and are short-lived (∼2.5 min). Interestingly, a subset of PCs appeared to transition into invadosomes (Fig. 2E-H), and we suspect that these two processes could be linked (Chan et al., 2009; Liu et al., 2010). It is noteworthy that the F-actin foci that we observed in growth cones share many similarities to actin patches that have been found to be responsible for collateral branching of chick DRG axons (Ketschek and Gallo, 2010; Spillane et al., 2011).

Two important questions are what extracellular cues and intracellular signals instruct invadosome formation versus PC formation in growth cones? In non-neuronal cells, the formation of invadosomes is stimulated by growth factors, such as platelet-derived growth factor (PDGF), transforming growth factor-β (TGFβ) and epidermal growth factor (EGF) (Murphy and Courtneidge, 2011). These growth factors activate Src tyrosine kinases, which phosphorylate Tks5, Cttn and other key regulatory proteins that target to invadosomes. Consistent with a role for Src in growth cone invadosomes, we found that active Src targeted to invadosomes and that inhibition of Src reduced invadosome formation (Fig. 5F,G). Other intracellular signaling proteins, such as protein kinase C (PKC) and PI3K cooperate with Src in regulating invadosome formation (Hoshino et al., 2012; Yamaguchi et al., 2011). PI3K might initiate the formation of invadosomes through local production of PIP₃ and PIP₂, which recruits Tks5 through association with its N-terminal PX domain (Hoshino et al., 2013; Yamaguchi and Oikawa, 2010). Consistent with a similar role in growth cones, we found that pharmacological inhibition of PI3K, as well as expression of ΔPX-Tks5-GFP, reduced and mobilized F-actin foci, respectively, and limited ECM degradation (Fig. 5G, Fig. 7L; supplementary material Movie 9). Because several growth factors and axon guidance cues, such as brain-derived growth factor (BDNF), netrin and ephrins are known to regulate Src, PKC and PI3K function in growth cones (Dent et al., 2011; Hall and Lalli, 2010), we hypothesize that guidance cues in combination with specific ECM proteins control PC-mediated substratum adhesion versus invadosome-mediated basal lamina penetration.

Another crucial open question is what roles do invadosomes have in the guidance of axons and dendrites to their synaptic targets? One possible function of invadosomes is the targeting of MMPs and other proteases, such as ADAMs, in order to promote local matrix degradation and the extension of axons through tissues (McFarlane, 2003; Yong et al., 2001), and we found that these MMPs target to growth cone invadosomes (Fig. 7). Neural crest cells form invadosomes that are necessary for their proper migration, the development of craniofacial structures and pigmentation in zebrafish embryos (Murphy et al., 2011). Matrix proteases might regulate axon extension in part by degrading the surrounding matrix to create a passage for axonal outgrowth (Gutierrez-Fernandez et al., 2009; Hayashita-Kinoh et al., 2001; Zuo et al., 1998), but more recent studies that were performed both in vitro and in vivo now suggest that proteolytic cleavage is directed toward specific ligands in the environment, as well as towards receptors on growth cones in order to activate or terminate motility (Chen et al., 2007; Fambrough et al., 1996; Hehr et al., 2005; Schimmelpfeng et al., 2001; Webber et al., 2002). For example, MMPs are necessary for the conversion of pro-neurotrophins into mature neurotrophins (Lee et al., 2001) and during ectodomain shedding of guidance cue receptors (Bai et al., 2011; Browne et al., 2012; Coleman et al., 2010; Galko and Tessier-Lavigne, 2000; Hattori et al., 2000; Kanning et al., 2003; Lin et al., 2008; Miller et al., 2008; Walmsley et al., 2004). We found that commissural interneurons near the midline formed robust F-actin foci (Fig. 1G) that could be invadosomes interacting with cues within the ventral floorplate. Also, we showed that invadosomes are necessary for motoneuron axon exit from the spinal cord into the peripheral myotome (Fig. 8). However, a specific role for MMPs has not been established at these choice points, nor has the morphogen that promotes invadosome formation by these growth cones been identified. Future studies will examine candidate morphogens, such as BDNF or pro-BDNF, that might be released by myocytes (Yang et al., 2009) to regulate peripheral sprouting of invadosomes.

Primers against Tks5, MMP2, MMP9 and MMP14 were designed from Xenopus laevis sequences (Xenbase; Bowes et al., 2010) using Primer3 (Rozen and Skaletsky, 2000), and were then synthesized at the Biotechnology Center (University of Wisconsin, Madison). Tks5 and MMP transcripts were amplified by using PCR from reverse transcribed mRNA isolated with TRIzol (Invitrogen) from spinal cord tissues of stage-24 to stage-26 embryos.

Xenopus laevis embryos were obtained and staged as described previously (Gomez et al., 2003). For protein expression experiments, eight-cell stage embryos were injected with 0.25 ng of in vitro-transcribed capped mRNA (mMessage Machine, Ambion) or 60-70 pg of DNA (constructs are detailed in the methods in the supplementary material). Neural tube and retinal cultures were prepared as described previously (Gomez et al., 2003; Woo and Gomez, 2006). For 3D cultures, explants were cultured onto 1.5 mg/ml collagen-I gels (BD Biosciences). Cultures were imaged or fixed 18-22 h after plating. All methods were approved by the University of Wisconsin School of Public Heath Animal Care and Use Committee.

Human neurons were differentiated from iPSCs as previously described (Gamm et al., 2008; Hu and Zhang, 2009; Liu et al., 2013; Pankratz et al., 2007). Neurospheres were plated onto glass coverslips coated with poly-D-lysine (PDL) and laminin (PDL-LN) and cultured in neural basal media with B27 supplements (Gibco).

Fluorescent gelatin-coated coverslips were prepared as described previously (Artym et al., 2009). Briefly, coverslips were coated with 50 µg/ml PDL and activated with 0.5% glutaraldehyde, followed by conjugation with Oregon Green 488 gelatin (Invitrogen) or Cy3 gelatin (Millipore) for 10 min at room temperature. Residual reactive groups were quenched with sodium borohydride. The matrix protease inhibitor GM6001 (10 µM) was applied overnight, whereas PP2, SU6656 and MβCD were applied for 4 h before fixation. Coverslips coated with DQ collagen-IV (Invitrogen) were prepared as described previously (Sloane et al., 2006). Briefly, coverslips were coated with a solution containing 25 µg/ml laminin, fibronectin and DQ collagen-IV. Cultures were fixed 18-22 h after plating and imaged by using confocal microscopy. Further details of the reagents and microscopy analysis used can be found in the methods in the supplementary material.

Tks5 and MMP14 were blotted from total protein extracts from stage-24 to stage-26 embryo spinal cords using the Novex NuPAGE SDS-PAGE gel system (Invitrogen). Tks5 (1:500) and MMP14 (1:1000) primary antibodies and horseradish peroxidase (HRP)-conjugated secondary antibodies (1:5000, Jackson ImmunoResearch) were used, and the blots were developed by using enhanced chemiluminescence (ThermoScientific Pierce ECL).

Neurons in 2D culture were immunolabeled as described previously (Myers and Gomez, 2011; Santiago-Medina et al., 2013). Three-dimensional cultures were fixed for 1 h and blocked in Ca²⁺- Mg²⁺-free (CMF)-PBS containing 0.5% fish gelatin and 0.2% Triton X-100 (GBT, gelatin block with Triton) overnight at 4°C. See methods in the supplementary material for antibody details.

Gelatin zymography was performed using stage-24 to stage-26 Xenopus embryo spinal cords. Lysates were run on a Novex zymogram gelatin gel and processed according to the manufacturer's instructions (Invitrogen).

To detect F-actin in motoneurons and commissural interneurons in vivo, mCh-UtrCH or GFP-UtrCH was expressed by using targeted blastomere injection of mRNA. For imaging of commissural interneurons, embryos were fixed at stage 23 to stage 25 and dissected to expose the spinal cord as described previously (Moon and Gomez, 2005). Peripheral Rohon-Beard and motoneuron growth cones were imaged within intact stage-24 to stage-25 embryos that were fixed in 4% PFA in sucrose, dehydrated in methanol and cleared, as described previously (Huang et al., 2007). Undissected embryos were immunolabeled in GBT solution for NCAM (HNK-1 antibody) and Cttn for Rohon-Beard neurons, or GFP, Znp-1 and acetylated tubulin for motoneurons. After immunolabeling and dehydration, embryos were cleared in Murray's solution (2:1 benzyl benzoate:benzyl alcohol) and imaged by using 3D confocal microscopy. Further details of the 3D microscopy are given in the methods in the supplementary material. To quantify the percentage of peripheral motoneuron process extensions, we counted the total number of tubulin-positive peripheral axons and the total number of GFP-positive (GFP or ΔPX-Tks5-GFP) peripheral axons (double labeled). The percentage of tubulin-positive peripheral axons that were also GFP-positive was calculated for each z-stack to determine variability and significance between groups. The extent of motoneuron labeling was qualitatively similar between experiment and control embryos.

We thank Erik Dent and members of the Gomez lab for comments on the manuscript; Tom Fothergill, Derrick McVicker and Diana Cowdrey for providing mouse embryonic hippocampal and cortical cultures; and Zhen Huang, Shang Ma and Hyo Jun Kwon for the gelatin zymography.