Rapid Remodeling of Invadosomes by Gi-coupled Receptors

Invadosomes are actin-rich membrane protrusions that degrade the extracellular matrix to drive tumor cell invasion. Key players in invadosome formation are c-Src and Rho family GTPases. Invadosomes can reassemble into circular rosette-like superstructures, but the underlying signaling mechanisms remain obscure. Here we show that Src-induced invadosomes in human melanoma cells (A375M and MDA-MB-435) undergo rapid remodeling into dynamic extracellular matrix-degrading rosettes by distinct G protein-coupled receptor agonists, notably lysophosphatidic acid (LPA; acting through the LPA1 receptor) and endothelin. Agonist-induced rosette formation is blocked by pertussis toxin, dependent on PI3K activity and accompanied by localized production of phosphatidylinositol 3,4,5-trisphosphate, whereas MAPK and Ca2+ signaling are dispensable. Using FRET-based biosensors, we show that LPA and endothelin transiently activate Cdc42 through Gi, concurrent with a biphasic decrease in Rac activity and differential effects on RhoA. Cdc42 activity is essential for rosette formation, whereas G12/13-mediated RhoA-ROCK signaling suppresses the remodeling process. Our results reveal a Gi-mediated Cdc42 signaling axis by which G protein-coupled receptors trigger invadosome remodeling, the degree of which is dictated by the Cdc42-RhoA activity balance.

Invadopodia and podosomes, collectively called invadosomes, consist of a core of F-actin and various actin-associated structural and regulatory proteins (1,2,4,5). One major player in the formation and maintenance of invadosomes is the Src tyrosine kinase, which phosphorylates invadopodial substrates, such as cortactin and the scaffold protein Tks5 (tyrosine kinase substrate 5) (2,11). Therefore, cells expressing active Src are a convenient system for studying the regulation of invadosomes.
Interestingly, individual invadosomes can assemble into higher-order "rosettes" consisting of giant circular arrays of F-actin. Rosettes are observed in some cancer cells (19,20), v-Src-transformed fibroblasts (21), osteoclasts (22), and endothelial cells (9,23). Invadosome rosettes may remodel the ECM more efficiently and in a more localized manner than do individual invadosomes (20). Evidence for invadosome rosettes in human tissues is emerging, for example, in the vasculature of lung tumors (9). However, the signal inputs and pathways that drive the remodeling of pre-existing invadosomes into rosettes remain largely unknown.
Here we examine how distinct GPCR agonists, notably lysophosphatidic acid (LPA) and endothelin, influence the behavior of Src-induced invadosomes in human A375M melanoma cells. LPA is a multifunctional lipid mediator and a major serum constituent that signals through six distinct GPCRs (LPA [1][2][3][4][5][6] ) (24,25). LPA is produced by autotaxin, a secreted lysophospholipase D originally identified as a motility factor for melanoma cells (26,27). Autotaxin-LPA signaling promotes invasive cell migration and experimental metastasis (28 -30), but little is known about how LPA may affect invadosome behavior. Endothelin is produced by stromal and tumor cells and signals in an autocrine or paracrine manner to promote malignant cell behavior; acting through the endothelin B receptor, endothelin is strongly implicated in melanoma progression (31)(32)(33).
We show here that LPA and endothelin induce the rapid transition of the ECM-degrading invadosome cluster into highly dynamic rosettes through G i , and we analyze the underlying signaling events with a focus on Rho family GTPases. By using FRET-based biosensors, we monitor and dissect the agonist-regulated activities of RhoA, Rac1, and Cdc42 and find a key role for G i -mediated Cdc42 activation with a likely modulatory role for Rac1 and an opposing role for RhoA. Our results provide new insights into how certain GPCRs remodel invadosomes, thereby rapidly redistributing ECM-degrading activity.
Hairpins were introduced using the pLKO lentiviral vector (empty pLKO vector as a negative control). Viral particles were produced in HEK293 cells transfected with calcium phosphate. Transduced cells were imaged or harvested for total mRNA extraction after 48 h.
Matrix Degradation Assay-Coverslips were coated with gelatin as described previously (59). To determine degradative capacity, 100,000 cells/coverslip were seeded in serum-free DMEM (with or without GM6001). After 48 h, coverslips were washed with PBS and fixed with 4% paraformaldehyde, and cells were stained with phalloidin. Gelatin degradation was determined from confocal images of Ͼ20 fields of view/coverslip, using 2 coverslips/condition in two independent experiments (4 coverslips/condition). For time lapse imaging, cells expressing GFP-actin were plated on coverslips coated with gelatin-ATTO-633 for 24 h.
Live Cell Imaging-Cells were seeded on glass coverslips (24 mm) for 24 -72 h in DMEM (10% FCS) and serum-starved Ͼ2 h before experiments. Cells were imaged in DMEM/F-12 at 37°C in a humidified chamber at 5% CO 2 using a TCS SP5 confocal microscope (Leica Microsystems) with a ϫ63, 1.4 numerical aperture oil immersion objective. The pinhole was set to 1.5 Airy units, and focus was set to the ventral membrane of the cells. We took great care to excite at minimal laser intensity to avoid photobleaching or phototoxicity.
Wide Field FRET Experiments-Experiments were performed in HEPES-buffered saline (containing 140 mM NaCl, 5 mM KCl, 1 mM MgCl 2 , 1 mM CaCl 2 , 10 mM glucose, 10 mM HEPES), pH 7.2, at 37°C. Cells were plated on uncoated coverslips and transfected 24 h before experiments with the indicated biosensors were placed on a thermostatted (37°C) inverted Nikon Diaphot microscope and excited at 425 nm. Donor and acceptor emission were detected simultaneously with two photomultipliers, using a 505-nm beam splitter and optical filters: 470 Ϯ 20 nm (CFP channel) and 530 Ϯ 25 nm (YFP channel). FRET was expressed as the ratio between acceptor and donor signals, set at 1 at the onset of the experiment.
Rho GTPase Biosensors-The design of FRET-based biosensors of Rac1, Cdc42, and RhoA was based on the design of the Raichu sensors (34,35) and FLARE-RhoA (36) (details to be described elsewhere). Briefly, the complete amino acid sequence of a given Rho GTPase was positioned at the C terminus of a single polypeptide chain to preserve its interaction with GDI and other regulatory proteins. A FRET pair consisting of Cerulean3 and circularly permutated Venus was used. The CRIB domain of PAK and HR1 region of PKN were used as the effector domain for activated Rac1/Cdc42 and RhoA, respectively. In control biosensors, point mutations (H83D/H86D in PAK and L59Q in PKN) were introduced to generate bindingdeficient effector domains, so that FRET ratios remained at the basal level regardless of the activation state of the Rho GTPases.
Confocal FRET Experiments-Cells transfected with a given biosensor were imaged on the TCS SP5 confocal microscope using a Leica ϫ63, 1.4 numerical aperture "lambda-blue" oil immersion objective. Excitation was at 442 nm, and the FRET ratio was determined from emission images acquired simultaneously at 448 -505 nm (CFP channel) and at 505-555 nm (YFP channel) and expressed as a ratio (YFP/CFP). In these cells, LifeAct-mCherry was imaged simultaneously in the range 568 -650 nm.
Ca 2ϩ Imaging-Intracellular [Ca 2ϩ ] was detected essentially as published (60). Experiments were done in serum-free DMEM/F-12, using a Leica SP5 confocal microscope with excitation at 488 nm and emission at two channels (495-550 nm and 560 -650 nm). The confocal pinhole was fully opened, and recordings were normalized by setting basal levels to 1.0.
Image Analysis-Invadosomes were detected by intensity and size segmentation of colocalizing actin and cortactin signals after manual thresholding using Fiji software (37) and normalized to control cells. At least three independent experiments were analyzed for every condition (18 fields of view/ condition, 5-10 cells/field of view, Ͼ100 cells/condition).

c-Src Induces Functional Invadosomes in Melanoma Cells-
Given the key role of c-Src in invadopodia formation, we examined various human tumor cells for their ability to produce invadopodia upon expression of active c-Src and for their responsiveness to selected GPCR agonists. On the basis of these criteria, we selected metastatic A375M melanoma cells as our main model system. A375M cells showed characteristic actinrich invadopodia that colocalized with cortactin at the ventral plasma membrane (Fig. 1A). Expression of constitutively active Src(Y530F) (Fig. 1D) had only minor effects on cell morphology ( Fig. 1B) but led to a marked increase in the number of invadosomes that clustered predominantly at the cell periphery (Fig. 1,  C and E). These actin-rich clusters contained the invadopodial marker Tks5, cortactin, vinculin, and focal adhesion kinase (FAK) (Fig. 1C). Along with focal adhesion kinase, high levels of phosphotyrosine (pY) were also found in these invadosomes (Fig. 1C). Of note, invadosome cluster formation was not unique for A375M cells, because very similar structures were also observed in Src(YF)-expressing MDA-MB-435 melanoma cells (see Fig. 2D). As expected, invadosome clusters disappeared upon the addition of the Src inhibitor PP2 (results not shown). The Src-induced invadosomes were stable and persisted in serum-free medium, indicating that their formation and maintenance is a cell-intrinsic mechanism, not requiring exogenous growth factors, at least in the presence of active c-Src. The invadosomes were functional in that they colocalized with sites of robust gelatin degradation, a process that was inhibited by the metalloprotease inhibitor GM6001 (Fig. 1, F  and G).
Rapid Formation of ECM-degrading Rosettes by LPA and Endothelin through G i -We tested a number of GPCR agonists for their ability to influence invadosome abundance and orga-nization in A375M(Src ϩ ) by monitoring actin remodeling using time lapse confocal microscopy. Strikingly, LPA and endothelin induced the rapid formation of highly dynamic invadosome rosettes (Fig. 2, A and B, and supplemental Video 1). Agonist-induced rosettes displayed various dynamic behaviors as they went through phases of expansion and contraction (supplemental Video 1), a behavior reminiscent of the formation of dynamic podosome rings in osteoclasts (38). Other agonists for which A375M cells express functional receptors (as determined by Ca 2ϩ mobilization), including sphingosine 1-phosphate (S1P) and thrombin, showed little or no rosetteinducing capacity. Receptor tyrosine kinase agonists, such as PDGF and VEGF, left invadosome organization similarly unaltered ( Fig. 2A). Thus, a subclass of GPCRs mediates the rapid remodeling of invadosome clusters into rosettes.
Rosettes appeared within 1 min after LPA or endothelin addition and typically evolved from already existing invadosome clusters at the cell periphery, containing Tks5, cortactin, vinculin, focal adhesion kinase, and enhanced phosphotyrosine (Fig. 2B). The newly formed rosettes degraded the ECM in a highly dynamic manner ( Fig. 2C and supplemental Video 2). LPA-induced rosette formation from pre-existing invadosome clusters was also observed in MDA-MB-435(Srcϩ) melanoma cells (Fig. 2D).  (39), which are known to couple to G i and G q (40); furthermore, these cells were found to co-express three distinct LPA receptors, namely LPA 1 , LPA 3 , and LPA 6 (Fig. 3A). The selective LPA 1 /LPA 3 antagonist Ki16425 inhibited LPA-induced rosette formation with an IC 50 value of ϳ30 nM (Fig. 3B), at which dose the inhibitor antagonizes activation of LPA 1 but not LPA 3 (41). Moreover, LPA 1 knockdown cells failed to form rosettes in response to LPA (Fig. 3C). We conclude that LPA-induced rosette formation is mediated by the LPA 1 receptor in a non-redundant manner. LPA 1 is known to couple to G i , G 12/13 , and G q (25,42). Pretreatment of the cells with pertussis toxin (PTX) blocked rosette formation by LPA and endothelin (Fig. 3D), indicating that invadosome remodeling critically depends on G i -linked signaling pathways.

A375M cells express endothelin B receptors
Dissection of Signaling Pathways: PI3K, ERK/MAPK, and Ca 2ϩ -In addition to Rho family GTPases, G protein-linked effectors and signals implicated in F-actin remodeling include the ERK/MAPK pathway, Ca 2ϩ mobilization, and PI3K. Like many melanoma cells, A375M cells express oncogenic B-RAF(V600E), resulting in constitutive activation of the MEK-ERK/MAPK pathway (43). Consistent with this, LPA could not further enhance basal ERK activity (Fig. 4A). Inhibitors of B-RAF and MEK (PLX4720 and U0126, respectively) strongly reduced MAPK activity, without affecting pre-existing invadosomes or LPA-induced rosette formation (Fig. 4, A and B). LPA and endothelin induced a rapid rise in cytosolic Ca 2ϩ , which was abrogated by cell-permeable BAPTA-AM (Fig. 4C) (results not shown). Ca 2ϩ -buffered cells showed fewer invadosome clusters (not shown), but the ability of LPA to induce rosette formation was not affected (Fig. 4D). Furthermore, raising cytosolic Ca 2ϩ by thapsigargin did not affect rosette formation; nor did the protein kinase C (PKC) inhibitor GÖ6983 (44) (Fig. 3, E  and F). These findings rule out a critical role for ERK/MAPK, Ca 2ϩ , and PKC in GPCR-induced invadosome remodeling. PI3K generates phosphatidyl 3,4,5-trisphosphate (PIP 3 ) to activate downstream effectors such as Akt. LPA phosphorylated Akt in a PTX-and wortmannin-sensitive manner (Fig. 4G). Wortmannin interfered with the maintenance of invadosomes and caused the disassembly of newly formed rosettes (Fig. 4H), indicating an indispensable role for basal PI3K activity. Using a PIP 3 -specific biosensor, pGRP1(PH)-EGFP (45,46), we found that LPA stimulates PIP 3 production at the ventral plasma membrane, specifically in the region of rosettes (Fig. 4I). We conclude that LPA activates PI3K through G i , resulting in localized PIP 3 accumulation, serving as an essential signal for rosette formation.
Monitoring Rho GTPase Activities-Rho family GTPases, particularly RhoA, Rac1, and Cdc42, are central regulators of the actin cytoskeleton and implicated in invasive cell migration (47). Cdc42 is known to govern invadosome formation through its downstream effector N-WASP, but much less is known about the role of RhoA and Rac1 in invadosome formation and remodeling (12). We measured the activation of RhoA, Rac1, and Cdc42 by GPCR agonists in real time, using newly developed FRET-based biosensors (see "Experimental Procedures").
As shown in Fig. 5A, LPA triggered the rapid co-activation of Cdc42 and RhoA, with very similar kinetics, concurrent with a transient decrease in Rac1 activity. Endothelin, the strongest inducer of invadasome remodeling, similarly enhanced Cdc42 activity with a concomitant reduction in Rac1 activity. Unlike LPA, however, endothelin did not affect RhoA activity (Fig. 5A). Peak values of Cdc42 activation and decreased Rac activity were reached at about 1 min (Fig. 5A), which coincides with the initiation of rosette formation. We also tested S1P, which showed little or no effect on invadosome remodeling ( Fig. 2A). When compared with LPA and endothelin, S1P evoked a strikingly robust activation of RhoA and a rather weak Cdc42 activation signal. In addition, S1P rapidly reduced Rac1 activity for a prolonged period of time (Ͼ10 min) (Fig. 5A). The RhoA response to S1P must be largely mediated by the S1P 2 receptor, whose coupling efficiency to the G 12/13 -RhoA pathway is particularly strong (48,49). Fig. 5B summarizes the distinct Rho GTPase responses, showing that Cdc42 activation and Rac deactivation are strongly associated with invadosome remodeling, whereas RhoA activation is inversely correlated and hence may exert an opposing effect. Of note, the pattern of activation and deactivation of RhoGTPases was independent of Src(YF) expression, because A375M cells responded to LPA stimulation in the exact same manner as did A375M(Src ϩ ) cells (Fig. 5C).
G i -mediated Activation of Cdc42 Is Essential for Rosette Formation-LPA-induced activation of Cdc42 was almost completely inhibited by PTX (Fig. 6, A and B), indicating a key role for G i . Wortmannin inhibited Cdc42 activation by about 40%, indicating that both PI3K-dependent and PI3K-independent pathways downstream of G i lead to Cdc42 activation. LPA-induced Cdc42 activation was inhibited by Ki16425, confirming LPA 1 involvement (Fig. 6, A and B). Upon stimulation, Cdc42 was activated predominantly within the rosettes themselves (Fig. 6C). Furthermore, rosette formation was impaired upon Cdc42 knockdown using shRNA or by expressing dominant-negative Cdc42(T17N) (Fig. 6, D and E). Expression of constitutively active Cdc42(Q61L) or Cdc42(F28L) induced the formation of many individual invadosomes, but it prevented LPA from reorganizing them into rosettes (Fig. 6, E and F). We therefore conclude that a tight spatiotemporal control of Cdc42 activity is critical for agonist-induced rosette formation.
Rac Deactivation Is Non-G i -mediated-The GPCR-mediated decrease in Rac activity is unexpected, because LPA and other G i -coupled receptor agonists normally enhance Rac activity as measured by pull-down assays (e.g. see Ref. 42). Remarkably, following PTX treatment, the decrease in Rac activity was more pronounced (Fig. 7A). It thus appears that the overall Rac signal consists of two components mediated by distinct G proteins: the decrease in Rac activity is non-G i -mediated and is superimposed by a G i -mediated increase in Rac activity. Recent evidence indicates that reduced Rac activity promotes invadosome stability, whereas elevated Rac activity drives invadosome dis- assembly (14). Consistent with this, expression of constitutively active Rac(QL) abrogated invadosome cluster formation and suppressed rosette formation by LPA (Fig. 7B).
RhoA-ROCK Signaling Antagonizes Invadosome Remodeling-The magnitude of agonist-induced RhoA activation showed a marked inverse correlation with rosette formation, suggesting that enhanced RhoA activity suppresses invadosome remodeling. Several lines of experimental evidence support this notion. First, cells expressing active RhoA(V14) lacked invadosome clusters and rosettes (Fig. 7C). Second, the ROCK inhibitor Y27632 boosted rosette formation by LPA and endothelin, whereas it conferred rosette-inducing capacity to S1P (Fig. 7D). Finally, prior stimulation of the cells with RhoA-acti-vating S1P attenuated the ability of LPA and endothelin to induce rosettes (Fig. 7E). From these results, we conclude that the G 12/13 -linked RhoA-ROCK pathway counteracts G i -Cdc42mediated rosette formation.

Discussion
Unraveling the signaling inputs and pathways that drive the formation, maintenance, and remodeling of invadopodia is essential to better understand tumor cell invasion into the ECM and surrounding tissues, which is a first step in the metastatic cascade. Numerous molecular components of invadosomes have been identified, and increasing evidence points to their importance in vivo (7-10). However, relatively little is still known about how pre-existing invadopodia are reassembled into giant rosettes by extracellular cues. Rosette formation is usually assumed to be a spontaneous self-assembly process, but our results indicate that this is not necessarily true.
Our results reveal a previously unknown role for G i -coupled receptors in driving rosette formation. The use of newly developed FRET-based biosensors allowed us to monitor the kinetics of Rho GTPase responses during agonist-induced rosette formation with high temporal resolution. In our melanoma cell system, active Src promotes invadosome formation and clustering but did not induce rosette formation by itself. Melanoma-relevant GPCR agonists, notably LPA (acting through the LPA 1 receptor) and endothelin (acting via the endothelin B receptor), signal through a G i -Cdc42 axis to remodel stable Srcinduced invadosomes in a highly dynamic manner (Fig. 8). Our results exclude a role for Ca 2ϩ mobilization and MAPK activity in the formation of rosettes. Active Cdc42(QL) produced individual invadosomes but failed to form organized rosettes, which emphasizes the importance of a tight spatiotemporal control of Cdc42 activity upon receptor stimulation. The newly formed rosettes rapidly redistribute the ECM-degrading activity, which may help tumor cells to invade the ECM and surrounding tissues in a more efficient and dynamic spatio-temporal manner than stable invadosomes can do. A recent study has implicated EGF as an inducer of rosettes in carcinoma cells (50), although those rosettes lacked ECM-degrading activity.
LPA stimulated PI3K-mediated PIP 3 production at the ventral membrane, specifically in the region of rosettes. Localized PIP 3 accumulation may serve as an essential signal for rosette formation by recruiting PH domain-containing proteins, including GDP/GTP exchange factors (GEFs) for Rho family GTPases. PI3K exists in distinct isoforms. GPCR agonists activate mainly the ␤-isoform, so it seems likely that PI3K-␤ is the main player in our cell system (51,52). We find that G i -mediated Cdc42 activation is regulated by both PI3K-dependent and PI3K-independent pathways. This is consistent with the fact that Cd42 activation occurs through multiple pathways, involving both PI3K/PIP 3 -driven recruitment of specific GEFs and direct interaction of G(␤␥) subunits with specific GEFs (52-  54). Given the multitude of Cdc42/Rac-specific GEFs in most cell types, it is too early to speculate about the identity of the GEF(s) involved.
G i -mediated activation of Cdc42 was accompanied by a rapid fall in Rac activity. This unexpected Rac response could be dissected into two components: a G i -mediated increase in Rac activity and a non-G i -mediated decrease in Rac-GTP. The latter phase was dominant over the first. RhoA and Rac are known to oppose each other at multiple levels, and their activity balance orchestrates cell shape, migration, and invasion (55)(56)(57). Active RhoA can inhibit Rac through activation of a GTPaseactivating protein (58). However, the decrease in Rac activity observed here cannot be attributed to RhoA activation, because endothelin lowered Rac activity without activating RhoA. Therefore, it is more likely that Rac is inhibited through G qmediated phospholipase C activation, a scenario that needs to be further explored. Whatever the mechanism of Rac deactivation, our results are consistent with recent evidence suggesting that decreased Rac1 activity is necessary for maintaining invadosome stability (14). The finding that active Rac(QL) prevents rosette formation lends further support to this view. We therefore propose that deactivated Rac cooperates with active Cdc42 to promote rosette assembly. Finally, we show that the well established G 12/13 -RhoA-ROCK signaling pathway antagonizes invadosome remodeling. Thus, GPCR agonists that do not activate the G 12/13 -linked RhoA activation are predicted to be the most efficient inducers of rosette formation, as we indeed found here for endothelin.
In conclusion, our study reveals G i -coupled receptor agonists, notably LPA and endothelin, as potent inducers of rosette formation in the context of active c-Src. The degree of invadosome remodeling is dictated by the Cdc42-RhoA activity balance, with a likely modulatory role for Rac deactivation. Determination of precisely how the activities of the distinct Rho GTPases are regulated and coordinated during agonist-induced invadosome remodeling remains a challenge for further studies.