PDGF-A suppresses contact inhibition during directional collective cell migration

Individually migrating cells often show contact inhibition of locomotion. When they collide, their forward movement ceases and the cells reorient and move apart. This process typically includes the collapse of locomotory protrusions upon cell-cell contact (Abercrombie and Dunn, 1975; Stramer and Mayor, 2016). Contact repulsion between individually translocating cells in vivo can regulate cell dispersal (Bard and Hay, 1975; Villar-Cerviño et al., 2013; Davis et al., 2015). During collective migration, loose streams, such as those formed by Xenopus neural crest cells, can use contact inhibition to orient cell motile behavior. Where cells are in contact, protrusions are inhibited, but when free space is available, for example at the leading edge of the cell mass, cells extend processes, and this orients the overall movement of the stream (Carmona-Fontaine et al., 2008).

Contact inhibition of protrusive activity is less understood when cells migrate as a coherent sheet. Despite their tight adhesion to neighbors, cells can extend ‘cryptic’ lamellipodia on the substratum, which contribute to the translocation of the cell mass (e.g. Farooqui and Fenteany, 2005; Vasilyev et al., 2009; Gutzeit, 1991; Lewellyn et al., 2013; Hayer et al., 2016). Moreover, cells can use the surface of adjacent cells as substratum for migration. For example, border cell clusters in the Drosophila ovary require cadherin-mediated adhesion to nurse cells for translocation (Niewiadomska et al., 1999). This could be explained by a basic absence of contact inhibition in the respective cells, as in malignant cells (Abercrombie, 1979) or in premigratory neural crest cells (Scarpa et al., 2015). Positive regulation of protrusive activity, for example by chemotactic factors, would then be sufficient to endow collectively migrating cells with locomotory processes. Alternatively, contact inhibition could be present, but be regulated and suppressed if collective migration required it.

This second possibility can be studied in the collectively migrating leading-edge mesendoderm (LEM) of the Xenopus gastrula. LEM cells show contact inhibition on artificial in vitro substrata. However, when translocating on their endogenous substratum as tight cell aggregates, all cells in contact with the substratum form lamellipodia that are oriented towards the animal pole of the embryo (Winklbauer and Nagel, 1991; Winklbauer and Selchow, 1992; Winklbauer et al., 1992). In the present work, we ask how contact inhibition of lamellipodia is established in LEM cells, how it is suppressed during collective migration, and how this suppression is related to LEM directional migration.

The LEM migrates on the ectodermal blastocoel roof (BCR), which is covered with an extracellular matrix (ECM) of fibronectin (FN) (Nakatsuji et al., 1985; Winklbauer and Stoltz, 1995; Winklbauer, 1998). FN interacts not only with other matrix components and cellular receptors, but also with itself to assemble into fibrils (Schwarzbauer and DeSimone, 2011). Interaction with the FN fibrils of the BCR, or with an artificial substratum of purified FN, is necessary and sufficient for lamellipodia formation in isolated LEM cells (Winklbauer, 1990; Winklbauer and Selchow, 1992; Ramos and DeSimone, 1996; Wacker et al., 1998; Luu et al., 2008). However, on pure FN, contact leads to the sudden retraction of lamellipodia in colliding LEM cells, which might explain the absence of protrusions inside LEM cell aggregates on FN (Winklbauer et al., 1992). The presence of lamellipodia in the embryo and in LEM aggregates migrating on BCR matrix in vitro (Winklbauer et al., 1992) suggests that factors within the matrix block contact inhibition.

Here, we show that a matrix-binding form of platelet-derived growth factor-A (PDGF-A), which is expressed in the BCR and guides LEM migration (Ataliotis et al., 1995; Nagel et al., 2004; Smith et al., 2009; Damm and Winklbauer, 2011), enables cells to overcome contact inhibition of lamellipodia. Contact inhibition is implemented in LEM cells through a control module that includes the FN receptor integrin α5β1 and the heparan sulfate proteoglycan syndecan 4. Both factors can bind FN, and we show that FN interaction is indeed involved. However, FN present as puncta on the surface of LEM cells, rather than the FN serving as external substratum for migration, is part of the control module (Winklbauer, 1998). In contrast to the well-studied FN fibrils, little is known about this cell surface FN (csFN). The most detailed description of its function so far relates to the dissemination of rat breast cancer: csFN puncta mediate attachment of the cancer cells to lung endothelium, which is essential for pulmonary metastasis (Cheng et al., 1998, 2003; Huang et al., 2008).

We show further that the csFN-integrin α5β1-syndecan 4 module is controlled by p21-activated kinase 1 (Pak1) to regulate, in turn, the Eph receptor tyrosine kinase ligand, ephrin B1, which eventually controls lamellipodial stability. Thus, the external FN substratum permits LEM cells to form lamellipodia; Pak1 acts through the csFN-integrin-syndecan module on ephrin B1 to induce the collapse of lamellipodia upon contact, implementing a contact inhibition mechanism; and PDGF-A signaling interferes with the control module to overcome inhibition and to stabilize lamellipodia. This stabilization is directional, consistent with a role of PDFG-A signaling in LEM cell orientation and guidance.

To reconstruct the collective migration of LEM cell aggregates in vitro, we first analyzed the contact behavior of single cells on purified FN, the minimal substratum for LEM migration. Free LEM cells translocated on FN while extending two to three lamellipodia, and a protrusion usually collapsed when a cell started to move in the opposite direction (Fig. 1A, Movie 1). During collapse, filopodia extended from the shrinking lamellipodia, to turn eventually into retraction fibers (Fig. 1B). The average lifetime of lamellipodia was 10 min (Fig. 1G). Movement away from an existing lamellipodium was largely prevented when cells were part of an aggregate (Fig. 1E), and lamellipodia protruding from the margin of such an aggregate survived for 35 min on average (Fig. 1G, Movie 2). Nonattached, isolated LEM cells extend protrusions from an active zone, which persists for a similar time (Winklbauer and Selchow, 1992), suggesting that the intrinsic lifespan of LEM protrusions is about 30 min.

When lamellipodia collide head-on with those of other cells, they immediately retract (Winklbauer et al., 1992). We observed that lamellipodia also collapsed when they underlapped a cell body (Fig. 1C). Small protrusions shrank and vanished, whereas larger ones retreated between filiform extensions that rapidly protruded, as though the lamellipodium were transiently changing into a bundle of filopodia. Eventually, retraction fibers attached to the target cell surface were left behind as the contacting cell moved away (Fig. 1C). The remaining lifetime after contact was only 1-2 min (Fig. 1G). Thus, lamellipodia are induced to collapse despite having access to the external FN substratum. The induced retraction appeared similar to not only a Slit-Robo-mediated collapse of growth cones (McConnell et al., 2016), but, notably, also the retraction of free lamellipodia. A basic mechanism of lamellipodia collapse that can be triggered by different molecular or mechanical cues, might be used here in the contact inhibition of lamellipodia.

In apposed cell pairs, tiny processes bridged narrow gaps between cells at different levels above the substratum (Fig. S1A). Directly at the substratum, small lamelliform protrusions of a cell attempted to underlap the adjacent cell, but soon retracted before growing to a significant size (Fig. 1D). As expected from this behavior, protrusion formation was attenuated in the interior of cell aggregates (Fig. 1F). Cells were tightly packed above the substratum level, yet separated by small gaps where attached to FN (Fig. S1B). As in cell pairs, lamelliform protrusions formed, but were small and short-lived because they immediately encountered an adjacent cell and collapsed, turning transiently into bundles of filopodia as they shrunk back and left retraction fibers behind (Fig. 1F). Most of the cells were devoid of lamellipodia at any given time, and the lifetime of protrusions was strongly reduced, to a quarter of that of free lamellipodia (Fig. 1G). Protrusive activity was sustained at the margin, where large lamellipodia extended onto the free substratum surface (Fig. 1E).

Overall, when lamellipodia retracted from a cell body, the short reaction time and narrow range of survival times suggest that the effect of contact is prompt and efficient, and we propose that it corresponds to the first step of a classical ‘contact inhibition of locomotion’. This usually involves a short lamellipodium-to-cell adhesion phase, followed by a retraction of the protrusion (Abercrombie and Dunn, 1975). The second step, where a cell repolarizes and moves away in a different direction, is observed when single LEM cells collide (Winklbauer et al., 1992) but is not expected to occur under the crowded conditions of an LEM cell aggregate.

Expression of kinase-dead Pak1 (KD-Pak1) affects LEM cell protrusions in the embryo (Nagel et al., 2009), and we asked whether it controls lamellipodia stability. In single cells expressing KD-Pak1, lamellipodia were more frequent and larger, and not serrate as in uninjected cells, but smooth-rimmed (Fig. 2A,B). KD-Pak1 induced similar lamellipodia in the interior of LEM aggregates (Fig. 2C,D). Cells were more mobile and less densely packed, with irregular gaps between cells (Fig. S1C, Movie 3). Occasionally, underlapping led to the spontaneous alignment of moving cells. A lamellipodium, once under the body of a cell ahead of it, kept this contact and the second cell followed the first one closely. In this way, chains of coordinately moving cells were generated that underlapped each other and pointed in the same direction (Fig. 2E, Movie 4), mimicking the shingle arrangement of directionally migrating LEM cells in the embryo (Winklbauer and Nagel, 1991; Winklbauer and Selchow, 1992; Nagel et al., 2004).

The lifespan of submarginal KD-Pak1-induced lamellipodia was comparable to that of single-cell lamellipodia (Fig. 2G). Protrusions became fragmented when pulled back by respective cell movements and disappeared, while occasionally leaving retraction fibers behind. When two lamellipodia collided, their contacting regions retreated locally, and either they split in two (Fig. 2F) or vanished as a whole. The increased lifetime of KD-Pak1 protrusions (Fig. 2G) contributed to an increased number of lamellipodia per cell (Fig. 2H) and a high protrusion density in aggregates. The number of protrusions per cell at the free margin was not affected (Fig. 2H). Induction of submarginal lamellipodia by KD-Pak1 was also detected by F-actin staining with fluorescent phalloidin, and the effect was reversed by co-expression of constitutively active CA-Pak1 (Fig. S2A-C). Altogether, KD-Pak1 expression induced a new, smooth-rimmed type of lamellipodium and, at the same time, rendered lamellipodia resistant to contact-induced collapse.

Ephrins are also candidates for the promotion of lamellipodial collapse (Astin et al., 2010; Lin et al., 2015). In Xenopus, ephrin B1 and B2 are expressed in the mesoderm and are involved in cell repulsion at Brachet's cleft (Rohani et al., 2011). In LEM explants, knockdown of ephrin B1, but not of ephrinB2, strongly stimulated submarginal protrusion formation. Lamellipodia were large and smooth-rimmed, similar to those induced by KD-Pak1, serial underlapping occurred (Fig. 3A, Movies 5 and 6) and lamellipodial lifespans were prolonged (Fig. 2G). Both in KD-Pak1 and ephrin B1-morpholino (MO) aggregates, gaps between cells were wider at the substratum level, but were closed above this level (Fig. S1C-E).

We asked whether ephrin B1 knockdown is required in lamellipodia or the underlapped cell body to stabilize protrusions. In mosaics, lamellipodia of ephrin B1-MO-injected cells were stable not only when in contact with ephrin B1-MO cells, but also when underlapping uninjected cells (Fig. 3B,F). By contrast, lamellipodia of uninjected cells collapsed when in contact with the bodies of ephrin B1-MO or uninjected cells (Fig. 3C,F). Thus, ephrin B1 appears to act in lamellipodia to induce their collapse. Consistent with this notion, ephrin B1 protein is expressed in lamellipodia of single cells (Fig. 3D). However, not all lamellipodia were labeled, and cells appeared polarized with regard to ephrin B1 expression. When mCherry-tagged ephrin B1 (Wen and Winklbauer, 2017) was overexpressed in LEM cells, most lamellipodia were unstable on FN (Fig. 3E). When encountering a cell body, they collapsed after initial spreading, as in control cells: filiform protrusions extended, whereas the lamellipodium shrunk. Retraction fibrils remained behind when the protrusion eventually retreated (Fig. 3F), showing that substratum attachment was not abolished by ephrin B1 overexpression.

Lamellipodia formation in LEM cells depends on an external FN substratum. To see whether csFN also affects lamellipodia stability, we first characterized its presence on LEM cells. Antibody staining showed that, on the surface of LEM aggregates, cells were covered with fine, submicron-sized csFN puncta (Fig. 4A). These puncta were also present at the interface between closely apposed cells within aggregates (Fig. 4B), and at free cell surfaces at interstitial gaps (Fig. 4C). An average-sized lamellipodium covered several puncta (Fig. 4D,D′).

After injection of MO (FN1-MO and FN2-MO) directed against the two isoforms of Xenopus FN, the number of FN puncta was strongly reduced, and some cells completely lacked puncta (Fig. 4E,F). Average FN staining intensity was lowered to 37% of controls (Fig. 4G). Integrin α5β1 is the main FN receptor of the gastrula (Whittaker and DeSimone, 1993), and knockdown of the β1 subunit reduced FN staining intensity (Fig. 4G-I). Another putative FN-binding protein, the transmembrane heparan sulfate proteoglycan syndecan 4, is ubiquitously expressed in the embryo, and required for gastrulation (Muñoz et al., 2006) and for the development of the LEM-derived foregut (Zhang et al., 2016). Its knockdown with MO (xSyn4.1-MO and xSyn4.2-MO) directed against the two Xenopus isoforms reduced syndecan 4 expression (Fig. 4G) but did not affect csFN density or distribution (Fig. 4G,J,K). Similarly, KD-Pak1 expression did not alter the csFN pattern (Fig. 4L,M).

Compared with the punctate FN staining, integrin β1 distribution was more diffuse (Fig. 4N). Syndecan 4 expression was punctate, but patches were larger and fewer compared with csFN puncta (Fig. 4O). Thus, neither of the two putative receptors for csFN showed an expression pattern identical to that of csFN, but integrin function was required for csFN puncta formation or retention, whereas syndecan 4 was dispensable. Although KD-Pak1 expression did not affect the csFN distribution on cell bodies, it increased the lamellipodial localization of integrin β1 and syndecan 4 in single cells (Fig. 4P,Q).

Knockdown of csFN with FN-MOs had no effect on the shape or lifetime of submarginal lamellipodia (Fig. S3A,B). In LEM aggregates co-injected with KD-Pak1 mRNA and FN-MOs, cells formed broad, smooth-rimmed submarginal lamellipodia, as expected from KD-Pak1 cells. However, lamellipodia collapsed consistently and rapidly, switching transiently to filopodia extension before retreating and leaving retraction fibers behind (Fig. 5A,F, Movie 7). Marginal lamellipodia were stable (Fig. 5B), indicating that external FN substratum could be used in csFN-depleted aggregates, but that it was not able to stabilize submarginal protrusions: csFN was required to overcome contact inhibition of lamellipodia. In FN-MO/KD-Pak1 cells, ephrin B1-MO fully rescued lamellipodia stability (Fig. 5C,F), placing csFN function upstream of ephrin B1 and downstream of Pak1 (Fig. 5G).

Given that csFN density is reduced upon integrin β1 knockdown, we asked whether the latter treatment also interfered with lamellipodia stabilization. Surprisingly, in KD-Pak1-expressing aggregates, knockdown of integrin β1 did not shorten lamellipodial lifespans (Fig. 5F). In fact, injection of integrin β1-MO alone promoted submarginal lamellipodia (Fig. 5D,F), and knockdown of integrin β1 in FN-MO explants also increased lamellipodial stability (Fig. 5F). These results imply that integrin interferes with lamellipodia stability, but that it is negatively regulated by FN (Fig. 5G).

Integrin α5β1 recognizes the RGD cell-binding site in FN, and an inhibitory RGD peptide prevents mesoderm cell spreading on FN in vitro (Winklbauer, 1990). This raises the question of how lamellipodia can be maintained in integrin β1-MO explants. We noted that single LEM cells extended lamellipodia on FN in vitro and migrated even when integrin β1 was knocked down. Interestingly, the smooth-rimmed shape of their protrusions conformed to that of KD-Pak1 or ephrin B1 (eB1)-MO-induced submarginal lamellipodia (Fig. S4). RGD peptide inhibited the spread of both uninjected and integrin β1-MO-injected cells (not shown). It might be that residual integrin α5β1 in the morphants is sufficient for spreading on external FN, but not for a role related to csFN. Alternatively, a different RGD-binding integrin is used in a compensatory manner, for example integrin αvβ3, the subunits of which are expressed in the gastrula (Whittaker and DeSimone, 1993).

In contrast to integrin β1-MO, syndecan 4 MOs, which, by themselves, had no effect on lamellipodia shape or lifetime (Fig. S3C), caused the collapse of lamellipodia in KD-Pak1 explants (Fig. 5F). Moreover, lamellipodia were not smooth-rimmed, but serrate and rich in filopodia. Knockdown of integrin β1 in syndecan 4 morphants did not prolong lamellipodia lifespan, suggesting that syndecan 4 acts downstream of integrin. By contrast, ephrin B1-MO increased lamellipodia stability in syndecan 4 morphants (Fig. 5F). Together with a similar effect of ephrin B1-MO in FN morphants, this places a tentative FN-integrin-syndecan module between Pak1 and ephrin B1 (Fig. 5G). Altogether, with the respective results combined, the control factors can be ordered in a linear epistatic interaction pathway. A step-by-step deduction of the pathway is provided in Fig. S5. Formally, all interactions are inhibitory, implying that components are alternatingly ‘on’ and ‘off’ along the pathway. With Pak1 active, ephrin B1 is ‘on’ and causes lamellipodia to collapse upon contact. KD-Pak1 reverses the on-off states of all components, inactivating ephrin B1 and, thus, stabilizing lamellipodia (Fig. 5G). However, the list of components is contingent and not necessarily complete and, with additional factors, further inhibitory or activating interactions could be introduced (Fig. S5).

C-cadherin is the main cell-cell adhesion molecule in the Xenopus gastrula (Kühl and Wedlich, 1996; Winklbauer, 2009). It contributes to the mutual attachment of cells, but it is not clear whether it also mediates the adhesion of lamellipodia to cell bodies. When we knocked down C-cadherin in LEM aggregates on FN with C-cad-MO (Ninomiya et al., 2012), cells were loosely packed at the substratum level, but were densely packed interiorly (Fig. S1F). Submarginal protrusions were serrate and fragmented (Fig. 5E), as in untreated cells, but their lifespans were slightly increased compared with controls (Fig. 5E,F). Strikingly, C-cad-MO rescued lamellipodia survival in aggregates where KD-Pak1-induced protrusions had been inhibited by FN-MO injection (Fig. 5F). Weakened cell-cell contact formation and, hence, reduced ephrin B1 signaling in cadherin-depleted aggregates could be responsible for delayed lamellipodia retraction (Zantek et al., 1999; Hess et al., 2006). In Xenopus neural crest cells, PDGF-A/PDGFRα signaling affects contact inhibition of locomotion by promoting N-cadherin expression (Bahm et al., 2017). However, this cadherin isoform is not expressed in gastrula-stage LEM (Kühl and Wedlich, 1996), and C-cadherin might have taken over part of the role of N-cadherin.

In the embryo or on conditioned substratum, matrix-bound PDGF-A is required for directional migration and oriented protrusion formation (Nagel et al., 2004). We asked whether PDGF-A also promotes submarginal lamellipodia formation in LEM aggregates. We added heparin to surface-adsorbed FN to interact with its heparin-binding domain, and also added a long form of PDGF-A (lf-PDGF-A), which binds to heparin (Smith et al., 2009; Damm and Winklbauer, 2011). Isolated LEM cells migrated and formed lamellipodia on FN-PDGF substrata (Fig. 6A,B). Lamellipodia were similar to those of KD-Pak1 cells, and readily underlapped other cells (Movie 8). The lifespan of free lamellipodia was similar to that of single cells on pure FN (Fig. 6F).

In aggregates on PDGF-FN, LEM cells showed extensive underlapping of large, smooth-rimmed submarginal lamellipodia, often in shingle arrangements, similar to KD-Pak1-expressing cells (Fig. 6C, Fig. S2D). The lifespans of protrusions were similarly increased (Fig. 6F, Movie 9). Cells appeared loosely packed at the substratum level, but gaps vanished interiorly (Fig. S1G). Submarginal protrusions were suppressed by the PDGFR inhibitor AG1296 and the PI3K inhibitor LY294002 (Fig. S2E,H) on PDGF-FN, but not in KD-Pak1 aggregates on FN (Fig. S2G), consistent with a PDGF signaling-dependent effect (Nagel et al., 2004). Likewise, expression of a kinase-dead PDGF receptor-α (KD-PDGFRα), which interferes with PDGF-A signaling in LEM cells (Nagel et al., 2004), diminished submarginal protrusion formation (Fig. 6D,F).

To see whether PDGF signaling affected contact inhibition by rendering cells nonrepulsive or lamellipodia nonresponsive, we generated mosaics of untreated and KD-PDGFRα-expressing LEM cells and observed contact behavior on FN-PDGF (Fig. 6E). Under these conditions, untreated cells underlapped each other, and lamellipodia survived as in untreated explants. However, lamellipodial lifetime was shortened when KD-PDGFRα cells underlapped untreated cells, untreated cells moved underneath KD-PDGFRα cells, or KD-PDGFRα cells moved under KD-PDGFRα cells (Fig. 6G). Thus, PDGF-A signaling is required in both of two interacting cells to prevent lamellipodial collapse.

The similarity of the effects of KD-Pak1, ephrin B1-MO and PDGF-A suggests that the Pak1-ephrin B1 axis normally destabilizes submarginal lamellipodia, and that PDGF-A signaling overcomes this inhibition of protrusion maintenance. To place PDGF-A relative to Pak1, we expressed constitutively active CA-Pak1 in the LEM. In such aggregates, long-lived submarginal protrusions formed on PDGF-FN (Fig. 7A,F), consistent with PDGF-A acting downstream of Pak1. In agreement with this, DN-Pak1 augmented lamellipodia formation on FN-PDGF (Fig. S2F).

Consistent with a requirement of the PDGF signal between Pak1 and ephrin B1, syndecan 4 knockdown reduced lamellipodial lifetimes on PDGF-FN to control levels, and simultaneous eB1 knockdown rescued lamellipodia stability (Fig. 7B,C,F). Integrin β1-MO did not affect protrusion stability on PDGF-FN (Fig. 7D,F). This suggests that PDGF-A signaling acts by inhibiting integrin activity, a function that is performed by csFN in KD-Pak1 cells. As predicted from this interpretation, knockdown of csFN did not interfere with lamellipodia stability on PDGF-FN (Fig. 7E,F). Thus, Pak1 activity indirectly prevents syndecan 4 from inhibiting ephrin B1 (Fig. 5G). PDGF-A signaling blocks this syndecan-inhibiting function and, consequently, syndecan 4 inhibits ephrin B1 and prevents lamellipodial collapse (see Fig. 10A).

PDGF-A controls the orientation of LEM cell lamellipodia towards the animal pole of the embryo (Nagel et al., 2004). We asked whether it also stabilizes protrusions on the FN fibril matrix of the BCR. LEM migration cannot be observed at high resolution in situ, and we examined lamellipodia by scanning electron microscopy (SEM) instead. We confirmed the ‘shingle arrangement’ of LEM cells (Winklbauer and Nagel, 1991) and identified chains of cells underlapping with wide lamellipodia, as observed in aggregates on PDGF-FN in vitro (Fig. 8A). When the matrix-binding lf-PDGF-A was overexpressed in the BCR, the fraction of cells with two or more lamellipodia increased compared with untreated embryos (Fig. 8B,D), and protrusions were disoriented (Fig. 8B), as shown previously (Nagel et al., 2004). By contrast, dominant-negative PDGF-A increased the fraction of cells with no protrusions, and remaining lamellipodia were small and serrate (Fig. 8C,D). These findings are consistent with PDGF-A stabilizing lamellipodia in LEM cells in the embryo. Pak1 polarizes single LEM cells in vitro by constraining the size and the number of lamellipodia per cell (Nagel et al., 2009); in addition, it promotes the contact-induced collapse of lamellipodia. In the embryo, KD-Pak1 expression increases the number of lamellipodia per cell, and the normal number is rescued by co-expression of Pak1 (Fig. 8D). Both functions of Pak1 in LEM cells could contribute to this effect.

In the Xenopus gastrula, PDGF-A is expressed in the BCR, and its receptor, PDGFRα, in the mesoderm (Ho et al., 1994). Interestingly, PDGFRα protein expression is polarized in LEM cells in situ (Fig. 8E). The receptor forms puncta on the LEM cell surface that are larger yet fewer than those of integrin β1 and are evenly distributed over much of the cell surface. However, puncta are concentrated into dense patches at animally oriented protrusion and are occasionally also enriched at the animally oriented sides of cells in the absence of protrusions. Integrin β1 is also dense at protrusions, but, although it overlaps, it is not strictly colocalized with PDGFRα (Fig. 8E,F). Protrusions pointing laterally do not accumulate the receptor (Fig. 8F). Likewise, single LEM cells on FN or on PDGF-FN do not enrich the receptor at or near lamellipodia (Fig. 8G,H), suggesting that the effect in the embryo results from orienting cues in the BCR matrix.

To directly observe the effects of matrix-bound PDGF-A, we examined LEM cells migrating directionally on the endogenous FN- and PDGF-A-containing ECM deposited in vitro by cultured BCR (Fig. 9A) (Nagel et al., 2004). On this conditioned substratum, LEM aggregates formed submarginal protrusions similar to those on PDGF-FN, although they were narrower and more serrate (Fig. 9B), perhaps because FN is present as discrete fibrils rather than as an even layer. Most cells extended one protrusion (Fig. 9D) in animal orientation (Fig. 9E), and the lifetime of lamellipodia was prolonged, as on PDGF-FN (Fig. 9F). The shape and size (Fig. 9C), frequency (Fig. 9D), orientation (Fig. 9E) and lifetime of lamellipodia (Fig. 9F) were all sensitive to treatment with the PDGFR inhibitor AG1296, confirming that both the stabilization and orientation of protrusions depend on PDGF-A present in the BCR matrix.

We next asked whether protrusions are stabilized and oriented by two parallel PDGF functions, or whether orientation comprises the selective stabilization of lamellipodia pointing animally. Indeed, in LEM explants on conditioned substratum, lamellipodia that pointed animally survived longer compared with protrusions pointing vegetally, which collapsed rapidly (Fig. 9G). Together with the polarized expression of the PDGF receptor, this suggests that, on conditioned substratum, PDGF-A stabilizes lamellipodia in an oriented fashion.

Integrins, cadherins and Eph/ephrin signaling have previously been implicated in contact inhibition of locomotion (see below). Concentrating on lamellipodial collapse as an essential component of contact inhibition, and by adding Pak1, syndecan 4 and csFN to the system, we were able to formally order these disparate factors in an epistatic interaction framework. Moreover, we found that PDGF-A signaling constitutes a branch of this network that relieves contact inhibition to increase lamellipodial lifetimes in aggregates (Fig. 10A). On the endogenous substratum of LEM migration, the PDGF-A-containing BCR matrix, lamellipodia stabilization was directionally biased, linking the contact inhibition-suppressing function of PDGF-A to its role in LEM guidance.

A role of integrins in contact inhibition of locomotion was first proposed two decades ago, based on observations that ectopic expression in myoblasts of integrin α5, integrin β1 or integrin effectors, such as paxillin and FAK, caused lamellipodial paralysis upon cell contact (Huttenlocher et al., 1998). In support of such a role for integrins, contact-induced lamellipodia collapse was prevented in LEM cells by integrin β1 knockdown, suggesting that the main FN binding integrin α5β1 of the gastrula (Whittaker and DeSimone, 1993) is involved. Moreover, we identified csFN as an inhibitor of this integrin function. csFN puncta have been ascribed a role in cancer cell adhesion (Cheng et al., 1998, 2003), and have been shown by transmission electron microscopy to physically link adjacent cells (Hedman et al., 1978; Chen and Singer, 1982). We described here a signaling function for csFN puncta. We also identified a putative FN and integrin interactor, syndecan 4 (Roper et al., 2012), as a target for inhibition by integrin α5β1. We combined these factors into a novel pathway module (Fig. 10A; Fig. S5).

Eph/ephrin signaling has been implicated in various roles in the control of contact inhibition of locomotion. In contacts between prostate cancer cells and fibroblasts, ephrinA-stimulated EphA receptors elicit contact repulsion, whereas EphB/ephrinB interaction suppresses it (Astin et al., 2010; Batson et al., 2014). Ephrin B1 also suppresses contact inhibition in glioblastoma cells (Tanaka et al., 2012). By contrast, ephrin B1/EphB3 signaling in breast adenocarcinoma cells promotes contact inhibition (Lin et al., 2015), as do EphA/ephrin A and EphB/ephrin B interactions in Cajal-Retzius neurons (Villar-Cerviño et al., 2013). In the Xenopus gastrula, ephrin B1 is involved in cell repulsion at the ectoderm-mesoderm boundary (Rohani et al., 2011). In LEM cells, it acted downstream of the FN-integrin-syndecan module to mediate contact inhibition of lamellipodia (Fig. 10A). It is required in the lamellipodium and could interact with an Eph receptor on an adjacent cell to sense cell-cell contact and trigger repulsion. Candidate receptors in the LEM would be EphB2 and EphB3 (Rohani et al., 2011).

Pak1 function is required in the LEM for its movement across the BCR (Nagel et al., 2009). In the present context, it antagonized lamellipodia stability by acting upstream of other factors. Based on their epistatic relationships, these factors can formally be placed in an interaction network (Fig. 10A). The central module mediates a series of inhibitory interactions, which implies that alternating pathway nodes are either ‘on’ or ‘off’ (Fig. 5G). In aggregates on FN, submarginal lamellipodia collapsed, supposedly because of Pak1 inhibiting the next node in the interaction network, csFN. This notion describes a formal placement of Pak1 in the pathway; the actual interaction between the cytoplasmic kinase and the extracellular FN will of course be indirect. Nevertheless, Pak1 activity determines the on-off state of each pathway component and induces contact inhibition by blocking the inhibition of ephrin B1 (Figs 5G and 10A).

Pak1 inhibition, knockdown of integrin β1 or ephrin B1, and also exposure to PDGF-A not only increased the stability, but also changed the shape of lamellipodia, from serrate and moderate-sized to smooth-rimmed and wide. Importantly, when FN-MO was injected into KD-Pak1 cells, lamellipodia were of the smooth-wide type, but collapsed upon contact, indicating that lamellipodia shape and lifespan are controlled independently. In syndecan 4 morphants, lamellipodia were similar to those of untreated cells, suggesting that syndecan 4 is required for the smooth shape as well as for the prolonged lifetime of protrusions. A similar change in lamellipodia shape in cultured epithelial cells is associated with a transition from integrin α5β1 to integrin αvβ3 (Danen et al., 2005), possibly because of syndecan 4 differential phosphorylation (Morgan et al., 2013). Such a transition would explain the RGD peptide-sensitive, yet integrin β1-independent, migration of LEM cells on FN in our experiments. This transition also implies that the FN-integrin-syndecan module not only regulates lamellipodia-cell body interactions, but also selects integrin α5β1 or integrin αvβ3, depending on its activity status, for adhesion to the external FN substratum.

Two types of mechanism could underlie the formally deduced interactions of pathway components: (1) the physical association of components at the membrane in cis or trans, modulated by the steric exclusion of some associations because of molecular size; and (2) signaling exchanges between the factors involved. To illustrate the first mechanism by a hypothetical example (Fig. 10B,C), we assume a state where a lamellipodia-destabilizing Eph/ephrin trans-interaction occurs (Fig. 10B), requiring membrane separation distances close to the 25 nm found at cadherin contacts (Lambert et al., 2005; Xu et al., 2013). Some of the integrin α5β1 molecules are assumed to bind csFN puncta >100 nm in size (Hedman et al., 1978; Chen and Singer, 1982), but others would associate with syndecan 4 (Fiore et al., 2014) and competitively block a putative ephrin-syndecan interaction. In a different state (Fig. 10C), integrins would preferentially bind to csFN, which would allow ephrin B1 to associate with syndecan 4, perhaps through the scaffold protein, syntenin (Grembecka et al., 2006). With a membrane separation distance of >40 nm at syndecan sites (Roper et al., 2012), Eph/ephrin trans-interaction would be excluded and signaling would be blocked, allowing lamellipodia to survive cell contact. The first state could depend on Pak1 activity, the second on either its absence, or on the presence of the PDGF-A signal.

The involvement of Pak1, a cytoplasmic kinase, indicates that the second mechanism, interaction by signaling, is involved at least at one point in the pathway, and indirect evidence suggests that integrin inhibition by PDGFRα is also transmitted through the cytoplasm. Moreover, in the previous description of the role of integrin in contact inhibition, signaling through paxillin and FAK was implicated (Huttenlocher et al., 1998). Multiple signaling pathways originate at syndecan 4, for example PKCα-based branches (Roper et al., 2012; Afratis et al., 2017), which could indirectly affect ephrin B1 activity. Future analysis must also address the possibility that different components of the mechanism function on different sides of the lamellipodium-cell body contact.

In the embryo or on conditioned substratum, PDGF-A overcame Pak1-dependent contact inhibition by blocking the csFN-integrin-syndecan module at the integrin level (Fig. 8A). Although syndecan 4 can physically bind PDGFRβ to integrin α5β1 (Veevers-Lowe et al., 2011; Tanaka et al., 2014), we propose that, in the present context, the interaction is indirect. While the PDGFRα ligand, PDGF-A, is present on the external substratum, integrin α5β1 must interact with csFN at the cell-cell interface. This implies a transmission of the signal across the cell (e.g. from the lower to the upper side of the lamellipodium to be stabilized). A candidate for the respective signal transduction would be PI3K, which affects directional LEM migration on PDGF-containing matrix (Nagel et al., 2004) and lamellipodial lifetime on PDGF-FN.

The control of lamellipodial stability by PDGF-A is linked to the directionality of LEM migration (Nagel et al., 2004). Previously, a PDGF-A gradient in the BCR matrix had been proposed to orient LEM cells (Nagel et al., 2004), although the gradient had not been demonstrated directly. However, regardless of whether PDGF-A distribution is graded, the leading edge of an aggregate cell at a given animal-vegetal position would encounter the same PDGF-A concentration as the rear end of a cell directly ahead of it. In this situation, the polarization of PDFGR-α expression in LEM cells could explain the selective stabilization of animally pointing lamellipodia. Increased receptor density at animally pointing protrusions would increase PDGF-A signal intensity relative to lateral and rear sides and, thus, differentially increase lamellipodia lifetime. PDFGR-α is not polarized on artificial PDGF-FN substratum, yet lamellipodia are stabilized. This could be simply explained by a lower PDGF-A concentration on the BCR matrix compared with the artificial FN-PDGF substratum. In the absence of PDFGR-α polarization, the matrix PDGF-A density would remain below the threshold for contact inhibition suppression.

Receptor polarization could itself be a response to a putative PDGF-A gradient, as part of a mechanism that augments the reaction to a shallow gradient through a positive feedback loop. Chemoattractant-stimulated polarization has been shown for GABA and BDNF receptors on nerve growth cones (Bouzigues et al., 2007; Guirland et al., 2004), and for the PDGF/VEGF receptor during border cell migration in the Drosophila ovary (Janssens et al., 2010). The chains of underlapping LEM cells that form on both nonoriented and oriented substratum suggest that spontaneously polarized cell-cell adhesion also contributes to cell alignment, for example through the polarized expression of ephrin B1 in single cells. Such chains of aligned cells could then be oriented towards the animal pole by the same mechanism as individual cells.

In summary, PDGF-A is involved in the suppression of contact inhibition of lamellipodia in the LEM. Suppression occurs in a polarized manner, which has the ultimate effect of orienting cells towards the animal pole, in the direction of overall LEM migration. Cell orientation based on the differential survival of lamellipodia has also been described for the chick neural crest (Genuth et al., 2018). Similar to LEM cells, chick neural crest cells do not exhibit contact inhibition of migration in situ, and a combination of increased formation and differential survival of protrusions pointing in the overall direction of movement effectively polarizes cells throughout the translocating cell stream. The averages and ranges of protrusion lifetimes are also strikingly similar in both systems. If PDFGR-α and PDGF-A are expressed at the chick neural crest as they are in other vertebrates (Hoch and Soriano, 2003), it would be interesting to see whether these cells show contact inhibition of movement in the absence of PDGF-A, for example on FN in vitro. This would suggest that differential survival of protrusions by the oriented suppression of contact inhibition is conserved not only as a cellular mechanism, but also molecularly.

Xenopus laevis embryos from in vitro fertilized eggs (Animal Use Protocol 20011765, Bioscience LACC) were dejellied with 2% cysteine in 1/10 Modified Barth's Solution [MBS; 88 mM NaCl, 1 mM KCl, 2.4 mM NaHCO₃, 0.82 mM MgSO₄, 0.33 mM Ca(NO3)₂, 0.41 mM CaCl₂, 10 mM Hepes (+NaOH), 1% streptomycin, 1% penicillin, pH 7.4], pH 8.0. Embryos were injected at the four-cell stage in both dorsal blastomeres using a Nanoinject II (Drummond Scientific Company) in 4% Ficoll and cultured at 15°C in 1/10 MBS until the gastrula stage.

Tissue culture dishes (35 mm) with a polymer coverslip bottom from ibidi GmbH were coated with bovine plasma FN (Sigma) at 200 ng/ml, or with 200 ng/ml of FN, 5 µg/ml of heparin and 100 ng/ml of human recombinant PDGF-AA (Calbiochem) (PDGF-FN) for 1 h and saturated with 1 mg/ml of bovine serum albumin (BSA) for 30 min. Substrata were conditioned according to Nagel and Winklbauer (1999): stage-10 BCR explants were held against the bottom of tissue culture dishes for 2 h. After removal of the BCR, substrata were saturated with 1 mg/ml BSA.

Embryos were staged according to Nieuwkoop and Faber (1967). At gastrula stage 10.5, the vitelline membrane was removed with forceps. Microsurgery was performed in MBS at room temperature under a MZ16F (Leica) stereomicroscope. The LEM was excised, cut into four or five pieces and placed on prepared substrata for imaging. Single cells and mosaics were obtained by dissociating mesendoderm aggregates in Dissociation Buffer [88 mM NaCl, 1 mM KCl, 2.4 mM NaHCO₃, 10 mM Hepes (+NaOH), 1% streptomycin, and 1% penicillin (pH 7.4)]. For mosaics, two dissociated populations, as indicated in the text, were mixed and reaggregated in MBS. Operation techniques were as described elsewhere (Winklbauer, 1990; Winklbauer and Schürfeld, 1999).

A kinase dead construct of the plasmid pCS2 wild-type Xenopus PDGFRα (Ataliotis et al., 1995) was linearized with NheI and transcribed with T7 polymerase. CS2+mbGFP (a gift from R. Harland, University of California at Berkeley, CA, USA) and mbRFP (a gift from A. Bruce, University of Toronto, ON, Canada) were linearized with NotI and transcribed with SP6 polymerase. A kinase-dead mutant K281A of xPak1 (KD-Pak1), a constitutively active Pak1 construct, L98F (CA-Pak1) (Poitras et al., 2003; Bisson et al., 2003) were linearized with XbaI and transcribed with T7 polymerase. CS2+memeGFP (a gift from R. Harland) and mbRFP (a gift from A. Bruce) were linearized with NotI and transcribed with SP6 polymerase. Plasmid pGHE2 containing the long form of Xenopus PDGFA (lf-PDGFA) (Mercola et al., 1988), and pGHE2 harboring a processing defective mutant of mouse PDGFA that acts as a dominant-negative mutant in Xenopus (mPDGFdn) (Mercola et al., 1990) were linearized with NheI and transcribed with T7 polymerase. Embryos were injected at the four-cell stage marginally into the dorsal two blastomeres. See Tables S1 and S2 for the amounts of each mRNA injected per blastomere and details of morpholinos used.

PDGFR inhibitor AG1296 (Calbiochem) (5 µM) and PI3K inhibitor LY294002 (Calbiochem) (30 µM) were dissolved in DMSO (Sigma-Aldrich) and used as previously described (Nagel et al., 2004). RGD-peptide H-Gly-Arg-Gly-Asp-Ser-Pro-OH (Calbiochem) was used at 4 mg/ml in MBS as previously described (Winklbauer and Keller, 1996).

Cells were injected with membrane-GFP or membrane-RFP mRNA to label the cell membrane for live imaging. The closely apposed upper and lower membranes of lamellipodia or filopodia showed these structures as more intensely stained regions, compared with the cell body. The identity of lamellipodia was confirmed by fluorescent phalloidin staining (Fig. S2).

F-actin was stained in specimens fixed in 4% paraformaldehyde (20 min, 0.01% Triton-X-100 added after 5 min) with Alexa Fluor 488 phalloidin or rhodamine phalloidin (Invitrogen) at 1:100 in PBS/BSA for 20 min. Rabbit antiserum against X. laevis plasma FN (Winklbauer, 1998) (1:1000), mouse monoclonal antibody 8C8 against X. laevis integrinβ1 (1:6; Developmental Studies Hybridoma Bank, University of Iowa, generated by P. Hausen and V. Gawantka), polyclonal rabbit antibody against syndecan 4 (SDC4) (1:200, TA 314520, Acris, OriGene), and polyclonal rabbit antibody against ephrin-B1 (1:200, A-20, sc-1011, Santa Cruz Biotechnology; Lee et al., 2009), and PDGFRα (1:2000, 3164, Cell Signaling Technology; Bahm et al., 2017) were used. Secondary antibodies were Cy3-goat-anti-rabbit IgG, FITC-goat-anti-rabbit IgG and Cy3-goat-anti-mouse IgG (Jackson ImmunoResearch Laboratories). Staining intensity was measured using the Leica Application Suite X software (free download from Leica). For image collection, an If-SP8-nonresonant confocal microscope (Leica) with 40× immersion oil objective and a Zeiss Axiovert 200M microscope (PlanNeofluar 20× and 40× oil objectives) with Leica Application SuiteX software or AxioVision LE64 software (free download from Zeiss) were used. Data were collected from at least two different batches of embryos for each experiment, and at least three time-lapse recordings (eight on average) per experiment. The n in the figures relates to the number of lamellipodia counted. Data were compared using the Student's t-test (two-tailed, unequal variances).

Embryos were fixed in 2.5% glutaraldehyde/0.1 M sodium cacodylate overnight at 4°C, post-fixed in osmium tetraoxide, and dehydrated in an ethanol/0.1 M cacodylate and ethanol/hexamethyl-disilizane series. Specimens were dried overnight, sputter coated with gold-palladium, and inspected using a Hitachi SU 3500 scanning electron microscope.

We thank K. Symes, R. Harland, A. Bruce and T. Moss for reagents.